jackkuo commited on
Commit
c8fa9a0
·
verified ·
1 Parent(s): cca3e42

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +3 -0
  2. 1dAzT4oBgHgl3EQf8v7L/content/tmp_files/2301.01910v1.pdf.txt +1373 -0
  3. 1dAzT4oBgHgl3EQf8v7L/content/tmp_files/load_file.txt +0 -0
  4. 4tFKT4oBgHgl3EQfRy39/content/tmp_files/2301.11773v1.pdf.txt +3963 -0
  5. 4tFKT4oBgHgl3EQfRy39/content/tmp_files/load_file.txt +0 -0
  6. 79FLT4oBgHgl3EQfAy4h/content/tmp_files/2301.11967v1.pdf.txt +1315 -0
  7. 79FLT4oBgHgl3EQfAy4h/content/tmp_files/load_file.txt +0 -0
  8. 9NE4T4oBgHgl3EQfdgy1/content/tmp_files/2301.05092v1.pdf.txt +3087 -0
  9. 9NE4T4oBgHgl3EQfdgy1/content/tmp_files/load_file.txt +0 -0
  10. AdFLT4oBgHgl3EQfEy_H/content/tmp_files/2301.11985v1.pdf.txt +0 -0
  11. AdFLT4oBgHgl3EQfEy_H/content/tmp_files/load_file.txt +0 -0
  12. BNE5T4oBgHgl3EQfTA98/content/tmp_files/2301.05533v1.pdf.txt +421 -0
  13. CdFJT4oBgHgl3EQftC3b/content/tmp_files/2301.11616v1.pdf.txt +1537 -0
  14. CdFJT4oBgHgl3EQftC3b/content/tmp_files/load_file.txt +0 -0
  15. GNFLT4oBgHgl3EQfGi_B/content/tmp_files/2301.11993v1.pdf.txt +3951 -0
  16. GNFLT4oBgHgl3EQfGi_B/content/tmp_files/load_file.txt +0 -0
  17. I9AyT4oBgHgl3EQfTfcL/content/tmp_files/2301.00104v1.pdf.txt +1865 -0
  18. I9AyT4oBgHgl3EQfTfcL/content/tmp_files/load_file.txt +0 -0
  19. KdE4T4oBgHgl3EQfiA17/content/tmp_files/2301.05130v1.pdf.txt +0 -0
  20. KdE4T4oBgHgl3EQfiA17/content/tmp_files/load_file.txt +0 -0
  21. LtAyT4oBgHgl3EQfgPig/content/tmp_files/2301.00356v1.pdf.txt +1175 -0
  22. LtAyT4oBgHgl3EQfgPig/content/tmp_files/load_file.txt +0 -0
  23. QtE3T4oBgHgl3EQfDAk3/content/2301.04281v1.pdf +3 -0
  24. TdE5T4oBgHgl3EQfAQ7r/content/tmp_files/2301.05378v1.pdf.txt +2484 -0
  25. TdE5T4oBgHgl3EQfAQ7r/content/tmp_files/load_file.txt +0 -0
  26. VdAyT4oBgHgl3EQfhfjt/content/tmp_files/2301.00380v1.pdf.txt +535 -0
  27. VdAyT4oBgHgl3EQfhfjt/content/tmp_files/load_file.txt +447 -0
  28. WNAyT4oBgHgl3EQfu_m7/vector_store/index.pkl +3 -0
  29. WtAyT4oBgHgl3EQfWPdu/content/tmp_files/2301.00159v1.pdf.txt +2049 -0
  30. WtAyT4oBgHgl3EQfWPdu/content/tmp_files/load_file.txt +0 -0
  31. XtFJT4oBgHgl3EQf5y0l/content/tmp_files/2301.11671v1.pdf.txt +0 -0
  32. XtFJT4oBgHgl3EQf5y0l/content/tmp_files/load_file.txt +0 -0
  33. _NE1T4oBgHgl3EQfUwPr/content/2301.03095v1.pdf +3 -0
  34. b9AyT4oBgHgl3EQfXPer/content/tmp_files/2301.00180v1.pdf.txt +700 -0
  35. b9AyT4oBgHgl3EQfXPer/content/tmp_files/load_file.txt +337 -0
  36. bdE_T4oBgHgl3EQfzRz6/content/2301.08323v1.pdf +3 -0
  37. d9FIT4oBgHgl3EQfoiuQ/content/tmp_files/2301.11319v1.pdf.txt +0 -0
  38. d9FIT4oBgHgl3EQfoiuQ/content/tmp_files/load_file.txt +0 -0
  39. dtE0T4oBgHgl3EQfoQFr/content/tmp_files/2301.02523v1.pdf.txt +1043 -0
  40. dtE0T4oBgHgl3EQfoQFr/content/tmp_files/load_file.txt +0 -0
  41. dtE1T4oBgHgl3EQfyAVc/content/tmp_files/2301.03428v1.pdf.txt +760 -0
  42. dtE1T4oBgHgl3EQfyAVc/content/tmp_files/load_file.txt +431 -0
  43. e9E2T4oBgHgl3EQfGgbf/content/tmp_files/2301.03659v1.pdf.txt +1307 -0
  44. e9E2T4oBgHgl3EQfGgbf/content/tmp_files/load_file.txt +0 -0
  45. edFST4oBgHgl3EQfFzg4/content/tmp_files/2301.13719v1.pdf.txt +815 -0
  46. edFST4oBgHgl3EQfFzg4/content/tmp_files/load_file.txt +0 -0
  47. f9AzT4oBgHgl3EQfavwP/content/tmp_files/2301.01372v1.pdf.txt +2306 -0
  48. f9AzT4oBgHgl3EQfavwP/content/tmp_files/load_file.txt +0 -0
  49. gdE2T4oBgHgl3EQfyQgz/content/tmp_files/2301.04118v1.pdf.txt +0 -0
  50. gdE2T4oBgHgl3EQfyQgz/content/tmp_files/load_file.txt +0 -0
.gitattributes CHANGED
@@ -172,3 +172,6 @@ JdAyT4oBgHgl3EQffviy/content/2301.00347v1.pdf filter=lfs diff=lfs merge=lfs -tex
172
  p9FLT4oBgHgl3EQfiS8Z/content/2301.12106v1.pdf filter=lfs diff=lfs merge=lfs -text
173
  7dAzT4oBgHgl3EQfgPzo/content/2301.01467v1.pdf filter=lfs diff=lfs merge=lfs -text
174
  FtE3T4oBgHgl3EQftQtg/content/2301.04674v1.pdf filter=lfs diff=lfs merge=lfs -text
 
 
 
 
172
  p9FLT4oBgHgl3EQfiS8Z/content/2301.12106v1.pdf filter=lfs diff=lfs merge=lfs -text
173
  7dAzT4oBgHgl3EQfgPzo/content/2301.01467v1.pdf filter=lfs diff=lfs merge=lfs -text
174
  FtE3T4oBgHgl3EQftQtg/content/2301.04674v1.pdf filter=lfs diff=lfs merge=lfs -text
175
+ bdE_T4oBgHgl3EQfzRz6/content/2301.08323v1.pdf filter=lfs diff=lfs merge=lfs -text
176
+ QtE3T4oBgHgl3EQfDAk3/content/2301.04281v1.pdf filter=lfs diff=lfs merge=lfs -text
177
+ _NE1T4oBgHgl3EQfUwPr/content/2301.03095v1.pdf filter=lfs diff=lfs merge=lfs -text
1dAzT4oBgHgl3EQf8v7L/content/tmp_files/2301.01910v1.pdf.txt ADDED
@@ -0,0 +1,1373 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.01910v1 [math.DS] 5 Jan 2023
2
+ Differentiability of the largest Lyapunov exponent for planar open
3
+ billiards
4
+ Amal Al Dowais a,b
5
+ Abstract
6
+ In this paper, we estimate the largest Lyapunov exponent for open billiards in the plane.
7
+ We show that the largest Lyapunov exponent is differentiable with respect to a billiard defor-
8
+ mation.
9
+ keywords. Open billiards; Lyapunov exponents; Non-wandering set; Billiard deformation
10
+ Mathematics Subject Classification (2010). 37D50, 37B10, 37D20, 34D08
11
+ 1
12
+ Introduction
13
+ The stability and instability of a dynamical system can be studied by means of Lyapunov expo-
14
+ nents. A dynamical system is considered chaotic if it has a positive Lyapunov exponent. Examples
15
+ of chaotic systems are the dispersing billiards or so-called Sinai billiards (see [15], [16]). Billiards
16
+ are dynamical systems in which a particle moves with constant speed and hits the billiard’s wall
17
+ (boundary of the billiard’s domain) according to the law of geometrical optics,“the angle of inci-
18
+ dence equals the angle of reflection”. Open billiards are a particular case of billiards in unbounded
19
+ domains. The domain is the exterior of finitely many strictly convex compact obstacles satisfying
20
+ the no-eclipse condition (H) of Ikawa [6]: the convex hull of any two obstacles does not intersect
21
+ with another obstacle; in other words, there does not exist a straight line that intersects more
22
+ than two obstacles. It follows from Sinai [15], [16] (see also [14]) that the non-wandering set of
23
+ the open billiard map is hyperbolic (i.e. there exist positive and negative Lyapunov exponents).
24
+ Many studies have investigated Lyapunov exponents for billiards (see [22], [1], [4], [9], [10]). In
25
+ this paper, we estimate the largest Lyapunov exponent for open billiard in R2. We demonstrate
26
+ that the Lyapunov exponent depends continuously on a parameter α related to a deformation of
27
+ the billiard as defined in [21]. Moreover, we prove that the Lyapunov exponent is differentiable
28
+ with respect to the deformation parameter α.
29
+ Here we state the main results:
30
+ In the following theorems, we denote the billiard deformation by K(α) where α ∈ [0, b].
31
+ See
32
+ Section 4 for the precise definition.
33
+ aDepartment of Mathematics and Statistics, School of Physics, Mathematics and Computing, University of
34
+ Western Australia, Perth, WA 6009, Australia
35
+ Email address: [email protected]
36
+ bDepartment of Mathematics, College of Science and Arts, Najran University, Najran, Saudi Arabia
37
+ Email address: [email protected]
38
+ 1
39
+
40
+ Theorem 1.1. (Continuity) Let K(α) be a C4,2 billiard deformation in R2. Let λ1(α) be the
41
+ largest Lyapunov exponent for K(α). Then the largest Lyapunov exponent is continuous as a
42
+ function of α.
43
+ Theorem 1.2. (Differentiability) Let K(α) be a C5,3 billiard deformation in R2. Let λ1(α) be
44
+ the largest Lyapunov exponent for K(α). Then λ1(α) is C1 with respect to α.
45
+ There are many works studying continuity properties of Lyapunov exponents (see e.g. [20], [3]).
46
+ However, to our knowledge, in all these continuity is established generically, i.e. with respect to
47
+ “most” (typical) values of the parameters/perturbations involved. In the case of the open billiard
48
+ considered in the present paper we establish continuity and even differentiability for all values
49
+ of the parameter that appears in the perturbation, which is a truly remarkable property of this
50
+ physical system.
51
+ 2
52
+ Preliminaries
53
+ This section provides some preliminary concepts for open billiards, billiard flow, symbolic coding
54
+ and stable/unstable manifolds. We also describe some notations related to curvatures, distances,
55
+ and collision angles. In the last part of this section, we state the Oseledets multiplicative ergodic
56
+ theorem and its consequence for open billiards.
57
+ 2.1
58
+ Open billiard
59
+ Let Ki, where i = 1, 2, 3, ..., z0, be strictly convex compact domains with smooth boundaries ∂Ki
60
+ in R2. In this paper, we assume that K = �
61
+ i Ki satisfies the following condition(H) of Ikawa [6]:
62
+ for any i ̸= j ̸= k the convex hull of Ki ∪Kk does not have any common points with Kj. Let Ω be
63
+ the exterior of K (i.e., Ω = R2\K). Let Φt, t ∈ R, be the billiard flow such that for any particle
64
+ x = (q, v), where q ∈ Ω represents the position of x and v is the unit velocity of the particle x,
65
+ then Φt(x) = (qt, vt) = (q + tv, v). When the particle hits the boundary, then the velocity follows
66
+ the collision law vnew = vold − 2 < vold, n > n, where n is the outwards unit normal vector to ∂K
67
+ at q, and φ the angle between n = n(q) and v.
68
+ We denote the time of the j-th reflection of x by tj(x) ∈ (−∞, ∞) for j ∈ Z. We say tj(q) = ∞
69
+ (tj(q) = −∞) if the forwards (backwards) trajectory of x has less than j reflections. We denote
70
+ the non-wandering set of the flow Φt by Λ = {x ∈ �Ω, |tj(x)| < ∞, for all j ∈ Z}, where
71
+ �Ω = {(q, v) ∈ int Ω × S1 or (q, v) ∈ ∂Ω × S1 : ⟨n(q), v⟩ ≥ 0}.
72
+ Now let M = {x = (q, v) ∈ ∂K × S1 : ⟨n(q), v⟩ ≥ 0} and let π : M → ∂K be the canonical
73
+ projection map defined by π(q, v) = q. Let t1(x) be the time of the first reflection of x and let
74
+ M1 = {x ∈ M : t1(x) < ∞}. Define the billiard ball map B : M1 → M by B(x) = Φt1(x)(x), (e.g.
75
+ if y = (p0, w0), where p0 lies on ∂Ki then B(y) = B(p0, w0) = (p1, w1) where p1 = p0+t1w0 ∈ ∂Kj
76
+ and w1 = w0 − 2 < w0, n > n). The non-wandering set of the open billiard map is M0 = {x ∈
77
+ M : |tj(x)| < ∞} which is a subset of Λ. Finally, let B : M0 → M0 be the restriction of the
78
+ open billiard map on the non-wandering set M0. It is obvious that the non-wandering set is an
79
+ invariant set. See [15], [16], [4], [5], [14], for general information about billiard dynamical systems.
80
+ 2
81
+
82
+ 2.2
83
+ Symbolic coding for open billiards
84
+ Each particular x ∈ M0 can be coded by a bi-infinite sequence
85
+ ξ(x) = (..., ξ−1, ξ0, ξ1, ...) ∈ {1, 2, ..., z0}Z,
86
+ in which ξi ̸= ξi+1, for all i ∈ Z, and ξj indicates the obstacle Kξj such that πBj(x) ∈ ∂Kξj. For
87
+ example, if there are three obstacles K1, K2 and K3 as above and a particular q repeatedly hits
88
+ K1, K3, K2, K1, K3, K2, then the bi-infinite sequence is (..., 1, 3, 2, 1, 3, 2, ...). Let Σ be the symbol
89
+ space which is defined as:
90
+ Σ = {ξ = (..., ξ−1, ξ0, ξ1, ...) ∈ {1, 2, ..., z0}Z : ξi ̸= ξi+1, ∀i ∈ Z}.
91
+ Define the representation map R : M0 → Σ by R(x) = ξ(x). Let σ : Σ → Σ be the two-sided
92
+ subshift map defined by σ(ξi) = ξi+1. Given θ ∈ (0, 1) define the metric dθ on Σ by:
93
+ dθ(ξ, η) =
94
+ � 0
95
+ if
96
+ ξi = ηi for all i ∈ Z
97
+ θn
98
+ if
99
+ n = max{j ≥ 0 : ξi = ηi for all |i| < j}
100
+ Then σ is continuous with respect to dθ ([2]).
101
+ It is also known that the representation map
102
+ R : M0 → Σ is a homeomorphism (see e.g. [14]). See [6], [8], [11], [14], [17], for topics related to
103
+ symbolic dynamics for open billiards.
104
+ 2.3
105
+ Lyapunov exponents
106
+ Here we state a consequence of Oseledets Multiplicative Ergodic Theorem for billiards (see e.g.
107
+ Ch. 2 in [13], also see [12], [20], [7]).
108
+ For the open billiard map B : M0 −→ M0 in R2 we will use the coding R : M0 −→ Σ from Section
109
+ 2.2, which conjugates B with the shift map σ : Σ −→ Σ, to define Lyapunov exponents. It is
110
+ well known that there are ergodic σ-invariant measures µ on Σ. Let µ be an ergodic σ-invariant
111
+ probability measure on Σ. The following is a consequence of Oseledets Multiplicative Ergodic
112
+ Theorem:
113
+ Theorem 2.1 (A Consequence of Oseledets Multiplicative Ergodic Theorem). There exist real
114
+ numbers λ1 > 0 > −λ1 and one-dimensional vector subspaces Eu(x) and Es(x) of Tx(∂K),
115
+ x ∈ M0, depending measurably on R(x) ∈ Σ such that:
116
+ (i) Eu(x) and Es(x) for almost all x ∈ M0;
117
+ (ii) DxB(Eu(x)) = Eu(B(x)) and DxB(Es(x)) = Es(B(x)) for almost all x ∈ M0, and
118
+ (iii) For almost all x ∈ M0 there exists
119
+ lim
120
+ n→∞
121
+ 1
122
+ n log ∥DxBn(w)∥ = λ1
123
+ whenever 0 ̸= w ∈ Eu(x).
124
+ Here ”for almost all x” means ”for almost all R(x)” with respect to µ. The numbers λ1 > 0 > −λ1
125
+ are called Lyapunov exponents, while the invariant subspaces Eu(x) and Es(x) are called Oseledets
126
+ subspaces.
127
+ 3
128
+
129
+ 2.4
130
+ Propagation of unstable manifolds for open billiards
131
+ We describe a formula which is useful in getting estimates for
132
+ lim
133
+ m→∞
134
+ 1
135
+ m log ∥DxBmw∥, (0 ̸= w ∈ Eu(x), x ∈ M0).
136
+ Let M0 be the non-wandering set of the billiard ball map B of an open billiard. Then
137
+ Λ = {Φt(x) : x ∈ M0 , t ∈ R},
138
+ is the non-wandering set for the billiard flow Φt. For x ∈ Λ and a sufficiently small ǫ > 0 let
139
+
140
+ W s
141
+ ǫ (x) = {y ∈ Λ : d(Φt(x), Φt(y)) ≤ ǫ for all t ≥ 0 , d(Φt(x), Φt(y)) →t→∞ 0 },
142
+
143
+ W u
144
+ ǫ (x) = {y ∈ Λ : d(Φt(x), Φt(y)) ≤ ǫ for all t ≤ 0 , d(Φt(x), Φt(y)) →t→−∞ 0 }
145
+ be the (strong) stable and unstable manifolds of size ǫ for the billiard flow. Then �Eu(x) = Tx�
146
+ W u
147
+ ǫ (x)
148
+ and �Es(x) = Tx�
149
+ W s
150
+ ǫ (x). In a similar way one defines stable/unstable manifolds for the billiard
151
+ ball map B. For any x = (q, v) ∈ M0 define
152
+ W s
153
+ ǫ (x) = {y ∈ M0 : d(Bn(x), Bn(y)) ≤ ǫ for all n ∈ N , d(Bn(x), Bn(y)) →n→∞ 0 },
154
+ W u
155
+ ǫ (x) = {y ∈ M0 : d(B−n(x), B−n(y)) ≤ ǫ for all n ∈ N , d(B−n(x), B−n(y)) →n→∞ 0 }.
156
+ In what follows we will just write W u(x) and W s(x) for W u
157
+ ǫ (x) and W s
158
+ ǫ (x), assuming some
159
+ appropriately chosen sufficiently small ǫ > 0 is involved. Similarly for �
160
+ W u and �
161
+ W s.
162
+ It is well-known that there is an one-to-one correspondence between the stable/unstable man-
163
+ ifolds for the billiard ball map and these for the flow.
164
+ Geometrically the easiest (and most
165
+ convenient way) to describe this is as follows.
166
+ Given x = (q, v) ∈ M0 (so q ∈ ∂K and v ∈ S1), and a small 0 < r < t1(x), set y = (q + rv, v).
167
+ Then there is a 1-1 correspondence
168
+ ϕ : W u(x) −→ �
169
+ W u(y)
170
+ such that ϕ(z, w) = (z + t w, w) for all (z, w) ∈ W u(x), where t = t(z, w) > 0. Similarly, there is
171
+ a correspondence between W s(x) and �
172
+ W s(y). Moreover
173
+ Dϕ(x) : TxM0 −→ TyΛ
174
+ is so that D��(x)(Eu(x)) = �Eu(y) and Dϕ(x)(Es(x)) = �Es(y).
175
+ It is known that �
176
+ W u(y) has the form �
177
+ W u(y) = �Y , where
178
+ �Y = {(p, νY (p)) : p ∈ Y }
179
+ for some smooth curve Y in R2 containing the point y such that Y is strictly convex with respect
180
+ to the unit normal field νY , i.e. the curvature of Y is strictly positive.
181
+ 4
182
+
183
+ Next, let x and y be as above and let x1 = (q1, v1) = B(x). Then q1 = q + t1 v. Define
184
+ y1 = (q1 + r′v1, v1) for some small 0 < r′ < t2(x) − t1(x), where 0 = t0(x) < t1(x) < t2(x). Then
185
+ there is a 1-1 correspondence
186
+ ϕ1 : W u(x1) −→ ˜W u(y1)
187
+ defined as above. Again, we can write �
188
+ W u(y1) = �Y1, where
189
+ �Y1 = {(p1, νY (p1)) : p1 ∈ Y1}
190
+ for some smooth curve Y1 in R2 containing the point y1 such that Y1 is strictly convex with respect
191
+ to the unit normal field νY1. Moreover the following diagram is commutative, where t = t1 + r′:
192
+ W u(x)
193
+ B
194
+ −→
195
+ W u(x1)
196
+ �ϕ
197
+ �ϕ1
198
+
199
+ W u(y) = �Y
200
+ Φt
201
+ −→
202
+
203
+ W u(y1) = �Y1
204
+ Similarly, the following diagram is commutative:
205
+ Eu(x)
206
+ DB(x)
207
+ −→
208
+ Eu(x1)
209
+ �Dϕ
210
+ �Dϕ1
211
+ �Eu(y)
212
+ DΦt(y)
213
+ −→
214
+ �Eu(y1)
215
+ Since the derivatives Dϕ and Dϕ1 are uniformly bounded, the above conjugacy can be used later
216
+ to calculate the Lyapunov exponents of the billiard ball map using propagation of appropriate
217
+ convex curves Y which we describe as follows.
218
+ Let x0 = (q0, v0) ∈ M0 and let W u
219
+ ǫ (x0) be the local unstable manifold for x0 for sufficiently
220
+ small ǫ > 0. Let t1(x0) be the time of the first reflection of x0. Then �
221
+ X = W u
222
+ ǫ (x0) = {(q, nX(q)) :
223
+ q ∈ X} for some C3 curve X in Ω such that q0 ∈ X and X is strictly convex curve with respect to
224
+ the outer unit normal field nX(q). Let X be parametrized by q(s), s ∈ [0, a], such that q(0) = q0,
225
+ and has unit normal field nX(q(s)).
226
+ Set q0(s) = q(s).
227
+ Let qj(s), j ≥ 1 be the jth-reflection
228
+ points of the forward billiard trajectory γ(s) generated by x(s) = (q(s), nX(q(s)). We assume
229
+ that a > 0 is sufficiently small so that the jth-reflection points qj(s) belong to the same boundary
230
+ component ∂Kξj for every s ∈ [0, a]. Let 0 = t0(x(s)) < t1(x(s)) < ... < tm+1(x(s)) be the times
231
+ of the reflections of the ray γ(s) at ∂K. Let κj(s) be the curvature of ∂Kξj at qj(s) and φj(s)
232
+ be the collision angle between the outward unit normal to ∂K and the reflection ray of γ(s) at
233
+ qj(s). Also, let dj(s) be the distance between two reflection points i.e. dj(s) = ∥qj+1(s) − qj(s)∥,
234
+ j = 0, 1, . . . , m.
235
+ Given a large m ≥ 1, let tm(x(s)) < t < tm+1(x(s)). Set Φt( �
236
+ X) = �
237
+ Xt. Let π(Φt(x(s))) = p(s).
238
+ Then p(s), s ∈ [0, a], is a parametrization of the C3 curve Xt = π(Φt( �
239
+ X).
240
+ Next, let k0(s) > 0 be the curvature of X at q(s).
241
+ Let tj(x(s)) < τ < tj+1(x(s)), j =
242
+ 1, 2, . . . , m. Denote by uτ(s) be the shift of (q(s), n(q(s))) along the forward billiard trajectory
243
+ γ(s) after time τ > 0. Then Xτ = {uτ(s) : s ∈ [0, a]} is a C3 convex curve with respect to the
244
+ outward normal field n(uτ(s)). Let kj(s) > 0 be the curvature of Xtj = limτցtj(s) Xτ at qj(s). It
245
+ follows from Sinai [15] that
246
+ kj+1(s) =
247
+ kj(s)
248
+ 1 + dj(s)kj(s) + 2
249
+ κj+1(s)
250
+ cos φj+1(s)
251
+ ,
252
+ 0 ≤ j ≤ m − 1 .
253
+ (2.1)
254
+ 5
255
+
256
+ Moreover, the curvature of Xτ at uτ(s) is
257
+ kτ(s) =
258
+ kj(s)
259
+ 1 + (τ − tj(s))kj(s).
260
+ (2.2)
261
+ Set
262
+ δj(s) =
263
+ 1
264
+ 1 + dj(s)kj(s)
265
+ ,
266
+ 1 ≤ j ≤ m .
267
+ (2.3)
268
+ Theorem 2.2. [18] For all s ∈ [0, a] we have
269
+ ∥ ˙q(s)∥ = ∥ ˙p(s)∥δ1(s)δ2(s) . . . δm(s) .
270
+ (2.4)
271
+ This was proved in [18] in the 2D case and in [19] in the general case.
272
+ Finally, we want to introduce some notation related to the maximum and minimum of previous
273
+ billiard characteristies dj(s),κj(s), φj(s) and kj(s).
274
+ For all j, we have dmin ≤ dj(s) ≤ dmax,
275
+ where dmax and dmin are constants independent of j such that dmax = max{d(Ki, Kk)} and
276
+ dmin = min{d(Ki, Kk)} for i ̸= k. Also, since ∂K is strictly convex, we have constants κmin > 0
277
+ and κmax > 0 independent of j such that κmin ≤ κj(s) ≤ κmax. And it follows from the condition
278
+ (H) that there exists a constant φmax ∈ (0, π
279
+ 2 ) such that 0 ≤ φj(s) ≤ φmax < π
280
+ 2 , (see e.g. [17]).
281
+ Let kj(s) be as in equation (2.1). It follows easily that kmin ≤ kj(s) ≤ kmax, where kmin = 2κmin
282
+ and kmax =
283
+ 1
284
+ dmin +
285
+ 2κmax
286
+ cos φmax .
287
+ 3
288
+ Estimation of the largest Lyapunov exponent for open billiards
289
+ A formula for the largest Lyapunov exponents for a rather general class of billiards can be found
290
+ in [5], see Theorem 3.41 there. In our case we derive this formula again (see (3.1) below) and then
291
+ we use Theorem 2.2 to derive important regularity properties of the largest Lyapunov exponent.
292
+ Assume that µ is an ergodic σ-invariant measure on Σ, and let x0 = (q0, v0) ∈ M0 correspond
293
+ to a typical point in Σ with respect to µ via the representation map R. That is as in Theorem
294
+ 2.1, we have
295
+ λ1 = lim
296
+ m→∞
297
+ 1
298
+ m log ∥Dx0Bm(w)∥,
299
+ with 0 ̸= w ∈ Eu(x0). As in Sect. 2.4, let X be a (small) C3 strictly convex curve containing q0 and
300
+ having a unit normal field nX so that nX(q0) = v0. As in Sect. 2.4 again, let X be parametrised
301
+ by arc length via q(s), s ∈ [0, a], such that q(0) = q0. Let again qj(s), j = 1, 2, . . . , m + 1, be the
302
+ consecutive reflection points of the billiard trajectory γ(s) determined by x(s) = (q(s), nX(q(s)).
303
+ Given an integer m > 0 and assuming the interval [0, a] is sufficiently small, the jth reflection
304
+ points qj(s) belong to the same boundary component ∂Kξj for all s ∈ [0, a]. Next, define dj(s),
305
+ tj(x(s)), etc. as in Sect. 2.4, let tm(x(0)) < t < tm+1(x(0)), and let p(s) be the parametrisation
306
+ of �
307
+ Xt corresponding to q(s). Then the formula (2.4) in Theorem 2.2 (holds with ∥ ˙q(s)∥ = 1 from
308
+ our assumptions). Now the discussion in Sect. 2.4 implies that there exist some global constants
309
+ c1 > c2 > 0, independent of x0, X, m, etc. such that
310
+ c2∥ ˙p(s)∥ ≤ ∥Dx0Bm(w)∥ ≤ c1∥ ˙p(s)∥
311
+ 6
312
+
313
+ for all s ∈ [0, a]. So, by (2.4),
314
+ c2
315
+ δ1(0)δ2(0) . . . δm(0) ≤ ∥Dx0Bm(w)∥ ≤
316
+ c1
317
+ δ1(0)δ2(0) . . . δm(0)
318
+ for all s ∈ [0, a]. Using this for s = 0, taking logarithms and limits as m → ∞, we obtain
319
+ − lim
320
+ m→∞
321
+ 1
322
+ m log (δ1(0)δ2(0) . . . δm(0)) ≤ lim
323
+ m→∞
324
+ 1
325
+ m log ∥Dx0Bm(w)∥
326
+ ≤ − lim
327
+ m→∞
328
+ 1
329
+ m log (δ1(0)δ2(0) . . . δm(0)) .
330
+ Hence,
331
+ λ1 = lim
332
+ m→∞ − 1
333
+ m
334
+ m
335
+
336
+ i=1
337
+ log δi(0).
338
+ This implies that the largest Lyapunov exponent at the initial point x0, so at almost every point
339
+ wilt respect to the given measure µ, is given by
340
+ λ1 = lim
341
+ m→∞
342
+ 1
343
+ m
344
+ m
345
+
346
+ i=1
347
+ log
348
+
349
+ 1 + di(0)ki(0)
350
+
351
+ .
352
+ (3.1)
353
+ From equation (3.1), we can estimate the largest Lyapunov exponent from below and above as
354
+ log (1 + dminkmin) ≤ λ1 ≤ log (1 + dmaxkmax).
355
+ 4
356
+ Billiard deformations
357
+ In this section, we consider some changes to the billiards in the plane, such as moving, rotating,
358
+ and changing the shape of one or multiple obstacles. This kind of billiard transformation is called a
359
+ billiard deformation as defined in [21]. We describe this deformation by adding an extra parameter
360
+ α ∈ [0, b] for some b ∈ R+, which is called the deformation parameter, to the parametrization of
361
+ the boundary of obstacles i.e., if the boundary of an obstacle parametrized by ϕ(u), it will become
362
+ ϕ(u, α). In this section, we provide the definition a billiard deformation as defined in [21]. In
363
+ addition, we describe the propagation of unstable manifolds for billiard deformations. We also
364
+ estimate the higher derivatives of some of the billiard characteristics such as distance, collision
365
+ angle and curvature, with respect to deformation parameter α.
366
+ Let α ∈ I = [0, b], for some b ∈ R+, be a deformation parameter and let ∂Ki(α) be
367
+ parametrized counterclockwise by ϕi(ui, α) and parametrized by arc-length ui. Let qi = ϕi(ui, α)
368
+ be a point that lies on ∂Ki(α). Denote the perimeter of ∂Ki(α) by Li(α), and let Pi = {(ui, α) :
369
+ α ∈ I, ui ∈ [0, Li(α)]}.
370
+ Definition 4.1. [21] For any α ∈ I = [0, b], let K(α) be a subset of R2. For integers r ≥ 4, r′ ≥ 2,
371
+ we call K(α) a Cr,r′-billiard deformation (i.e. Cr with respect to u and Cr′ with respect to α) if
372
+ the following conditions hold for all α ∈ I:
373
+ 1. K(α) = �z0
374
+ i=1 Ki(α) satisfies the no-eclipse condition (H).
375
+ 7
376
+
377
+ 2. Each Ki(α) is a compact, strictly convex set with Cr boundary and total arc length Li(α).
378
+ 3. Each Ki is parametrized counterclockwise by arc-length with Cr,r′ functions ϕi : Pi → R2.
379
+ 4. For all integers 0 ≤ l ≤ r, 0 ≤ l′ ≤ r′ (apart from l = l′ = 0), there exist constants C(l,l′)
380
+ ϕ
381
+ depending only on the choice of the billiard deformation and the parametrizations ϕi, such
382
+ that for all integers i = 1, 2, 3, ..., z0,
383
+ ��� ∂l+l′ϕi
384
+ ∂ul
385
+ i∂αl′
386
+ ��� ≤ C(l,l′)
387
+ ϕ
388
+ .
389
+ Let Bα be the open billiard map on a non-wandering set Mα for K(α). Let Σ defined in Sec.
390
+ 2.2, we defined Rα : Mα → Σ by Rα(x(α)) = ξ(x(α)). We can write the points that correspond
391
+ to the billiard trajectories according to the parameterization in previous definition as follows,
392
+ π(Bj(x(α))) = qξj(α) = ϕξj(uξj(α), α) ∈ ∂Kξj(α), where uξj(α) ∈ [0, Lξj(α)]. For brevity, we will
393
+ write qj(α) = ϕj(uj(α), α).
394
+ The next corollary shows that uj(α) = uξj(α) for a fixed ξ ∈ Σ, is differentiable with respect
395
+ to α. This corollary is proved in [21].
396
+ Theorem 4.2. [21] Let K(α) be a Cr,r′ billiard deformation with r, r′ ≥ 2.
397
+ Then uj(α) is
398
+ Cmin{r−1,r′−1} with respect to α, and there exist constants C(n)
399
+ u
400
+ > 0 such that
401
+ ���dnuj(α)
402
+ dαn
403
+ ��� ≤ C(n)
404
+ u .
405
+ The next corollary follows from Definition 4.1 and Theorem 4.2.
406
+ Corollary 4.3. Let K(α) be a Cr,r′ billiard deformation with r, r′ ≥ 2. Let qj(α) belongs to ∂Kξj.
407
+ Then qj(α) is Cn, where n = min{r − 1, r′ − 1}, with respect to α, and there exist constants
408
+ C(n)
409
+ q
410
+ > 0 such that
411
+ ���dnqj(α)
412
+ dαn
413
+ ��� ≤ C(n)
414
+ q
415
+ .
416
+ 4.1
417
+ Propagation of unstable manifolds for billiard deformations
418
+ We described the unstable manifolds propagation in Section 2.4 for open billiards. Here in this
419
+ section, we describe it for billiard deformations.
420
+ Let K(α), α ∈ [0, b] be a Cr,r′ billiard deformation as in Definition 4.1 with r ≥ 3, r′ ≥ 1.
421
+ x0(α) = (q0(α), v0(α)) ∈ Mα and let W u
422
+ ǫ (x0(α)) be the local unstable manifold for x0(α) for
423
+ sufficiently small ǫ > 0. Take a curve Xα containing q0(α) such that Xα = {q0(s, α) : s ∈ [0, a]}
424
+ is a convex curve with outer unit normal field nX(q0(s, α)) = v0(α) and C3 with respect to s.
425
+ It follows from Sinai [15], [16] that W u
426
+ ǫ (x0(α)) = {(q0(s, α), nX(q0)) : s ∈ [0, a]}.
427
+ Set �
428
+ Xα =
429
+ W u
430
+ ǫ (x0(α)). Let a ∈ R+ be small enough such that all reflection points qj(s, α), j = 1, 2, ..., m,
431
+ that are generated by x0(s, α) = (q0(s, α), nX(q0(s, α))) belong to the same boundary ∂Kξj(α).
432
+ Let dj(s, α) = ∥qj+1(s, α) − qj(s, α)∥ be the distance between two reflection points qj+1(s, α) and
433
+ qj(s, α). Denote the curvature of ∂K(α) at qj(s, α) by κj(s, α), the collision angle between the
434
+ unit normal to ∂K(α) and the reflection vector at qj(s, α) by φj(s, α), and the curvature of X at
435
+ q0(s, α) by k0(s, α) .
436
+ 8
437
+
438
+ Let tj(x(s, α)) = tj(s, α) be the time of the j-th reflection. Given t with tj < t < tj+1 for
439
+ some j = 1, 2, ..., m, set π(Φt( �
440
+ Xα)) = Xαt. Then Xαt = {uαt(s, α) : s ∈ [0, a]} is C3 with respect
441
+ to s and a convex curve with outer unit normal field nXαt(uαt(s, α)). Denote the curvature of
442
+ Xαtj(s,α) at qj(s, α) by kj(s, α), where Xαtj(s,α) = limtցtj(s,α) Xαt. As in equation (2.1), we can
443
+ define kj(s, α) as follows:
444
+ kj+1(s, α) =
445
+ kj(s, α)
446
+ 1 + dj(s, α)kj(s, α) + 2
447
+ κj+1(s, α)
448
+ cos φj+1(s, α)
449
+ ,
450
+ 0 ≤ j ≤ m − 1 .
451
+ (4.1)
452
+ From now on, we will need to use previous characteristics in the case s = 0, so for brevity,
453
+ we will write dj(α) = dj(0, α), etc. Also, we denote the billiard deformation by K(α), so all of
454
+ its characteristics will be denoted dj(α), kj(α), etc. The initial open billiard is K(0) so all of its
455
+ characteristics will be denoted dj(0), etc.
456
+ 4.2
457
+ The higher derivatives of billiard characteristics
458
+ Let K(α) be a Cr,r′ billiard deformation as in
459
+ Definition 4.1 with r ≥ 4, r′ ≥ 2. Recall that
460
+ ∂Kξj(α) is parametrized by arc-length ujand qj(α) = ϕj(uj(α), α) ∈ ∂Kξj(α). Here, we state
461
+ some corollaries related to bounds of the higher derivatives of curvature, distance and collision
462
+ angle of a billiard deformation. These corollaries are forthright consequences of condition 4 in
463
+ Definition 4.1.
464
+ Corollary 4.4. Let K(α) be a Cr,r′ billiard deformation with r ≥ 4, r′ ≥ 2. Then the curvature
465
+ κj(α) at qj(α) is Cn, where n = min{r − 3, r′ − 1} with respect to α and there exist constants
466
+ C(n)
467
+ κ
468
+ > 0 depending only on n such that
469
+ ���dnκ
470
+ dαn
471
+ ��� ≤ C(n)
472
+ κ .
473
+ Proof. Suppose K(α) is a a Cr,r′ billiard deformation with r ≥ 3, r′ ≥ 1.
474
+ Since ∂Kj(α) is
475
+ paramitrized by arc-length uj, then the curvature of ∂Kj(α) at qj(α) = ϕj(uj(α), α) is κj = ∂2ϕj
476
+ ∂u2
477
+ j ,
478
+ for j = 0, 1, ..., m. Then κj(α) is Cmin{r−3,r′−1} with respect to α.
479
+ For the first derivative, we have
480
+ ���dκj
481
+
482
+ ��� =
483
+ ���∂3ϕj
484
+ ∂u3
485
+ j
486
+ ∂uj
487
+ ∂α + ∂3ϕj
488
+ ∂u2
489
+ j∂α
490
+ ��� ≤ C(1)
491
+ κ ,
492
+ this estimate was obtained in [21]. Next, we continue to estimate the second derivative, so we
493
+ have
494
+ ���d2κj
495
+ dα2
496
+ ��� =
497
+ ���∂4ϕj
498
+ ∂u4
499
+ j
500
+ �∂uj
501
+ ∂α
502
+ �2 + ∂3ϕj
503
+ ∂u3
504
+ j
505
+ ∂u2
506
+ j
507
+ ∂α2 + 2 ∂4ϕj
508
+ ∂u3
509
+ j∂α
510
+ ∂uj
511
+ ∂α +
512
+ ∂4ϕj
513
+ ∂u2
514
+ j∂α2
515
+ ���.
516
+ By using condition 4 in Definition 4.1 and Theorem 4.2, there exists a constant C(2)
517
+ κ
518
+ > 0 such
519
+ that
520
+ 9
521
+
522
+ ���d2κj
523
+ dα2
524
+ ��� ≤ C(4,0)
525
+ ϕ
526
+ (C(1)
527
+ u )2 + C(3,0)
528
+ ϕ
529
+ C(2)
530
+ u
531
+ + 2Cϕ(3,1)C(1)
532
+ u
533
+ + C(2,2)
534
+ ϕ
535
+ = C(2)
536
+ κ .
537
+ Continuing by induction we see that the n-th derivative, where n = min{r −3, r′ −1}, is bounded
538
+ by a constant C(n)
539
+ κ
540
+ > 0 which depends only on n such that
541
+ ���dnκ
542
+ dαn
543
+ ��� ≤ C(n)
544
+ κ .
545
+ Corollary 4.5. Let K(α) be a Cr,r′ billiard deformation with r ≥ 3, r′ ≥ 1. Then the distance
546
+ dj(α) between two points qj+1(α) and qj(α) is Cn, where n = min{r − 1, r′ − 1} with respect to α
547
+ and there exist constants C(n)
548
+ d
549
+ > 0 depending only on n such that
550
+ ���dndj
551
+ dαn
552
+ ��� ≤ C(n)
553
+ d .
554
+ Proof. Since dj = ∥qj+1(α) − qj(α)∥ = ∥ϕj+1(uj+1(α), α) − ϕj(uj(α), α)∥ for j = 0, 1, ..., m, then
555
+ dj is Cmin{r−1,r′−1}. The first derivative is
556
+ ddj
557
+ dα =
558
+ � ϕj+1(uj+1(α), α) − ϕj(uj(α), α)
559
+ ∥ϕj+1(uj+1(α), α) − ϕj(uj(α), α)∥, ∂ϕj+1
560
+ ∂uj+1
561
+ ∂uj+1
562
+ ∂α
563
+ + ∂ϕj+1
564
+ ∂α
565
+ + ∂ϕj
566
+ ∂uj
567
+ ∂uj
568
+ ∂α + ∂ϕj
569
+ ∂α
570
+
571
+ .
572
+ And then
573
+ ���ddj
574
+
575
+ ��� =
576
+ ���∂ϕj+1
577
+ ∂uj+1
578
+ ∂uj+1
579
+ ∂α
580
+ + ∂ϕj+1
581
+ ∂α
582
+ + ∂ϕj
583
+ ∂uj
584
+ ∂uj
585
+ ∂α + ∂ϕj
586
+ ∂α
587
+ ��� ≤ C(1)
588
+ d ,
589
+ which was estimated in [21]. For the second derivative, using condition 4 in Definition 4.1 and
590
+ Theorem 4.2 it follows that
591
+ ���d2dj
592
+ dα2
593
+ ��� =
594
+ ���∂2ϕj+1
595
+ ∂u2
596
+ j+1
597
+ �∂uj+1
598
+ ∂α
599
+ �2
600
+ + ∂ϕj+1
601
+ ∂uj+1
602
+ ∂2uj+1
603
+ ∂α2
604
+ + 2 ∂2ϕj+1
605
+ ∂uj+1∂α
606
+ ∂uj+1
607
+ ∂α
608
+ + ∂2ϕj+1
609
+ ∂α2
610
+ + ∂2ϕj
611
+ ∂u2
612
+ j
613
+ �∂uj
614
+ ∂α
615
+ �2
616
+ + ∂ϕj
617
+ ∂uj
618
+ ∂2uj
619
+ ∂α2 + 2 ∂2ϕj
620
+ ∂uj∂α
621
+ ∂uj
622
+ ∂α + ∂2ϕj
623
+ ∂α2
624
+ ���.
625
+ By using condition 4 in Definition 4.1 and Theorem 4.2, there exists a constant C(2)
626
+ κ
627
+ > 0 such
628
+ that
629
+ ���d2dj
630
+ dα2
631
+ ��� ≤ 2C(2,0)
632
+ ϕ
633
+ (C(1)
634
+ u )2 + 2C(2)
635
+ u
636
+ + 4C(1,1)
637
+ ϕ
638
+ (C(1)
639
+ u )2 + 2C(0,2)
640
+ ϕ
641
+ = C(2)
642
+ d .
643
+ Continuing by induction, we can see that there exists a constant C(n)
644
+ d
645
+ > 0 depends only on n such
646
+ that
647
+ ���dndj
648
+ dαn
649
+ ��� ≤ C(n)
650
+ d
651
+ .
652
+ Corollary 4.6. Let K(α) be a Cr,r′ billiard deformation with r ≥ 4, r′ ≥ 2. Then cos φj(α) is
653
+ Cmin{r−1,r′−1} and there exists a constant C(n)
654
+ φ
655
+ > 0 depending only on n such that
656
+ ���dn cos φj
657
+ dαn
658
+ ��� ≤ C(n)
659
+ φ .
660
+ 10
661
+
662
+ Proof. We can write
663
+ cos 2φj =
664
+
665
+ qj+1(α) − qj(α)
666
+
667
+ ·
668
+
669
+ qj(α) − qj−1(α)
670
+
671
+ |qj+1(α) − qj(α)||qj(α) − qj−1(α)|
672
+ =
673
+
674
+ ϕj+1(, uj+1, α) − ϕj(uj, α)
675
+
676
+ ·
677
+
678
+ ϕj(uj, α) − ϕj−1(uj−1α)
679
+
680
+ |ϕj+1(uj+1, α) − ϕj(uj, α)||ϕj(uj, α) − ϕj−1(uj−1, α)|
681
+ .
682
+ And then, cos φj(α) =
683
+
684
+ cos 2φj(α)+1
685
+ 2
686
+ . Therefore, the statement follows from condition 4 in Defi-
687
+ nition 4.1 and Corollary 4.2.
688
+ The next corollary follows from Corollaries 4.4, 4.6.
689
+ Corollary 4.7. Let K(α) be a Cr,r′ billiard deformation with r ≥ 4, r′ ≥ 2. Then the expression
690
+ gj(α) =
691
+ 2κj
692
+ cos φj is Cmin{r−3,r′−1} and there exist constants C(n)
693
+ g
694
+ > 0 depending only on n such that
695
+ ���dngj
696
+ dαn
697
+ ��� ≤ C(n)
698
+ g
699
+ .
700
+ The next corollary concerning the curvature kj, defined in (4.1), follows from Corollaries 4.5 and
701
+ 4.7.
702
+ Corollary 4.8. Let K(α) be a Cr,r′ billiard deformation with r ≥ 4, r′ ≥ 2. Then the curvature
703
+ kj(α) is Cn, where n = min{r − 3, r′ − 1} and here exist constants C(n)
704
+ k
705
+ depending only on n such
706
+ that
707
+ ���dnkj
708
+ dαn
709
+ ��� ≤ C(n)
710
+ k .
711
+ Proof. First, we recall
712
+ kj+1(α) =
713
+ kj(α)
714
+ 1 + dj(α)kj(α) + 2
715
+ κj+1(α)
716
+ cosφj+1(α)
717
+ ,
718
+ 0 ≤ j ≤ m − 1 .
719
+ We will write kj+1(α) simply as follows
720
+ kj+1(α) =
721
+ kj(α)
722
+ 1 + dj(α)kj(α) + gj+1(α),
723
+ where gj+1(α) =
724
+ 2κj+1
725
+ cos φj+1 . [21] contains an estimate that the first derivative of kj(α) with respect
726
+ to α is bounded by a constant C(1)
727
+ k . Here, we use the same argument in [21] and show that the
728
+ second derivative of kj(α) with respect to α is also bounded. These estimates are useful and will
729
+ be used later in Section 5.
730
+ Next, we start with the first derivative of kj+1 with respect to α and we will use the notation
731
+ ˙k, ¨k,...etc. to simplify equations. So, we have
732
+ ˙kj+1 =
733
+ ˙kj
734
+ (1 + djkj)2 −
735
+ ˙djk2
736
+ j
737
+ (1 + djkj)2 + ˙gj+1.
738
+ And for the second derivative, we have
739
+ ¨kj+1 =
740
+ ¨kj
741
+ (1 + djkj)2 −
742
+ k2
743
+ j ( ¨dj + ¨djdjkj − 2 ˙d2
744
+ jkj) + 2˙kj(˙kjdj + 2 ˙djkj)
745
+ (1 + djkj)3
746
+ + ¨gj+1.
747
+ 11
748
+
749
+ Let
750
+ βj =
751
+ 1
752
+ (1 + djkj)2 ,
753
+ ηj = −
754
+ k2
755
+ j( ¨dj + ¨djdjkj − 2 ˙d2
756
+ jkj) + 2˙kj(˙kjdj + 2 ˙djkj)
757
+ (1 + djkj)3
758
+ + ¨gj+1
759
+ ,
760
+ 0 ≤ j ≤ m − 1 .
761
+ From Corollaries 4.5 and 4.7, and the estimate of ˙kj, we have
762
+ |βj| ≤ βmax =
763
+ 1
764
+ (1 + dminkmin)2 ,
765
+ |ηj| ≤ ηmax = k2
766
+ max(C(2)
767
+ d
768
+ + C(2)
769
+ d dmaxkmax + 2(C(1)
770
+ d )2kmax)
771
+ (1 + dminkmin)3
772
+ + 2C(1)
773
+ k (C(1)
774
+ k dmax + 2C(1)
775
+ d kmax)
776
+ (1 + dminkmin)3
777
+ + C(2)
778
+ g .
779
+ Then, we have
780
+ ¨km(α) = ηm−1 + βm−1¨km−1(α)
781
+ = ηm−1 + βm−1 ηm−2 + .... + βm−1....β1 η0 + βm−1....β0 ¨k0(α).
782
+ To solve this equation, we assume that (q(α), v(α)) is periodic such that Bm
783
+ α (q(α), v(α)) =
784
+ (q(α), v(α)). Then km(α) = k0(α). From this, we can solve the previous equation as follows
785
+ ¨km(α) − βm−1....β0 ¨k(α) = ηm−1 + βm−1 ηm−2 + .... + βm−1....β1 η0
786
+ ¨km(α) =
787
+ 1
788
+ 1 − βm−1....β0
789
+
790
+ ηm−1 + βm−1 ηm−2 + .... + βm−1....β1
791
+
792
+ By the maximum value of ηj and βi, we have
793
+ |¨km(α)| ≤
794
+ ηmax
795
+ 1 − βm
796
+ max
797
+
798
+ 1 + βmax + .... + βm−1
799
+ max
800
+
801
+ =
802
+ ηmax
803
+ 1 − βm
804
+ max
805
+ �1 − βm
806
+ max
807
+ 1 − βmax
808
+
809
+ =
810
+ ηmax
811
+ 1 − βmax
812
+ .
813
+ This means there exists a constant C(2)
814
+ k
815
+ > 0 does not depend on m or α such that |¨kj(α)| ≤ C(2)
816
+ k ,
817
+ for every j = 0, 1, ..., m. Continuing by induction we can see that the n-th derivative of kj(α)
818
+ with respect to α is bounded by constant C(n)
819
+ k
820
+ > 0 that depending only on n.
821
+ 5
822
+ Continuity of the largest Lyapunov exponent
823
+ In this section, we show that the largest Lyapunov exponent λ1 depends continuously on a planar
824
+ billiard deformation. Let K(α) be a billiard deformation as defined in Definition 4.1 and let K(0)
825
+ be the initial open billiard. Let kj(α), kj(0) and dj(α), di(0) be the curvatures and the distances
826
+ that are described in section 4.1.
827
+ 12
828
+
829
+ For every α ∈ [0, b], let Mα be the non-wandering set for the billiard map and let Rα :
830
+ Mα −→ Σ be the analogue of the conjugacy map R : M0 −→ Σ, so that the following diagram is
831
+ commutative:
832
+
833
+
834
+ −→
835
+
836
+ �Rα
837
+ �Rα
838
+ Σ
839
+ σ
840
+ −→
841
+ Σ
842
+ where Bα is the billiard ball map on Mα. By Theorem 2.1 there exists a subset Aα of Σ with
843
+ µ(Aα) = 1 so that
844
+ λ1(α) = lim
845
+ m→∞
846
+ 1
847
+ m log ∥Dx0Bm
848
+ α (w)∥
849
+ (5.1)
850
+ for all x ∈ Mα with Rα(x) ∈ Aα. Similarly, let A0 be the set with µ(A0) = 1 which we get from
851
+ Theorem 2.1 for α = 0.
852
+ Lemma 5.1. Given an arbitrary sequence
853
+ α1, α2, . . . , αp, . . .
854
+ of elements of [0, b], for µ-almost all ξ ∈ Σ the formula (5.1) is valid for α = αp and x = R−1
855
+ α (ξ)
856
+ for all p = 1, 2, . . . and also for α = 0 and x = R−1(ξ).
857
+ Proof. The set A = A0 ∩ ∩∞
858
+ p=1Aαp has µ(A) = 1 since
859
+ Σ \ A = (Σ \ A0) ∪ ∪∞
860
+ p (Σ \ Aαp)
861
+ has measure zero as a countable union of sets of measure zero. If α = αp for some p and Rα(x) ∈ A,
862
+ then Rα(x) ∈ Aαp so formula (5.1) holds. Similarly (5.1) holds for α = 0 as well.
863
+ Thus, using the notation x(0, α) ∈ Mα, we can choose ξ ∈ Σ so that formula (5.1) applies for
864
+ α = αp and x = x(0, αp) for all p = 1, 2, . . ., and also for α = 0 and x = (0, 0).
865
+ From the formula for the largest Lyapunov exponent (3.1), we can write the Lyapunov expo-
866
+ nents for K(α) and K(0) as follows:
867
+ λ1(α) = lim
868
+ m→∞
869
+ 1
870
+ m
871
+ m
872
+
873
+ j=1
874
+ log
875
+
876
+ 1 + dj(α)kj(α)
877
+
878
+ = lim
879
+ m→∞ λ(m)
880
+ 1
881
+ (α),
882
+ λ1(0) = lim
883
+ m→∞
884
+ 1
885
+ m
886
+ m
887
+
888
+ j=1
889
+ log
890
+
891
+ 1 + dj(0)kj(0)
892
+
893
+ = lim
894
+ m→∞ λ(m)
895
+ 1
896
+ (0),
897
+ where
898
+ λ(m)
899
+ 1
900
+ (α) = 1
901
+ m
902
+ m
903
+
904
+ j=1
905
+ log
906
+
907
+ 1 + dj(α)kj(α)
908
+
909
+ and
910
+ λ(m)
911
+ 1
912
+ (0) = 1
913
+ m
914
+ m
915
+
916
+ j=1
917
+ log
918
+
919
+ 1 + dj(0)kj(0)
920
+
921
+ .
922
+ (5.2)
923
+ Now, we prove Theorem 1.1
924
+ 13
925
+
926
+ Proof of Theorem 1.1: Let K(α) be a C4,2 billiard deformation in R2, and let
927
+ α ∈ [0, b]. Assume that λ1(α) is not continuous at α = 0. Then there exists ε > 0 and a sequence
928
+ α1 > α2 > ... > αp > ... → 0 in [0, b] with αp → 0 such that |λm
929
+ 1 (αk) − λm
930
+ 1 (0)| ≥ ε for all p ≥ 1.
931
+ By using Lemma 5.1 and the previous expressions of λm
932
+ 1 (α) for α = αp and λm
933
+ 1 (0) in (5.2), we
934
+ have
935
+ �����λm
936
+ 1 (αp) − λm
937
+ 1 (0)
938
+ ����� =
939
+ �����
940
+ 1
941
+ m
942
+ m
943
+
944
+ j=1
945
+ (log δj(αp) − log δj(0))
946
+ �����
947
+ =
948
+ �����
949
+ −1
950
+ m
951
+ m
952
+
953
+ j=1
954
+ (log(1 + dj(αp)kj(αp)) − log(1 + dj(0)kj(0)))
955
+ �����
956
+ ≤ 1
957
+ m
958
+ m
959
+
960
+ j=1
961
+ ����� log(1 + dj(αp)kj(αp)) − log(1 + dj(0)kj(0))
962
+ �����
963
+ ≤ 1
964
+ m
965
+ m
966
+
967
+ j=1
968
+ �����
969
+ 1 + dj(αp)kj(αp) − (1 + dj(0)kj(0))
970
+ 1 + min{dj(αp)kj(αp), dj(0)kj(0)}
971
+ �����
972
+ = 1
973
+ m
974
+ m
975
+
976
+ j=1
977
+ �����
978
+ dj(αp)kj(αp) − dj(0)kj(0)
979
+ 1 + dminkmin
980
+ �����
981
+ = 1
982
+ m C0
983
+ m
984
+
985
+ j=1
986
+ �����dj(αp)kj(αp) − dj(0)kj(0)
987
+ �����
988
+ = 1
989
+ m C0
990
+ m
991
+
992
+ j=1
993
+ �����(dj(αp) − dj(0))kj(αp) + dj(0)(kj(αp) − kj(0))
994
+ �����,
995
+ where C0 =
996
+ 1
997
+ 1+dminkmin > 0 is a global constant independent of αp.
998
+ Fix a small δ > 0; we will state later how small δ > 0 should be. Next consider p sufficiently large so
999
+ that αp < δ. For all p, we have |kj(αp)−kj(0)| = αp|˙kj(s(αp))| and |dj(αp)−dj(0)| = αp| ˙dj(r(αp))|,
1000
+ for some s(αp), r(αp) ∈ [0, αp]. From Corollaries 4.5 and 4.8 , there exist constants Ck and Cd
1001
+ such that |˙kj(s(αp))| ≤ Ck and | ˙dj(s(αp))| ≤ Cd. Therefore for all j,
1002
+ |kj(αp) − kj(0)| ≤ αpCk < δCk, and |dj(αp) − dj(0)| ≤ αpCd < δCd. Then
1003
+ ���λm
1004
+ 1 (αp) − λm
1005
+ 1 (0)
1006
+ ��� ≤ 1
1007
+ m C0
1008
+ m
1009
+
1010
+ j=1
1011
+ ����dj(αp) − dj(0)
1012
+ ���kj(αp) + dj(0)
1013
+ ���kj(αp) − kj(0)
1014
+ ���
1015
+
1016
+ < 1
1017
+ m C0
1018
+ m
1019
+
1020
+ j=1
1021
+ δ(Cdkmax + Ckdmax)
1022
+ = C0δ(Cdkmax + Ckdmax) < ε,
1023
+ if we take δ <
1024
+ ε
1025
+ Cdkmax+Ckdmax . We now have a contradiction because with the choice of the sequence
1026
+ α1 > α2 > ... > αp > ... → 0 in [0, b]. Therefore the statement is proved.
1027
+ 14
1028
+
1029
+ 6
1030
+ Differentiability of the largest Lyapunov exponent
1031
+ Here we prove Theorem 1.2
1032
+ Proof of Theorem 1.2: We will prove differentiability at α = 0. From this differentiability at any
1033
+ α ∈ [0, b] follows. To prove the differentiability at α = 0, we have to show that there exists
1034
+ lim
1035
+ α→0
1036
+ λ1(α) − λ1(0)
1037
+ α
1038
+ .
1039
+ Equivalently, there exists a number F such that
1040
+ lim
1041
+ p→∞
1042
+ λ1(αp) − λ1(0)
1043
+ αp
1044
+ = F,
1045
+ for any sequence α1 > α2 > ... > αp > ... → 0 as p → ∞ in [0, b].
1046
+ Let K(α) ⊂ R2 be a C5,3 billiard deformation and α ∈ [0, b] for a positive number b. Let λ1(α) be
1047
+ the largest Lyapunov exponent for K(α) and λ1(0) be the largest Lyapunov exponent for K(0).
1048
+ By using Lemma 5.1 and the expressions of λm
1049
+ 1 (α) for α = αp and λm
1050
+ 1 (0) in (5.2), we have
1051
+ λ(m)
1052
+ 1
1053
+ (αp) → λ1(αp) and λ(m)
1054
+ 1
1055
+ (0) → λ1(0) when m → ∞. Also,
1056
+ λ(m)
1057
+ 1
1058
+ (αp) − λ(m)
1059
+ 1
1060
+ (0)
1061
+ αp
1062
+ = − 1
1063
+ m
1064
+ m
1065
+
1066
+ j=1
1067
+ log δj(αp) − log δj(0)
1068
+ αp
1069
+ = − 1
1070
+ m
1071
+ m
1072
+
1073
+ j=1
1074
+ log
1075
+
1076
+ 1 + dj(αp)kj(αp)
1077
+
1078
+ − log
1079
+
1080
+ 1 + dj(0)kj(0)
1081
+
1082
+ αp
1083
+ .
1084
+ Set fj(αp) = log
1085
+
1086
+ 1 + dj(αp)kj(αp)
1087
+
1088
+ and fj(0) = log
1089
+
1090
+ 1 + dj(0)kj(0)
1091
+
1092
+ . Then
1093
+ λ(m)
1094
+ 1
1095
+ (αp) − λ(m)
1096
+ 1
1097
+ (0)
1098
+ αp
1099
+ = − 1
1100
+ m
1101
+ m
1102
+
1103
+ j=1
1104
+ fj(αp) − fj(0)
1105
+ αp
1106
+ .
1107
+ Taylor’s formula gives
1108
+ fj(αp) = fj(0) + αp ˙fj(0) + α2
1109
+ p
1110
+ 2
1111
+ ¨fj(rj(αp))
1112
+ for some rj(αp) ∈ [0, αp]. Then
1113
+ fj(αp) − fj(0)
1114
+ αp
1115
+ − ˙fj(0) = αp
1116
+ 2
1117
+ ¨fj(rj(αp)).
1118
+ Let
1119
+ Fm = 1
1120
+ m
1121
+ m
1122
+
1123
+ j=1
1124
+ ˙fj(0).
1125
+ Summing up the above for j = 1, 2, ..., m, we get
1126
+ λ(m)
1127
+ 1
1128
+ (αp) − λ(m)
1129
+ 1
1130
+ (0)
1131
+ αp
1132
+ − Fm = − 1
1133
+ m
1134
+ m
1135
+
1136
+ j=1
1137
+ �fj(αp) − fj(0)
1138
+ αp
1139
+ − ˙fj(0)
1140
+
1141
+ .
1142
+ 15
1143
+
1144
+ From the definition of fj(αp),
1145
+ ˙fj(αp) =
1146
+ ˙dj(αp)kj(αp) + dj(αp)˙kj(αp)
1147
+ 1 + dj(αp)kj(αp)
1148
+ ,
1149
+ and therefore,
1150
+ ¨fj(αp) =
1151
+ � ¨dj(αp)kj(αp) + 2 ˙dj(αp)˙kj(αp) + dj(αp)¨kj(αp)
1152
+ ��
1153
+ 1 + dj(αp)kj(αp)
1154
+
1155
+
1156
+ 1 + dj(αp)kj(αp)
1157
+ �2
1158
+
1159
+ � ˙dj(αp)k(αp) + dj(αp)˙kj(αp)
1160
+ �2
1161
+
1162
+ 1 + dj(αp)kj(αp)
1163
+ �2
1164
+ .
1165
+ Then from Corollaries 4.5 and 4.8, we get
1166
+ ��� ˙fj(αp)
1167
+ ��� ≤ C(1)
1168
+ d kmax + dmaxC(1)
1169
+ k
1170
+ 1 + dminkmin
1171
+ = C1,
1172
+ ��� ¨fj(αp)
1173
+ ��� ≤
1174
+
1175
+ C(2)
1176
+ d kmax + 2C(1)
1177
+ d C(1)
1178
+ k
1179
+ + dmaxC(2)
1180
+ k
1181
+ ��
1182
+ 1 + dmaxkmax
1183
+
1184
+
1185
+ 1 + dminkmin
1186
+ �2
1187
+ +
1188
+
1189
+ C(1)
1190
+ d kmax + dmaxC(1)
1191
+ k
1192
+ �2
1193
+
1194
+ 1 + dminkmin
1195
+ �2
1196
+ = C2.
1197
+ Therefore
1198
+ | ¨fj(rj(αp))| ≤ C2,
1199
+ for some constant C2 > 0 independent of rj(αp) and j. This implies
1200
+ ���λ(m)
1201
+ 1
1202
+ (αp) − λ(m)
1203
+ 1
1204
+ (0)
1205
+ αp
1206
+ − Fm
1207
+ ��� ≤ 1
1208
+ m
1209
+ m
1210
+
1211
+ j=1
1212
+ αp
1213
+ 2
1214
+ ��� ¨fj(tj(αp))
1215
+ ���
1216
+ ≤ C2
1217
+ 2 αp.
1218
+ Since | ˙fj(αp)| ≤ C1, we have |Fm| ≤
1219
+ 1
1220
+ m
1221
+ �m
1222
+ j=1 | ˙fj(0)| ≤ C1, for all m. Therefore, the sequence
1223
+ {Fm} has convergent subsequences. Let for example Fmh → F, for some sub-sequence {mh}.
1224
+ Then
1225
+ ���λ(mh)
1226
+ 1
1227
+ (αp) − λ(mh)
1228
+ 1
1229
+ (0)
1230
+ αp
1231
+ − Fmh
1232
+ ��� ≤ C2
1233
+ 2 αp,
1234
+ for all h ≥ 1. So, letting h → ∞, we get
1235
+ ���λ1(αp) − λ1(0)
1236
+ αp
1237
+ − F
1238
+ ��� ≤ C2
1239
+ 2 αp,
1240
+ and letting αp → 0 as p → ∞ we get that there exists
1241
+ lim
1242
+ p→∞
1243
+ λ1(αp) − λ1(0)
1244
+ αp
1245
+ = F.
1246
+ for every sequence α1 > α2 > ... > αp > ... → 0 as p → ∞ in [0, b].
1247
+ Thus, there exists
1248
+ F = limm→∞ 1
1249
+ m
1250
+ �m
1251
+ j=1 ˙fj(0). This is true for every subsequence {mh}, so for any subsequence we
1252
+ have Fmh → F. Hence, Fm converges to F as well. This implies that there exists
1253
+ 16
1254
+
1255
+ lim
1256
+ α→0
1257
+ λ1(α) − λ1(0)
1258
+ α
1259
+ = F,
1260
+ so λ1 is differentiable at α = 0 and ˙λ1(0) = F.
1261
+ Corollary 6.1. Let K(α) be a C5,3 billiard deformation. Then there exists a constant Cλ1 > 0
1262
+ such that
1263
+ ���dλ1(α)
1264
+
1265
+ ��� ≤ Cλ1,
1266
+ for all α ∈ [0, b].
1267
+ Proof. We have
1268
+ λ1(α) = lim
1269
+ m→∞
1270
+ 1
1271
+ m
1272
+ m
1273
+
1274
+ j=1
1275
+ log(1 + dj(α)kj(α)).
1276
+ By Theorem 1.2, λ1(α) is C1. So, from the formula in the previous proof that
1277
+ ˙λ1(0) = limm→∞ 1
1278
+ m
1279
+ �m
1280
+ j=1 ˙fj(0), we have
1281
+ dλ1
1282
+ dα = lim
1283
+ m→∞
1284
+ 1
1285
+ m
1286
+ m
1287
+
1288
+ j=1
1289
+ ddj
1290
+ dα kj(α) + dj(α)dkj
1291
+
1292
+ 1 + dj(α)kj(α)
1293
+ .
1294
+ From Corollaries 4.5 and 4.8, there exist constants C(1)
1295
+ d , C(1)
1296
+ k
1297
+ > 0 such that
1298
+ ���ddj
1299
+
1300
+ ��� ≤ C(1)
1301
+ d
1302
+ and
1303
+ ���dkj
1304
+
1305
+ ��� ≤ C(1)
1306
+ k . Then, we have
1307
+ ���dλ1
1308
+
1309
+ ��� ≤ lim
1310
+ m→∞
1311
+ 1
1312
+ m
1313
+ m
1314
+
1315
+ j=1
1316
+ C(1)
1317
+ d kmax + dmaxC(1)
1318
+ k
1319
+ 1 + dminkmin
1320
+ = C(1)
1321
+ d kmax + dmaxC(1)
1322
+ k
1323
+ 1 + dminkmin
1324
+ = Cλ1.
1325
+ This proves the statement.
1326
+ Acknowledgment
1327
+ The author would like to thank Prof. Luchezar Stoyanov for his suggestions, comments, and help.
1328
+ This work was supported by a scholarship from Najran University, Saudi Arabia.
1329
+ References
1330
+ [1] L. Barreira and Ya. Pesin, Lyapunov exponents and smooth ergodic theory. Univ. Lect. Series 23,
1331
+ American Mathematical Society, Providence, RI, 2001.
1332
+ [2] R. Bowen, Symbolic dynamics for hyperbolic flows. Amer. J. Math. 95 (1973), 429-460.
1333
+ [3] P. Duarte, S. Klein and M. Poletti, H¨older continuity of the Lyapunov exponents of linear cocycles over
1334
+ hyperbolic maps. Math. Z. 302 (2022), 2285–2325.
1335
+ 17
1336
+
1337
+ [4] N. Chernov, Entropy, Lyapunov exponents, and mean free path for billiards. Journal of Statistical
1338
+ Physics, 88 (1997), 1-29.
1339
+ [5] N. Chernov and R. Markarian, Chaotic Billiards. Math. Surveys and Monographs Vol. 127, Amer.
1340
+ Math. Soc. 2006.
1341
+ [6] M. Ikawa, Decay of solutions of the wave equation in the exterior of several strictly convex bodies. Ann.
1342
+ Inst. Fourier 38 (1988), 113-146.
1343
+ [7] A. Katok and J. M. Strelcyn, Invariant Manifolds, Entropy and Billiards; Smooth Maps with Singular-
1344
+ ities. Lecture Notes in Mathematics 1222, Springer, 1986.
1345
+ [8] A. Lopes and R. Markarian, Open billiards: invariant and conditionally invariant probabilities on
1346
+ Cantor sets. SIAM J. Appl. Math. 56 (1996), 651-680.
1347
+ [9] R. Markarian, Billiards with Pesin Region of Measure one. Comm. in Math Phys. 118 (1988), 87-97.
1348
+ [10] R. Markarian, New ergodic Billiards: exact results. Nonlinearity 6. (1993), 819-841
1349
+ [11] T. Morita, The symbolic representation of billiards without boundary condition. Trans. Amer. Math.
1350
+ Soc. 325 (1991), 819-828.
1351
+ [12] V. I. Oseledets, A multiplicative ergodic theorem. Lyapunov characteristic numbers for dynamical
1352
+ systems. Trans. Moscow Math. Soc. 19 (1968), 197-221.
1353
+ [13] M. Pollicott, Lectures on ergodic theory and Pesin theory on compact manifolds. Cambridge Univ.
1354
+ Press, Cambridge 1993.
1355
+ [14] V. Petkov and L. Stoyanov, Geometry of Reflecting Rays and Inverse Spectral Problems. Wiley, Chich-
1356
+ ester, (1992).
1357
+ [15] Ya. Sinai, Dynamical systems with elastic reflections. Russian Math. Surveys 25 (1970), 137-190.
1358
+ [16] Ya. Sinai, Development of Krylov’s ideas, An addendum to: N.S.Krylov ”Works on the foundations
1359
+ of statistical physics”. Princeton Univ. Press, Princeton 1979, 239-281.
1360
+ [17] L. Stoyanov, Exponential instability and entropy for a class of dispersing billiards. Ergod. Th. &
1361
+ Dynam. Sys. 19 (1999), 201-226.
1362
+ [18] L. Stoyanov, Spectrum of the Ruelle operator and exponential decay of correlation for open billiard
1363
+ flows. Amer. J. Math. 123 (2001), 715-759.
1364
+ [19] L. Stoyanov, Non-integrability of open billiard flows and Dolgopyat-type estimates. Ergodic Th. & Dyn.
1365
+ Systems 32 (2012), 295-313.
1366
+ [20] M. Viana, Lectures on Lyapunov exponents, Cambridge Studies in Adv. Math. vol.145, Cambridge
1367
+ Univ. Press 2014.
1368
+ [21] P. Wright, Differentiability of the Hausdorff dimension of the non-wandering set in a planar open
1369
+ billiard, Discrete & Continuous Dynamical Systems 36(7) (2016), 3993-4014.
1370
+ [22] M, P. Wojtkowski, Principles for the design of billiards with nonvanishing Lyapunov exponents. Com-
1371
+ mun. Math. Phys. 105 (1986), 391-414.
1372
+ 18
1373
+
1dAzT4oBgHgl3EQf8v7L/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
4tFKT4oBgHgl3EQfRy39/content/tmp_files/2301.11773v1.pdf.txt ADDED
@@ -0,0 +1,3963 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+ Automatic Modulation Classification with Deep
3
+ Neural Networks
4
+ Clayton A. Harper, Mitchell A. Thornton, and Eric C. Larson
5
+ Darwin Deason Institute for Cyber Security
6
+ {caharper, mitch, eclarson}@smu.edu
7
+ Abstract—Automatic modulation classification is a desired
8
+ feature in many modern software-defined radios. In recent years,
9
+ a number of convolutional deep learning architectures have
10
+ been proposed for automatically classifying the modulation used
11
+ on observed signal bursts. However, a comprehensive analysis
12
+ of these differing architectures and importance of each design
13
+ element has not been carried out. Thus it is unclear what
14
+ tradeoffs the differing designs of these convolutional neural
15
+ networks might have. In this research, we investigate numerous
16
+ architectures for automatic modulation classification and perform
17
+ a comprehensive ablation study to investigate the impacts of
18
+ varying hyperparameters and design elements on automatic
19
+ modulation classification performance. We show that a new
20
+ state of the art in performance can be achieved using a subset
21
+ of the studied design elements. In particular, we show that
22
+ a combination of dilated convolutions, statistics pooling, and
23
+ squeeze-and-excitation units results in the strongest performing
24
+ classifier. We further investigate this best performer according
25
+ to various other criteria, including short signal bursts, common
26
+ misclassifications, and performance across differing modulation
27
+ categories and modes.
28
+ Index Terms—Automatic modulation classification, deep learn-
29
+ ing, convolutional neural network.
30
+ I. INTRODUCTION
31
+ A
32
+ UTOMATIC modulation classification (AMC) is of par-
33
+ ticular interest for radio frequency (RF) analysis and in
34
+ modern software-defined radios to perform numerous tasks
35
+ including “spectrum interference monitoring, radio fault detec-
36
+ tion, dynamic spectrum access, opportunistic mesh network-
37
+ ing, and numerous regulatory and defense applications” [1].
38
+ Upon detection of an RF signal with unknown characteristics,
39
+ AMC is a crucial initial procedure in order to demodulate the
40
+ signal. Efficient AMC allows for maximal usage of transmis-
41
+ sion mediums and can provide resilience in modern cognitive
42
+ radios. Systems capable of adaptive modulation schemes can
43
+ monitor current channel conditions with AMC and adjust
44
+ exercised modulation schemes to maximize usage across the
45
+ transmission medium.
46
+ Moreover, for receivers that have a versatile demodulation
47
+ capability, AMC is a requisite task. The correct demodulation
48
+ scheme must be applied to recover the modulated message
49
+ within a detected signal. In systems where the modulation
50
+ scheme is not known a priori, AMC allows for efficient predic-
51
+ tion of the employed modulation scheme. Higher performing
52
+ AMC can increase the throughput and accuracy of these
53
+ systems; therefore, AMC is currently an important research
54
+ topic in the fields of machine learning and communication
55
+ systems, specifically for software-defined radios.
56
+ Typical benchmarks are constructed on the premise that the
57
+ AMC model must classify not only the mode of modulation
58
+ (e.g., QAM), but the exact variant of that mode of modulation
59
+ (e.g., 32QAM). While many architectures have proven to be
60
+ effective at high signal to noise ratios (SNRs), performance
61
+ degrades significantly at lower SNRs that often occur in real-
62
+ world applications. Other works have investigated increasing
63
+ classification performance at lower SNR levels through the
64
+ use of SNR-specific modulation classifiers [2] and clustering
65
+ based on SNR ranges [3]. To perform classification, a variety
66
+ of signal features have been investigated. Historically, AMC
67
+ has relied upon statistical moments and higher order cumulants
68
+ [4]–[6] derived from the received signal. Recent approaches
69
+ [1], [7]–[9] use raw time-domain in-phase (I) and quadrature
70
+ (Q) components as features to predict the modulation variant
71
+ of a signal. Further works have investigated additional features
72
+ including I/Q constellation plots [10]–[12].
73
+ After selecting the signal input features, machine learning
74
+ models are used to determine statistical patterns in the data
75
+ for the classification task. Support vector machines, decision
76
+ trees, and neural networks are commonly used classifiers for
77
+ this application [1], [3], [7]–[10], [13], [14]. Residual neural
78
+ networks (ResNets), along with convolutional neural networks
79
+ (CNNs), have been shown to achieve high classification perfor-
80
+ mance for AMC [1], [3], [7]–[10]. Thus, deep learning based
81
+ methods in AMC have become more prevalent due to their
82
+ promising performance and their ability to generalize to large,
83
+ complex datasets.
84
+ While other works have contributed to increased AMC
85
+ performance, the importance of many design elements for
86
+ AMC remains unclear and a number of architectural elements
87
+ have yet to be investigated. Therefore, in this work, we aim
88
+ to formalize the impact of a variety of architectural changes
89
+ and model design decisions on AMC performance. Numerous
90
+ modifications to architectures from previous works, including
91
+ our own [7], and novel combinations of elements applied to
92
+ AMC are considered. After an initial investigation, we provide
93
+ a comprehensive ablation study in this work to investigate
94
+ the performance impact of various architectural modifications.
95
+ Additionally, we achieve new state-of-the-art classification
96
+ performance on the RadioML 2018.01A dataset [15]. Using
97
+ the best performing model, we provide additional analyses
98
+ that characterize its performance across modulation modes and
99
+ arXiv:2301.11773v1 [cs.LG] 27 Jan 2023
100
+
101
+ 2
102
+ Fig. 1. ResNet architecture used in [1]. Each block represents a unit in the network, which may be comprised of several layers and connections as shown
103
+ on the right of the figure. Dimensions of the tensors on the output of each block are also shown where appropriate.
104
+ signal burst duration.
105
+ II. RELATED WORK
106
+ The area of AMC has been investigated by several research
107
+ groups. We provide a summary of results in AMC to provide
108
+ context and motivation for our contributions to AMC and the
109
+ corresponding ablation study described in this paper.
110
+ Corgan et al. [8] illustrate that deep convolutional neural
111
+ networks are able to achieve high classification performance
112
+ particularly at low SNRs on a dataset comprising 11 different
113
+ types of modulation. It was found that CNNs exceeded perfor-
114
+ mance over expertly crafted features. Comparing results with
115
+ architectures in [8] and [1], [16] improved AMC performance
116
+ utilizing self-supervised contrastive learning. First, an encoder
117
+ is pre-trained in a self-supervised manner through creating
118
+ contrastive pairs with data augmentation. By creating different
119
+ views of the input data through augmentation, contrastive loss
120
+ is used to maximize the cosine similarity between positive
121
+ pairs (augmented views of the same input). Once converged,
122
+ the encoder is frozen (i.e., the weights are set to fixed
123
+ values) and two fully-connected layers are added following the
124
+ encoder to form the classifier. The classifier is trained using
125
+ supervised learning to predict the 11 different modulation
126
+ schemes. Chen et al. applied a novel architecture to the
127
+ same dataset where the input signal is sliced and transformed
128
+ into a square matrix and apply a residual network to predict
129
+ the modulation schemes [17]. Other work has investigated
130
+ empirical and variational mode decomposition to improve few-
131
+ shot learning for AMC [18]. In our work, we utilize a larger,
132
+ more complex dataset consisting of 24 modulation schemes,
133
+ as well as modeling improvements.
134
+ Spectrograms and I/Q constellation plots in [19] were found
135
+ to be effective input features to a traditional CNN achieving
136
+ nearly equivalent performance as the baseline CNN network
137
+ in [1] which used raw I/Q signals.
138
+ Further, [10]–[12] also used I/Q constellations as an input
139
+ feature in their machine learning models on a smaller scale
140
+ of four or eight modulation types. Other features have been
141
+ used in AMC— [20], [21] utilized statistical features and
142
+ support vector machines while [22], [23] used fusion methods
143
+ in CNN classifiers. Mao et al. utilized various constellation
144
+ diagrams at varying symbol timings alleviating symbol timing
145
+ synchronization concerns [24]. A squeeze-and-excitation [25]
146
+ inspired architecture was used as an attention mechanism to
147
+ focus on the most important diagrams.
148
+ Although spectrograms and constellation plots have shown
149
+ promise, they require additional processing overhead and have
150
+ had comparable performance to raw I/Q signals. In addition,
151
+ models that use raw I/Q signals could be more adept at
152
+ handling varying-length signals than constellation plots be-
153
+ cause they are not limited by periodicity constraints for short
154
+ duration signals (i.e., burst transmissions). Consequently, we
155
+ utilize raw I/Q signals in our work.
156
+ Tridgell, in his dissertation [26], builds upon these works by
157
+ investigating these architectures when deployed on resource-
158
+ limited Field Programmable Gate Arrays (FGPAs). His work
159
+ stresses the importance of reducing the number of parameters
160
+ for modulation classifiers because they are typically deployed
161
+ in resource-constrained embedded systems.
162
+ Fig. 2. X-Vector architecture overview. The convolutional activations imme-
163
+ diately before pooling are shown. These activations are fed into two statistical
164
+ pooling layers that collapse the activations over time, creating a fixed-length
165
+ tensor that can be further processed by fully connected dense layers.
166
+
167
+ ResNet Architecture
168
+ Residual Stack
169
+ Residual Unit
170
+ Input ↓ Batch size ×1024 ×2
171
+ Batch size x 128 x 32
172
+ Batch size x 512
173
+ Residual Stack
174
+ Residual Stack
175
+ Dense + SeLU (128)
176
+ Input
177
+ Conv1D + Linear (32, 1)
178
+ Conv1D + ReLU (32, 3)
179
+ Batch size × 512 × 32
180
+ ↓ Batch size x× 64 × 32
181
+ Batch size x 128
182
+ Residual Stack
183
+ Residual Stack
184
+ Dense + SeLU (128)
185
+ Residual Unit
186
+ Conv1D + Linear (32, 3)
187
+ ↓ Batch size × 256 × 32
188
+ Batch size x 32 x 32
189
+ Batch size x 128
190
+ Dense + Softmax (24)
191
+ Residual Unit
192
+ Residual Stack
193
+ Residual Stack
194
+ Batch size x 16 × 32
195
+ Batch size x 24
196
+ Max Pooling (stride=2)
197
+ ndno
198
+ Flatten
199
+ Prediction
200
+ ↑ andno
201
+ Conv1D
202
+ (number of filters, filter size)Time
203
+ μ
204
+ Mean
205
+ Statistics Pooling
206
+ Dense
207
+ Channels
208
+ Across Channels
209
+ Layers
210
+ 0
211
+ Variance
212
+ Fixed-length
213
+ Convolutional Activations
214
+ X-Vector
215
+ Pooled
216
+ Statistics3
217
+ Fig. 3. Proposed CNN Architecture in [7]. This is the first work to employ an X-Vector inspired architecture for AMC showing strong performance. This
218
+ architecture is used as a baseline for the modifications investigated in this paper. The f and k variables shown designate the number of kernels and size of
219
+ each kernel, respectively, in each layer. These parameters are investigated for optimal sizing in our initial investigation.
220
+ In [1], Oshea et al. created a dataset with 24 different
221
+ types of modulation, known as RadioML 2018.01A, and
222
+ achieved high classification performance using convolutional
223
+ neural networks—specifically using residual connections (see
224
+ Figure 1) within the network (ResNet). A total of 6 residual
225
+ stacks were used in the architecture. A residual stack is defined
226
+ as a series of a convolutional layers, residual units, and a max
227
+ pooling operation as shown in Figure 1. The ResNet employed
228
+ by [1] attained approximately 95% classification accuracy at
229
+ high SNR values.
230
+ Harper et al. proposed the use of X-Vectors [27] to increase
231
+ classification performance using CNNs [7]. X-Vectors are tra-
232
+ ditionally used in speaker recognition and verification systems
233
+ making use of aggregate statistics. X-Vectors employ statistical
234
+ moments, specifically mean and variance, across convolutional
235
+ filter outputs. It can be theorized that taking the mean and
236
+ variance of the embedding layer helps to eliminate signal-
237
+ specific information, leaving global, modulation-specific char-
238
+ acteristics. Figure 2 illustrates the X-Vector architecture where
239
+ statistics are computed over the activations from a convolu-
240
+ tional layer producing a fixed-length vector.
241
+ Additionally,
242
+ this
243
+ architecture
244
+ maintains
245
+ a
246
+ fully-
247
+ convolutional structure enabling variable size inputs into
248
+ the network. Using statistical aggregations allows for this
249
+ property to be exploited. When using statistical aggregations,
250
+ the input to the first dense layer is dependent upon the
251
+ number of filters in the final convolutional layer. The number
252
+ of filters is a hyperparameter, independent of the length in
253
+ time of the input signal into the neural network.
254
+ Without the statistical aggregations, the input signals into
255
+ a traditional CNN or ResNet would need to be resampled,
256
+ cropped or padded to a fixed-length in time such that there is
257
+ not a size mismatch with the final convolutional output and
258
+ the first dense layer. While the dataset used in this work has
259
+ uniformly sized signals in terms of duration, (1024 × 2), this
260
+ is an architectural advantage in our deployment as received
261
+ signals may vary in duration. Instead of modifying the inputs
262
+ to the network via sampling, cropping, padding, etc., the X-
263
+ Vector architecture can directly operate with variable-length
264
+ inputs without modifications to the network or input signal.
265
+ Figure 3 outlines the employed X-Vector architecture in [7]
266
+ where F = [f1, f2, ..., f7] = 64 and K = [k1, k2, ..., k7] = 3.
267
+ Mean and variance pooling are performed on the final con-
268
+ volutional outputs, concatenated, and fed through a series of
269
+ dense layers creating the fixed-length X-Vector. A maximum
270
+ of 98% accuracy was achieved at high SNR levels.
271
+ Fig. 4.
272
+ Accuracy comparison of the reproduced ResNet in [1] and the X-
273
+ Vector inspired model from [7] over varying SNRs. This accuracy comparison
274
+ shows the superior performance of the X-Vector architecture, especially at
275
+ higher SNRs, and supports using this architecture as a baseline for the
276
+ improvements investigated in this paper.
277
+ The work of [7] replicated the ResNet architecture from
278
+ [1] and compared the results with the X-Vector architectures
279
+ as seen in Figure 4. Harper et al. [7] were able to reproduce
280
+ this architecture achieving a maximum of 93.7% accuracy. The
281
+ authors attribute the difference in performance to differences in
282
+ the train and test set separation they used since these parame-
283
+ ters were unavailable. As expected, the classifiers perform with
284
+ a higher accuracy as the SNR value increases. In signals with a
285
+ low SNR value, noise becomes more dominant and the signal
286
+ is harder to distinguish. In modern software-defined radio
287
+
288
+ Input
289
+ Batch size x 1024 x 2
290
+ Batch size x 1024 × f4
291
+ Batch size x 1024 x fz
292
+ Conv1D + ReLU (f5, k5, 1)
293
+ Statistics Pooling
294
+ Conv1D + ReLU (f1, k1, 1)
295
+ Batch size x 1024 x f1
296
+ I Batch size x 1024 x fs
297
+ Batch size x (f*2)
298
+ Conv1D + ReLU (f2, k2, 1)
299
+ Conv1D (f6, k6, 1)
300
+ Dense + SeLU (128)
301
+ Conv1D (number of filters,
302
+ filter size, dilation rate)
303
+ Batch size x 1024 x f2
304
+ + Batch size x 1024 x fe
305
+ Batch size x 128
306
+ Conv1D + ReLU (f3, k3, 1)
307
+ Conv1D + ReLU (f7, k7, 1)
308
+ Dense + SeLU (128)
309
+ Batch size x 1024 x f3
310
+ Batch size x 128
311
+ Conv1D + ReLU (f4, k4, 1)
312
+ Dense + Softmax (24)
313
+ Batch size x 24
314
+ Prediction1
315
+ 0.8
316
+ Accuracy
317
+ 0.6
318
+ 0.4
319
+ 0.2
320
+ 0
321
+ -20
322
+ -10
323
+ 10
324
+ 20
325
+ 30
326
+ 0
327
+ SNR (dB)4
328
+ applications, a high SNR value is not always a given. However,
329
+ there is still significant improvement compared to random
330
+ chance, even at low SNR values. Moreover, in systems where
331
+ the modulation type must be classified quickly, this could
332
+ become crucially important as fewer demodulation schemes
333
+ would need to be applied in a trial and error manner to discover
334
+ the correct scheme.
335
+ One challenge of AMC is that performance is desired to
336
+ work well across a large range of SNRs. For instance, Figure 4
337
+ illustrates modulation classification performance plateaued in
338
+ peak performance beyond +8dB SNR and approached chance
339
+ classification performance below −8dB SNR on the RadioML
340
+ 2018.01A dataset. This range is denoted by the shaded region.
341
+ Harper et al. investigated methods to improve classification
342
+ performance in this range by employing an SNR regression
343
+ model to aid separate modulation classifiers (MCs). While
344
+ other works have trained models to be as resilient as possible
345
+ under varying SNR conditions, Harper et al. employed SNR-
346
+ specific MCs [2].
347
+ TABLE I
348
+ SNR GROUPINGS FOR TRAINING SNR-SPECIFIC CLASSIFIERS AND
349
+ DEMULTIPLEXED CLASSIFICATION RANGES FOR EACH PREDICTED SNR.
350
+ Training Range (dB)
351
+ Demultiplexed Classification Range (dB)
352
+ [-20, -8]
353
+ (−∞, -8)
354
+ [-8, -4]
355
+ [-8, -4)
356
+ [-4, 0]
357
+ [-4, 0)
358
+ [0, 4]
359
+ [0, 4)
360
+ [4, 8]
361
+ [4, 8)
362
+ [8, 30]
363
+ [8, ∞)
364
+ Six MCs were created by discretizing the SNR range to
365
+ ameliorate performance between −8dB to +8dB SNR (see
366
+ Figure 5). These groupings were chosen in order to provide
367
+ sufficient training data to avoid overfitting the MCs and
368
+ provide enough resolution so that combining MCs provided
369
+ more value than a single classifier.
370
+ By first predicting the SNR of the received signal with
371
+ a regression model, an SNR-specific MC that was trained
372
+ on signals with the predicted SNR is applied to make the
373
+ final prediction. Although the SNR values in the dataset
374
+ are discrete, SNR is measured on a continuous scale in a
375
+ deployment scenario and can vary over time. As a result,
376
+ regression is used over classification to model SNR. Using this
377
+ approach, different classifiers can tune their feature processing
378
+ for differing SNR ranges. Each MC in this approach uses the
379
+ same architecture as that proposed in [7]; however, each MC
380
+ is trained with signals within each MC’s SNR training range
381
+ (see Table I).
382
+ Highlighting improvements across varying SNR values, Fig-
383
+ ure 6 shows the overall performance improvement (in percent-
384
+ age accuracy) using the SNR-assisted architecture compared to
385
+ the baseline classification architecture described in [7]. While
386
+ a slight decrease in performance was observed for −8dB and
387
+ a larger decrease for −2dB, improvement is shown under most
388
+ SNR conditions—particularly in the target range of −8dB to
389
+ +8dB. A possible explanation for the decrease in performance
390
+ at particular SNRs is that the optimization for a particular
391
+ MC helped overall performance for a grouping at the expense
392
+ of a single value in the group. That is, the MC for [−4, 0)
393
+ Fig. 5. The architecture using SNR regression and SNR-specific classifiers
394
+ from [2]. Each MC block shown employs the same architecture as the baseline
395
+ from [7], but specifically trained to perform AMC within a more narrow range
396
+ of SNRs (denoted as dB ranges in each block).
397
+ boosted the overall performance by performing well at −4
398
+ and 0dB at the expense of −2dB. Due to the large size of
399
+ the testing set, these small percentage gains are impactful
400
+ because thousands more classifications are correct. All results
401
+ are statistically significant based on a McNemar’s test [28],
402
+ therefore achieving new state-of-the-art performance at the
403
+ time.
404
+ Soltani et al. [3] found SNR regions of [−10, −2]dB,
405
+ [0, 8]dB, and [10, 30]dB having similar classification patterns.
406
+ Instead of predicting exact modulation variants, the authors
407
+ group commonly confused variants into a more generic,
408
+ coarse-grained label. This grouping increases performance of
409
+ AMC by combining modulation variants that are commonly
410
+ confused. However, it also decreases the sensitivity of the
411
+ model to the numerous possible variants.
412
+ Cai et al. utilized a transformer based architecture to aid
413
+ performance at low SNR levels with relatively few training pa-
414
+ rameters (approximately 265,0000 parameters) [29]. A multi-
415
+ scale network along with center loss [30] was used in [31].
416
+ It was found that larger kernel sizes improved AMC perfor-
417
+ mance. We further explore kernel size performance impacts
418
+ in this work. Zhang et al. proposed a high-order attention
419
+ mechanism using the covariance matrix achieving a maximum
420
+ accuracy of 95.49% [32].
421
+ Although many discussed works use the same RadioML
422
+ 2018.01A dataset, there is a lack of a uniform dataset split
423
+ to establish a benchmark for papers to report performance.
424
+ In an effort to make AMC work more reproducible and
425
+ comparable across publications, we have made our dataset split
426
+ and accompanying code available on GitHub.1
427
+ While numerous works have investigated architectural im-
428
+ provements, we aim to improve upon these works by intro-
429
+ ducing additional modifications as well as a comprehensive
430
+ ablation study that illustrates the improvement of each mod-
431
+ ification. With the new modifications, we achieve new state-
432
+ of-the-art AMC performance.
433
+ III. DATASET
434
+ To evaluate different machine learning architectures, we
435
+ use the RadioML 2018.01A dataset that is comprised of 24
436
+ 1https://github.com/caharper/Automatic-Modulation-Classification-with-
437
+ Deep-Neural-Networks
438
+
439
+ SNR Regression
440
+ Model
441
+ DEMUX
442
+ MC
443
+ MC
444
+ MC
445
+ MC
446
+ MC
447
+ MC
448
+ (-8, -8)
449
+ [-8, -4)
450
+ [-4, 0]
451
+ [0, 4]
452
+ [4, 8]
453
+ (8, 8)5
454
+ Fig. 6. Summary of residual improvement in accuracy over [7] that was first
455
+ published in [2]. This work showed how the baseline architecture could be
456
+ tuned to specific SNR ranges. Positive improvement is observed for most SNR
457
+ ranges.
458
+ different modulation types [1], [15]. Due to the complexity
459
+ and variety of modulation schemes in the dataset, it is fairly
460
+ representative of typically encountered modulation schemes.
461
+ Moreover, this variety increases the likelihood that AMC
462
+ models will generalize to more exotic or non-existing modu-
463
+ lation schemes in the training data that are derived from these
464
+ traditional variants.
465
+ There are a total of 2.56 million labeled signals, S(T),
466
+ each consisting of 1024 time domain digitized intermediate
467
+ frequency (IF) samples of in-phase (I) and quadrature (Q)
468
+ signal components where S(T) = I(T) + jQ(T). The data
469
+ was collected at a 900MHz IF with an assumed sampling
470
+ rate of 1MS/sec such that each 1024 time domain digitized
471
+ I/Q sample is 1.024 ms [33]. The 24 modulation types and
472
+ the representative groups that we chose for each are listed as
473
+ follows:
474
+ • Amplitude: OOK, 4ASK, 8ASK, AM-SSB-SC, AM-
475
+ SSB-WC, AM-DSB-WC, and AM-DSB-SC
476
+ • Phase: BPSK, QPSK, 8PSK, 16PSK, 32PSK, and
477
+ OQPSK
478
+ • Amplitude and Phase: 16APSK, 32APSK, 64APSK,
479
+ 128APSK, 16QAM, 32QAM, 64QAM, 128QAM, and
480
+ 256QAM
481
+ • Frequency: FM and GMSK
482
+ Each modulation type includes a total of 106, 496 obser-
483
+ vations ranging from −20dB to +30dB SNR in 2dB steps
484
+ for a total of 26 different SNR values. SNR is assumed
485
+ to be consistent over the same window length as the I/Q
486
+ sample window. For evaluation, we divided the dataset into 1
487
+ million different training observations and 1.5 million testing
488
+ observations under a random shuffle split, stratified across
489
+ modulation type and SNR. Because of this balance, the
490
+ expected performance for a random chance classifier is 1/24
491
+ or 4.2%. With varying SNR levels across the dataset, it is
492
+ expected that the classifier would perform with a higher degree
493
+ of accuracy as the SNR value is increased. For consistency,
494
+ each model investigated in this work was trained and evaluated
495
+ on the same train and test set splits.
496
+ IV. INITIAL INVESTIGATION
497
+ In this work, we use the architecture described in [7] as
498
+ the baseline architecture. We note that [2] improved upon the
499
+ baseline; however, each individual MC used the baseline archi-
500
+ tecture except trained on specific SNR ranges. Therefore, the
501
+ base architectural elements were similar to [7], but separated
502
+ for different SNRs. In this work, our focus is to improve upon
503
+ the employed CNN architecture for an individual MC rather
504
+ than the use of several MCs. Therefore, we use the architecture
505
+ from [7] as our baseline.
506
+ Before exploring an ablation study, we make a few notable
507
+ changes from the baseline architecture in an effort to increase
508
+ AMC performance. This initial exploration is for clarity as
509
+ it reserves the ablation study that follows from requiring an
510
+ inordinate number of models. It also introduces the general
511
+ training procedures that assist and orient the reader in fol-
512
+ lowing the ablation study—the ablation study mirrors these
513
+ procedures. We first provide an initial investigation exploring
514
+ these notable changes.
515
+ We train each model using the Adam optimizer [34] with
516
+ an initial learning rate lr = 0.0001, a decay factor of 0.1 if
517
+ the validation loss does not decrease for 12 epochs, and a
518
+ minimum learning rate of 1e-7. If the validation loss does not
519
+ decrease after 20 epochs, training is terminated and the models
520
+ are deemed converged. For all experiments, mini-batches of
521
+ size 32 are used. As has been established in most programming
522
+ packages for neural networks, we refer to fully connected
523
+ neural network layers as dense layers, which are typically
524
+ followed by an activation function.
525
+ A. Architectural Changes
526
+ A common property of neural networks is using fewer but
527
+ larger kernels in the early layers of the network, and an
528
+ increase of smaller kernels are used in the later layers than
529
+ the baseline architecture. This is commonly referred to as the
530
+ information distillation pipeline [35]. By utilizing a smaller
531
+ number of large kernels in early layers, we are able to increase
532
+ the temporal context of the convolutional features without
533
+ dramatically increasing the number of trainable parameters.
534
+ Numerous, but smaller kernels are used in later convolu-
535
+ tional layers to create more abstract features. Configuring
536
+ the network in this manner is especially popular in image
537
+ classification where later layers represent more abstract, class-
538
+ specific features.
539
+ We investigate this modification in three stages, using the
540
+ baseline architecture described in Figure 3 [7]. We denote
541
+ number of filters in the network and the filter sizes as
542
+ F = [f1, f2, ..., f7] and K = [k1, k2, ...k7] in Figure 3. The
543
+ baseline architecture used f = 64 (for all layers) and k = 3
544
+ (consistent kernel size for all layers). Our first modification to
545
+ the baseline architecture is F = [32, 48, 64, 72, 84, 96, 108],
546
+ but keeping k = 3 for all layers. Second, we use the baseline
547
+ architecture, but change the size of filters in the network where
548
+ f = 64 (same as baseline) and K = [7, 5, 7, 5, 3, 3, 3]. Third,
549
+ we make both modifications and compare the result to the
550
+ baseline model where F = [32, 48, 64, 72, 84, 96, 108] and
551
+ K = [7, 5, 7, 5, 3, 3, 3]. These modifications are not exhaustive
552
+
553
+ 0.318
554
+ 0.306
555
+ 0.3 -
556
+ 0.2600.259
557
+ Residual Improvement (0-100%)
558
+ 0.235
559
+ 0.229
560
+ 0.222
561
+ 0.216
562
+ 0.2
563
+ 0.192
564
+ 0.172
565
+ 0.170
566
+ 0.165
567
+ 0.1540.157
568
+ 0.142
569
+ 0.149
570
+ 0.134
571
+ 0.124
572
+ 0.1
573
+ 0.020
574
+ 0.008
575
+ 0.0
576
+ -0.012 -0.008
577
+ -0.008
578
+ -0.060
579
+ -0.1
580
+ 0.094
581
+ -0.111
582
+ 20-18-16-14-12-10
583
+ -8
584
+ -6
585
+ -4
586
+ 0
587
+ 2
588
+ 4
589
+ 6
590
+ 8
591
+ 10
592
+ 12
593
+ 16
594
+ 20
595
+ 22
596
+ 24
597
+ 26
598
+ 2830
599
+ SNR (dB)6
600
+ searches; rather, these modifications are meant to guide future
601
+ changes to the network by understanding the influence of filter
602
+ quantity and filter size in a limited context.
603
+ TABLE II
604
+ INITIAL INVESTIGATION PERFORMANCE OVERVIEW. ALL
605
+ ARCHITECTURES EMPLOY THE BASELINE WITH VARYING NUMBERS OF
606
+ KERNELS AND KERNEL SIZES.
607
+ Notes
608
+ # Params
609
+ Avg.
610
+ Accuracy
611
+ Max
612
+ Accuracy
613
+ Reproduced ResNet [1]
614
+ 165,144
615
+ 59.2%
616
+ 93.7%
617
+ X-Vector in [7]
618
+ 110,680
619
+ 61.3%
620
+ 98.0%
621
+ More Filters
622
+ (Same Filter Sizes)
623
+ 149,168
624
+ 61.0%
625
+ 96.1%
626
+ Larger Filter Sizes
627
+ (Same # Filters)
628
+ 143,960
629
+ 62.6%
630
+ 98.2%
631
+ Combined
632
+ 174,000
633
+ 62.9%
634
+ 98.6%
635
+ B. Initial Investigation Results
636
+ As shown in Table II, increasing the size of the filters
637
+ in earlier layers increases both average and maximum test
638
+ accuracy over [7]; but, at the cost of additional parameters.
639
+ A possible explanation for the increase in performance is the
640
+ increase in temporal context due to the larger kernel sizes.
641
+ Increasing the number of filters without increasing temporal
642
+ context decreases performance. This is possibly because it in-
643
+ creases the complexity of the model without adding additional
644
+ signal context.
645
+ Fig. 7.
646
+ SNR vs. accuracy comparison of the initial investigation using the
647
+ baseline architecture. Noticeable improvements can be observed across all
648
+ SNRs.
649
+ Figure 7 illustrates the change in accuracy with varying
650
+ SNR. The combined model, utilizing various kernel sizes
651
+ and numbers of filters, consistently outperforms the other
652
+ architectures across changing SNR conditions.
653
+ Although increasing the number of filters decreases per-
654
+ formance alone, combining the approach with larger kernel
655
+ sizes yields the best performance in our initial investigation.
656
+ Increasing the temporal context may have allowed additional
657
+ filters to better characterize the input signal.
658
+ Because increased temporal context improves AMC perfor-
659
+ mance, we are inspired to investigate additional methods such
660
+ as squeeze-and-excitation blocks and dilated convolutions that
661
+ can increase global and local context [25], [36].
662
+ V. ABLATION STUDY ARCHITECTURE BACKGROUND
663
+ Building upon our findings from our initial investigation,
664
+ we make additional modifications to the baseline architecture.
665
+ For the MCs, we introduce dilated convolutions, squeeze-
666
+ and-excitation blocks, self-attention, and other architectural
667
+ changes. We also investigate various kernel sizes and the
668
+ quantity of kernels employed from the initial investigation.
669
+ Our goal is to improve upon existing architectures while
670
+ investigating the impact of each modification on classification
671
+ accuracy through an ablation study. In this section, we describe
672
+ each modification performed.
673
+ A. Squeeze-and-Excitation Networks
674
+ Fig. 8.
675
+ Squeeze-and-Excitation block proposed in [25]. One SE block is
676
+ shown applied to a single layer convolutional output activation. Two paths
677
+ are shown, a scaling path and an identity path. The scaling vector is applied
678
+ across channels to the identity path of the activations.
679
+ Squeeze-and-Excitation (SE) blocks introduce a channel-
680
+ wise attention mechanism first proposed in [25]. Due to the
681
+ limited receptive field of each convolutional filter, SE blocks
682
+ propose a recalibration step based on global statistics across
683
+ channels (average pooling) to provide global context. Although
684
+ initially utilized for image classification tasks [25], [37], [38],
685
+ we argue the use of SE blocks can provide meaningful global
686
+ context to the convolutional network used for AMC over the
687
+ time domain.
688
+ Figure 8 depicts an SE block. The squeeze operation is de-
689
+ fined as temporal global average pooling across convolutional
690
+ filters. For an individual channel, c, the squeeze operation is
691
+ defined as:
692
+ zc = Fsq(xc) = 1
693
+ T
694
+ T
695
+
696
+ i=1
697
+ xi,c
698
+ (1)
699
+ where X
700
+
701
+ RT ×C
702
+ =
703
+ [x1, x2, ..., xC], Z
704
+
705
+ R1×C
706
+ =
707
+ [z1, z2, ..., zC], T is the number of samples in time, and C is
708
+ the total number of channels. To model nonlinear interactions
709
+ between channel-wise statistics, Z is fed into a series of dense
710
+ layers followed by nonlinear activation functions:
711
+ s = Fex(z, W) = σ(g(z, W)) = σ(W2δ(W1z))
712
+ (2)
713
+ where δ is the rectified linear (ReLU) activation function,
714
+ W1 ∈ R
715
+ C
716
+ r ×C, W2 ∈ RC× C
717
+ r , r is a dimensionality reduction
718
+ ratio, and σ is the sigmoid activation function. The sigmoid
719
+ function is chosen as opposed to the softmax function so that
720
+ multiple channels can be accentuated and are not mutually-
721
+ exclusive. That is, the normalization term in the softmax
722
+ can cause dependencies among channels, so the sigmoid
723
+ activation is preferred. W1 imposes a bottleneck to improve
724
+ generalization performance and reduce parameter counts while
725
+ W2 increases the dimensionality back to the original number
726
+ of channels for the recalibration operation. In our work, we
727
+
728
+ 0.8
729
+ 2
730
+ 0.6
731
+ Accurac
732
+ 0.4
733
+ 0.2
734
+ 0
735
+ -20
736
+ -10
737
+ 0
738
+ 10
739
+ 20
740
+ 30
741
+ SNR (dB)
742
+ Model - X-Vector Model from [7] - More Filters (Same Filter Sizes)
743
+ - Larger Filter Sizes (Same # Filters) → Combined - - Random ChanceX
744
+ Fex(·, W)
745
+ Time (T)
746
+ Fsq(·)
747
+ 1 ×C
748
+ 1 × C
749
+ Channels (C)
750
+ Fscale(·, ·)
751
+ T×C
752
+ T×C7
753
+ Fig. 9. Proposed architecture with modifications including SENets, dilated convolutions, optional ReLU activation before statistics pooling, and self-attention.
754
+ The output tensor sizes are also shown for each unit in the diagram. An * denotes where the sizes differ from the baseline architecture.
755
+ use r = 2 for all SE blocks to ensure a reasonable number
756
+ of trainable parameters without over-squashing the embedding
757
+ size.
758
+ The final operation in the SE block, scaling or recalibration,
759
+ is obtained by scaling the the input X by s:
760
+ ˆxc = Fscale(xc, sc) = scxc
761
+ (3)
762
+ where ˆX ∈ RT ×C = [ ˆx1, ˆx2, ..., ˆ
763
+ xC].
764
+ B. Dilated Convolutions
765
+ Fig. 10.
766
+ Dilated convolutions diagram. The top shows a traditional kernel
767
+ applied to sequential time series points. The middle and bottom diagram
768
+ illustrate dilation rates of two and three, respectively. These dilations serve
769
+ to increase the receptive field of the filter without increasing the number of
770
+ trainable variables in the kernel.
771
+ Proposed in [36], Figure 10 depicts dilated convolutions
772
+ where the convolutional kernels are denoted by the colored
773
+ components. In a traditional convolution, the dilation rate
774
+ is equal to 1. Dilated convolutions build temporal context
775
+ by increasing the receptive field of the convolutional kernels
776
+ without increasing parameter counts as the number of entries
777
+ in the kernel remains the same.
778
+ Dilated convolutions also do not downsample the signals
779
+ like strided convolutions. Instead, the output of a dilated
780
+ convolution can be the exact size of the input after properly
781
+ handling edge effects at the beginning and end of the signal.
782
+ C. Final Convolutional Activation
783
+ We also investigate the impact of using an activation func-
784
+ tion (ReLU) after the last convolutional layer, just before
785
+ statistics pooling. Because ReLU transforms the input se-
786
+ quence to be non-negative, the distribution characterized by
787
+ the pooling statistics may become skewed. In [7] and [2],
788
+ no activation was applied after the final convolutional layer
789
+ as shown in Figure 3. We investigate if this transformation
790
+ impacts classification performance.
791
+ D. Self-Attention
792
+ Self-attention allows the convolutional outputs to interact
793
+ with one another enabling the network to learn to focus on
794
+ important outputs. Self-attention before statistics pooling es-
795
+ sentially creates a weighted summation over the convolutional
796
+ outputs weighting their importance similarly to [39]–[41].
797
+ We use the attention mechanism described by Vaswani et
798
+ al. in [42] where each output element is a weighted sum of
799
+ the linearly transformed input where the dimensionality of K
800
+ is dk as seen in Equation (4).
801
+ Attention(Q, K, V ) = softmax
802
+ � QKT
803
+ |√dk|
804
+
805
+ V
806
+ (4)
807
+ In the case of self-attention, Q, K, and V are equal. A scaling
808
+ factor of
809
+ 1
810
+ |√dk| is applied to counteract vanishing gradients in
811
+ the softmax output when dk is large.
812
+ VI. ABLATION STUDY ARCHITECTURE
813
+ Applying the specified modifications to the architecture in
814
+ [7], Figure 9 illustrates the proposed architecture with every
815
+ modification included in the graphic. Each colored block
816
+ represents an optional change to the architecture that will be
817
+ investigated in the ablation study. That is, each combination
818
+ of network modifications are analyzed to aid understanding of
819
+ each modification’s impact on the network.
820
+ Each convolutional layer has the following parameters:
821
+ number of filters, kernel size, and dilation rate. The asterisk
822
+ next to each dilation rate represents the changing of dilation
823
+ rates in the ablation study. If dilated convolutions are used,
824
+
825
+ Time
826
+ Dilation rate = 1
827
+ Dilation rate = 2
828
+ Dilation rate = 3Input +
829
+ Batch size × 1024 x 2
830
+ 1 Batch size × 1024 × 64
831
+ 1 Batch size x 1024 x 84
832
+ IBatch size x 1024 x 108
833
+ Conv1D + ReLU (32, 7, 1*)
834
+ SE Block
835
+ Conv1D + ReLU (96, 3, 2*)
836
+ Statistics Pooling
837
+ + Batch size × 1024 x 32
838
+ I Batch size x 1024 × 64
839
+ ↓ Batch size × 1024 x 96
840
+ Batch size x 216
841
+ Conv1D (number of filters,
842
+ SE Block
843
+ Conv1D + ReLU (72, 5, 2*)
844
+ SE Block
845
+ Dense + SeLU (128)
846
+ filter size, dilation rate)
847
+ Batch size × 1024 × 32
848
+ I Batch size x 1024 ×72
849
+ I Batch size x 1024 × 96
850
+ Batch size × 128
851
+ Equals 1 for the
852
+ initial investigation
853
+ Conv1D + ReLU (48, 5, 2*)
854
+ SE Block
855
+ Conv1D (108, 3, 1*)
856
+ Dense + SeLU (128)
857
+ Not included in the
858
+ initial investigation
859
+ Batch size × 1024 × 48
860
+ I Batch size x 1024 x 72
861
+ Batch size x 1024 × 108
862
+ Batch size x 128
863
+ SE Block
864
+ Conv1D + ReLU (84, 3, 2*)
865
+ ReLU
866
+ Dense + Softmax (24)
867
+ + Batch size x 1024 × 48
868
+ I Batch size x 1024 × 84
869
+ I Batch size × 1024 × 108
870
+ Batch size x 24
871
+ Conv1D + ReLU (64, 7, 3*)
872
+ SE Block
873
+ Self-Attention
874
+ Prediction8
875
+ TABLE III
876
+ ABLATION STUDY PERFORMANCE OVERVIEW.
877
+ Model Name
878
+ Notes
879
+ SENet
880
+ Dilated
881
+ Convolutions
882
+ Final
883
+ Activation
884
+ Attention
885
+ # Params
886
+ Avg.
887
+ Accuracy
888
+ Max
889
+ Accuracy
890
+
891
+ Reproduced ResNet [1]
892
+
893
+
894
+
895
+
896
+ 165,144
897
+ 59.2%
898
+ 93.7%
899
+
900
+ X-Vector in [7]
901
+
902
+
903
+
904
+
905
+ 110,680
906
+ 61.3%
907
+ 98.0%
908
+ 0000
909
+ Best performing model from
910
+ the initial investigation
911
+
912
+
913
+
914
+
915
+ 174,000
916
+ 62.9%
917
+ 98.6%
918
+ 0001
919
+
920
+
921
+
922
+ 
923
+ 221,088
924
+ 62.3%
925
+ 97.6%
926
+ 0010
927
+
928
+
929
+ 
930
+
931
+ 174,000
932
+ 62.8%
933
+ 98.6%
934
+ 0011
935
+
936
+
937
+ 
938
+ 
939
+ 221,088
940
+ 62.3%
941
+ 97.5%
942
+ 0100
943
+
944
+ 
945
+
946
+
947
+ 174,000
948
+ 63.2%
949
+ 98.9%
950
+ 0101
951
+
952
+ 
953
+
954
+ 
955
+ 221,088
956
+ 63.1%
957
+ 97.9%
958
+ 0110
959
+
960
+ 
961
+ 
962
+
963
+ 174,000
964
+ 63.2%
965
+ 98.9%
966
+ 0111
967
+
968
+ 
969
+ 
970
+ 
971
+ 221,088
972
+ 63.0%
973
+ 98.0%
974
+ 1000
975
+ 
976
+
977
+
978
+
979
+ 202,880
980
+ 62.9%
981
+ 98.5%
982
+ 1001
983
+ 
984
+
985
+
986
+ 
987
+ 249,968
988
+ 62.6%
989
+ 98.2%
990
+ 1010
991
+ 
992
+
993
+ 
994
+
995
+ 202,880
996
+ 62.6%
997
+ 98.3%
998
+ 1011
999
+ 
1000
+
1001
+ 
1002
+ 
1003
+ 249,968
1004
+ 62.8%
1005
+ 98.1%
1006
+ 1100
1007
+ 
1008
+ 
1009
+
1010
+
1011
+ 202,880
1012
+ 62.8%
1013
+ 98.2%
1014
+ 1101
1015
+ 
1016
+ 
1017
+
1018
+ 
1019
+ 249,968
1020
+ 63.0%
1021
+ 97.7%
1022
+ 1110
1023
+ Overall best performing model
1024
+ 
1025
+ 
1026
+ 
1027
+
1028
+ 202,880
1029
+ 63.7%
1030
+ 98.9%
1031
+ 1111
1032
+ 
1033
+ 
1034
+ 
1035
+ 
1036
+ 249,968
1037
+ 63.0%
1038
+ 97.8%
1039
+ then the dilation rate value in the graphic is used. If dilated
1040
+ convolutions are not used, each dilation rate is set to 1. That
1041
+ is, a traditional convolution is applied. All convolutions use a
1042
+ stride of 1, and the same training procedure from the initial
1043
+ investigation is used.
1044
+ VII. EVALUATION METRICS
1045
+ We present several evaluation metrics to compare the dif-
1046
+ ferent architectures considered in the ablation study. In this
1047
+ section, we will discuss each evaluation technique used in the
1048
+ results section.
1049
+ Due to the varying levels of SNRs in the employed dataset,
1050
+ we plot classification accuracy over each true SNR value. This
1051
+ allows for a visualization of the tradeoff in performance as
1052
+ noise becomes more or less dominant in the received signals.
1053
+ Additionally, we report average accuracy and maximum ac-
1054
+ curacy across the entire test set for each model. While we
1055
+ note that average accuracy is not indicative of the model’s
1056
+ performance, as accuracy is highly correlated to the SNR of
1057
+ the input signal, we share this result to give other researchers
1058
+ the ability to reproduce and compare works.
1059
+ As discussed in [26], AMC is often implemented on
1060
+ resource-constrained devices. In these systems, using larger
1061
+ models in terms of parameter counts may not be feasible. We
1062
+ report the number of parameters for each model in the ablation
1063
+ study to examine the tradeoff in AMC performance and model
1064
+ size.
1065
+ Additional analyses are also carried out. However, due to
1066
+ the large number of models investigated in this study, we
1067
+ will select the best performing model from the ablation study
1068
+ for brevity and analyze the performance of this model in
1069
+ greater detail. For example, confusion matrices for the best
1070
+ performing model from the ablation study are provided to
1071
+ show common misclassifications for each modulation type.
1072
+ Additionally, there exist several use-cases where relatively
1073
+ short signal bursts are received. For example, a wide-band
1074
+ scanning receiver may only detect a short signal burst. There-
1075
+ fore, signal duration in the time domain versus AMC perfor-
1076
+ mance is investigated to determine the robustness of the best
1077
+ performing model when short signal bursts are received.
1078
+ VIII. ABLATION RESULTS
1079
+ A. Overall Performance
1080
+ Table III lists the maximum and average accuracy perfor-
1081
+ mance for each model in the ablation study. A binary naming
1082
+ convention is used to indicate the various methods used for
1083
+ each architecture. Similarly to the result found in Section IV,
1084
+ increasing the temporal context typically results in increased
1085
+ performance. Models that incorporate dilated convolutions
1086
+ tended to have higher average accuracies than models without
1087
+ dilated convolutions.
1088
+ The best performing model, in terms of average accuracy
1089
+ across all SNR conditions included SE blocks, dilated convolu-
1090
+ tions, and a ReLU activation prior to statistics pooling (model
1091
+ 1110) with an average accuracy of approximately 63.7%. This
1092
+ model also achieved the highest maximum accuracy of about
1093
+ 98.9% at a 22dB level.
1094
+ SE blocks did not increase performance compared to model
1095
+ 0000 with the exception of models 1110 and 1111. However,
1096
+ SE blocks were incorporated in the best performing model,
1097
+ 1110. Self-attention was not found to aid classification perfor-
1098
+ mance in general with the proposed architecture. Self-attention
1099
+ introduces a large number of trainable parameters possibly
1100
+ forming a complex loss space.
1101
+ Table IV lists the performances of single modification (from
1102
+ baseline) architectures. Each component of the ablation study,
1103
+ with the exception of dilated convolutions, decreased perfor-
1104
+ mance when applied individually. When combined, however,
1105
+ the best performing model was found. Therefore, we conclude
1106
+ that each component could possibly aid the optimization of
1107
+
1108
+ 9
1109
+ Fig. 11. Ablation study results in terms of classification accuracy across SNR ranges. The best performing model is in the second to last row and displays
1110
+ strong performance across SNR values.
1111
+ TABLE IV
1112
+ INDIVIDUAL NETWORK MODIFICATION PERFORMANCE OVERVIEW.
1113
+ ENTRIES ARE REPEATED FROM TABLE III FOR CLARITY.
1114
+ Model Name
1115
+ Notes
1116
+ SENet
1117
+ Dilated
1118
+ Convolutions
1119
+ Final
1120
+ Activation
1121
+ Attention
1122
+ # Params
1123
+ Avg.
1124
+ Accuracy
1125
+ Max
1126
+ Accuracy
1127
+
1128
+ X-Vector in [7]
1129
+
1130
+
1131
+
1132
+
1133
+ 110,680
1134
+ 61.3%
1135
+ 98.0%
1136
+ 0000
1137
+
1138
+
1139
+
1140
+
1141
+ 174,000
1142
+ 62.9%
1143
+ 98.6%
1144
+ 0001
1145
+
1146
+
1147
+
1148
+ 
1149
+ 221,088
1150
+ 62.3%
1151
+ 97.6%
1152
+ 0010
1153
+
1154
+
1155
+ 
1156
+
1157
+ 174,000
1158
+ 62.8%
1159
+ 98.6%
1160
+ 0100
1161
+
1162
+ 
1163
+
1164
+
1165
+ 174,000
1166
+ 63.2%
1167
+ 98.9%
1168
+ 1000
1169
+ 
1170
+
1171
+
1172
+
1173
+ 202,880
1174
+ 62.9%
1175
+ 98.5%
1176
+ 1110
1177
+ Best performer
1178
+ 
1179
+ 
1180
+ 
1181
+
1182
+ 202,880
1183
+ 63.7%
1184
+ 98.9%
1185
+ each other—and, in general, dilated convolutions tend to have
1186
+ the most dramatic performance increases.
1187
+ B. Accuracy Over Varying SNR
1188
+ Figure 11 summarizes the ablation study in terms of classi-
1189
+ fication accuracy over varying SNR levels. We add this figure
1190
+ for completeness and reproducibility for other researchers.
1191
+ The accuracy within each SNR band is shown along with the
1192
+ modifications used, similar to Table III. The coloring in the
1193
+ figure denotes the accuracy in each SNR band. Performance
1194
+ follows a trend similar to that of a sigmoid function, where
1195
+ the rate at which peak classification accuracy is achieved is
1196
+ the most distinguishing feature between the different models.
1197
+ With the improved architectures, a maximum of 99% accuracy
1198
+ is achieved at high SNR levels (starting around 12dB SNR).
1199
+ While the proposed changes to the architectures gener-
1200
+ ally improve performance at higher SNR levels, the largest
1201
+ improvements occur between −12dB and 12dB compared
1202
+ to the baseline model in [7]. For example, at 4dB, the
1203
+ performance increases from 75% up to 82%. Incorporating
1204
+ these modifications to the network may prove to be critical
1205
+ in real-world situations where noisy signals are likely to be
1206
+ obtained. Improving AMC performance at lower SNR ranges
1207
+ (< −12dB) is still an open research topic, with accuracies
1208
+ near chance level.
1209
+ One observation is the best performing model can vary with
1210
+ SNR. In systems that have available memory and processing
1211
+ power, an approach similar to [2] may be used to utilize several
1212
+ models and intelligently chose predictions based on estimated
1213
+ SNR conditions. That is, if the SNR of the signal of interest is
1214
+ known, a model can be tuned to increase performance slightly,
1215
+ as shown in [2]. Using the results presented here, researchers
1216
+ could also choose the architecture differences that perform best
1217
+ for a given SNR range (although performance differences are
1218
+ subtle).
1219
+ C. Parameter Count Tradeoff
1220
+ Fig. 12.
1221
+ Ablation study parameter count tradeoff. The x-axis shows the
1222
+ number of trainable variables in each model and the y-axis shows max or
1223
+ average accuracy. The callout for each point denotes the model name as shown
1224
+ in Table III.
1225
+ An overview of each model’s complexity and overall per-
1226
+ formance across the entire testing set is shown in Table III.
1227
+ This information is also shown graphically in Figure 12 for
1228
+ the maximum accuracy over SNR and the average accuracy
1229
+ across all SNRs. Whether looking at the maximum or the
1230
+ average measures of performance, the conclusions are similar.
1231
+ The previously described binary model name also appears in
1232
+ the figure. We found a slight correlation between the number
1233
+ of model parameters and overall model performance; however,
1234
+ with the architectures explored, there was a general parameter
1235
+ count where performance peaked. Models with parameter
1236
+ counts between approximately 170k to 205k generally per-
1237
+ formed better than smaller and larger models. We note that the
1238
+
1239
+ Dilated
1240
+ Final
1241
+ Model Name
1242
+ Notes
1243
+ SENet
1244
+ Activation
1245
+ Attention
1246
+ Modulation Classification Results
1247
+ Convolutions
1248
+
1249
+ Reproduced ResNet [1]
1250
+ -
1251
+ -
1252
+
1253
+ 0.04
1254
+ 15
1255
+ 0.93
1256
+ 0.94
1257
+ 0.94
1258
+ 0.94
1259
+ 0.93
1260
+ -
1261
+ X-Vector in [7]
1262
+ -
1263
+ -
1264
+ -
1265
+ -
1266
+ 0.04
1267
+ 0.91
1268
+ 0.94
1269
+ 0.97
1270
+ 0.98
1271
+ 0.98
1272
+ 0.98
1273
+ 0.98
1274
+ 0.98
1275
+ 0.98
1276
+ 0.98
1277
+ 0.98
1278
+ 000
1279
+ -
1280
+ -
1281
+ -
1282
+ -
1283
+
1284
+ 0.04
1285
+ 0.95
1286
+ 0.97
1287
+ 0.98
1288
+ 0.98
1289
+ 86°0
1290
+ 66°0
1291
+ 66°0
1292
+ 66°0
1293
+ 66°0
1294
+ 0.99
1295
+ 0.99
1296
+ 0001
1297
+ -
1298
+
1299
+ -
1300
+ -
1301
+ 0.0
1302
+ 0.94
1303
+ 0.97
1304
+ 0.97
1305
+ 0.97
1306
+ 0.97
1307
+ 0.98
1308
+ 0.98
1309
+ 0.98
1310
+ 0.98
1311
+ 0010
1312
+
1313
+ -
1314
+ V
1315
+ 0.98
1316
+ 0.98
1317
+ 0.99
1318
+ 0.99
1319
+ 0011
1320
+ -
1321
+ V
1322
+ 0.89
1323
+ 0.94
1324
+ 0.96
1325
+ 0.97
1326
+ 0.97
1327
+ 0.97
1328
+ 0.97
1329
+ 0.97
1330
+ 0.97
1331
+ 0.97
1332
+ 0100
1333
+
1334
+ V
1335
+ -
1336
+
1337
+ 0.0
1338
+ 0.92
1339
+ 0.97
1340
+ 0.98
1341
+ 0.99
1342
+ 0.99
1343
+ 0.99
1344
+ 0.99
1345
+ 0.99
1346
+ 0.99
1347
+ 0.99
1348
+ 0.99
1349
+ 0.99
1350
+ 0.99
1351
+ 0101
1352
+
1353
+ -
1354
+
1355
+ 0.04
1356
+ 0.92
1357
+ 0.97
1358
+ 0.98
1359
+ 0.98
1360
+ 0.98
1361
+ 0.98
1362
+ 0.98
1363
+ 0110
1364
+ -
1365
+ -
1366
+ 0.04
1367
+ 0.97
1368
+ 0.99
1369
+ 66:0
1370
+ 0.99
1371
+ 0.99
1372
+ 0.99
1373
+ 0.99
1374
+ 0111
1375
+ -
1376
+
1377
+ V
1378
+ 0.04
1379
+ 0.97
1380
+ 0.98
1381
+ 0.98
1382
+ 0.98
1383
+ 0.98
1384
+ 0.98
1385
+ 0.98
1386
+ 0.98
1387
+ 1000
1388
+
1389
+ v
1390
+ -
1391
+ -
1392
+
1393
+ 0.04
1394
+ 0.98
1395
+ 0.98
1396
+ 0.98
1397
+ 0.98
1398
+ 0.98
1399
+ 1001
1400
+ -
1401
+ v
1402
+ -
1403
+ -
1404
+ 0.04
1405
+ 0.98
1406
+ 1010
1407
+ -
1408
+ -
1409
+ v
1410
+ -
1411
+ 0.98
1412
+ 0.98
1413
+ 0.98
1414
+ 0.98
1415
+ 0.98
1416
+ 0.98
1417
+ 1011
1418
+ -
1419
+ v
1420
+ -
1421
+ v
1422
+ v
1423
+ 0.98
1424
+ 0.98
1425
+ 0.96
1426
+ 0.98
1427
+ 0.98
1428
+ 0.98
1429
+ 0.98
1430
+ 0.98
1431
+ 0.98
1432
+ 0.98
1433
+ 1100
1434
+ -
1435
+ v
1436
+ -
1437
+
1438
+ 0.04
1439
+ 0.91
1440
+ 0.96
1441
+ 0.97
1442
+ 0.98
1443
+ 0.98
1444
+ 0.98
1445
+ 86°0
1446
+ 0.98
1447
+ 0.98
1448
+ 0.98
1449
+ 0.98
1450
+ 0.98
1451
+ 0.98
1452
+ 1101
1453
+ -
1454
+ -
1455
+ 0.04
1456
+ 0.81
1457
+ 0.91
1458
+ 0.95
1459
+ 0.97
1460
+ 0.97
1461
+ 0.98
1462
+ 86°0
1463
+ 0.98
1464
+ 86°0
1465
+ 0.98
1466
+ 0.98
1467
+ 0.98
1468
+ 0.98
1469
+ 0.98
1470
+ 1110
1471
+ Overall best performing model
1472
+ V
1473
+ 0.28
1474
+ 0.36
1475
+ 0.46
1476
+ 0.58
1477
+ 0.69
1478
+ 0.82
1479
+ 0.92
1480
+ 0.97
1481
+ 0.98
1482
+ 0.99
1483
+ 66°0
1484
+ 66°0
1485
+ 66°0
1486
+ 66°0
1487
+ 66°0
1488
+ 66°0
1489
+ 0.99
1490
+ 0.99
1491
+ 66:0
1492
+ 1111
1493
+ -
1494
+ 0.81
1495
+ 0.91
1496
+ 0.96
1497
+ 0.97
1498
+ 0.97
1499
+ 0.98
1500
+ 0.98
1501
+ 0.98
1502
+ 86°0
1503
+ 0.98
1504
+ 0.98
1505
+ 0.98
1506
+ 0.98
1507
+ 20
1508
+ -18
1509
+ -16
1510
+ -14
1511
+ -12
1512
+ -10
1513
+ -6
1514
+ -2
1515
+ 10
1516
+ 12
1517
+ 14
1518
+ 16
1519
+ 18
1520
+ 20
1521
+ 22
1522
+ 24
1523
+ 4
1524
+ 6
1525
+ 8
1526
+ 28
1527
+ 30
1528
+ SNR (dB)1.00
1529
+ Overall Accuracy
1530
+ _0110
1531
+ 0100-
1532
+ ←1110
1533
+ 1000→
1534
+ Max Accuracy
1535
+ 0.98
1536
+ LLO
1537
+ 0101
1538
+ X-Vector Model from [7]
1539
+ 0001
1540
+ 0011
1541
+ 0.96
1542
+ 0.94
1543
+ %)
1544
+ Reproduced ResNet from [1]
1545
+ acy
1546
+ 0.92
1547
+ ra
1548
+ 0.64
1549
+ iccur
1550
+ ^—1110
1551
+ 0100-
1552
+ →-0110
1553
+ 0101
1554
+ 1101.
1555
+ 1000+1100
1556
+ toiii
1557
+ 0.63
1558
+ 1111
1559
+ A
1560
+ 0000-
1561
+ —0010
1562
+ ←1011
1563
+ -i010
1564
+ 1001→
1565
+ 0.62
1566
+ X-Vector Model from [7]
1567
+ 0.61
1568
+ 0.60
1569
+ Reproduced ResNet from [1]
1570
+ 0.59
1571
+ # Params10
1572
+ (a)
1573
+ (b)
1574
+ (c)
1575
+ Fig. 13. Accuracy over varying SNR conditions for model 1110 with (a), (b), and (c) showing the top-1, top-2, and top-5 accuracy respectively. Random
1576
+ chance for each is defined as 1/24, 2/24, and 5/24.
1577
+ models with more than 205k parameters included self-attention
1578
+ which was found to decrease model performance with the
1579
+ proposed architectures. This implies that one possible reason
1580
+ self-attention did not perform as well as other modifications
1581
+ is because of the increase in parameters, resulting in a more
1582
+ difficult loss space from which to optimize.
1583
+ IX. BEST PERFORMING MODEL INVESTIGATION
1584
+ Due to the large volume of models, we focus upon the
1585
+ best performing model, (model 1110), for the remainder of
1586
+ this work. As previously mentioned, this model employs all
1587
+ modifications except self-attention.
1588
+ A. Top-K Accuracy
1589
+ As discussed, in systems where the modulation schemes
1590
+ must be classified quickly, it is advantageous to apply fewer
1591
+ demodulation schemes in a trial and error fashion. This is
1592
+ particularly significant at lower SNR values where accuracy is
1593
+ mediocre. Top-k accuracy allows an in-depth view on the ex-
1594
+ pected number of trials before finding the correct modulation
1595
+ scheme. Although traditional accuracy (top-1 accuracy) char-
1596
+ acterizes the performance of the model in terms of classifying
1597
+ the exact variant, top-k accuracy characterizes the percentage
1598
+ of the classifier predicting the correct variant among the top-
1599
+ k predictions (sorted by descending class probabilities). We
1600
+ plot the top-1, top-2, and top-5 classification accuracy over
1601
+ varying SNR conditions for each modulation grouping defined
1602
+ in Section III in Figure 13.
1603
+ Although performance decays to approximately random
1604
+ chance for the overall (all modulation schemes) performance
1605
+ curves for each top-k accuracy, it is notable that some modu-
1606
+ lation group performances drop below random chance. The
1607
+ models are trained to maximize the overall model perfor-
1608
+ mance. This could explain why certain modulation groups dip
1609
+ below random chance but the overall performance and other
1610
+ modulation groups remain at or above random chance.
1611
+ Using the proposed method greatly reduces the correct
1612
+ modulation scheme search space. While high performance in
1613
+ top-1 accuracy is increasingly difficult to achieve with low
1614
+ SNR signals, top-2 and top-5 accuracy converge to higher
1615
+ values at a much faster rate. This indicates our proposed
1616
+ method greatly reduces the search space from 24 modulation
1617
+ candidates to fewer candidate types when employing trial and
1618
+ error methods to determine the correct modulation scheme.
1619
+ Further, if the group of modulation is known (e.g., FM), one
1620
+ can view a more specific tradeoff curve in terms of SNR and
1621
+ top-k accuracy given in Figure 13.
1622
+ B. Short Duration Signal Bursts
1623
+ Due to the rapid scanning characteristic of some modern
1624
+ software-defined radios, we investigate the performance trade-
1625
+ off of varying signal duration and AMC performance. This
1626
+ analysis is meant to emulate the situation wherein a receiver
1627
+ only detects a short RF signal burst. We investigate signal
1628
+ burst durations of 1.024 ms (full length signal from original
1629
+ dataset), 512 µs, 256 µs, 128 µs, 64 µs, 32 µs, and 16 µs.
1630
+ We assume the same 1MS/sec sampling rate as in the previous
1631
+ analyses such that 16 µs burst is captured in 16 I/Q samples.
1632
+ Fig. 14. Tradeoff in accuracy for various signal lengths across SNR, grouped
1633
+ by modulation category for the best performing model 1110. The top plot
1634
+ shows the baseline performance using the full sequence. Subsequent plots
1635
+ show the same information using increasingly smaller signal lengths for
1636
+ classification.
1637
+ In this section, we use the same test set as our other
1638
+ investigations; however, a uniformly random starting point is
1639
+
1640
+ 0.8
1641
+ 0.6
1642
+ 0.4
1643
+ Overall
1644
+
1645
+ Amplitude
1646
+ ←Phase
1647
+ - Amplitude and Phase
1648
+ 0.2
1649
+ Frequency
1650
+ - - Random Chance
1651
+ 0
1652
+ -20
1653
+ -10
1654
+ 0
1655
+ 10
1656
+ 20
1657
+ 30
1658
+ SNR (dB)0.8
1659
+ 0.6
1660
+ 0.4
1661
+ 0.2
1662
+ 0
1663
+ -20
1664
+ -10
1665
+ 0
1666
+ 10
1667
+ 20
1668
+ 30
1669
+ SNR (dB)0.8
1670
+ 0.6
1671
+ 0.4
1672
+ 0.2
1673
+ 0
1674
+ -20
1675
+ -10
1676
+ 0
1677
+ 10
1678
+ 20
1679
+ 30
1680
+ SNR (dB)1.024 ms (n=1024)
1681
+ 1
1682
+ 0.8
1683
+ Overall
1684
+ 0.6
1685
+ Amplitude
1686
+ Phase
1687
+ 0.4
1688
+ Amplitude and Phase
1689
+ Frequency
1690
+ 0.2
1691
+ Random Chance
1692
+ 0
1693
+ -20
1694
+ -10
1695
+ 0
1696
+ 10
1697
+ 20
1698
+ 30
1699
+ 512 μs (n=512)
1700
+ 256 μs (n=256)
1701
+ 1
1702
+ 1
1703
+ Accuracy
1704
+ 0.8
1705
+ 0.8
1706
+ 0.6
1707
+ 0.6
1708
+ 0.4
1709
+ 0.4
1710
+ 0.2
1711
+ 0.2
1712
+ 0
1713
+ 0
1714
+ -20
1715
+ -10
1716
+ 0
1717
+ 10
1718
+ 20
1719
+ 30
1720
+ -20
1721
+ -10
1722
+ 0
1723
+ 10
1724
+ 20
1725
+ 30
1726
+ 128 μs (n=128)
1727
+ 64 μs (n=64)
1728
+ 1
1729
+ 1
1730
+ 0.8
1731
+ 0.8
1732
+ 0.6
1733
+ 0.6
1734
+ 0.4
1735
+ 0.4
1736
+ 0.2
1737
+ 0.2
1738
+ 0
1739
+ 0
1740
+ -20
1741
+ -10
1742
+ 0
1743
+ 10
1744
+ 20
1745
+ 30
1746
+ -20
1747
+ -10
1748
+ 0
1749
+ 10
1750
+ 20
1751
+ 30
1752
+ 32 μs (n=32)
1753
+ 16 μs (n=16)
1754
+ 1
1755
+ 1
1756
+ 0.8
1757
+ 0.8
1758
+ 0.6
1759
+ 0.6
1760
+ 0.4
1761
+ 0.4
1762
+ 0.2
1763
+ 0.2
1764
+ 0
1765
+ 0
1766
+ 20
1767
+ -10
1768
+ 0
1769
+ 10
1770
+ 20
1771
+ 30
1772
+ -20
1773
+ 0
1774
+ 10
1775
+ 20
1776
+ -10
1777
+ 30
1778
+ SNR11
1779
+ (a)
1780
+ (b)
1781
+ (c)
1782
+ Fig. 15. Confusion matrices for (a) model 1110 (best performing model from this work), (b) the reproduced ResNet model from [1], and (c) the X-Vector
1783
+ inspired model from [19] with SNR ≥ 0dB.
1784
+ determined for each signal such that a contiguous sample of
1785
+ the desired duration, starting at the random point, is chosen.
1786
+ Thus, the chosen segment from a test set sample is randomly
1787
+ assigned.
1788
+ We also note that, although the sample length for the evalu-
1789
+ ation is changed, the best performing model is the same archi-
1790
+ tecture with the exact same trained weights because this model
1791
+ uses statistics pooling from the X-Vector inspired modification.
1792
+ A significant benefit to the X-Vector inspired architecture is
1793
+ its ability to handle variable-length inputs without the need
1794
+ of padding, retraining, or other network modifications. This
1795
+ is achieved by taking global statistics across convolutional
1796
+ channels producing a fixed-length vector, regardless of signal
1797
+ duration. Due to this flexibility, the same model (model 1110)
1798
+ weights are used for each duration experiment. This fact also
1799
+ emphasizes the desirability of using X-vector inspired AMC
1800
+ architectures for receivers that are deployed in an environment
1801
+ where short-burst and variable duration signals are anticipated
1802
+ to be present.
1803
+ For each signal duration in the time domain, we plot the
1804
+ overall classification accuracy over varying SNR conditions
1805
+ as well as the accuracy for each modulation grouping de-
1806
+ fined in Section III. Figure 14 demonstrates the tradeoff for
1807
+ various signal durations where n is the number of samples
1808
+ from the time domain I/Q signal. The first observation is,
1809
+ as we would expect, that classification performance degrades
1810
+ with decreased signal duration. For example, the maximum
1811
+ accuracy begins to degrade at 256 µs and is more noticeable
1812
+ at 128 µs. This is likely a result of using sample statistics
1813
+ that result in unstable or biased estimates for short signal
1814
+ lengths since the number of received signal data points are
1815
+ insufficient to characterize the sample statistics used during
1816
+ training. Random classification accuracy is approximately 4%
1817
+ and is shown in the black dotted line in Figure 14. Although
1818
+ classification performance decreases with decreased duration,
1819
+ we are still able to achieve significantly higher classification
1820
+ accuracy than random chance down to 16 µs of signal capture.
1821
+ FM (frequency modulation) signals were typically more
1822
+ resilient to noise interference than AM (amplitude modulation)
1823
+ and AM-PM (amplitude and phase modulation) signals in our
1824
+ AMC. This was observed across all signal burst durations and
1825
+ our top-k accuracy analysis. This behavior indicates that the
1826
+ performance of our AMC for short bursts, in the presence
1827
+ of increasing amounts of noise, is more robust for signals
1828
+ modulated by changes in the carrier frequency and is more
1829
+ sensitive to signals modulated by varying the carrier amplitude.
1830
+ We attribute this behavior to our AMC architecture, the
1831
+ architecture of the receiver, or a combination of both of the
1832
+ AMC and receiver.
1833
+ C. Confusion Matrices
1834
+ While classification accuracy provides a holistic view of
1835
+ model performance, it lacks the granularity to investigate
1836
+ where misclassifications are occurring. Confusion matrices are
1837
+ used to analyze the distribution of classifications for each given
1838
+ class. For each true label, the proportion of correctly classified
1839
+ samples is calculated along with the proportion of incorrect
1840
+ predictions for each opposing class. In this way, we can see
1841
+ which classes the model is struggling to distinguish from
1842
+ one another. A perfect classifier would be the identity matrix
1843
+ where the diagonal values represent the true class matches the
1844
+ predicted class. Each matrix value represents the percentage
1845
+ of classifications for the true label and each row sums to 1
1846
+ (100%).
1847
+ Figure 15 illustrates the class confusion matrices for SNR
1848
+ levels greater than or equal to 0dB for models 1110, the
1849
+ reproduced ResNet architecture from [1], and the baseline X-
1850
+ Vector architecture from [7] respectively. Shown in [7], the
1851
+ X-Vector architecture was able to distinguish PSK and AM-
1852
+ SSB variants to a higher degree and performed better overall
1853
+ than [1]. Both architectures struggled to differentiate QAM
1854
+ variants.
1855
+ Model 1110 improved upon these prior results for QAM
1856
+ signals and in general has higher diagonal components than
1857
+ the other architectures. This again supports a conclusion that
1858
+ model 1110 achieves a new state-of-the-art in AMC perfor-
1859
+ mance.
1860
+ X. CONCLUSION
1861
+ A comprehensive ablation study was carried out with regard
1862
+ to AMC architectural features using the extensive RadioML
1863
+ 2018.01A dataset. This ablation study built upon a strong
1864
+ performance of a new baseline model that was also intro-
1865
+ duced in the initial investigation of this study. This initial
1866
+ investigation informed the design of a number of AMC ar-
1867
+ chitecture modifications—specifically, the use of X-Vectors,
1868
+ dilated convolutions, and SE blocks. With the combined
1869
+
1870
+ 1.0
1871
+ OOK
1872
+ 0.0
1873
+ 0.0
1874
+ 0.0
1875
+ 0.0
1876
+ 0.0
1877
+ 0.0
1878
+ 0.0
1879
+ 0.0
1880
+ 0.0
1881
+ 0.0
1882
+ 0.0
1883
+ 0.0
1884
+ 0.0
1885
+ 0.0
1886
+ 0.0
1887
+ 0.0
1888
+ 0.0
1889
+ 0.0
1890
+ 0.0
1891
+ 0.0
1892
+ 0.0
1893
+ 0.0
1894
+ 0.0
1895
+ 0.0
1896
+ 0.97
1897
+ 0.03
1898
+ 4ASK
1899
+ 0.0
1900
+ 0.0
1901
+ 0.0
1902
+ 0.0
1903
+ 0.0
1904
+ 0.0
1905
+ 0.0
1906
+ 0.0
1907
+ 0.0
1908
+ 0.0
1909
+ 0.0
1910
+ 0.0
1911
+ 0.0
1912
+ 0.0
1913
+ 0.0
1914
+ 0.0
1915
+ 0.0
1916
+ 0.0
1917
+ 0.0
1918
+ 0.0
1919
+ 0.0
1920
+ 0.03
1921
+ 0.97
1922
+ 8ASK
1923
+ 0.0
1924
+ 0.0
1925
+ 0.0
1926
+ 0.0
1927
+ 0.0
1928
+ 0.0
1929
+ 0.0
1930
+ 0.0
1931
+ 0.0
1932
+ 0.0
1933
+ 0.0
1934
+ 0.0
1935
+ 0.0
1936
+ 0.0
1937
+ 0.0
1938
+ 0.0
1939
+ 0.0
1940
+ 0.0
1941
+ 0.0
1942
+ 0.0
1943
+ 0.0
1944
+ 0.0
1945
+ 1.0
1946
+ BPSK
1947
+ 0.0
1948
+ 0.0
1949
+ 0.0
1950
+ 0.0
1951
+ 0.0
1952
+ 0.0
1953
+ 0.0
1954
+ 0.0
1955
+ 0.0
1956
+ 0.0
1957
+ 0.0
1958
+ 0.0
1959
+ 0.0
1960
+ 0.0
1961
+ 0.0
1962
+ 0.0
1963
+ 0.0
1964
+ 0.0
1965
+ 0.0
1966
+ 0.0
1967
+ 0.0
1968
+ 0.0
1969
+ 0.0
1970
+ 0.0
1971
+ 1.0
1972
+ 0.0
1973
+ QPSK
1974
+ 0.0
1975
+ 0.0
1976
+ 0.0
1977
+ 0.0
1978
+ 0.0
1979
+ 0.0
1980
+ 0.0
1981
+ 0.0
1982
+ 0.0
1983
+ 0.0
1984
+ 0.0
1985
+ 0.0
1986
+ 0.0
1987
+ 0.0
1988
+ 0.0
1989
+ 0.0
1990
+ 0.0
1991
+ 0.0
1992
+ 0.0
1993
+ 0.0
1994
+ 0.0
1995
+ S6°0
1996
+ 8PSK
1997
+ 0.0
1998
+ 0.0
1999
+ 0.0
2000
+ 0.0
2001
+ 0.0
2002
+ 0.01
2003
+ 0.03
2004
+ 0.0
2005
+ 0.0
2006
+ 0.0
2007
+ 0.0
2008
+ 0.0
2009
+ 0.0
2010
+ 0.0
2011
+ 0.0
2012
+ 0.0
2013
+ 0.0
2014
+ 0.0
2015
+ 0.0
2016
+ 0.0
2017
+ 0.0
2018
+ 0.0
2019
+ 0.0
2020
+ 16PSK
2021
+ 0.0
2022
+ 0.03
2023
+ 0.87
2024
+ 0.0
2025
+ 0.0
2026
+ 0.0
2027
+ 0.0
2028
+ 0.1
2029
+ 0.0
2030
+ 0.0
2031
+ 0.0
2032
+ 0.0
2033
+ 0.0
2034
+ 0.0
2035
+ 0.0
2036
+ 0.0
2037
+ 0.0
2038
+ 0.0
2039
+ 0.0
2040
+ 0.0
2041
+ 0.0
2042
+ 0.0
2043
+ 0.0
2044
+ 0.0
2045
+ 0.0
2046
+ 0.0
2047
+ 0.0
2048
+ 0.0
2049
+ 0.05
2050
+ 0.92
2051
+ 32PSK
2052
+ 0.0
2053
+ 0.03
2054
+ 0.0
2055
+ 0.0
2056
+ 0.0
2057
+ 0.0
2058
+ 0.0
2059
+ 0.0
2060
+ 0.0
2061
+ 0.0
2062
+ 0.0
2063
+ 0.0
2064
+ 0.0
2065
+ 0.0
2066
+ 0.0
2067
+ 0.0
2068
+ 0.0
2069
+ 0.0
2070
+ 0.0
2071
+ 0.0
2072
+ 0.0
2073
+ 0.0
2074
+ 0.0
2075
+ 0.0
2076
+ 0.0
2077
+ 0.0
2078
+ 0.98
2079
+ 16APSK
2080
+ 0.0
2081
+ 0.0
2082
+ 0.0
2083
+ 0.01
2084
+ 0.01
2085
+ 0.0
2086
+ 0.0
2087
+ 0.0
2088
+ 0.0
2089
+ 0.0
2090
+ 0.0
2091
+ 0.0
2092
+ 0.0
2093
+ 0.0
2094
+ 0.0
2095
+ 0.0
2096
+ 0.0
2097
+ 0.0
2098
+ 0.0
2099
+ 0.0
2100
+ 0.0
2101
+ 0.0
2102
+ 0.0
2103
+ 0.96
2104
+ 32APSK
2105
+ 0.0
2106
+ 0.01
2107
+ 0.01
2108
+ 0.0
2109
+ 0.0
2110
+ 0.0
2111
+ 0.0
2112
+ 0.01
2113
+ 0.0
2114
+ 0.0
2115
+ 0.0
2116
+ 0.0
2117
+ 0.0
2118
+ 0.0
2119
+ 0.0
2120
+ Label
2121
+ 0.01
2122
+ 0.87
2123
+ 64APSK
2124
+ 0.0
2125
+ 0.0
2126
+ 0.0
2127
+ 0.0
2128
+ 0.0
2129
+ 0.0
2130
+ 0.01
2131
+ 0.03
2132
+ 0.0
2133
+ 0.0
2134
+ 0.01
2135
+ 0.03
2136
+ 0.01
2137
+ 0.02
2138
+ 0.01
2139
+ 0.0
2140
+ 0.0
2141
+ 0.0
2142
+ 0.0
2143
+ 0.0
2144
+ 0.0
2145
+ 0.0
2146
+ 0.03
2147
+ 0.88
2148
+ 128APSK
2149
+ 0.0
2150
+ 0.0
2151
+ 0.0
2152
+ 0.0
2153
+ 0.0
2154
+ 0.0
2155
+ 0.0
2156
+ 0.0
2157
+ 0.0
2158
+ 0.02
2159
+ 0.01
2160
+ 0.01
2161
+ 0.01
2162
+ 0.01
2163
+ 0.02
2164
+ 0.0
2165
+ 0.0
2166
+ 0.0
2167
+ 0.0
2168
+ 0.0
2169
+ 0.0
2170
+ 0.0
2171
+ True
2172
+ 0.01
2173
+ 0.92
2174
+ 16QAM
2175
+ 0.0
2176
+ 0.0
2177
+ 0.0
2178
+ 0.0
2179
+ 0.0
2180
+ 0.0
2181
+ 0.0
2182
+ 0.0
2183
+ 0.01
2184
+ 0.01
2185
+ 0.01
2186
+ 0.01
2187
+ 0.02
2188
+ 0.01
2189
+ 0.01
2190
+ 0.0
2191
+ 0.0
2192
+ 0.0
2193
+ 0.0
2194
+ 0.0
2195
+ 0.0
2196
+ 0.0
2197
+ 0.0
2198
+ 0.02
2199
+ 0.88
2200
+ 32QAM
2201
+ 0.0
2202
+ 0.0
2203
+ 0.0
2204
+ 0.0
2205
+ 0.0
2206
+ 0.0
2207
+ 0.0
2208
+ 0.02
2209
+ 0.01
2210
+ 0.03
2211
+ 0.01
2212
+ 0.01
2213
+ 0.02
2214
+ 0.01
2215
+ 0.0
2216
+ 0.0
2217
+ 0.0
2218
+ 0.0
2219
+ 0.0
2220
+ 0.0
2221
+ 0.0
2222
+ 0.83
2223
+ 64QAM
2224
+ 0.0
2225
+ 0.0
2226
+ 0.0
2227
+ 0.0
2228
+ 0.0
2229
+ 0.0
2230
+ 0.0
2231
+ 0.0
2232
+ 0.0
2233
+ 0.01
2234
+ 0.01
2235
+ 0.01
2236
+ 0.02
2237
+ 0.01
2238
+ 0.02
2239
+ 0.08
2240
+ 0.0
2241
+ 0.0
2242
+ 0.0
2243
+ 0.0
2244
+ 0.0
2245
+ 0.0
2246
+ 0.0
2247
+ 0.81
2248
+ 128QAM
2249
+ 0.0
2250
+ 0.0
2251
+ 0.0
2252
+ 0.0
2253
+ 0.0
2254
+ 0.0
2255
+ 0.0
2256
+ 0.0
2257
+ 0.01
2258
+ 0.01
2259
+ 0.02
2260
+ 0.02
2261
+ 0.02
2262
+ 0.03
2263
+ 0.03
2264
+ 0.05
2265
+ 0.0
2266
+ 0.0
2267
+ 0.0
2268
+ 0.0
2269
+ 0.0
2270
+ 0.0
2271
+ 0.0
2272
+ 0.01
2273
+ 0.83
2274
+ 256QAM
2275
+ 0.0
2276
+ 0.0
2277
+ 0.0
2278
+ 0.0
2279
+ 0.0
2280
+ 0.0
2281
+ 0.0
2282
+ 0.0
2283
+ 0.0
2284
+ 0.01
2285
+ 0.01
2286
+ 0.01
2287
+ 0.01
2288
+ 0.08
2289
+ 0.04
2290
+ 0.0
2291
+ 0.0
2292
+ 0.0
2293
+ 0.0
2294
+ 0.0
2295
+ 0.0
2296
+ 0.0
2297
+ 0.94
2298
+ AM-SSB-WC
2299
+ 0.0
2300
+ 0.0
2301
+ 0.0
2302
+ 0.0
2303
+ 0.0
2304
+ 0.0
2305
+ 0.0
2306
+ 0.0
2307
+ 0.0
2308
+ 0.0
2309
+ 0.0
2310
+ 0.0
2311
+ 0.0
2312
+ 0.0
2313
+ 0.0
2314
+ 0.0
2315
+ 0.0
2316
+ 0.06
2317
+ 0.0
2318
+ 0.0
2319
+ 0.0
2320
+ 0.0
2321
+ 0.0
2322
+ 0.0
2323
+ 0.0
2324
+ 0.0
2325
+ 0.0
2326
+ 0.0
2327
+ 0.0
2328
+ 0.0
2329
+ 0.0
2330
+ 0.0
2331
+ 0.0
2332
+ 0.0
2333
+ 0.0
2334
+ 0.0
2335
+ 0.0
2336
+ 0.0
2337
+ 0.0
2338
+ 0.0
2339
+ 0.16
2340
+ 0.84
2341
+ AM-SSB-SC
2342
+ 0.0
2343
+ 0.0
2344
+ 0.0
2345
+ 0.0
2346
+ 0.0
2347
+ 0.0
2348
+ 0.96
2349
+ 0.04
2350
+ AM-DSB-WC
2351
+ 0.0
2352
+ 0.0
2353
+ 0.0
2354
+ 0.0
2355
+ 0.0
2356
+ 0.0
2357
+ 0.0
2358
+ 0.0
2359
+ 0.0
2360
+ 0.0
2361
+ 0.0
2362
+ 0.0
2363
+ 0.0
2364
+ 0.0
2365
+ 0.0
2366
+ 0.0
2367
+ 0.0
2368
+ 0.0
2369
+ 0.0
2370
+ 0.0
2371
+ 0.0
2372
+ 0.89
2373
+ AM-DSB-SC
2374
+ 0.0
2375
+ 0.0
2376
+ 0.0
2377
+ 0.0
2378
+ 0.0
2379
+ 0.0
2380
+ 0.0
2381
+ 0.0
2382
+ 0.0
2383
+ 0.0
2384
+ 0.0
2385
+ 0.0
2386
+ 0.0
2387
+ 0.0
2388
+ 0.0
2389
+ 0.0
2390
+ 0.0
2391
+ 0.0
2392
+ 0.11
2393
+ 0.0
2394
+ 0.0
2395
+ 0.0
2396
+ 0.0
2397
+ 0.0
2398
+ 1.0
2399
+ 0.0
2400
+ 0.0
2401
+ 0.0
2402
+ 0.0
2403
+ 0.0
2404
+ 0.0
2405
+ 0.0
2406
+ 0.0
2407
+ 0.0
2408
+ 0.0
2409
+ 0.0
2410
+ 0.0
2411
+ 0.0
2412
+ 0.0
2413
+ 0.0
2414
+ 0.0
2415
+ 0.0
2416
+ 0.0
2417
+ 0.0
2418
+ FM
2419
+ 0.0
2420
+ 0.0
2421
+ 0.0
2422
+ 0.0
2423
+ 0.0
2424
+ 0.0
2425
+ 0.0
2426
+ 0.0
2427
+ 1.0
2428
+ GMSK
2429
+ 0.0
2430
+ 0.0
2431
+ 0.0
2432
+ 0.0
2433
+ 0.0
2434
+ 0.0
2435
+ 0.0
2436
+ 0.0
2437
+ 0.0
2438
+ 0.0
2439
+ 0.0
2440
+ 0.0
2441
+ 0.0
2442
+ 0.0
2443
+ 0.0
2444
+ 0.0
2445
+ 0.0
2446
+ 0.0
2447
+ OQPSK
2448
+ 0.0
2449
+ 0.0
2450
+ 0.0
2451
+ 0.0
2452
+ 0.0
2453
+ 0.0
2454
+ 0.0
2455
+ 0.0
2456
+ 0.0
2457
+ 0.0
2458
+ 0.0
2459
+ 0.0
2460
+ 0.0
2461
+ 0.0
2462
+ 0.0
2463
+ 0.0
2464
+ 0.0
2465
+ 0.0
2466
+ 0.0
2467
+ 0.0
2468
+ 0.0
2469
+ 0.0
2470
+ 0.0
2471
+ 1.0
2472
+ YSds
2473
+ 16PSK
2474
+ YO0
2475
+ BPSK
2476
+ Sdo
2477
+ 32AP
2478
+ 64AA
2479
+ 16QA
2480
+ GMSK
2481
+ 8ASK
2482
+ 16Ap
2483
+ 1284
2484
+ 2560
2485
+ FM
2486
+ 4ASK
2487
+ SSB-WC
2488
+ APSK
2489
+ PSK
2490
+ Wvo.
2491
+ Wvo.
2492
+ PSK
2493
+ Wka1.0
2494
+ OOK
2495
+ 0.0
2496
+ 0.0
2497
+ 0.0
2498
+ 0.0
2499
+ 0.0
2500
+ 0.0
2501
+ 0.0
2502
+ 0.0
2503
+ 0.0
2504
+ 0.0
2505
+ 0.0
2506
+ 0.0
2507
+ 0.0
2508
+ 0.0
2509
+ 0.0
2510
+ 0.0
2511
+ 0.0
2512
+ 0.0
2513
+ 0.0
2514
+ 0.0
2515
+ 0.0
2516
+ 0.0
2517
+ 0.0
2518
+ 0.0
2519
+ 0.96
2520
+ 0.04
2521
+ 4ASK
2522
+ 0.0
2523
+ 0.0
2524
+ 0.0
2525
+ 0.0
2526
+ 0.0
2527
+ 0.0
2528
+ 0.0
2529
+ 0.0
2530
+ 0.0
2531
+ 0.0
2532
+ 0.0
2533
+ 0.0
2534
+ 0.0
2535
+ 0.0
2536
+ 0.0
2537
+ 0.0
2538
+ 0.0
2539
+ 0.0
2540
+ 0.0
2541
+ 0.0
2542
+ 0.0
2543
+ 0.02
2544
+ 0.98
2545
+ 8ASK
2546
+ 0.0
2547
+ 0.0
2548
+ 0.0
2549
+ 0.0
2550
+ 0.0
2551
+ 0.0
2552
+ 0.0
2553
+ 0.0
2554
+ 0.0
2555
+ 0.0
2556
+ 0.0
2557
+ 0.0
2558
+ 0.0
2559
+ 0.0
2560
+ 0.0
2561
+ 0.0
2562
+ 0.0
2563
+ 0.0
2564
+ 0.0
2565
+ 0.0
2566
+ 0.0
2567
+ 0.0
2568
+ 1.0
2569
+ BPSK
2570
+ 0.0
2571
+ 0.0
2572
+ 0.0
2573
+ 0.0
2574
+ 0.0
2575
+ 0.0
2576
+ 0.0
2577
+ 0.0
2578
+ 0.0
2579
+ 0.0
2580
+ 0.0
2581
+ 0.0
2582
+ 0.0
2583
+ 0.0
2584
+ 0.0
2585
+ 0.0
2586
+ 0.0
2587
+ 0.0
2588
+ 0.0
2589
+ 0.0
2590
+ 0.0
2591
+ 0.0
2592
+ 0.0
2593
+ 0.99
2594
+ 0.0
2595
+ 0.0
2596
+ QPSK
2597
+ 0.0
2598
+ 0.0
2599
+ 0.0
2600
+ 0.0
2601
+ 0.0
2602
+ 0.0
2603
+ 0.0
2604
+ 0.0
2605
+ 0.0
2606
+ 0.0
2607
+ 0.0
2608
+ 0.0
2609
+ 0.0
2610
+ 0.0
2611
+ 0.0
2612
+ 0.0
2613
+ 0.0
2614
+ 0.0
2615
+ 0.0
2616
+ 0.0
2617
+ 0.0
2618
+ 0.9
2619
+ 8PSK
2620
+ 0.0
2621
+ 0.0
2622
+ 0.0
2623
+ 0.0
2624
+ 0.0
2625
+ 0.09
2626
+ 0.0
2627
+ 0.0
2628
+ 0.0
2629
+ 0.0
2630
+ 0.0
2631
+ 0.0
2632
+ 0.0
2633
+ 0.0
2634
+ 0.0
2635
+ 0.0
2636
+ 0.0
2637
+ 0.0
2638
+ 0.0
2639
+ 0.0
2640
+ 0.0
2641
+ 0.0
2642
+ 0.0
2643
+ 0.0
2644
+ 0.97
2645
+ 16PSK
2646
+ 0.0
2647
+ 0.0
2648
+ 0.0
2649
+ 0.0
2650
+ 0.02
2651
+ 0.0
2652
+ 0.0
2653
+ 0.0
2654
+ 0.0
2655
+ 0.0
2656
+ 0.0
2657
+ 0.0
2658
+ 0.0
2659
+ 0.0
2660
+ 0.0
2661
+ 0.0
2662
+ 0.0
2663
+ 0.0
2664
+ 0.0
2665
+ 0.0
2666
+ 0.0
2667
+ 0.0
2668
+ 0.0
2669
+ 0.0
2670
+ 0.0
2671
+ 0.0
2672
+ 0.01
2673
+ 0.42
2674
+ 0.56
2675
+ 32PSK
2676
+ 0.0
2677
+ 0.0
2678
+ 0.0
2679
+ 0.0
2680
+ 0.0
2681
+ 0.0
2682
+ 0.0
2683
+ 0.0
2684
+ 0.0
2685
+ 0.0
2686
+ 0.0
2687
+ 0.0
2688
+ 0.0
2689
+ 0.0
2690
+ 0.0
2691
+ 0.0
2692
+ 0.0
2693
+ 0.96
2694
+ 16APSK
2695
+ 0.0
2696
+ 0.0
2697
+ 0.0
2698
+ 0.0
2699
+ 0.0
2700
+ 0.0
2701
+ 0.0
2702
+ 0.0
2703
+ 0.0
2704
+ 0.0
2705
+ 0.0
2706
+ 0.01
2707
+ 0.02
2708
+ 0.0
2709
+ 0.0
2710
+ 0.0
2711
+ 0.0
2712
+ 0.0
2713
+ 0.0
2714
+ 0.0
2715
+ 0.0
2716
+ 0.0
2717
+ 0.0
2718
+ 0.0
2719
+ 0.0
2720
+ 0.0
2721
+ 0.0
2722
+ 0.0
2723
+ 0.0
2724
+ 0.0
2725
+ 0.0
2726
+ 0.97
2727
+ 32APSK
2728
+ 0.0
2729
+ 0.0
2730
+ 0.01
2731
+ 0.0
2732
+ 0.01
2733
+ 0.0
2734
+ 0.01
2735
+ 0.01
2736
+ 0.0
2737
+ 0.0
2738
+ 0.0
2739
+ 0.0
2740
+ 0.0
2741
+ 0.0
2742
+ 0.0
2743
+ 0.82
2744
+ 64APSK
2745
+ 0.0
2746
+ 0.0
2747
+ 0.0
2748
+ 0.0
2749
+ 0.0
2750
+ 0.0
2751
+ 0.0
2752
+ 0.0
2753
+ 0.01
2754
+ 0.03
2755
+ 0.03
2756
+ 0.01
2757
+ 0.05
2758
+ 0.0
2759
+ 0.04
2760
+ 0.01
2761
+ 0.0
2762
+ 0.0
2763
+ 0.0
2764
+ 0.0
2765
+ 0.0
2766
+ 0.0
2767
+ 0.0
2768
+ 0.03
2769
+ 0.83
2770
+ 128APSK
2771
+ 0.0
2772
+ 0.0
2773
+ 0.0
2774
+ 0.0
2775
+ 0.0
2776
+ 0.0
2777
+ 0.0
2778
+ 0.0
2779
+ 0.0
2780
+ 0.05
2781
+ 0.01
2782
+ 0.02
2783
+ 0.0
2784
+ 0.03
2785
+ 0.02
2786
+ 0.0
2787
+ 0.0
2788
+ 0.0
2789
+ 0.0
2790
+ 0.0
2791
+ 0.0
2792
+ 0.0
2793
+ 0.0
2794
+ 0.0
2795
+ 0.0
2796
+ 0.0
2797
+ 0.0
2798
+ 0.01
2799
+ 0.89
2800
+ 16QAM
2801
+ 0.0
2802
+ 0.0
2803
+ 0.0
2804
+ 0.0
2805
+ 0.01
2806
+ 0.02
2807
+ 0.02
2808
+ 0.01
2809
+ 0.02
2810
+ 0.01
2811
+ 0.0
2812
+ 0.0
2813
+ 0.0
2814
+ 0.0
2815
+ 0.0
2816
+ 0.0
2817
+ 0.0
2818
+ 0.0
2819
+ 0.0
2820
+ 0.02
2821
+ 0.87
2822
+ 0.0
2823
+ 0.04
2824
+ 0.01
2825
+ 0.0
2826
+ 32QAM
2827
+ 0.0
2828
+ 0.0
2829
+ 0.0
2830
+ 0.0
2831
+ 0.0
2832
+ 0.0
2833
+ 0.02
2834
+ 0.02
2835
+ 0.01
2836
+ 0.01
2837
+ 0.0
2838
+ 0.0
2839
+ 0.0
2840
+ 0.0
2841
+ 0.0
2842
+ 0.0
2843
+ 64QAM
2844
+ 0.0
2845
+ 0.0
2846
+ 0.0
2847
+ 0.0
2848
+ 0.0
2849
+ 0.0
2850
+ 0.0
2851
+ 0.0
2852
+ 0.03
2853
+ 0.01
2854
+ 0.02
2855
+ 0.02
2856
+ 0.65
2857
+ 0.0
2858
+ 0.0
2859
+ 0.06
2860
+ 0.21
2861
+ 0.0
2862
+ 0.0
2863
+ 0.0
2864
+ 0.0
2865
+ 0.0
2866
+ 0.0
2867
+ 0.0
2868
+ 0.75
2869
+ 128QAM
2870
+ 0.0
2871
+ 0.0
2872
+ 0.0
2873
+ 0.0
2874
+ 0.0
2875
+ 0.0
2876
+ 0.0
2877
+ 0.0
2878
+ 0.01
2879
+ 0.02
2880
+ 0.01
2881
+ 0.02
2882
+ 0.02
2883
+ 0.05
2884
+ 0.02
2885
+ 0.1
2886
+ 0.0
2887
+ 0.0
2888
+ 0.0
2889
+ 0.0
2890
+ 0.0
2891
+ 0.0
2892
+ 0.0
2893
+ 0.76
2894
+ 256QAM
2895
+ 0.0
2896
+ 0.0
2897
+ 0.0
2898
+ 0.0
2899
+ 0.0
2900
+ 0.0
2901
+ 0.0
2902
+ 0.0
2903
+ 0.0
2904
+ 0.03
2905
+ 0.0
2906
+ 0.01
2907
+ 0.02
2908
+ 0.02
2909
+ 0.07
2910
+ 0.09
2911
+ 0.0
2912
+ 0.0
2913
+ 0.0
2914
+ 0.0
2915
+ 0.0
2916
+ 0.0
2917
+ 0.0
2918
+ 0.94
2919
+ AM-SSB-WC
2920
+ 0.0
2921
+ 0.0
2922
+ 0.0
2923
+ 0.0
2924
+ 0.0
2925
+ 0.0
2926
+ 0.0
2927
+ 0.0
2928
+ 0.0
2929
+ 0.0
2930
+ 0.0
2931
+ 0.0
2932
+ 0.0
2933
+ 0.0
2934
+ 0.0
2935
+ 0.0
2936
+ 0.0
2937
+ 0.06
2938
+ 0.0
2939
+ 0.0
2940
+ 0.0
2941
+ 0.0
2942
+ 0.0
2943
+ 0.0
2944
+ 0.0
2945
+ 0.0
2946
+ 0.0
2947
+ 0.0
2948
+ 0.0
2949
+ 0.0
2950
+ 0.0
2951
+ 0.0
2952
+ 0.0
2953
+ 0.0
2954
+ 0.0
2955
+ 0.0
2956
+ 0.0
2957
+ 0.0
2958
+ 0.0
2959
+ 0.0
2960
+ 0.0
2961
+ 1.0
2962
+ AM-SSB-SC
2963
+ 0.0
2964
+ 0.0
2965
+ 0.0
2966
+ 0.0
2967
+ 0.0
2968
+ 0.0
2969
+ 0.97
2970
+ 0.03
2971
+ AM-DSB-WC
2972
+ 0.0
2973
+ 0.0
2974
+ 0.0
2975
+ 0.0
2976
+ 0.0
2977
+ 0.0
2978
+ 0.0
2979
+ 0.0
2980
+ 0.0
2981
+ 0.0
2982
+ 0.0
2983
+ 0.0
2984
+ 0.0
2985
+ 0.0
2986
+ 0.0
2987
+ 0.0
2988
+ 0.0
2989
+ 0.0
2990
+ 0.0
2991
+ 0.0
2992
+ 0.0
2993
+ 0.84
2994
+ AM-DSB-SC
2995
+ 0.0
2996
+ 0.0
2997
+ 0.0
2998
+ 0.0
2999
+ 0.0
3000
+ 0.0
3001
+ 0.0
3002
+ 0.0
3003
+ 0.0
3004
+ 0.0
3005
+ 0.0
3006
+ 0.0
3007
+ 0.0
3008
+ 0.0
3009
+ 0.0
3010
+ 0.0
3011
+ 0.0
3012
+ 0.0
3013
+ 0.16
3014
+ 0.0
3015
+ 0.0
3016
+ 0.0
3017
+ 0.0
3018
+ 1.0
3019
+ 0.0
3020
+ 0.0
3021
+ 0.0
3022
+ 0.0
3023
+ 0.0
3024
+ 0.0
3025
+ 0.0
3026
+ 0.0
3027
+ 0.0
3028
+ 0.0
3029
+ 0.0
3030
+ 0.0
3031
+ 0.0
3032
+ 0.0
3033
+ 0.0
3034
+ 0.0
3035
+ 0.0
3036
+ 0.0
3037
+ 0.0
3038
+ 0.0
3039
+ 0.0
3040
+ FM
3041
+ 0.0
3042
+ 0.0
3043
+ GMSK
3044
+ 0.0
3045
+ 0.0
3046
+ 0.0
3047
+ 0.0
3048
+ 0.0
3049
+ 0.0
3050
+ 0.0
3051
+ 0.0
3052
+ 1.0
3053
+ 0.0
3054
+ 0.0
3055
+ 0.0
3056
+ 0.0
3057
+ 0.0
3058
+ 0.0
3059
+ 0.0
3060
+ 0.0
3061
+ 0.0
3062
+ 0.0
3063
+ 0.0
3064
+ 0.0
3065
+ 0.0
3066
+ 0.0
3067
+ 0.0
3068
+ 0.0
3069
+ 1.0
3070
+ OQPSK
3071
+ 0.0
3072
+ 0.0
3073
+ 0.0
3074
+ 0.0
3075
+ 0.0
3076
+ 0.0
3077
+ 0.0
3078
+ 0.0
3079
+ 0.0
3080
+ 0.0
3081
+ 0.0
3082
+ 0.0
3083
+ 0.0
3084
+ 0.0
3085
+ 0.0
3086
+ 0.0
3087
+ 0.0
3088
+ 0.0
3089
+ 0.0
3090
+ 0.0
3091
+ 0.0
3092
+ 0.0
3093
+ Sd8
3094
+ 16PSK
3095
+ YO0
3096
+ BPSK
3097
+ Sdo
3098
+ 32Ah
3099
+ 64AA
3100
+ GMSK
3101
+ 8ASK
3102
+ 32Ps
3103
+ 16AA
3104
+ 128A
3105
+ 16Q
3106
+ 1280
3107
+ 2560
3108
+ FM
3109
+ 4ASK
3110
+ PSK
3111
+ PSK
3112
+ Wvo.
3113
+ Wvo.
3114
+ Predicted
3115
+ Label1.0
3116
+ OOK
3117
+ 0.0
3118
+ 0.0
3119
+ 0.0
3120
+ 0.0
3121
+ 0.0
3122
+ 0.0
3123
+ 0.0
3124
+ 0.0
3125
+ 0.0
3126
+ 0.0
3127
+ 0.0
3128
+ 0.0
3129
+ 0.0
3130
+ 0.0
3131
+ 0.0
3132
+ 0.0
3133
+ 0.0
3134
+ 0.0
3135
+ 0.0
3136
+ 0.0
3137
+ 0.0
3138
+ 0.0
3139
+ 0.0
3140
+ 1
3141
+ 0.0
3142
+ 0.98
3143
+ 4ASK
3144
+ 0.02
3145
+ 0.0
3146
+ 0.0
3147
+ 0.0
3148
+ 0.0
3149
+ 0.0
3150
+ 0.0
3151
+ 0.0
3152
+ 0.0
3153
+ 0.0
3154
+ 0.0
3155
+ 0.0
3156
+ 0.0
3157
+ 0.0
3158
+ 0.0
3159
+ 0.0
3160
+ 0.0
3161
+ 0.0
3162
+ 0.0
3163
+ 0.0
3164
+ 0.0
3165
+ 0.0
3166
+ 8ASK
3167
+ 0.04
3168
+ 0.96
3169
+ 0.0
3170
+ 0.0
3171
+ 0.0
3172
+ 0.0
3173
+ 0.0
3174
+ 0.0
3175
+ 0.0
3176
+ 0.0
3177
+ 0.0
3178
+ 0.0
3179
+ 0.0
3180
+ 0.0
3181
+ 0.0
3182
+ 0.0
3183
+ 0.0
3184
+ 0.0
3185
+ 0.0
3186
+ 0.0
3187
+ 0.0
3188
+ 0.0
3189
+ 0.0
3190
+ 0.0
3191
+ 1.0
3192
+ BPSK
3193
+ 0.0
3194
+ 0.0
3195
+ 0.0
3196
+ 0.0
3197
+ 0.0
3198
+ 0.0
3199
+ 0.0
3200
+ 0.0
3201
+ 0.0
3202
+ 0.0
3203
+ 0.0
3204
+ 0.0
3205
+ 0.0
3206
+ 0.0
3207
+ 0.0
3208
+ 0.0
3209
+ 0.0
3210
+ 0.0
3211
+ 0.0
3212
+ 0.0
3213
+ 0.0
3214
+ 0.0
3215
+ 0.0
3216
+ 66°0
3217
+ QPSK
3218
+ 0.0
3219
+ 0.0
3220
+ 0.0
3221
+ 0.0
3222
+ 0.0
3223
+ 0.0
3224
+ 0.0
3225
+ 0.0
3226
+ 0.0
3227
+ 0.0
3228
+ 0.0
3229
+ 0.0
3230
+ 0.0
3231
+ 0.0
3232
+ 0.0
3233
+ 0.0
3234
+ 0.0
3235
+ 0.0
3236
+ 0.0
3237
+ 0.0
3238
+ 0.0
3239
+ 0.0
3240
+ 0.0
3241
+ 0.8
3242
+ 0.94
3243
+ 8PSK
3244
+ 0.0
3245
+ 0.0
3246
+ 0.0
3247
+ 0.0
3248
+ 0.01
3249
+ 0.01
3250
+ 0.04
3251
+ 0.0
3252
+ 0.0
3253
+ 0.0
3254
+ 0.0
3255
+ 0.0
3256
+ 0.0
3257
+ 0.0
3258
+ 0.0
3259
+ 0.0
3260
+ 0.0
3261
+ 0.0
3262
+ 0.0
3263
+ 0.0
3264
+ 0.0
3265
+ 0.0
3266
+ 0.0
3267
+ 0.05
3268
+ 0.83
3269
+ 16PSK
3270
+ 0.0
3271
+ 0.0
3272
+ 0.0
3273
+ 0.0
3274
+ 0.01
3275
+ 0.11
3276
+ 0.0
3277
+ 0.0
3278
+ 0.0
3279
+ 0.0
3280
+ 0.0
3281
+ 0.0
3282
+ 0.0
3283
+ 0.0
3284
+ 0.0
3285
+ 0.0
3286
+ 0.0
3287
+ 0.0
3288
+ 0.0
3289
+ 0.0
3290
+ 0.0
3291
+ 0.0
3292
+ 0.0
3293
+ 0.0
3294
+ 0.0
3295
+ 0.01
3296
+ 0.05
3297
+ 0.06
3298
+ 0.88
3299
+ 32PSK
3300
+ 0.0
3301
+ 0.0
3302
+ 0.0
3303
+ 0.0
3304
+ 0.0
3305
+ 0.0
3306
+ 0.0
3307
+ 0.0
3308
+ 0.0
3309
+ 0.0
3310
+ 0.0
3311
+ 0.0
3312
+ 0.0
3313
+ 0.0
3314
+ 0.0
3315
+ 0.0
3316
+ 0.0
3317
+ 0.0
3318
+ 0.96
3319
+ 16APSK
3320
+ 0.0
3321
+ 0.0
3322
+ 0.0
3323
+ 0.0
3324
+ 0.0
3325
+ 0.0
3326
+ 0.0
3327
+ 0.0
3328
+ 0.0
3329
+ 0.0
3330
+ 0.01
3331
+ 0.01
3332
+ 0.0
3333
+ 0.0
3334
+ 0.0
3335
+ 0.0
3336
+ 0.0
3337
+ 0.0
3338
+ 0.0
3339
+ 0.0
3340
+ 0.0
3341
+ 0.0
3342
+ 0.0
3343
+ 0.0
3344
+ 0.0
3345
+ 0.0
3346
+ 0.0
3347
+ 0.0
3348
+ 0.0
3349
+ 0.0
3350
+ 0.95
3351
+ 32APSK
3352
+ 0.0
3353
+ 0.01
3354
+ 0.02
3355
+ 0.01
3356
+ 0.01
3357
+ 0.0
3358
+ 0.0
3359
+ 0.0
3360
+ 0.0
3361
+ 0.0
3362
+ 0.0
3363
+ 0.0
3364
+ 0.0
3365
+ 0.0
3366
+ 0.0
3367
+ 0.6
3368
+ 0.82
3369
+ 64APSK
3370
+ 0.0
3371
+ 0.0
3372
+ 0.0
3373
+ 0.0
3374
+ 0.0
3375
+ 0.0
3376
+ 0.01
3377
+ 0.02
3378
+ 0.04
3379
+ 0.0
3380
+ 0.0
3381
+ 0.03
3382
+ 0.04
3383
+ 0.01
3384
+ 0.03
3385
+ 0.01
3386
+ 0.0
3387
+ 0.0
3388
+ 0.0
3389
+ 0.0
3390
+ 0.0
3391
+ 0.0
3392
+ 0.0
3393
+ 0.05
3394
+ 0.82
3395
+ 128APSK
3396
+ 0.0
3397
+ 0.0
3398
+ 0.0
3399
+ 0.0
3400
+ 0.0
3401
+ 0.0
3402
+ 0.0
3403
+ 0.0
3404
+ 0.0
3405
+ 0.03
3406
+ 0.02
3407
+ 0.02
3408
+ 0.01
3409
+ 0.03
3410
+ 0.02
3411
+ 0.0
3412
+ 0.0
3413
+ 0.0
3414
+ 0.0
3415
+ 0.0
3416
+ 0.0
3417
+ 0.0
3418
+ 0.0
3419
+ 0.0
3420
+ 0.0
3421
+ 0.0
3422
+ 0.01
3423
+ 0.91
3424
+ 16QAM
3425
+ 0.0
3426
+ 0.0
3427
+ 0.0
3428
+ 0.0
3429
+ 0.0
3430
+ 0.02
3431
+ 0.01
3432
+ 0.02
3433
+ 0.01
3434
+ 0.01
3435
+ 0.01
3436
+ 0.0
3437
+ 0.0
3438
+ 0.0
3439
+ 0.0
3440
+ 0.0
3441
+ 0.0
3442
+ 0.0
3443
+ 0.0
3444
+ 0.0
3445
+ 0.03
3446
+ 0.83
3447
+ 32QAM
3448
+ 0.0
3449
+ 0.0
3450
+ 0.0
3451
+ 0.0
3452
+ 0.0
3453
+ 0.0
3454
+ 0.02
3455
+ 0.01
3456
+ 0.03
3457
+ 0.02
3458
+ 0.01
3459
+ 0.04
3460
+ 0.01
3461
+ 0.0
3462
+ 0.0
3463
+ 0.0
3464
+ 0.0
3465
+ 0.0
3466
+ 0.0
3467
+ 0.0
3468
+ 0.75
3469
+ 0.4
3470
+ 64QAM
3471
+ 0.0
3472
+ 0.0
3473
+ 0.0
3474
+ 0.0
3475
+ 0.0
3476
+ 0.0
3477
+ 0.0
3478
+ 0.0
3479
+ 0.01
3480
+ 0.01
3481
+ 0.01
3482
+ 0.02
3483
+ 0.04
3484
+ 0.02
3485
+ 0.05
3486
+ 0.09
3487
+ 0.0
3488
+ 0.0
3489
+ 0.0
3490
+ 0.0
3491
+ 0.0
3492
+ 0.0
3493
+ 0.0
3494
+ 0.74
3495
+ 128QAM
3496
+ 0.0
3497
+ 0.0
3498
+ 0.0
3499
+ 0.0
3500
+ 0.0
3501
+ 0.0
3502
+ 0.0
3503
+ 0.0
3504
+ 0.01
3505
+ 0.01
3506
+ 0.02
3507
+ 0.03
3508
+ 0.03
3509
+ 0.05
3510
+ 0.04
3511
+ 0.07
3512
+ 0.0
3513
+ 0.0
3514
+ 0.0
3515
+ 0.0
3516
+ 0.0
3517
+ 0.0
3518
+ 0.0
3519
+ 0.02
3520
+ 0.67
3521
+ 256QAM
3522
+ 0.0
3523
+ 0.0
3524
+ 0.0
3525
+ 0.0
3526
+ 0.0
3527
+ 0.0
3528
+ 0.0
3529
+ 0.0
3530
+ 0.0
3531
+ 0.01
3532
+ 0.01
3533
+ 0.03
3534
+ 0.02
3535
+ 0.14
3536
+ 0.09
3537
+ 0.0
3538
+ 0.0
3539
+ 0.0
3540
+ 0.0
3541
+ 0.0
3542
+ 0.0
3543
+ 0.0
3544
+ 0.96
3545
+ AM-SSB-WC
3546
+ 0.0
3547
+ 0.0
3548
+ 0.0
3549
+ 0.0
3550
+ 0.0
3551
+ 0.0
3552
+ 0.0
3553
+ 0.0
3554
+ 0.0
3555
+ 0.0
3556
+ 0.0
3557
+ 0.0
3558
+ 0.0
3559
+ 0.0
3560
+ 0.0
3561
+ 0.0
3562
+ 0.0
3563
+ 0.04
3564
+ 0.0
3565
+ 0.0
3566
+ 0.0
3567
+ 0.0
3568
+ 0.0
3569
+ 0.0
3570
+ 0.0
3571
+ 0.0
3572
+ 0.0
3573
+ 0.0
3574
+ 0.0
3575
+ 0.0
3576
+ 0.0
3577
+ 0.0
3578
+ 0.0
3579
+ 0.0
3580
+ 0.0
3581
+ 0.0
3582
+ 0.0
3583
+ 0.0
3584
+ 0.0
3585
+ 0.0
3586
+ 0.25
3587
+ 0.75
3588
+ AM-SSB-SC
3589
+ 0.0
3590
+ 0.0
3591
+ 0.0
3592
+ 0.0
3593
+ 0.0
3594
+ 0.2
3595
+ 0.0
3596
+ 0.0
3597
+ 0.97
3598
+ 0.03
3599
+ AM-DSB-WC
3600
+ 0.0
3601
+ 0.0
3602
+ 0.0
3603
+ 0.0
3604
+ 0.0
3605
+ 0.0
3606
+ 0.0
3607
+ 0.0
3608
+ 0.0
3609
+ 0.0
3610
+ 0.0
3611
+ 0.0
3612
+ 0.0
3613
+ 0.0
3614
+ 0.0
3615
+ 0.0
3616
+ 0.0
3617
+ 0.0
3618
+ 0.0
3619
+ 0.0
3620
+ 0.86
3621
+ AM-DSB-SC
3622
+ 0.0
3623
+ 0.0
3624
+ 0.0
3625
+ 0.0
3626
+ 0.0
3627
+ 0.0
3628
+ 0.0
3629
+ 0.0
3630
+ 0.0
3631
+ 0.0
3632
+ 0.0
3633
+ 0.0
3634
+ 0.0
3635
+ 0.0
3636
+ 0.0
3637
+ 0.0
3638
+ 0.0
3639
+ 0.0
3640
+ 0.0
3641
+ 0.14
3642
+ 0.0
3643
+ 0.0
3644
+ 0.0
3645
+ 1.0
3646
+ 0.0
3647
+ 0.0
3648
+ 0.0
3649
+ 0.0
3650
+ 0.0
3651
+ 0.0
3652
+ 0.0
3653
+ 0.0
3654
+ 0.0
3655
+ 0.0
3656
+ 0.0
3657
+ 0.0
3658
+ 0.0
3659
+ 0.0
3660
+ 0.0
3661
+ 0.0
3662
+ 0.0
3663
+ 0.0
3664
+ 0.0
3665
+ 0.0
3666
+ 0.0
3667
+ FM
3668
+ 0.0
3669
+ 0.0
3670
+ 0.0
3671
+ 0.0
3672
+ 0.0
3673
+ 0.0
3674
+ 0.0
3675
+ 0.0
3676
+ 1.0
3677
+ GMSK
3678
+ 0.0
3679
+ 0.0
3680
+ 0.0
3681
+ 0.0
3682
+ 0.0
3683
+ 0.0
3684
+ 0.0
3685
+ 0.0
3686
+ 0.0
3687
+ 0.0
3688
+ 0.0
3689
+ 0.0
3690
+ 0.0
3691
+ 0.0
3692
+ 0.0
3693
+ 0.0
3694
+ 0.0
3695
+ OQPSK
3696
+ 0.0
3697
+ 0.0
3698
+ 0.0
3699
+ 0.0
3700
+ 0.0
3701
+ 0.0
3702
+ 0.0
3703
+ 0.0
3704
+ 0.0
3705
+ 0.0
3706
+ 0.0
3707
+ 0.0
3708
+ 0.0
3709
+ 0.0
3710
+ 0.0
3711
+ 0.0
3712
+ 0.0
3713
+ 0.0
3714
+ 0.0
3715
+ 0.0
3716
+ 0.0
3717
+ 0.0
3718
+ 0.0
3719
+ VSd8
3720
+ 16PSK
3721
+ YO0
3722
+ 8ASK
3723
+ BPSK
3724
+ Sdo
3725
+ 16AP
3726
+ 32AP
3727
+ 64AA
3728
+ 1284
3729
+ 16QA
3730
+ AM-D
3731
+ 4ASK
3732
+ 2560
3733
+ AM.S
3734
+ hy
3735
+ APSK
3736
+ Wvo.
3737
+ Wvoo
3738
+ -DSB.
3739
+ PSK
3740
+ PSK12
3741
+ modifications, we achieved a new state-of-the-art in AMC
3742
+ performance. Among these modifications, dilated convolutions
3743
+ were found to be the most critical architectural feature for
3744
+ model performance. Self-attention was also investigated but
3745
+ was not found to increase performance—although increased
3746
+ temporal context improved upon prior works.
3747
+ REFERENCES
3748
+ [1] T. O’Shea, T. Roy, and T. C. Clancy, “Over-the-air deep learning based
3749
+ radio signal classification,” in IEEE Journal of Selected Topics in Signal
3750
+ Processing, vol. 12:1.
3751
+ IEEE, 2018, pp. 168–179.
3752
+ [2] C. A. Harper, A. Sinha, M. A. Thornton, and E. C. Larson, “SNR-
3753
+ boosted automatic modulation classification,” in 2021 55th Asilomar
3754
+ Conference on Signals, Systems, and Computers.
3755
+ IEEE, 2021, pp.
3756
+ 372–375.
3757
+ [3] N. Soltani, K. Sankhe, S. Ioannidis, D. Jaisinghani, and K. Chowdhury,
3758
+ “Spectrum awareness at the edge: Modulation classification using smart-
3759
+ phones,” in 2019 IEEE International Symposium on Dynamic Spectrum
3760
+ Access Networks (DySPAN).
3761
+ IEEE, 2019, pp. 1–10.
3762
+ [4] S. Soliman and S.-Z. Hsue, “Signal classification using statistical mo-
3763
+ ments,” IEEE Transactions on Communications, vol. 40, no. 5, pp. 908–
3764
+ 916, 1992.
3765
+ [5] A. Swami and B. Sadler, “Hierarchical digital modulation classification
3766
+ using cumulants,” IEEE Transactions on Communications, vol. 48, no. 3,
3767
+ pp. 416–429, 2000.
3768
+ [6] M. Abdelbar, W. H. Tranter, and T. Bose, “Cooperative cumulants-based
3769
+ modulation classification in distributed networks,” IEEE Transactions on
3770
+ Cognitive Communications and Networking, vol. 4, no. 3, pp. 446–461,
3771
+ 2018.
3772
+ [7] C. A. Harper, L. Lyons, M. A. Thornton, and E. C. Larson, “Enhanced
3773
+ automatic modulation classification using deep convolutional latent
3774
+ space pooling,” in 2020 54th Asilomar Conference on Signals, Systems,
3775
+ and Computers, 2020.
3776
+ [8] T. O’Shea, J. Corgan, and T. Clancy, “Convolutional radio modulation
3777
+ recognition networks,” in International Conference on Engineering
3778
+ Applications of Neural Networks., 2016, p. 213–226.
3779
+ [9] T. Huynh-The, C.-H. Hua, J.-W. Kim, S.-H. Kim, and D.-S. Kim,
3780
+ “Exploiting a low-cost cnn with skip connection for robust automatic
3781
+ modulation classification,” in 2020 IEEE Wireless Communications and
3782
+ Networking Conference (WCNC), 2020, pp. 1–6.
3783
+ [10] S. Peng, H. Jiang, H. Wang, H. Alwageed, and Y. Yao, “Modulation
3784
+ classification using convolutional neural network based deep learning
3785
+ model,” in 2017 26th Wireless and Optical Communication Conference
3786
+ (WOCC), 2017, pp. 1–5.
3787
+ [11] S. Peng, H. Jiang, H. Wang, H. Alwageed, Y. Zhou, M. M. Sebdani,
3788
+ and Y. Yao, “Modulation classification based on signal constellation
3789
+ diagrams and deep learning,” IEEE Transactions on Neural Networks
3790
+ and Learning Systems, vol. 30, no. 3, pp. 718–727, 2019.
3791
+ [12] F. Wang, Y. Wang, and X. Chen, “Graphic constellations and DBN
3792
+ based automatic modulation classification,” in 2017 IEEE 85th Vehicular
3793
+ Technology Conference (VTC Spring), 2017, pp. 1–5.
3794
+ [13] K. Hassan, I. Dayoub, W. Hamouda, C. N. Nzeza, and M. Berbineau,
3795
+ “Blind digital modulation identification for spatially-correlated mimo
3796
+ systems,” IEEE Transactions on Wireless Communications, vol. 11,
3797
+ no. 2, pp. 683–693, 2012.
3798
+ [14] M. Abdelbar, B. Tranter, and T. Bose, “Cooperative modulation classi-
3799
+ fication of multiple signals in cognitive radio networks,” in 2014 IEEE
3800
+ International Conference on Communications (ICC), 2014, pp. 1483–
3801
+ 1488.
3802
+ [15] DeepSig
3803
+ Incorporated,
3804
+ “RF
3805
+ datasets
3806
+ for
3807
+ machine
3808
+ learning,”
3809
+ https://www.deepsig.ai/datasets, 2018, accessed: 2021-
3810
+ 04-29.
3811
+ [16] D. Liu, P. Wang, T. Wang, and T. Abdelzaher, “Self-contrastive learning
3812
+ based semi-supervised radio modulation classification,” in MILCOM
3813
+ 2021-2021 IEEE Military Communications Conference (MILCOM).
3814
+ IEEE, 2021, pp. 777–782.
3815
+ [17] Z. Chen, H. Cui, J. Xiang, K. Qiu, L. Huang, S. Zheng, S. Chen,
3816
+ Q. Xuan, and X. Yang, “Signet: A novel deep learning framework for
3817
+ radio signal classification,” IEEE Transactions on Cognitive Communi-
3818
+ cations and Networking, vol. 8, no. 2, pp. 529–541, 2022.
3819
+ [18] T. Chen, S. Gao, S. Zheng, S. Yu, Q. Xuan, C. Lou, and X. Yang, “Emd
3820
+ and vmd empowered deep learning for radio modulation recognition,”
3821
+ IEEE Transactions on Cognitive Communications and Networking, pp.
3822
+ 1–1, 2022.
3823
+ [19] A. J. Uppal, M. Hegarty, W. Haftel, P. A. Sallee, H. B. Cribbs, and
3824
+ H. H. Huang, “High-performance deep learning classification for radio
3825
+ signals,” in 2019 53rd Asilomar Conference on Signals, Systems, and
3826
+ Computers, 2019.
3827
+ [20] C. Park, J. Choi, S. Nah, W. Jang, and D. Y. Kim, “Automatic modulation
3828
+ recognition of digital signals using wavelet features and SVM,” in 2008
3829
+ 10th International Conference on Advanced Communication Technology,
3830
+ vol. 1, 2008, pp. 387–390.
3831
+ [21] X. Teng, P. Tian, and H. Yu, “Modulation classification based on spectral
3832
+ correlation and SVM,” in 2008 4th International Conference on Wireless
3833
+ Communications, Networking and Mobile Computing, 2008, pp. 1–4.
3834
+ [22] Z. Zhang, C. Wang, C. Gan, S. Sun, and M. Wang, “Automatic
3835
+ modulation classification using convolutional neural network with fea-
3836
+ tures fusion of SPWVD and BJD,” IEEE Transactions on Signal and
3837
+ Information Processing over Networks, vol. 5, no. 3, pp. 469–478, 2019.
3838
+ [23] S. Zheng, P. Qi, S. Chen, and X. Yang, “Fusion methods for CNN-based
3839
+ automatic modulation classification,” IEEE Access, vol. 7, pp. 66 496–
3840
+ 66 504, 2019.
3841
+ [24] Y. Mao, Y.-Y. Dong, T. Sun, X. Rao, and C.-X. Dong, “Attentive siamese
3842
+ networks for automatic modulation classification based on multitiming
3843
+ constellation diagrams,” IEEE Transactions on Neural Networks and
3844
+ Learning Systems, 2021.
3845
+ [25] J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in
3846
+ Proceedings of the IEEE conference on computer vision and pattern
3847
+ recognition, 2018, pp. 7132–7141.
3848
+ [26] S. Tridgell, “Low latency machine learning on fpgas,” Ph.D. dissertation,
3849
+ The University of Sydney Australia, 2019.
3850
+ [27] D. Snyder, D. Garcia-Romero, G. Sell, D. Povey, and S. Khudanpur,
3851
+ “X-vectors: Robust DNN embeddings for speaker recognition,” in 2018
3852
+ IEEE International Conference on Acoustics, Speech and Signal Pro-
3853
+ cessing (ICASSP).
3854
+ IEEE, 2018, pp. 5329–5333.
3855
+ [28] Q. McNemar, “Note on the sampling error of the difference between
3856
+ correlated proportions or percentages,” Psychometrika, vol. 12, no. 2,
3857
+ pp. 153–157, 1947.
3858
+ [29] J. Cai, F. Gan, X. Cao, and W. Liu, “Signal modulation classification
3859
+ based on the transformer network,” IEEE Transactions on Cognitive
3860
+ Communications and Networking, vol. 8, no. 3, pp. 1348–1357, 2022.
3861
+ [30] Y. Wen, K. Zhang, Z. Li, and Y. Qiao, “A discriminative feature
3862
+ learning approach for deep face recognition,” in European conference
3863
+ on computer vision.
3864
+ Springer, 2016, pp. 499–515.
3865
+ [31] H. Zhang, F. Zhou, Q. Wu, W. Wu, and R. Q. Hu, “A novel automatic
3866
+ modulation classification scheme based on multi-scale networks,” IEEE
3867
+ Transactions on Cognitive Communications and Networking, vol. 8,
3868
+ no. 1, pp. 97–110, 2022.
3869
+ [32] D. Zhang, Y. Lu, Y. Li, W. Ding, and B. Zhang, “High-order convo-
3870
+ lutional attention networks for automatic modulation classification in
3871
+ communication,” IEEE Transactions on Wireless Communications, pp.
3872
+ 1–1, 2022.
3873
+ [33] T. J. O’Shea, J. Corgan, and T. C. Clancy, “Convolutional radio modula-
3874
+ tion recognition networks,” in International conference on engineering
3875
+ applications of neural networks.
3876
+ Springer, 2016, pp. 213–226.
3877
+ [34] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,”
3878
+ CoRR, vol. abs/1412.6980, 2015.
3879
+ [35] F. Chollet, Deep learning with Python.
3880
+ Simon and Schuster, 2021.
3881
+ [36] F. Yu and V. Koltun, “Multi-scale context aggregation by dilated
3882
+ convolutions,” in International Conference on Learning Representations
3883
+ (ICLR), May 2016.
3884
+ [37] X. Zhang, X. Zhou, M. Lin, and J. Sun, “Shufflenet: An extremely effi-
3885
+ cient convolutional neural network for mobile devices,” in Proceedings
3886
+ of the IEEE conference on computer vision and pattern recognition,
3887
+ 2018, pp. 6848–6856.
3888
+ [38] M. Tan and Q. Le, “Efficientnet: Rethinking model scaling for con-
3889
+ volutional neural networks,” in International conference on machine
3890
+ learning.
3891
+ PMLR, 2019, pp. 6105–6114.
3892
+ [39] K. Okabe, T. Koshinaka, and K. Shinoda, “Attentive Statistics Pooling
3893
+ for Deep Speaker Embedding,” in Proc. Interspeech 2018, 2018, pp.
3894
+ 2252–2256.
3895
+ [40] P. Safari, M. India, and J. Hernando, “Self-attention encoding and
3896
+ pooling for speaker recognition,” in Proc. Interspeech 2020, 2020, pp.
3897
+ 941–945.
3898
+ [41] G. Sammit, Z. Wu, Y. Wang, Z. Wu, A. Kamata, J. Nese, and E. C.
3899
+ Larson, “Automated prosody classification for oral reading fluency
3900
+ with quadratic kappa loss and attentive x-vectors,” in ICASSP 2022-
3901
+ 2022 IEEE International Conference on Acoustics, Speech and Signal
3902
+ Processing (ICASSP).
3903
+ IEEE, 2022.
3904
+
3905
+ 13
3906
+ [42] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez,
3907
+ Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in
3908
+ neural information processing systems, vol. 30, 2017.
3909
+ Clayton A. Harper received his B.S. in mathematics
3910
+ and computer engineering in 2019 and M.S. in
3911
+ data engineering in 2021 from Southern Methodist
3912
+ University in Dallas, TX, where he specialized in
3913
+ machine learning and audio signal processing. His
3914
+ main research area is the analysis of time series
3915
+ signal processing in computer systems, especially
3916
+ pertaining to security and privacy. He is a student
3917
+ member of IEEE and is currently pursuing his Ph.D.
3918
+ in computer science at Southern Methodist Univer-
3919
+ sity with the co-advisors Dr. Eric C. Larson and Dr.
3920
+ Mitchell A. Thornton.
3921
+ Mitchell (Mitch) A. Thornton is currently the Cecil
3922
+ H. Green Chair of Engineering and Professor in
3923
+ the Department of Electrical and Computer Engi-
3924
+ neering at Southern Methodist University in Dallas,
3925
+ Texas. He also serves as the Executive Director of
3926
+ the Darwin Deason Institute for Cyber Security, a
3927
+ research-only unit, and as Program Director for the
3928
+ interdisciplinary M.S. in Data Engineering degree
3929
+ program within the Lyle School of Engineering at
3930
+ SMU. His main research interests are in the areas
3931
+ of cyber security and quantum informatics. His past
3932
+ industrial experience includes full-time employment at the Amoco Research
3933
+ Center, E-Systems, Inc. (now L3Harris Technologies Inc.), and the Cyrix
3934
+ Corporation. Dr. Thornton is a member of several professional and honor
3935
+ societies including the IEEE and the ACM where he is a senior member in
3936
+ each organization. He was elected as Chair of the IEEE Technical Community
3937
+ on Multiple-Valued Logic (TCMVL, 2010-11) and has served in various roles
3938
+ for other IEEE/ACM committees. He is an author or co-author of five books
3939
+ and more than 300 technical articles. He is a named inventor on over 20
3940
+ US/PCT/WIPO patents and patents pending. He holds P.E. licenses in the
3941
+ states of Texas, Mississippi and Arkansas. He received the Ph.D. in computer
3942
+ engineering from SMU in 1995, M.S. in computer science from SMU in 1993,
3943
+ M.S. in electrical engineering from the University of Texas at Arlington in
3944
+ 1990, and B.S. in electrical engineering from Oklahoma State University in
3945
+ 1985.
3946
+ Eric C. Larson is an Associate Professor in the de-
3947
+ partment of Computer Science in the Bobby B. Lyle
3948
+ School of Engineering, Southern Methodist Univer-
3949
+ sity. His main research interests are in machine
3950
+ learning, sensing, and signal / image processing
3951
+ for various applications, in particular, for healthcare
3952
+ and security applications. His work in both areas
3953
+ has been commercialized and he holds a variety of
3954
+ patents for sustainability sensing and mobile phone-
3955
+ based health sensing. Dr. Larson has authored one
3956
+ textbook and over 70 technical articles. He is active
3957
+ in signal processing education for computer scientists and is an active member
3958
+ of IEEE and the ACM. He received his Ph.D. in 2013 from the University
3959
+ of Washington, where he was co-advised by Shwetak N. Patel and Les Atlas.
3960
+ He received his B.S. and M.S. in Electrical Engineering in 2006 and 2008,
3961
+ respectively, at Oklahoma State University, where he was advised by Damon
3962
+ Chandler.
3963
+
4tFKT4oBgHgl3EQfRy39/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
79FLT4oBgHgl3EQfAy4h/content/tmp_files/2301.11967v1.pdf.txt ADDED
@@ -0,0 +1,1315 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique
2
+ for Intermittent Computing
3
+ SATYAJASWANTH BADRI, MUKESH SAINI, and NEERAJ GOEL, Indian Institute of Technology,
4
+ Ropar
5
+ Battery-less technology evolved to replace battery usage in space, deep mines, and other environments to
6
+ reduce cost and pollution. Non-volatile memory (NVM) based processors were explored for saving the system
7
+ state during a power failure. Such devices have a small SRAM and large non-volatile memory. To make the
8
+ system energy efficient, we need to use SRAM efficiently. So we must select some portions of the application
9
+ and map them to either SRAM or FRAM. This paper proposes an ILP-based memory mapping technique for
10
+ Intermittently powered IoT devices. Our proposed technique gives an optimal mapping choice that reduces
11
+ the system’s Energy-Delay Product (EDP). We validated our system using a TI-based MSP430FR6989 and
12
+ MSP430F5529 development boards. Our proposed memory configuration consumes 38.10% less EDP than
13
+ the baseline configuration and 9.30% less EDP than the existing work under stable power. Our proposed
14
+ configuration achieves 15.97% less EDP than the baseline configuration and 21.99% less EDP than the existing
15
+ work under unstable power. This work supports intermittent computing and works efficiently during frequent
16
+ power failures.
17
+ Additional Key Words and Phrases: NVM, MSP430FR6989, ILP, Intermittent power, Memory-Mapping
18
+ 1
19
+ INTRODUCTION
20
+ The Internet of Things (IoT) is a network of sensors and nodes that allows nearby objects to
21
+ communicate and collaborate easily. Batteries are the most common source of power for IoT devices.
22
+ Because of the battery’s limited capacity and short lifespan [15], replacement is costly. IoT may
23
+ consist of billions of sensors and systems by the end of 2050 [9]. Replacing and disposing billions
24
+ of battery-operated devices is expensive and hazardous to the environment. As a result, we need
25
+ battery-free IoT devices.
26
+ Energy harvesters are a promising alternative to battery-powered devices. The energy harvester
27
+ collects energy from the environment and stores energy in capacitors. Energy harvesting is un-
28
+ reliable, power failures are unavoidable, and the application’s execution is irregular. This type of
29
+ computing is known as intermittent computing [14, 27, 34].
30
+ For intermittently powered IoT devices, energy harvesting is the primary energy source. Energy
31
+ harvesting sources like piezo-electric materials and radio-frequency devices extract a small amount
32
+ of energy from their surroundings. We must use energy efficiently in both stable and unstable
33
+ power supply scenarios.
34
+ In order to utilize energy efficiently and to make the system energy efficient, we primarily have
35
+ two choices. The first choice is to reduce energy consumption by proposing new techniques that
36
+ use energy efficiently. The second choice is to increase the number of different energy harvesters,
37
+ which will accumulate more energy while increasing maintenance costs. We need to maintain
38
+ these many energy harvesters, which is not a feasible solution. Thus, our main concern is to reduce
39
+ energy consumption by proposing new techniques which help to design an energy-efficient system.
40
+ Gonzalez et al. [10] mentioned energy as not an ideal metric for evaluating system efficiency. By
41
+ simply reducing supply voltage or load capacitance, energy can be reduced. Instead of using energy
42
+ as a metric, they suggested using the Energy-Delay Product (EDP) as the energy-efficient design
43
+ Authors’ address: SatyaJaswanth Badri, [email protected]; Mukesh Saini, [email protected]; Neeraj Goel, neeraj@
44
+ iitrpr.ac.in, Indian Institute of Technology, Ropar, S.Ramanujan Block, IIT Ropar Main Campus, Ropar, Punjab, India, 140001.
45
+ arXiv:2301.11967v1 [cs.AR] 27 Jan 2023
46
+
47
+ 2
48
+ S.J Badri, et al.
49
+ metric. The EDP considers both performance and energy simultaneously in a design. If a design
50
+ minimizes the EDP, we can call such a design energy-efficient. We define EDP in the equation 1.
51
+ 𝐸𝐷𝑃 = 𝐸𝑠𝑦𝑠𝑡𝑒𝑚 × 𝑁𝑢𝑚_𝑐𝑦𝑐𝑙𝑒𝑠
52
+ (1)
53
+ Where 𝐸𝑠𝑦𝑠𝑡𝑒𝑚 is the system’s energy consumption, 𝑁𝑢𝑚_𝑐𝑦𝑐𝑙𝑒𝑠 is the number of CPU cycles.
54
+ During these frequent power failures, executing IoT applications becomes more difficult because
55
+ all computed data may be lost, and the application’s execution must restart from the beginning.
56
+ During power failures, we need an additional procedure to backup/checkpoint the volatile memory
57
+ contents to non-volatile memory (NVM).
58
+ Flash memory was the prior NVM technology used by modern microcontrollers at the main
59
+ memory level, such as MSP430F5529 [24]. Flash is ineffective for frequent backups and checkpointing
60
+ because its erase/write operations require a lot of energy. Emerging NVMs outperform flash,
61
+ including spin-transfer-torque RAM (STT-RAM) [4, 28], phase-change memory (PCM) [25], resistive
62
+ RAM (ReRAM), and ferroelectric RAM (FRAM) [16]. Previous works have been demonstrated by
63
+ incorporating these emerging NVMs into low-power-based microcontrollers (MCUs) [16, 18, 24].
64
+ Recent non-volatile processors (NVPs), such as the flash-based MSP430F5529 and the FRAM-based
65
+ MSP430FR6989, encourage the use of hybrid main memory. The flash-based NVP, MSP430F5529, is
66
+ made up of SRAM and flash, while the FRAM-based NVP, MSP430FR6989, is made up of SRAM
67
+ and FRAM at the main memory level. The challenges associated with hybrid main memory-based
68
+ NVPs, such as MSP430FR6989, are as follows.
69
+ (1) FRAM consumes 2x times more energy and latency than SRAM. This design degrades system
70
+ performance and consumes extra energy even during normal operations.
71
+ (2) SRAM loses contents during a power failure and needs to execute the application from the
72
+ beginning, which consumes extra energy and time. For large-size applications, this design
73
+ will not be helpful. Anyway, using only SRAM performs better during regular operations.
74
+ (3) We can design a hybrid main memory to get the benefits from both SRAM and FRAM. The
75
+ following questions need to be answered and analyzed to use the hybrid main memory design.
76
+ (a) How do we choose the appropriate sections of a program and map them to either SRAM
77
+ or FRAM regions? A significant challenge is mapping a program’s stack, code, and data
78
+ sections to either SRAM or FRAM.
79
+ (b) How and where should volatile contents be backed up to the NVM region during frequent
80
+ power failures?
81
+ The main question is which section of an application should be placed in which memory region;
82
+ this is essentially a memory mapping problem. Concerning all of the challenges mentioned earlier,
83
+ this article makes the following contributions:
84
+ • To the best of our knowledge, this is the first work on the Integer-Linear Programming (ILP)
85
+ based memory mapping technique for intermittently powered IoT devices.
86
+ • We formulated the memory mapping problem to cover all the possible design choices. We
87
+ also formulated our problem in such a way that it supports large-size applications.
88
+ • We proposed a framework that efficiently consumes low energy during regular operation
89
+ and frequent power failures. Our proposed framework supports intermittent computing.
90
+ • We evaluated the proposed techniques and frameworks in actual hardware boards.
91
+ Our proposed ILP model recommends placing each section in either SRAM or FRAM. We com-
92
+ pared the proposed memory configuration and techniques with the baseline memory configurations
93
+ under both stable and unstable power scenarios. Our proposed memory configuration consumes
94
+ 38.10% less EDP than baseline-1 and 9.30% less EDP than the existing work under stable power.
95
+
96
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
97
+ 3
98
+ Our proposed configuration achieves 15.97% less EDP than baseline-1 and 21.99% less EDP than the
99
+ existing work under unstable power.
100
+ Paper organization: Section 2 discusses the background and related works. Section 3 explains
101
+ the motivation behind the proposed framework. Section 4 explains the system model and gives an
102
+ overview of the problem definition. Section 5 explains about proposed ILP-based memory mapping
103
+ technique and framework that supports during frequent power failures. The experimental setup
104
+ and results are described in section 6. We conclude this work in section 7.
105
+ 2
106
+ BACKGROUND AND RELATED WORKS
107
+ SRAM and DRAM are used to design registers, caches, and main memory in traditional processors.
108
+ For an intermittently aware design, we replace a regular processor’s volatile memory model
109
+ with an NVM. STT-RAM, PCM, flash, and FRAM are all relatively new NVM technologies [4–
110
+ 6, 11, 17, 25, 28, 29, 31, 37]. FRAM consumes less energy than other NVM technologies, such as flash.
111
+ FRAM can be helpful for IoT devices that are operating at low power. These NVM technologies
112
+ motivated researchers because of their appealing characteristics, such as non-volatility, low cost,
113
+ and high density [2, 3].
114
+ Researchers started using real-time NVPs for intermittent computing [16, 24, 30, 32]. Researchers
115
+ observed that using only NVMs at the cache or main memory level degrades the system’s perfor-
116
+ mance and consumes more energy, which gives an idea to explore hybrid memories. Recent NVPs
117
+ such as MSP430FR6989 [16] consists of both SRAM and FRAM. We need to utilize the SRAM and
118
+ FRAM efficiently and correctly; otherwise, we may degrade system performance and consume
119
+ extra energy. To make the system more efficient, we need to map the application contents to either
120
+ SRAM or FRAM. This is actually a memory mapping problem, similar to scratch-pad memories.
121
+ Researchers explored a similar mapping problem in scratch-pad memories (SPMs) [12, 26, 33].
122
+ Chakraborty et al. [1] documented the existing and standard memory mapping techniques on SPMs.
123
+ In earlier works, memory mapping was done mainly between SPMs and main memory. Memory
124
+ mapping can be done statically and dynamically [21, 22]. In static memory mapping, either ILP
125
+ or the compiler can assist in determining the best placement [12, 26, 33]. ILP-solver takes inputs
126
+ obtained from profilers and memory sizes as constraints in ILP-based memory mapping works. The
127
+ ILP-solver provides the best placement option based on the objective function. In dynamic memory
128
+ allocation [7, 8, 35, 36], either the user-defined program or the compiler will decide on an optimal
129
+ placement choice at run time.
130
+ However, our problem differs from the memory mapping techniques in SPMs because intermittent
131
+ computing brings new constraints. During intermittent computation, the challenges were the
132
+ forward progress of an application, data consistency, environmental consistency, and concurrency
133
+ between the tasks. Due to these challenges, the execution model and development environment
134
+ differ from the SPM-based memory mapping techniques. As a result, we require a memory mapping
135
+ technique that supports intermittent computation.
136
+ Researchers have explored memory mapping techniques and analysis for the MSP430FR6989
137
+ MCU. In FRAM-based MCUs, Jayakumar et al. [18] implement a checkpointing policy. They save the
138
+ system state to FRAM during a power failure. Jayakumar et al. [19, 20] propose an energy-efficient
139
+ memory mapping technique for TI-based applications in FRAM-based MCUs. Kim et al. [23] present
140
+ a detailed analysis of energy consumption for all memory sections in FRAM-based MCUs under
141
+ different memory mappings.
142
+ Earlier works investigated this problem by analyzing the possibilities to make the system efficient.
143
+ The authors [19, 20, 23] have not covered all the design choices and possibilities. In addition, there
144
+ is significantly less contribution towards memory mappings in FRAM-based MCUs that supports
145
+
146
+ 4
147
+ S.J Badri, et al.
148
+ intermittent computation. Our work proposes an energy-efficient memory mapping technique for
149
+ intermittently powered IoT devices that experience frequent power failures.
150
+ 3
151
+ MOTIVATION
152
+ This section discusses the advantages of using hybrid SRAM and FRAM for these MSP430-based
153
+ MCUs over unified SRAM or unified FRAM designs, as well as the importance of an efficient
154
+ memory allocation.
155
+ SRAM is 2KB, and FRAM is 128KB in a FRAM-based MCU, MSP430FR6989. The first naive
156
+ approach is to use the entire 128KB of FRAM in both stable and unstable power scenarios, resulting
157
+ in longer execution cycles and higher energy consumption. Similarly, we have a second naive
158
+ approach to use the entire 2KB SRAM for small applications (whichever fits within the SRAM size),
159
+ which has advantages during regular operation. Unfortunately, it loses all 2KB SRAM data during
160
+ a power failure and takes more time to backup 2KB contents to FRAM during a power failure.
161
+ These two approaches are treated as baselines 1 and 2 for this work. As shown in figure 1, for
162
+ the baseline-1 design, we map all three sections to FRAM and all three sections to SRAM for the
163
+ baseline-2 design.
164
+ int glob1, glob2,..., globn;
165
+ func_1(){
166
+ local_variables
167
+ }
168
+ func_2(){
169
+ local_variables
170
+ }
171
+ func_n(){
172
+ local_variables
173
+ }
174
+ Text
175
+ Data
176
+ Stack
177
+ For func_1 ()
178
+ Text
179
+ Data
180
+ Stack
181
+ For func_2 ()
182
+ For func_n ()
183
+ Program
184
+ Global_Variables
185
+ Functions
186
+ Consists of
187
+ Local Variables
188
+ For global_vars
189
+ Data
190
+ .bss
191
+ Text
192
+ Data
193
+ Stack
194
+ SRAM
195
+ SRAM (2 KB)
196
+ FRAM (128 KB)
197
+ Memory
198
+ Stack(func_1)
199
+ Stack(func_n)
200
+ Text(func_1)
201
+ Data(func_1)
202
+ Data(func_n)
203
+ Text(func_n)
204
+ Map to SRAM
205
+ FRAM
206
+ SRAM (2 KB)
207
+ FRAM (128 KB)
208
+ Memory
209
+ Stack(func_1)
210
+ Stack(func_n)
211
+ Text(func_1)
212
+ Data(func_1)
213
+ Data(func_n)
214
+ Text(func_n)
215
+ Map to FRAM
216
+ Baseline-1 Design
217
+ Baseline-2 Design
218
+ global_vars
219
+ global_vars
220
+ Fig. 1. Overview of the Baseline-1 and Baseline-2 memory mappings in MSP430FR6989
221
+
222
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
223
+ 5
224
+ We compared baseline-1 and baseline-2 in both stable and unstable power scenarios. Baseline-1
225
+ performs better during frequent power failures, while baseline-2 performs better during regular
226
+ operations (without any power failures), as shown in figure 2. On average, baseline-1 consumes
227
+ 47.9% more energy than baseline-2 during a stable power, as shown in figure 2 (a). On average,
228
+ baseline-2 consumes 32.7% more energy than baseline-1 during an unstable power, as shown in
229
+ figure 2 (b). We also observed that MCU would pitch an error to either increase the SRAM space or
230
+ use FRAM space for any computations. For large-size applications will not run using only SRAM, it
231
+ requires FRAM as well. Thus, large applications consume more energy in baseline-2 during a stable
232
+ power scenario.
233
+ These two designs motivate us to propose a hybrid memory design that effectively uses both
234
+ SRAM and FRAM. We also encountered that baseline-2 is ineffective for larger applications. As a
235
+ result, we had to use a hybrid memory and figure out how and where to place the sections. To the
236
+ best of our knowledge, only one work explored the memory mapping issue for these MCUs [20].
237
+ We analyzed the mapping decisions using their empirical model. Jayakumar et al. [20] calculated
238
+ the energy consumption values for each configuration. The authors suggested that allocate the
239
+ sections to either SRAM or FRAM based on the energy values.
240
+ 0
241
+ 0.2
242
+ 0.4
243
+ 0.6
244
+ 0.8
245
+ 1
246
+ 16bit_2dim
247
+ aes
248
+ basicmath_small
249
+ basicmath_large
250
+ bf
251
+ crc
252
+ dhrystone
253
+ dijkstra
254
+ fft
255
+ fir
256
+ matrix_mult
257
+ patricia
258
+ qsort_small
259
+ qsort_large
260
+ sha
261
+ susan
262
+ Normalized Energy Consumption
263
+ (Normalized with Baseline-1)
264
+ Benchmarks
265
+ Baseline-1
266
+ Baseline-2
267
+ (a) Under Stable Power
268
+ 0
269
+ 0.2
270
+ 0.4
271
+ 0.6
272
+ 0.8
273
+ 1
274
+ 16bit_2dim
275
+ aes
276
+ basicmath_small
277
+ basicmath_large
278
+ bf
279
+ crc
280
+ dhrystone
281
+ dijkstra
282
+ fft
283
+ fir
284
+ matrix_mult
285
+ patricia
286
+ qsort_small
287
+ qsort_large
288
+ sha
289
+ susan
290
+ Normalized Energy Consumption
291
+ (Normalized with Baseline-1)
292
+ Benchmarks
293
+ Baseline-1
294
+ Baseline-2
295
+ (b) Under Unstable Power
296
+ Fig. 2. Comparison between Baseline-1 and 2 configurations under Stable and Unstable Power Scenarios
297
+ Table 1. Analysis of the Empirical Methods Used by Jayakumar et al. [20] for qsort_small under stable and
298
+ unstable power supply scenarios
299
+ Configuration
300
+ Text
301
+ Data
302
+ Stack
303
+ 𝐸𝑛𝑒𝑟𝑔𝑦𝑠𝑡𝑎𝑏𝑙𝑒 (𝑚𝐽)
304
+ 𝐸𝑛𝑒𝑟𝑔𝑦𝑢𝑛𝑠𝑡𝑎𝑏𝑙𝑒 (𝑚𝐽)
305
+ 1 {SSS}
306
+ SRAM
307
+ SRAM
308
+ SRAM
309
+ 16.70
310
+ 79.56
311
+ 2 {SSF}
312
+ SRAM
313
+ SRAM
314
+ FRAM
315
+ 21.08
316
+ 66.34
317
+ 3 {SFS}
318
+ SRAM
319
+ FRAM
320
+ SRAM
321
+ 28.75
322
+ 33.79
323
+ 4 {SFF}
324
+ SRAM
325
+ FRAM
326
+ FRAM
327
+ 35.97
328
+ 52.10
329
+ 5 {FSS}
330
+ FRAM
331
+ SRAM
332
+ SRAM
333
+ 39.48
334
+ 68.24
335
+ 6 {FSF}
336
+ FRAM
337
+ SRAM
338
+ FRAM
339
+ 57.64
340
+ 54.75
341
+ 7 {FFS}
342
+ FRAM
343
+ FRAM
344
+ SRAM
345
+ 64.14
346
+ 45.83
347
+ 8 {FFF}
348
+ FRAM
349
+ FRAM
350
+ FRAM
351
+ 92.09
352
+ 36.07
353
+ The empirical method used by the authors is as follows. The authors considered functions as the
354
+ basic unit. They explored all configurations and calculated the energy values, as shown in table 1.
355
+ The authors have eight configurations because they have two memory regions (SRAM or FRAM)
356
+ and need to map three sections (stack, data, text). Using the author’s model, we calculated the
357
+
358
+ 6
359
+ S.J Badri, et al.
360
+ energy values for the qsort_small application. For instance, the SSS configuration performs better
361
+ during a stable power supply, and during a power failure, SFS consumes less energy than all other
362
+ configurations. As a result, authors allocate text and stack sections to SRAM and data sections to
363
+ FRAM.
364
+ We observed that this empirical method becomes ineffective as the number of configurations
365
+ increases. The authors considered all global variables, arrays, and constants as data sections. Instead,
366
+ why can’t we map each global variable or array to either SRAM or FRAM? This increases the
367
+ number of configurations, and calculating/tracking energy values is challenging. Our design space
368
+ grows enormously and makes our mapping problem challenging.
369
+ This new set of challenges motivated us to propose an energy-efficient memory mapping tech-
370
+ nique. Our proposed memory mapping framework supports large-size applications and covers all
371
+ possible configurations.
372
+ 4
373
+ SYSTEM MODEL AND PROBLEM DEFINITION
374
+ This section discusses the system model for embedded MCUs and defines the mapping problem for
375
+ these MCUs.
376
+ 4.1
377
+ System Model
378
+ We consider a simple, customized RISC instruction set with a Von-Neumann architecture, where
379
+ the instructions and data share the same address space that supports at least 16-bit addressing. Base
380
+ architecture doesn’t have a cache to avoid uncertainty. To make things simple, we assume single
381
+ cycle execution of the processor. Base architecture has a small SRAM memory and a larger NVM.
382
+ The MSP430 is an example of such a processor. Non-volatile memory sizes range from 1 kilobyte
383
+ (KB) to 256 KB, while volatile RAM sizes range from 256 bytes to 2KB. Both SRAM and NVM can
384
+ be accessed by instructions using a compiler/linker script. We can modify the linker script to map
385
+ memory according to the memory ranges specified by the user. MSP430 doesn’t have any operating
386
+ system.
387
+ 4.2
388
+ Problem Definition
389
+ Definition 4.1: Optimal Memory Mapping Problem: Given a program that consists of various
390
+ functions and global variables, sizes of SRAM and FRAM, the number of reads and writes for each
391
+ function/variable, and the energy required per read/write to the SRAM/FRAM. What is the optimal
392
+ memory mapping for these functions/variables in order to reduce the system’s EDP?
393
+ The inputs are : Number of functions; number of global variables; energy per write to SRAM
394
+ and FRAM; energy per read to SRAM and FRAM; SRAM and FRAM sizes; Number of CPU cycles
395
+ per each function; the number of reads; the number of writes.
396
+ The output is: Mapping information for all functions and global variables, under which the
397
+ system’s EDP is minimized.
398
+ Definition 4.2: Support for Intermittent Computing: During power failures, we must safely
399
+ backup the volatile contents to NVM. As previously stated, we must use SRAM efficiently for
400
+ energy savings; but again, how can we save the contents of SRAM? There are two significant issues
401
+ with intermittent computation. First, during a power failure, all SRAM’s mapping information
402
+ and register contents are lost, causing the system to become inconsistent. Second, how do we
403
+ backup/restore the mapping information and register contents to ensure system consistency?
404
+
405
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
406
+ 7
407
+ 5
408
+ MAPI-PRO: AN ENERGY EFFICIENT MEMORY MAPPING FOR INTERMITTENT
409
+ COMPUTING
410
+ In this section, we discuss the details of the proposed mapping technique. Our main objective is to
411
+ pick the optimal mapping choice among all the design choices, which reduces the system’s EDP. To
412
+ achieve this, we proposed an ILP-based mapping technique. The overview of the proposed mapping
413
+ technique is shown in figure 3. We also discuss how we support intermittent computing for these
414
+ MCUs.
415
+ int glob_1, glob_2,..., glob_n;
416
+ func_1(){
417
+ local_variables
418
+ }
419
+ func_2(){
420
+ local_variables
421
+ }
422
+ func_n(){
423
+ local_variables
424
+ }
425
+ Text
426
+ Data
427
+ Stack
428
+ For func_1 ()
429
+ Text
430
+ Data
431
+ Stack
432
+ For func_2 ()
433
+ For func_n ()
434
+ Program
435
+ Global_Variables
436
+ Functions
437
+ Consists of
438
+ Local Variables
439
+ Proposed ILP based
440
+ Mapping Technique
441
+ For global_vars
442
+ Data
443
+ .bss
444
+ Text
445
+ Data
446
+ Stack
447
+ Placement
448
+ Decision
449
+ for SRAM
450
+ Placement
451
+ Decision
452
+ for FRAM
453
+ Stack(func_2)
454
+ Stack(func_n)
455
+ Text(func_1)
456
+ Data(func_2)
457
+ Text(func_n)
458
+ Data(func_n)
459
+ Data(func_1)
460
+ Data(func_2)
461
+ Stack(func_1)
462
+ glob_3,....,glob_n
463
+ Text(func_1)
464
+ Text(func_2)
465
+ Data(func_n)
466
+ SRAM (2 KB)
467
+ FRAMn (125 KB)
468
+ Memory
469
+ Backup Region
470
+ FRAMb (3 KB)
471
+ glob_1, glob_2
472
+ Fig. 3. Overview of the proposed memory mappings in MSP430FR6989
473
+ 5.1
474
+ ILP Formulation for Data Mapping
475
+ We present the ILP formulation for the memory mapping problem mentioned in definition 4.1. We
476
+ divide this ILP formulation into two parts, one is for global variables, and the second is for the
477
+ functions. We have shown the overview block diagram of the proposed ILP framework in figure 4.
478
+ Application
479
+ Profiling and one-time
480
+ characterization
481
+ Assembly
482
+ Code
483
+ Number of reads & writes to
484
+ each variable
485
+ Number of reads & writes to
486
+ each function
487
+ Energy per read/write to
488
+ SRAM
489
+ Energy per read/write to
490
+ FRAM
491
+ ILP Solver
492
+ Number of CPU cycles
493
+ required for eachfunctions
494
+ and variable
495
+ Number of Functions
496
+ Number of Global variables
497
+ SRAM and FRAM sizes
498
+ Mapping Information for each
499
+ Variable and Function
500
+ MSP430FR6989
501
+ Fig. 4. Overview of the Proposed ILP Framework
502
+
503
+ 8
504
+ S.J Badri, et al.
505
+ For Global Variables: Let the number of global variables in a program be ‘G’. Let the number
506
+ of reads and writes to variable ‘i’ are 𝑟𝑖 and 𝑤𝑖. We divided FRAM’s 128 KB into two regions, i.e.,
507
+ 𝐹𝑅𝐴𝑀𝑛 and 𝐹𝑅𝐴𝑀𝑏, 𝐹𝑅𝐴𝑀𝑛 memory region has 125 KB, and the 𝐹𝑅𝐴𝑀𝑏 memory region has 3 KB.
508
+ We have two memory regions represented as 𝑀𝑒𝑚𝑗 as shown in the equation 2; when j=1, we
509
+ select the memory region as SRAM, and we use 𝐹𝑅𝐴𝑀𝑛 for j=2.
510
+ 𝑀𝑒𝑚𝑗 =
511
+
512
+ 𝑗 = 1
513
+ ; SRAM
514
+ 𝑗 = 2
515
+ ; 𝐹𝑅𝐴𝑀𝑛
516
+ (2)
517
+ Let the sizes of SRAM/FRAM as 𝑆𝑖𝑧𝑒(𝑀𝑒𝑚𝑗) as shown in equation 3, when j=1, we refer as
518
+ SRAM memory size in bytes, and when j=2, we refer as 𝐹𝑅𝐴𝑀𝑛 memory size in bytes.
519
+ 𝑆𝑖𝑧𝑒(𝑀𝑒𝑚𝑗) =
520
+
521
+ 𝑗 = 1
522
+ ; SRAM
523
+ 𝑗 = 2
524
+ ; 𝐹𝑅𝐴𝑀𝑛
525
+ (3)
526
+ Let the energy required for each read/write to 𝑀𝑒𝑚𝑗 is 𝐸𝑟_𝑗 and 𝐸𝑤_𝑗. Let the number of CPU
527
+ cycles required to execute a global variable 𝑣𝑖 be 𝑁𝐶𝑣𝑖, where ∀𝑖 ∈ [1,𝐺]). Using one-time charac-
528
+ terization and static profiling, we gathered data such as per read/write energy to SRAM/FRAM and
529
+ the number of cycles.
530
+ We define a binary variable (BV); 𝐼𝑗 (𝑣𝑖), which refers to a variable 𝑣𝑖 is allocated to memory
531
+ region 𝑗. If 𝐼𝑗 (𝑣𝑖)=1 then the variable 𝑣𝑖 is allocated and 𝐼𝑗 (𝑣𝑖)=0 indicates that the variable 𝑣𝑖 is
532
+ not allocated. 𝐼𝑗 (𝑣𝑖), where (∀𝑗 ∈ [1, 𝑀𝑒𝑚𝑗], ∀𝑖 ∈ [1,𝐺]) is defined as shown in the equation 4.
533
+ 𝐼𝑗 (𝑣𝑖) =
534
+
535
+ 1
536
+ 𝑣𝑖 is allocated to memory region 𝑗
537
+ 0
538
+ otherwise
539
+ (4)
540
+ Constraints: There are two constraints, one is for BV; 𝐼𝑗 (𝑣𝑖) and one is a memory size constraint.
541
+ In any case, a variable 𝑣𝑖 is allocated to only one memory region, which means 𝑣𝑖 is allocated to
542
+ either SRAM or FRAM but not both. This constraint is defined in the equation 5.
543
+ 𝑀𝑒𝑚𝑗
544
+ ∑︁
545
+ 𝑗=1
546
+ 𝐼𝑗 (𝑣𝑖) = 1
547
+ (∀𝑖 ∈ [1,𝐺])
548
+ (5)
549
+ The other constraint is related to memory sizes. The allocated variables 𝑣𝑖 and its 𝑆𝑖𝑧𝑒(𝑣𝑖);
550
+ ∀𝑖 ∈ [1,𝐺]) should not be greater than the 𝑆𝑖𝑧𝑒(𝑀𝑒𝑚𝑗). This constraint is defined in the equation
551
+ 6.
552
+ 𝐺
553
+ ∑︁
554
+ 𝑖=1
555
+ 𝐼𝑗 (𝑣𝑖) ∗ 𝑆𝑖𝑧𝑒(𝑣𝑖) ≤ 𝑆𝑖𝑧𝑒(𝑀𝑒𝑚𝑗)
556
+ (∀𝑗 ∈ [1, 𝑀𝑒𝑚𝑗])
557
+ (6)
558
+ Objective 4.1: The challenge of mapping global variables in a program to either SRAM or FRAM
559
+ is to reduce EDP and improve system performance. 𝐸𝑔𝑙𝑜𝑏𝑎𝑙 is defined in the equation 7. Where
560
+ 𝐸𝑔𝑙𝑜𝑏𝑎𝑙 is the energy required to allocate global variables to either SRAM or FRAM.
561
+ 𝐸𝑔𝑙𝑜𝑏𝑎𝑙 =
562
+ 𝑀𝑒𝑚𝑗
563
+ ∑︁
564
+ 𝑗=1
565
+ 𝐺
566
+ ∑︁
567
+ 𝑖=1
568
+ [𝐸𝑟_𝑗 × 𝑟𝑖 + 𝐸𝑤_𝑗 × 𝑤𝑖]
569
+ (7)
570
+ 𝐸𝐷𝑃𝑔𝑙𝑜𝑏𝑎𝑙 is defined in the equation 8. Where 𝐸𝐷𝑃𝑔𝑙𝑜𝑏𝑎𝑙 is the energy-delay product required to
571
+ allocate global variables to either SRAM or FRAM.
572
+
573
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
574
+ 9
575
+ 𝐸𝐷𝑃𝑔𝑙𝑜𝑏𝑎𝑙 =
576
+ 𝑀𝑒𝑚𝑗
577
+ ∑︁
578
+ 𝑗=1
579
+ 𝐺
580
+ ∑︁
581
+ 𝑖=1
582
+ 𝐼𝑗 (𝑣𝑖) [𝐸𝑔𝑙𝑜𝑏𝑎𝑙 × 𝑁𝐶𝑣𝑖]
583
+ (8)
584
+ For Functions: Let the number of functions in a program be ‘𝑁 ′
585
+ 𝑓 . Let the number of reads and
586
+ writes to 𝑖𝑡ℎ function are 𝑟 (𝐹𝑖) and 𝑤(𝐹𝑖), where ∀𝑖 ∈ [1, 𝑁𝑓 ]. Functions consist of procedural
587
+ parameters, local variables, and return variables. Internally the code/data of functions are divided
588
+ into the text, data, and stack sections. We map at least one section among these three sections to
589
+ either SRAM or FRAM regions, i.e., 𝑀𝑒𝑚𝑗 and 𝑆𝑒𝑐𝑘 (𝑖) defines section ‘k’ of 𝑖𝑡ℎ function as shown
590
+ in the equation 9, when k=1, we refer to the text section of 𝑖𝑡ℎ function, when k=2, we refer to the
591
+ data section of 𝑖𝑡ℎ function, and when k=3, we refer to the stack section of 𝑖𝑡ℎ function.
592
+ 𝑆𝑒𝑐𝑘 (𝑖) =
593
+ 
594
+ 
595
+ 𝑘 = 1
596
+ ; Text
597
+ 𝑘 = 2
598
+ ; Data
599
+ 𝑘 = 3
600
+ ; Stack
601
+ ;∀𝑖 ∈ [1, 𝑁𝑓 ]
602
+ (9)
603
+ We define a BV; 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖)), which refers to a section 𝑆𝑒𝑐𝑘 of 𝑖𝑡ℎ function is allocated to only
604
+ one memory region 𝑗. If 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖))=1 then the section 𝑆𝑒𝑐𝑖 is allocated and 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖))=0 that
605
+ indicates the section 𝑆𝑒𝑐𝑖 is not allocated. 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖)), where (∀𝑗 ∈ [1, 𝑀𝑒𝑚𝑗], ∀𝑖 ∈ [1, 𝑁𝑓 ]),
606
+ ∀𝑘 ∈ [1,𝑆𝑒𝑐𝑘 (𝑖)]) is defined as shown in the equation 10.
607
+ 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖)) =
608
+
609
+ 1
610
+ 𝑆𝑒𝑐𝑘 of 𝑖𝑡ℎ function is allocated to 𝑗
611
+ 0
612
+ otherwise
613
+ (10)
614
+ Constraints: There are two constraints, one is for BV; 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖)) and one is a memory size
615
+ constraint. In any case, a 𝑆𝑒𝑐𝑘 of 𝑖𝑡ℎ function is allocated to only one memory region, which means
616
+ 𝑆𝑒𝑐𝑘 of 𝑖𝑡ℎ function is either allocated to either SRAM or FRAM but not both. This constraint is
617
+ defined in the equation 11.
618
+ 3
619
+ ∑︁
620
+ 𝑘=1
621
+ 𝑀𝑒𝑚𝑗
622
+ ∑︁
623
+ 𝑗=1
624
+ 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖))) = 1
625
+ (∀𝑖 ∈ [1, 𝑁𝑓 ])
626
+ (11)
627
+ The other constraint is related to memory sizes. The allocated sections 𝑆𝑒𝑐𝑘 (𝑖) and its 𝑆𝑖𝑧𝑒(𝐹𝑖);
628
+ ∀𝑘 ∈ [1,𝑆𝑒𝑐𝑘 (𝑖)]), ∀𝑗 ∈ [1, 𝑀𝑒𝑚𝑗], ∀𝑖 ∈ [1, 𝑁𝑓 ] should not be greater than the 𝑆𝑖𝑧𝑒(𝑀𝑒𝑚𝑗). This
629
+ constraint is defined in the equation 12.
630
+ 𝐺
631
+ ∑︁
632
+ 𝑖=1
633
+ 𝐼𝑗 (𝑣𝑖) ∗ 𝑆𝑖𝑧𝑒(𝑣𝑖) +
634
+ 3
635
+ ∑︁
636
+ 𝑘=1
637
+ 𝑁𝑓
638
+ ∑︁
639
+ 𝑖=1
640
+ 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖)) ∗ 𝑆𝑖𝑧𝑒(𝐹𝑖) ≤ 𝑆𝑖𝑧𝑒(𝑀𝑒𝑚𝑗)
641
+ (12)
642
+ Objective 4.2: The challenge of mapping sections of these functions in a program to either
643
+ SRAM or FRAM is to minimize EDP and improve system performance. 𝐸𝑓 𝑢𝑛𝑐 is defined in the
644
+ equation 13, where 𝑀𝑐𝑖 is the number of the times 𝑖𝑡ℎ functions called.
645
+ 𝐸𝑓 𝑢𝑛𝑐 =
646
+ 𝑀𝑒𝑚𝑗
647
+ ∑︁
648
+ 𝑗=1
649
+ 𝑁𝑓
650
+ ∑︁
651
+ 𝑖=1
652
+ [𝐸𝑟_𝑗 × 𝑟 (𝐹𝑖) + 𝐸𝑤_𝑗 × 𝑤(𝐹𝑖)] × 𝑀𝑐𝑖
653
+ (13)
654
+ 𝐸𝐷𝑃𝑓 𝑢𝑛𝑐 is defined in the equation 14. Where 𝐸𝐷𝑃𝑓 𝑢𝑛𝑐 is the energy-delay product required to
655
+ allocate all functions to either SRAM or FRAM. Where 𝐸𝑓 𝑢𝑛𝑐 is the energy required to allocate
656
+
657
+ 10
658
+ S.J Badri, et al.
659
+ functions to either SRAM or FRAM. Where 𝑁𝐶𝐹𝑖 is the number of CPU cycles required to execute
660
+ a function 𝐹𝑖.
661
+ 𝐸𝐷𝑃𝑓 𝑢𝑛𝑐 =
662
+ 3
663
+ ∑︁
664
+ 𝑘=1
665
+ 𝑀𝑒𝑚𝑗
666
+ ∑︁
667
+ 𝑗=1
668
+ 𝑁𝑓
669
+ ∑︁
670
+ 𝑖=1
671
+ 𝐼𝑗 (𝑆𝑒𝑐𝑘 (𝑖)) [𝐸𝑓 𝑢𝑛𝑐 × 𝑁𝐶𝐹𝑖]
672
+ (14)
673
+ The overall system EDP, 𝐸𝐷𝑃𝑠𝑦𝑠𝑡𝑒𝑚, is the sum of both 𝐸𝐷𝑃𝑔𝑙𝑜𝑏𝑎𝑙 and 𝐸𝐷𝑃𝑓 𝑢𝑛𝑐 as shown in the
674
+ equation 15.
675
+ 𝐸𝐷𝑃𝑠𝑦𝑠𝑡𝑒𝑚 = 𝐸𝐷𝑃𝑔𝑙𝑜𝑏𝑎𝑙 + 𝐸𝐷𝑃𝑓 𝑢𝑛𝑐
676
+ (15)
677
+ Our objective function is shown in the equation 16. Our main objective is to minimize the
678
+ system’s EDP by choosing the optimal placement choice.
679
+ Objective Function: Minimize 𝐸𝐷𝑃𝑠𝑦𝑠𝑡𝑒𝑚
680
+ (16)
681
+ 5.2
682
+ Implementing Mapping Technique in MSP430FR6989
683
+ Once we obtain the placement information from the 𝐼𝐿𝑃_𝑠𝑜𝑙𝑣𝑒𝑟, we map the respective variables
684
+ and the sections of a function to either SRAM or FRAM. We modify the linker script accordingly
685
+ for mapping the sections or variables to either SRAM or FRAM. In our proposed mapping policy,
686
+ placing global variables is straightforward, i.e., mapping the respective variable to either SRAM or
687
+ FRAM based on the ILP decision.
688
+ We observed that from the linker script, we can map the whole stack section of each function
689
+ to either SRAM or FRAM. We analyzed the mappings of the stack section for each function by
690
+ modifying the linker script. We used the inbuilt attributes to differentiate mappings between SRAM
691
+ and FRAM; for instance, we used the inbuilt attribute (__𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒__((𝑟𝑎𝑚𝑓𝑢𝑛𝑐)) that maps that
692
+ function to SRAM. If we want to place the stack section to SRAM, we modify the linker script by
693
+ replacing the default setting with " .stack: {} > RAM (HIGH) ". If we want to place the stack section
694
+ to FRAM, we modify the linker script by replacing the default setting with " .stack: {} > FRAM".
695
+ Similarly, for the text section, we observed that placing the text section in either SRAM or FRAM
696
+ shows an impact on EDP. This effect is because the majority of access in the text section are read
697
+ accesses, as we observed that the energy consumption for each read access to SRAM/FRAM differs.
698
+ Table 3 shows that approximately FRAM consumes 2x more read energy than SRAM. Thus, we
699
+ analyzed each application where to map the text section based on the free space available. If we
700
+ have enough space available in SRAM, we place the text section in SRAM itself; otherwise, we
701
+ place the text section in FRAM. We included the following four lines in our linker script to check
702
+ the above condition and map the text section.
703
+ (1) #𝑖𝑓 𝑛𝑑𝑒𝑓 __𝐿𝐴𝑅𝐺𝐸_𝐶𝑂𝐷𝐸_𝑀𝑂𝐷𝐸𝐿__
704
+ (2) .text : {} > FRAM
705
+ (3) #else
706
+ (4) .text : {} » SRAM
707
+ We modified the linker script for mapping the data section by using the inbuilt compiler directives.
708
+ We followed the below three steps.
709
+ (1) Allocate a new memory block, for instance, 𝑁𝐸𝑊 _𝐷𝐴𝑇𝐴𝑆𝐸𝐶𝑇𝐼𝑂𝑁. We can declare the start
710
+ address and size of the data section in the linker script.
711
+ (2) Define a segment (.Localvars) which stores in this memory block (𝑁𝐸𝑊 _𝐷𝐴𝑇𝐴𝑆𝐸𝐶𝑇𝐼𝑂𝑁).
712
+
713
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
714
+ 11
715
+ (3) Use #pragma 𝐷𝐴𝑇𝐴_𝑆𝐸𝐶𝑇𝐼𝑂𝑁 (𝑓𝑢𝑛𝑐𝑡_𝑛𝑎𝑚𝑒,𝑠𝑒𝑔_𝑛𝑎𝑚𝑒) in the program to define functions
716
+ in this segment. Where 𝑓𝑢𝑛𝑐𝑡_𝑛𝑎𝑚𝑒 is the function name and𝑠𝑒𝑔_𝑛𝑎𝑚𝑒 is the created segment
717
+ name. For instance, #pragma 𝐷𝐴𝑇𝐴_𝑆𝐸𝐶𝑇𝐼𝑂𝑁 (𝑓𝑢𝑛𝑐_1, .𝐿𝑜𝑐𝑎𝑙𝑣𝑎𝑟𝑠)
718
+ Once we are done with creating the different sections, we can allocate these sections to either
719
+ SRAM or FRAM based on ILP decisions. For instance, placing " 𝑁𝐸𝑊 _𝐷𝐴𝑇𝐴𝑆𝐸𝐶𝑇𝐼𝑂𝑁: {} > FRAM"
720
+ in the linker script, which maps the 𝑁𝐸𝑊 _𝐷𝐴𝑇𝐴𝑆𝐸𝐶𝑇𝐼𝑂𝑁 to FRAM.
721
+ 5.3
722
+ Support for Intermittent Computing
723
+ When the power is stable, everything works properly. Because of the static allocation scheme, we
724
+ map all functions/variables to SRAM/FRAM for the first time. During a power failure, SRAM and
725
+ registers lose all of their contents, including mapping information. When power is restored, we
726
+ don’t know what functions/variables were allocated to SRAM before the failure. As a result, we
727
+ must either restart the execution from the beginning or end up with incorrect results. Restarting
728
+ the application consumes extra energy and time, making our system inefficient in terms of energy
729
+ consumption and performance.
730
+ We propose a backup strategy during frequent power failures. FRAM was divided into 𝐹𝑅𝐴𝑀𝑛
731
+ and 𝐹𝑅𝐴𝑀𝑏 as shown in the figure 3. 𝐹𝑅𝐴𝑀𝑛 has a size of 125 KB and is used for regular mappings.
732
+ 𝐹𝑅𝐴𝑀𝑏 has a size of 3 KB that serves as a backup region (BR) during power failures. So, during a
733
+ power failure, we back up all register and SRAM contents to FRAM. Whenever power is restored,
734
+ we restore the register and SRAM contents from 𝐹𝑅𝐴𝑀𝑏 to SRAM and resume the application
735
+ execution. The proposed backup strategy reduces extra energy consumption and makes the system
736
+ more energy efficient.
737
+ 6
738
+ EXPERIMENTAL SETUP AND RESULTS
739
+ 6.1
740
+ Experimental Setup
741
+ We used TI’s MSP430FR6989 for all experiments. We experimented on mixed benchmarks, which
742
+ have both Mi-Bench [13] and TI-based benchmarks. We have shown the experimental setup in the
743
+ table 2. The development platform and experimental setup are shown in figure 5. We performed
744
+ experiments to determine the energy required for a single read/write to SRAM/FRAM, as shown in
745
+ the table 3. We collected the number of reads/writes for each global variable and functions as part
746
+ of a one-time characterization. We also used TI’s MSP430F5529 for comparing flash with FRAM.
747
+ We performed experiments to determine the energy required for a single read/write to flash, as
748
+ shown in the table 3.
749
+ Table 2. Experimental Setup
750
+ Component
751
+ Description
752
+ Target Board
753
+ TI MSP430FR6989 Launchpad
754
+ Core
755
+ MSP430 (1.8-3.6 V; 16 MHz)
756
+ Memory
757
+ 2KB SRAM and 128KB FRAM
758
+ IDE
759
+ Code Composer Studio
760
+ Energy Profiling
761
+ Energy Trace++
762
+ ILP Solver
763
+ LPSolve_IDE
764
+ Benchmarks
765
+ Mixed benchmarks (MiBench and TI-based)
766
+ MCU, which we experimented has MSP430 architecture, which is more suitable for IoT devices.
767
+ The majority of MSP430 software is written in C and compiled with one of TI’s recommended
768
+
769
+ 12
770
+ S.J Badri, et al.
771
+ compilers ( IAR Embedded Code Bench, Code-Composer Studio (CCS), or msp430-gcc). The IAR
772
+ Embedded Code Bench and CCS compilers are part of integrated development environments (IDEs).
773
+ We used the widely used, freely available, and easily extended tool, i.e., CCS, for all experiments
774
+ in this article. EnergyTrace++ technology allows us to calculate energy and power consumption
775
+ directly. According to the datasheet for the MSP430FR6989, the number of cycles required to
776
+ read/write in FRAM is twice that of SRAM.
777
+ Table 3. Energy Values for each read/write to SRAM and FRAM
778
+ Memory
779
+ Per Read Energy (nJ)
780
+ Per Write Energy (nJ)
781
+ SRAM
782
+ 5500
783
+ 5600
784
+ FRAM
785
+ 10325
786
+ 13125
787
+ Flash
788
+ 23876
789
+ 31198
790
+ Fig. 5. (a) TI-based MSP430 Launchpad Development Boards (b) Working with EnergyTrace++ on CCS
791
+ 6.2
792
+ Evaluation Benchmarks
793
+ We chose benchmarks from both the MiBench suite and TI benchmarks. One of the primary
794
+ motivations for using the MiBench suite is that most of the TI-based benchmarks were small in size
795
+ and easily fit into either SRAM or FRAM. In these cases, we don’t require any hybrid memory design.
796
+ Most of the TI-based benchmarks have only one or two functions and 3-4 global variables, which is
797
+ not useful for the hybrid main-memory design. Thus we used mixed benchmarks consisting of 4
798
+ TI-based benchmarks and 12 from the MiBench suite.
799
+ For the MiBench suite, we first make MCU-compatible benchmarks by adding MCU-related
800
+ header files and watchdog timers. All benchmarks may not be compatible with the MCU. Thus, we
801
+ need to choose the benchmarks from the MiBench suite, which are compatible with the MSP430
802
+ boards. Once we have benchmarks, we execute them on board for the machine code. Using the
803
+ .asm file, we calculate the inputs that are required by the ILP solver, as shown in figure 4.
804
+ 6.3
805
+ Baseline Configurations
806
+ We chose five different memory configurations to compare with the proposed memory configuration.
807
+ We directly map all the functions/variables to FRAM in the baseline configuration 1, as shown in
808
+ figure 1. We use configuration-1 to compare our proposed memory configuration during stable and
809
+ unstable power scenarios.
810
+
811
+ Code
812
+ Composer
813
+ Studio
814
+ MSP430F5529MSP430FR6989
815
+ EnergyTrace++Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
816
+ 13
817
+ We directly map all the functions/variables to SRAM in the baseline configuration 2, as shown in
818
+ figure 1. We use configuration-2 to compare our proposed memory configuration during stable and
819
+ unstable power scenarios.
820
+ In baseline configuration 3, we used the empirical method of Jayakumar et al. [20]. We compare
821
+ this configuration-3 with our proposed configuration during stable and unstable power scenarios
822
+ to observe the importance of the proposed than the existing work.
823
+ In baseline configuration 4, we used the proposed ILP technique for the flash-based msp430
824
+ board [24]. We compare this configuration-4 with our proposed configuration during stable and
825
+ unstable power scenarios to observe the difference between FRAM and Flash technologies.
826
+ In baseline configuration 5, we only have a proposed memory mapping technique and no BR.
827
+ We compare this configuration-5 with our proposed configuration during frequent power failures
828
+ to observe the importance of BR. The overview of all baseline configurations is shown in table 4.
829
+ The experimental setup for all baseline configurations is the same as the one proposed.
830
+ Table 4. Overview of the Baseline Configurations
831
+ Configuration
832
+ FRAM
833
+ SRAM
834
+ Flash
835
+ Backup Region (BR)
836
+ ILP
837
+ Baseline-1
838
+
839
+
840
+
841
+
842
+
843
+ Baseline-2
844
+
845
+
846
+
847
+
848
+
849
+ Baseline-3 ( Jayakumar et al. [20])
850
+
851
+
852
+
853
+
854
+
855
+ Baseline-4
856
+
857
+
858
+
859
+
860
+
861
+ Baseline-5
862
+
863
+
864
+
865
+
866
+
867
+ Proposed
868
+
869
+
870
+
871
+
872
+
873
+ ✓- Supported , ✗- Not Supported
874
+ 6.4
875
+ Results
876
+ The proposed memory configuration is evaluated in this section under stable and unstable power.
877
+ The proposed memory configuration is compared with five baseline memory configurations as
878
+ discussed in the section 6.3.
879
+ 6.4.1
880
+ Under Stable Power: Our main objective of the proposed memory configuration is to
881
+ minimize the system’s EDP. All values shown in figure 6 are normalized with baseline-1. Compared
882
+ to baseline-1, the proposed gets 38.10% lesser EDP, as shown in figure 6. Because there are no
883
+ power interruptions in this scenario, this improvement is totally from the proposed ILP model. In
884
+ configuration-1, we place everything to FRAM, where FRAM consumes more energy and the number
885
+ of cycles than SRAM, as shown in the table 3. Our proposed memory configuration incorporates
886
+ the placement recommendation from the proposed ILP model and suggests utilizing both SRAM
887
+ and FRAM.
888
+ Under a stable power scenario, the proposed gets 9.30% less EDP than baseline-3, as shown in
889
+ figure 6. We discussed the author’s empirical model and assumptions in the previous section 3. The
890
+ authors assumed that the data section included all global variables, constants, and arrays. As a
891
+ result, our proposed ILP-based mapping differs from the author’s mapping in that our proposed
892
+ mapping outperforms the existing work. Under stable power, baseline-3 receives 24.57% less EDP
893
+ than baseline-1, as shown in figure 6. This advantage is primarily due to baseline-3’s hybrid memory.
894
+ In comparison to baseline-4, the proposed reduces EDP by 18.55%, as shown in figure 6. We
895
+ used flash+SRAM with our proposed ILP framework in baseline-4. As shown in table 3, the above
896
+ benefit is primarily due to FRAM because flash consumes more energy. Baseline-3 outperforms
897
+
898
+ 14
899
+ S.J Badri, et al.
900
+ 0
901
+ 0.2
902
+ 0.4
903
+ 0.6
904
+ 0.8
905
+ 1
906
+ 16bit_2dim
907
+ aes
908
+ basicmath_small
909
+ basicmath_large
910
+ bf
911
+ crc
912
+ dhrystone
913
+ dijkstra
914
+ fft
915
+ fir
916
+ matrix_mult
917
+ patricia
918
+ qsort_small
919
+ qsort_large
920
+ sha
921
+ susan
922
+ Normalized EDP (Normalized with
923
+ Baseline-1)
924
+ Benchmarks
925
+ Baseline-2
926
+ Jayakumar et al. [20]
927
+ Baseline-4
928
+ Proposed
929
+ Fig. 6. Comparison between Baseline configurations and the Proposed under Stable Power
930
+ baseline-4 during stable power. Because of FRAM in baseline-3, even our proposed ILP model is
931
+ ineffective in this case. We encountered that baseline-3 achieves 9.19% less EDP than baseline-4, and
932
+ this benefit is because of smaller applications. From figure 6, baseline-4 performs better for large
933
+ applications than baseline-4. Jayakumar et al. [20] empirical method suggests placing more content
934
+ on SRAM because SRAM is sufficient for placing the entire small-size application. As a result, the
935
+ performance of baseline 3 is dependent on the application size, as for large-size applications, even
936
+ FRAM does not outperform flash.
937
+ Baseline 2 outperforms the proposed and all other baselines under stable power conditions.
938
+ We noticed that this benefit is primarily due to SRAM, but it only applies to smaller applications.
939
+ Baseline 2 achieves 36.19% less EDP than the proposed for smaller applications, as shown in figure
940
+ 6. We also looked at large applications where the proposed outperforms the baseline-2 by a small
941
+ margin. When the SRAM is full, the MCU must wait for the space to be released, which consumes
942
+ extra energy and cycles. For more extensive applications, baseline-2 achieves 2.94% more EDP than
943
+ proposed.
944
+ We also evaluated our proposed framework with another MSP430F5529 MCU with flash and
945
+ SRAM for completeness. This comparison assists the user in selecting the most appropriate NVM
946
+ technology, such as FRAM or flash, as needed. To be fair, we used the same sizes of SRAM (2 KB)
947
+ and Flash (128 KB) in this comparison. We compared FRAM-based and flash-based MCUs under
948
+ stable power conditions. We used the proposed frameworks and techniques in both MCUs. We
949
+ discovered that the proposed FRAM-based configuration outperforms the flash-based configuration.
950
+ Flash-based configurations consume 26.03% more EDP than FRAM-based configurations, as shown
951
+ in figure 7. Flash consumes more energy, as shown in table 3.
952
+ 6.4.2
953
+ Under Unstable power: We used the default TI-based compute through power loss (ctpl)
954
+ tool for migration. During a power failure, we need to migrate the SRAM contents to a FRAM-based
955
+ backup region (𝐹𝑅𝐴𝑀𝑏), i.e., the backup process. Whenever power comes back, we need to migrate
956
+ the (𝐹𝑅𝐴𝑀𝑏) contents to SRAM, i.e., the restoration process. So, all these migrations are done using
957
+ ctpl() functions. We introduce a power failure by changing the low power modes mentioned in
958
+ the MSP430FR6989 design document. We used ctpl() for creating power failures. We assume that
959
+ the number of power failures is spread equally within the execution period. For instance, if the
960
+
961
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
962
+ 15
963
+ 0
964
+ 0.2
965
+ 0.4
966
+ 0.6
967
+ 0.8
968
+ 1
969
+ 16bit_2dim
970
+ aes
971
+ basicmath_small
972
+ basicmath_large
973
+ bf
974
+ crc
975
+ dhrystone
976
+ dijkstra
977
+ fft
978
+ fir
979
+ matrix_mult
980
+ patricia
981
+ qsort_small
982
+ qsort_large
983
+ sha
984
+ susan
985
+ Normalized EDP (Normalized with
986
+ MSP430F5529)
987
+ Benchmarks
988
+ MSP430F5529
989
+ MSP430FR6989
990
+ Fig. 7. Comparison between MSP430FR6989 (FRAM-based MCU) and MSP430F5529 (Flash-based MCU)
991
+ under Stable Power
992
+ total execution period for an application is 20 milliseconds (ms), and let’s say the number of power
993
+ failures is four, then for every 5 ms, we experience a power failure.
994
+ We performed experiments under unstable power to compare the proposed memory configuration
995
+ with baseline configurations. All values shown in figure 8 are normalized with baseline-2. Compared
996
+ to baseline-2, the proposed gets 15.97% lesser EDP, as shown in figure 8. We observed that migration
997
+ overhead is less than the energy consumed to execute the application from FRAM, and this migration
998
+ overhead depends on the number of power failures. For instance, one backup migration consumes
999
+ approximately 16.88 mJ of energy, and one restore migration consumes approximately 11.606 mJ of
1000
+ energy in a qsort application. The above benefit to our proposed configuration is using a hybrid
1001
+ memory.
1002
+ Under an unstable power scenario, the proposed gets 21.99% less EDP than baseline-3, as shown
1003
+ in figure 8. We discussed the author’s empirical model and assumptions in the previous section 3.
1004
+ As already stated, the Jayakumar et al. empirical method is more beneficial for small applications.
1005
+ In contrast, the author’s empirical method suggests placing more content on SRAM because SRAM
1006
+ is sufficient for placing the entire small-size application. Thus, for [20] work, backup/restore
1007
+ operations take more energy during a power failure. Our proposed mapping outperforms the
1008
+ existing work. During frequent power failures, baseline-3 receives 6.91% less EDP than baseline-1,
1009
+ as shown in figure 8. This advantage is primarily due to baseline-3’s hybrid memory.
1010
+ Compared to baseline-4, the proposed reduces EDP by 23.05%, as shown in figure 8. We used
1011
+ flash+SRAM with our proposed ILP framework in baseline-4. As shown in table 3, the above benefit
1012
+ is primarily due to FRAM because flash consumes more energy. Baseline-3 outperforms baseline-4
1013
+ during stable power. Because of FRAM in baseline-3, even our proposed ILP model is ineffective
1014
+ for this comparison. We encountered that baseline-3 achieves 6.28% less EDP than baseline-4 for
1015
+ smaller applications. The above benefit for baseline-3 is minimal because the size of backup/restores
1016
+ increases, which even neutralizes the flash for some applications, as shown in figure 8. Baseline-4
1017
+ achieves 2.69% less EDP than baseline-3 for large applications, as shown in figure 8. As a result, the
1018
+
1019
+ 16
1020
+ S.J Badri, et al.
1021
+ 0
1022
+ 0.2
1023
+ 0.4
1024
+ 0.6
1025
+ 0.8
1026
+ 1
1027
+ 16bit_2dim
1028
+ aes
1029
+ basicmath_small
1030
+ basicmath_large
1031
+ bf
1032
+ crc
1033
+ dhrystone
1034
+ dijkstra
1035
+ fft
1036
+ fir
1037
+ matrix_mult
1038
+ patricia
1039
+ qsort_small
1040
+ qsort_large
1041
+ sha
1042
+ susan
1043
+ Normalized EDP (Normalized with
1044
+ Baseline-2)
1045
+ Benchmarks
1046
+ Baseline-1
1047
+ Jayakumar et al. [20]
1048
+ Baseline-4
1049
+ Baseline-5
1050
+ Proposed
1051
+ Fig. 8. Comparison between Baseline configurations and the Proposed under Unstable Power
1052
+ performance of baseline 3 is dependent on the application size, as for large-size applications, even
1053
+ FRAM does not outperform flash.
1054
+ The proposed outperforms all baselines under unstable power conditions. This benefit is primarily
1055
+ due to a hybrid memory and the proposed mapping technique. Baseline 2 achieves 42.98% less EDP
1056
+ than the proposed, as shown in figure 8.
1057
+ When we remove BR, all the mapping information of SRAM is lost because our model is static.
1058
+ We introduce a BR in the FRAM memory region to save this mapping information. During a power
1059
+ failure, we migrate the SRAM contents to 𝐹𝑅𝐴𝑀𝑏, and whenever power comes back, we restore
1060
+ the 𝐹𝑅𝐴𝑀𝑏 contents to the SRAM.
1061
+ We experimented to know the importance of BR, where we compared the proposed memory
1062
+ configuration with baseline-5. Compared to baseline-5, the proposed gets 23.94% lesser EDP, as
1063
+ shown in figure 8. This benefit is because we need to re-execute the application four times from
1064
+ the beginning, which consumes extra time and energy. The number of times re-executing the
1065
+ application is equal to the number of power failures.
1066
+ We also evaluated our proposed framework with another MSP430F5529 MCU, which consists of
1067
+ flash and SRAM for completeness. This comparison assists the user in selecting the most appropriate
1068
+ NVM technology, such as FRAM or flash, as needed. To be fair, we used the same sizes of SRAM (2
1069
+ KB) and Flash (128 KB) in this comparison. We also used BR for both baselines; the only difference
1070
+ is that we replaced FRAM with the flash in the proposed configurations, and everything is the
1071
+ same. We compared FRAM-based and flash-based MCUs under unstable power conditions. We
1072
+ used the proposed frameworks and techniques in both MCUs. We discovered that the proposed
1073
+ FRAM-based configuration outperforms the flash-based configuration. Flash-based configurations
1074
+ consume 16.50% more EDP than FRAM-based configurations, as shown in figure 9. Flash consumes
1075
+ more energy, as shown in table 3.
1076
+ 6.5
1077
+ Summary of the Proposed Mapping Technique
1078
+ We outline the proposed ILP-based memory mapping technique in this section. Following all
1079
+ of these analyses, we observed that the mappings shown below consume less EDP than other
1080
+ design choices, as shown in the table. To keep things simple, we only showed the final mapping
1081
+
1082
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
1083
+ 17
1084
+ 0
1085
+ 0.2
1086
+ 0.4
1087
+ 0.6
1088
+ 0.8
1089
+ 1
1090
+ 16bit_2dim
1091
+ aes
1092
+ basicmath_small
1093
+ basicmath_large
1094
+ bf
1095
+ crc
1096
+ dhrystone
1097
+ dijkstra
1098
+ fft
1099
+ fir
1100
+ matrix_mult
1101
+ patricia
1102
+ qsort_small
1103
+ qsort_large
1104
+ sha
1105
+ susan
1106
+ Normalized EDP (Normalized with
1107
+ MSP430F5529)
1108
+ Benchmarks
1109
+ MSP430F5529
1110
+ MSP430FR6989
1111
+ Fig. 9. Comparison between MSP430FR6989 (FRAM-based MCU) and MSP430F5529 (Flash-based MCU)
1112
+ under Unstable Power
1113
+ configurations for each application’s stack, data, and text sections, keeping out the final mappings
1114
+ for global variables.
1115
+ Table 5. Optimal Placement for different Applications in MSP430FR6989
1116
+ Benchmarks
1117
+ Stack
1118
+ Text
1119
+ Data
1120
+ 16bit_2dim
1121
+ SRAM
1122
+ SRAM
1123
+ SRAM
1124
+ aes
1125
+ SRAM
1126
+ FRAM
1127
+ FRAM
1128
+ basicmath_small
1129
+ SRAM
1130
+ SRAM
1131
+ FRAM
1132
+ basicmath_large
1133
+ SRAM
1134
+ FRAM
1135
+ FRAM
1136
+ bf
1137
+ SRAM
1138
+ SRAM
1139
+ FRAM
1140
+ crc
1141
+ SRAM
1142
+ FRAM
1143
+ SRAM
1144
+ dhrystone
1145
+ FRAM
1146
+ SRAM
1147
+ FRAM
1148
+ dijkstra
1149
+ SRAM
1150
+ FRAM
1151
+ SRAM
1152
+ fft
1153
+ SRAM
1154
+ SRAM
1155
+ FRAM
1156
+ fir
1157
+ SRAM
1158
+ SRAM
1159
+ FRAM
1160
+ matrix_mult
1161
+ SRAM
1162
+ SRAM
1163
+ SRAM
1164
+ patricia
1165
+ SRAM
1166
+ FRAM
1167
+ SRAM
1168
+ qsort_small
1169
+ SRAM
1170
+ SRAM
1171
+ FRAM
1172
+ qsort_large
1173
+ SRAM
1174
+ FRAM
1175
+ FRAM
1176
+ sha
1177
+ SRAM
1178
+ FRAM
1179
+ FRAM
1180
+ susan
1181
+ SRAM
1182
+ FRAM
1183
+ FRAM
1184
+ Table 5 shows that, with the exception of the dhrystone application, the remaining three TI
1185
+ benchmark applications (fir, matrix, and 16bit_2dim) are very small and can easily be placed in SRAM.
1186
+ We don’t need FRAM for these types of smaller applications, but there is a disadvantage during
1187
+ frequent power failures. Backup and restore sizes to FRAM are larger for these applications during
1188
+ frequent power failures. As a result, our proposed backup/restore strategy should be intelligent
1189
+
1190
+ 18
1191
+ S.J Badri, et al.
1192
+ enough to reduce EDP. The dhrystone application, on the other hand, has a larger stack section
1193
+ that requires FRAM to accommodate the entire stack section.
1194
+ As we can see from the table 5, many applications used both SRAM and FRAM for the Mi-
1195
+ Bench applications. As a result, we can conclude that a hybrid main memory design is required
1196
+ for many applications. Using a hybrid main memory design helps to reduce EDP during stable
1197
+ power scenarios. Even so, determining how and where to backup the volatile contents can be
1198
+ difficult during frequent power outages. However, our proposed memory mapping technique and
1199
+ the framework suggest using a hybrid main memory design that supports intermittent computing.
1200
+ 7
1201
+ CONCLUSIONS
1202
+ This paper proposed an ILP-based memory mapping technique that reduces the system’s energy-
1203
+ delay product. For both global variables and functions, we formulated an ILP model. Functions
1204
+ consist of data, stack, and code sections. Our ILP model suggests placing each section on either SRAM
1205
+ or FRAM. Under both stable and unstable power scenarios, we compared the proposed memory
1206
+ configuration to the baseline memory configurations. We evaluated our proposed frameworks and
1207
+ techniques on actual boards. We added a backup region in FRAM to support intermittent computing.
1208
+ We compared the proposed framework with the recent related work.
1209
+ Under stable power, our proposed memory configuration consumes 38.10% less EDP than baseline-
1210
+ 1 and 9.30% less EDP than the existing work. Under unstable power, our proposed configuration
1211
+ achieves 15.97% less EDP than baseline-1 and 21.99% less EDP than the existing work. Under stable
1212
+ power, our proposed memory configuration consumes 18.55% less EDP than baseline-4. We also
1213
+ compared FRAM-based MSP430FR6989 with flash-based MSP430F5529. Compared to the flash,
1214
+ the FRAM-based hybrid main memory design consumes less EDP. FRAM-based design consumes
1215
+ 26.03% less EDP than flash-based design during stable power and 16.50% less EDP than flash based
1216
+ during frequent power failures.
1217
+ REFERENCES
1218
+ [1] Prasenjit Chakraborty, Preeti Ranjan Panda, and Sandeep Sen. 2016. Partitioning and Data Mapping in Reconfigurable
1219
+ Cache and Scratchpad Memory–Based Architectures. ACM Transactions on Design Automation of Electronic Systems
1220
+ (TODAES) 22, 1 (2016), 1–25.
1221
+ [2] Yiran Chen, Weng-Fai Wong, Hai Li, and Cheng-Kok Koh. 2011. Processor caches built using multi-level spin-transfer
1222
+ torque ram cells. In IEEE/ACM International Symposium on Low Power Electronics and Design. IEEE, 73–78.
1223
+ [3] Yu-Ting Chen, Jason Cong, Hui Huang, Bin Liu, Chunyue Liu, Miodrag Potkonjak, and Glenn Reinman. 2012. Dynami-
1224
+ cally reconfigurable hybrid cache: An energy-efficient last-level cache design. In 2012 Design, Automation & Test in
1225
+ Europe Conference & Exhibition (DATE). IEEE, 45–50.
1226
+ [4] Yu-Der Chih, Yi-Chun Shih, Chia-Fu Lee, Yen-An Chang, Po-Hao Lee, Hon-Jarn Lin, Yu-Lin Chen, Chieh-Pu Lo,
1227
+ Meng-Chun Shih, Kuei-Hung Shen, et al. 2020. 13.3 a 22nm 32Mb embedded STT-MRAM with 10ns read speed, 1M
1228
+ cycle write endurance, 10 years retention at 150 c and high immunity to magnetic field interference. In 2020 IEEE
1229
+ International Solid-State Circuits Conference-(ISSCC). IEEE, 222–224.
1230
+ [5] Ju-Hee Choi and Gi-Ho Park. 2017. NVM way allocation scheme to reduce NVM writes for hybrid cache architecture
1231
+ in chip-multiprocessors. IEEE Transactions on Parallel and Distributed Systems 28, 10 (2017), 2896–2910.
1232
+ [6] Tim Daulby, Anand Savanth, Alex S Weddell, and Geoff V Merrett. 2020. Comparing NVM technologies through
1233
+ the lens of Intermittent computation. In Proceedings of the 8th International Workshop on Energy Harvesting and
1234
+ Energy-Neutral Sensing Systems. 77–78.
1235
+ [7] Jean-Francois Deverge and Isabelle Puaut. 2007. WCET-directed dynamic scratchpad memory allocation of data. In
1236
+ 19th Euromicro Conference on Real-Time Systems (ECRTS’07). IEEE, 179–190.
1237
+ [8] Bernhard Egger, Jaejin Lee, and Heonshik Shin. 2008. Dynamic scratchpad memory management for code in portable
1238
+ systems with an MMU. ACM Transactions on Embedded Computing Systems (TECS) 7, 2 (2008), 1–38.
1239
+ [9] Hêriş Golpîra, Syed Abdul Rehman Khan, and Sina Safaeipour. 2021. A review of logistics internet-of-things: Current
1240
+ trends and scope for future research. Journal of Industrial Information Integration (2021), 100194.
1241
+ [10] Ricardo Gonzalez and Mark Horowitz. 1996. Energy dissipation in general purpose microprocessors. IEEE Journal of
1242
+ solid-state circuits 31, 9 (1996), 1277–1284.
1243
+
1244
+ Mapi-Pro: An Energy Efficient Memory Mapping Technique for Intermittent Computing
1245
+ 19
1246
+ [11] Laura M Grupp, Adrian M Caulfield, Joel Coburn, Steven Swanson, Eitan Yaakobi, Paul H Siegel, and Jack K Wolf. 2009.
1247
+ Characterizing flash memory: Anomalies, observations, and applications. In Proceedings of the 42nd Annual IEEE/ACM
1248
+ International Symposium on Microarchitecture. 24–33.
1249
+ [12] Yibo Guo, Qingfeng Zhuge, Jingtong Hu, Juan Yi, Meikang Qiu, and Edwin H-M Sha. 2013. Data placement and
1250
+ duplication for embedded multicore systems with scratch pad memory. IEEE Transactions on Computer-Aided Design of
1251
+ Integrated Circuits and Systems 32, 6 (2013), 809–817.
1252
+ [13] Matthew R Guthaus et al. 2001. MiBench: A free, commercially representative embedded benchmark suite. In Proceedings
1253
+ of the fourth annual IEEE international workshop on workload characterization. IEEE, IEEE, 3–14.
1254
+ [14] Josiah Hester and Jacob Sorber. 2017. The future of sensing is batteryless, intermittent, and awesome. In Proceedings of
1255
+ the 15th ACM conference on embedded network sensor systems. 1–6.
1256
+ [15] Xiaosong Hu, Le Xu, Xianke Lin, and Michael Pecht. 2020. Battery lifetime prognostics. Joule 4, 2 (2020), 310–346.
1257
+ [16] Texas Instruments. 2018. MSP430FR5969 launchpad development kit.
1258
+ [17] Pulkit Jain, Umut Arslan, Meenakshi Sekhar, Blake C Lin, Liqiong Wei, Tanaya Sahu, Juan Alzate-Vinasco, Ajay
1259
+ Vangapaty, Mesut Meterelliyoz, Nathan Strutt, et al. 2019. 13.2 A 3.6 Mb 10.1 Mb/mm 2 embedded non-volatile ReRAM
1260
+ macro in 22nm FinFET technology with adaptive forming/set/reset schemes yielding down to 0.5 V with sensing time
1261
+ of 5ns at 0.7 V. In 2019 IEEE International Solid-State Circuits Conference-(ISSCC). IEEE, 212–214.
1262
+ [18] Hrishikesh Jayakumar, Arnab Raha, and Vijay Raghunathan. 2014. QuickRecall: A low overhead HW/SW approach for
1263
+ enabling computations across power cycles in transiently powered computers. In 2014 27th International Conference on
1264
+ VLSI Design and 2014 13th International Conference on Embedded Systems. IEEE, 330–335.
1265
+ [19] Hrishikesh Jayakumar, Arnab Raha, and Vijay Raghunathan. 2016. Energy-aware memory mapping for hybrid FRAM-
1266
+ SRAM MCUs in IoT edge devices. In 2016 29th International Conference on VLSI Design and 2016 15th International
1267
+ Conference on Embedded Systems (VLSID). IEEE, 264–269.
1268
+ [20] Hrishikesh Jayakumar, Arnab Raha, Jacob R Stevens, and Vijay Raghunathan. 2017. Energy-aware memory mapping
1269
+ for hybrid FRAM-SRAM MCUs in intermittently-powered IoT devices. ACM Transactions on Embedded Computing
1270
+ Systems (TECS) 16, 3 (2017), 1–23.
1271
+ [21] Morteza Mohajjel Kafshdooz and Alireza Ejlali. 2015. Dynamic shared SPM reuse for real-time multicore embedded
1272
+ systems. ACM Transactions on Architecture and Code Optimization (TACO) 12, 2 (2015), 1–25.
1273
+ [22] Mahmut Kandemir, J Ramanujam, Mary Jane Irwin, Narayanan Vijaykrishnan, Ismail Kadayif, and Amisha Parikh.
1274
+ 2001. Dynamic management of scratch-pad memory space. In Proceedings of the 38th Design Automation Conference
1275
+ (IEEE Cat. No. 01CH37232). IEEE, 690–695.
1276
+ [23] Mirae Kim, Jungkeol Lee, Youngil Kim, and Yong Ho Song. 2018. An analysis of energy consumption under various
1277
+ memory mappings for FRAM-based IoT devices. In 2018 IEEE 4th World Forum on Internet of Things (WF-IoT). IEEE,
1278
+ 574–579.
1279
+ [24] Markus Koesler and Franz Graf. 2002. Programming a flash-based MSP430 using the JTAG Interface. SLAA149, TEXAS
1280
+ INSTRUMENTS (2002), 4–13.
1281
+ [25] Benjamin C Lee, Engin Ipek, Onur Mutlu, and Doug Burger. 2009. Architecting phase change memory as a scalable
1282
+ dram alternative. In Proceedings of the 36th annual international symposium on Computer architecture. 2–13.
1283
+ [26] Lian Li, Hui Feng, and Jingling Xue. 2009. Compiler-directed scratchpad memory management via graph coloring.
1284
+ ACM Transactions on Architecture and Code Optimization (TACO) 6, 3 (2009), 1–17.
1285
+ [27] Brandon Lucia, Vignesh Balaji, Alexei Colin, Kiwan Maeng, and Emily Ruppel. 2017. Intermittent computing: Challenges
1286
+ and opportunities. 2nd Summit on Advances in Programming Languages (SNAPL 2017) (2017).
1287
+ [28] Sheel Sindhu Manohar and Hemangee K Kapoor. 2022. CAPMIG: Coherence Aware block Placement and MIGration in
1288
+ Multi-retention STT-RAM Caches. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
1289
+ (2022).
1290
+ [29] Sparsh Mittal and Jeffrey S Vetter. 2015. A survey of software techniques for using non-volatile memories for storage
1291
+ and main memory systems. IEEE Transactions on Parallel and Distributed Systems 27, 5 (2015), 1537–1550.
1292
+ [30] Keni Qiu, Mengying Zhao, Zhenge Jia, Jingtong Hu, Chun Jason Xue, Kaisheng Ma, Xueqing Li, Yongpan Liu, and
1293
+ Vijaykrishnan Narayanan. 2020. Design insights of non-volatile processors and accelerators in energy harvesting
1294
+ systems. In Proceedings of the 2020 on Great Lakes Symposium on VLSI. 369–374.
1295
+ [31] Arindam Sarkar, Newton Singh, Varun Venkitaraman, and Virendra Singh. 2021. DAM: Deadblock Aware Migration
1296
+ Techniques for STT-RAM-Based Hybrid Caches. IEEE Computer Architecture Letters 20, 1 (2021), 62–4.
1297
+ [32] Sophiane Senni, Lionel Torres, Gilles Sassatelli, and Abdoulaye Gamatie. 2016. Non-volatile processor based on MRAM
1298
+ for ultra-low-power IoT devices. ACM Journal on Emerging Technologies in Computing Systems (JETC) 13, 2 (2016),
1299
+ 1–23.
1300
+ [33] Vivy Suhendra, Chandrashekar Raghavan, and Tulika Mitra. 2006. Integrated scratchpad memory optimization and
1301
+ task scheduling for MPSoC architectures. In Proceedings of the 2006 international conference on Compilers, architecture
1302
+ and synthesis for embedded systems. 401–410.
1303
+
1304
+ 20
1305
+ S.J Badri, et al.
1306
+ [34] Milijana Surbatovich, Brandon Lucia, and Limin Jia. 2020. Towards a formal foundation of intermittent computing.
1307
+ Proceedings of the ACM on Programming Languages 4, OOPSLA (2020), 1–31.
1308
+ [35] Sumesh Udayakumaran, Angel Dominguez, and Rajeev Barua. 2006. Dynamic allocation for scratch-pad memory
1309
+ using compile-time decisions. ACM Transactions on Embedded Computing Systems (TECS) 5, 2 (2006), 472–511.
1310
+ [36] Manish Verma, Lars Wehmeyer, and Peter Marwedel. 2004. Dynamic overlay of scratchpad memory for energy
1311
+ minimization. In Proceedings of the 2nd IEEE/ACM/IFIP international conference on Hardware/software codesign and
1312
+ system synthesis. 104–109.
1313
+ [37] Mimi Xie, Chen Pan, and Chun Jason Xue. 2018. A Novel STT-RAM-Based Hybrid Cache for Intermittently Powered
1314
+ Processors in IoT Devices. IEEE Micro 39, 1 (2018), 24–32.
1315
+
79FLT4oBgHgl3EQfAy4h/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
9NE4T4oBgHgl3EQfdgy1/content/tmp_files/2301.05092v1.pdf.txt ADDED
@@ -0,0 +1,3087 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.05092v1 [gr-qc] 12 Jan 2023
2
+ ,
3
+ Slowly rotating Kerr metric derived from the Einstein equations in affine-null
4
+ coordinates
5
+ Thomas M¨adler
6
+
7
+ Escuela de Obras Civiles and Instituto de Estudios Astrof´ısicos,
8
+ Facultad de Ingenier´ıa y Ciencias, Universidad Diego Portales,
9
+ Avenida Ej´ercito Libertador 441, Casilla 298-V, Santiago, Chile.
10
+ Emanuel Gallo
11
+
12
+ FaMAF, UNC; Instituto de F´ısica Enrique Gaviola (IFEG), CONICET,
13
+ Ciudad Universitaria, (5000) C´ordoba, Argentina.
14
+ Using a quasi-spherical approximation of an affine-null metric adapted to an asymptotic Bondi
15
+ inertial frame, we present high order approximations of the metric functions in terms of the specific
16
+ angular momentum for a slowly rotating stationary and axi-symmetric vacuum spacetime.
17
+ The
18
+ metric is obtained by following the procedure of integrating the hierarchy of Einstein equations in a
19
+ characteristic formulation utilizing master functions for the perturbations. It is further verified its
20
+ equivalence with the Kerr metric in the slowly rotation approximation by carrying out an explicit
21
+ transformation between the Boyer-Lindquist coordinates to the employed affine-null coordinates.
22
+ PACS numbers:
23
+ I.
24
+ INTRODUCTION
25
+ At the dawn of the ’Golden Era of General Relativity’
26
+ in the 60ties of the last century, two important space-
27
+ time metrics were found, the Bondi-Sachs metric [1–3]
28
+ and the Kerr metric [4, 5]. The first settled the question
29
+ that an isolated system looses mass via gravitational radi-
30
+ ation and that this effect is a non-linear effect of General
31
+ Relativity; while the second describes a stationary and
32
+ rotating isolated black hole that is expected to be the
33
+ end product of a gravitational collapse of a massive star
34
+ or a merger of two compact objects.
35
+ One of the defining features of the Bondi-Sachs metric
36
+ is that one coordinate is constant along a family of null
37
+ hypersurfaces while a radial coordinate along these null
38
+ hypersurfaces is an areal distance that can be related to a
39
+ luminosity distance [6]. As such, the first long term sta-
40
+ ble evolution of black hole space times were made using a
41
+ Bondi-Sachs metric in a null cone-world tube formalism
42
+ [7], also see [8, 9] for review. Apart from usage in numer-
43
+ ical relativity simulations, the Bondi-Sachs metric is now
44
+ frequently used in high energy physics addressing ques-
45
+ tions of the AdS/CFT correspondence [10] (and citations
46
+ thereof). It also became popular to discuss gravitational
47
+ wave memory effects [11–15]. A pleasant property of the
48
+ Bondi-Sachs formalism is that the Einstein equation can
49
+ be solved in a hierarchical manner when initial data on
50
+ a null hypersurface and boundary conditions at a world
51
+ tube or vertex are given. However, the radial coordinate
52
+ of the Bondi-Sachs metric has the unpleasant property
53
+ ∗Electronic address: thomas.maedler˙.at.˙mail.udp.cl
54
+ †Electronic address: egallo˙.at.˙unc.edu.ar
55
+ that it breaks down when an apparent horizon forms due
56
+ to the focusing of the surface-forming null rays and their
57
+ vanishing expansion.
58
+ This can be overcome in choos-
59
+ ing an affine parameter as radial coordinate, because an
60
+ affine parameter only becomes singular at a caustic. But,
61
+ the Einstein equations resulting from an affine-null met-
62
+ ric do not provide the hierarchical structure as the Bondi
63
+ Sachs metric [9] and the hierarchical structure needs to
64
+ be reestablished by various new definitions of variables
65
+ [16–18]. Moreover, it turns out that also the hierarchy
66
+ of equations in the affine-null metric formulation breaks
67
+ down in the events of apparent horizon formation, but
68
+ fortunately the equations can be regularized so that it is
69
+ possible to follow up the formation of black holes up to
70
+ singularity [19, 20].
71
+ Despite the success and popularity of the Bondi-Sachs
72
+ metric in the various areas, an explicit closed analytical
73
+ representation of the Kerr metric in Bondi-Sachs form
74
+ without bad behaviour in the exterior region or related
75
+ metrics with one or two null coordinates is missing. Var-
76
+ ious attempts have been made to derive a null metric
77
+ representation, numerically [21, 22] as well as analyti-
78
+ cally [23–25]. In all of the approaches, the authors start
79
+ out with the Kerr metric and then calculate the respec-
80
+ tive null metric via a coordinate transformation. After
81
+ these transformations the resulting metric can still posses
82
+ a conical singularity at the axis of symmetry (see [22]
83
+ for a complete discussion). In addition, the final met-
84
+ ric is determined by integrals of non-elementary func-
85
+ tions. Arga˜naraz and Moreschi’s approach [22] differs to
86
+ the aforementioned ones that the authors aim to find a
87
+ double–null representation of the Kerr metric by geomet-
88
+ rically adopting the coordinates to in- and outgoing null
89
+ geodesics adapted to the center of mass [27]. In this way,
90
+ the authors were successful in finding null coordinates
91
+
92
+ 2
93
+ that are not only regular at every point of the external
94
+ communication region (unlike the previous formulations)
95
+ but also that they are regular at the event horizon, thus
96
+ allowing a way to study the evolution of different matter
97
+ fields (as scalar fields) in such background even when they
98
+ cross the event horizon[26]. Unfortunately, even in their
99
+ construction arises a differential equation that needs to
100
+ be solved numerically and an explicit closed form repre-
101
+ sentation of the double null version of the Kerr metric is
102
+ not possible. The work of Bai and collaborators [28, 29]
103
+ also starts with the Kerr metric (in Boyer-Lindquist co-
104
+ ordinates) and then makes coordinate transformation to
105
+ a Bondi-Sachs metric valid near future null infinity (in a
106
+ compactified version of the metric). The authors are able
107
+ to calculate the Newman-Penrose quantities and multi-
108
+ poles at large distances and show the peeling property of
109
+ the Weyl tensor at large radii and the vanishing of the
110
+ so-called Newman-Penrose constants.
111
+ In this article, in contrast to all the previous works
112
+ which start with the Kerr metric expressed in Boyer-
113
+ Lindquist coordinates and attempt to find a null coordi-
114
+ nate version of it, we will directly solve the Einstein equa-
115
+ tions in a characteristic formulation based on an affine-
116
+ null metric formulation of the Einstein equations. In ad-
117
+ dition, inspired by the Hartle-Thorne methods for obtain-
118
+ ing solutions for slowly rotating compact stars [30], we
119
+ will employ a quasi-spherical approximation of the field
120
+ equations to find a high order approximation of the Kerr
121
+ metric in out-going polar null coordinates.
122
+ To obtain
123
+ our solution, we assume stationarity and axial symme-
124
+ try. We further require an asymptotic inertial observer
125
+ as well as that that Weyl scalar Ψ0 is regular everywhere
126
+ where the background solution is regular. A study of vac-
127
+ uum stationary metrics with a smooth future null infinity
128
+ in affine-null coordinates has recently be carried out by
129
+ Tafel in [31] by considering power series of the metric
130
+ components in terms of the inverse affine distance.
131
+ Throughout the article, we will use signature +2, units
132
+ G = c = 1 and the Einstein sum convention for indices
133
+ as well as products of associated Legendre polynomials.
134
+ The article is organised as follows:
135
+ Sec. II recalls
136
+ the affine-null metric formulation, makes the necessary
137
+ symmetry assumptions for archiving our goal and de-
138
+ fines the perturbative variables; in Sec. III, we determine
139
+ the background model (Sec. III A), define useful recur-
140
+ sively re-appearing functions in the perturbation analysis
141
+ (Sec. III B), solve the perturbation equations (Sec. III C-
142
+ III F) and in Sec. III G the affine-null metric functions
143
+ for the null are expressed in terms of the mass and spe-
144
+ cific angular momentum, in Sec. IV, to verify our re-
145
+ sults, we calculate the affine-null version of Kerr metric
146
+ in a Bondi frame via a coordinate transformation with a
147
+ method adopted from [28], in Sec. V position of the outer
148
+ ergosphere and event (past) horizon of the black hole we
149
+ discuss the and Sec. VI contains the final discussion of our
150
+ work. The article finishes with two appendices: App. A
151
+ lists relations between associated Legendre polynomials
152
+ and App. B presents a derivation of the expression of the
153
+ Komar charges relevant for this work.
154
+ II.
155
+ AFFINE-NULL METRIC FORMULATION
156
+ FOR STATIONARY AND AXIAL SYMMETRIC
157
+ SPACETIMES
158
+ Here we review the necessary properties of character-
159
+ istic initial value formulation of the Einstein equations
160
+ in affine-null coordinates, discuss the implications of the
161
+ imposed symmetry assumptions and present the notation
162
+ used in our analysis.
163
+ Taking coordinates xa = (u, λ, xA), where u is an out–
164
+ going null coordinate, λ an affine parameter, and xA are
165
+ angular coordinates, a generic line element for an affine-
166
+ null metric defined with respect to a family of outgoing
167
+ null hypersurfaces u = const is [16–18, 32]
168
+ gabdxadxb = −Wdu2 − 2dudλ
169
+ +R2hAB(dxA − W Adu)(dxB − W Bdu). (2.1)
170
+ The determinant det(hAB) = det(qAB) = sin2 θ is the
171
+ determinant of a round unit sphere metric qAB.
172
+ Con-
173
+ sequently hAB is transverse-traceless and has only two
174
+ degrees of freedom. Thus, the function R relates to the
175
+ area of cuts du = dλ = 0. The inverse metric is given by
176
+ guλ = −1 , gλλ = W , gλA = −W A , gAB = hAB
177
+ R2 ,
178
+ (2.2)
179
+ where W A = (W θ, W φ) and hABhBC = δC
180
+ A and in par-
181
+ ticular [33]
182
+ hABdxAdxB =
183
+
184
+ e2γdθ2 + sin2 θ
185
+ e2γ dφ2�
186
+ cosh(2δ)
187
+ +2 sinθ sinh(2δ)dθdφ .
188
+ (2.3)
189
+ A complex null dyad to represent the 2-metric hAB like
190
+ hAB = m(A ¯mB) with mAmBhAB = mA ¯mBhAB − 1 = 0
191
+ is
192
+ mA∂A =
193
+ 1
194
+
195
+ 2eγ
196
+
197
+ cosh δ − i sinh δ
198
+
199
+ ∂y
200
+ +
201
+ ieγ
202
+
203
+ 2 sin θ
204
+
205
+ cosh δ + i sinh δ
206
+
207
+ ∂φ,
208
+ (2.4)
209
+ Like in any Bondi-Sachs type metric [9], the vacuum field
210
+ equations Rab = 0 with Rab being the Ricci tensor can
211
+ be grouped into supplementary equations Si = 0 with
212
+ Si = (Ruu, Ruθ, Ruφ),
213
+ (2.5)
214
+ one trivial equation Ruλ = 0 and the six main equations
215
+ H(γ)
216
+ K
217
+ = 0, K ∈ (1, 2, 3, 4)) and H(δ)
218
+ k
219
+ = 0, k ∈ (1, 2)) with
220
+ H(γ)
221
+ K
222
+ =
223
+
224
+ Rλλ, Rλθ, hABRAB, ℜe(mAmBRAB)
225
+
226
+ ,
227
+ H(δ)
228
+ k
229
+ =
230
+
231
+ Rλφ, ℑm(mAmBRAB)
232
+
233
+ ,
234
+ (2.6)
235
+ with ℜe(x) and ℑm(x) the real an imaginary part of x re-
236
+ spectively. We assume that the spacetime is axisymmet-
237
+ ric and stationary with associated Killing vectors fields
238
+
239
+ 3
240
+ ∂u and ∂φ. Therefore the metric functions do not depend
241
+ on u and φ. The Killing symmetries imply two conserved
242
+ quantities, the Komar mass, Km, and the Komar angu-
243
+ lar momentum, KL, which can be calculated from their
244
+ respective integrals (also see App. B)
245
+ Km := K(∂u) = 1
246
+ 8π lim
247
+ λ→∞
248
+ � �
249
+ W,λ−R2hABW AW B
250
+
251
+
252
+ R2d2q
253
+ (2.7)
254
+ while for the axial Killing vector we have
255
+ KL := K(∂φ) = − 1
256
+ 16π lim
257
+ λ→∞
258
+ � �
259
+ R4hφBW B
260
+
261
+
262
+ d2q
263
+ (2.8)
264
+ where dq = sin θdθdφ is the surface area element of the
265
+ unit sphere.
266
+ Let us assume there is a smooth one parameter family
267
+ of stationary and axially symmetric metrics gab(ε), where
268
+ ε is a small dimensionless parameter such that ε = 0 is a
269
+ corresponds to a (static) spherically symmetric spacetime
270
+ solution of the vacuum Einstein equations. Then there is
271
+ an expansion of the metric fields like
272
+ R(λ, θ) = r(λ) + R[1](λ, θ)ε + R[2](λ, θ)ε2 + R[3](λ, θ)ε3 + O(ε4),
273
+ (2.9a)
274
+ W(λ, θ) = V (λ) + W[1](λ, θ)ε + W[2](λ, θ)ε2 + W[3](λ, θ)ε3 + O(ε4),
275
+ (2.9b)
276
+ W A(λ, θ) = W A
277
+ [1](λ, θ)ε + W A
278
+ [2](λ, θ)ε2 + W A
279
+ [3](λ, θ)ε3 + O(ε4),
280
+ (2.9c)
281
+ γ(λ, θ) = γ[1](λ, θ)ε + γ[2](λ, θ)ε2 + γ[3](λ, θ)ε3 + O(ε4),
282
+ (2.9d)
283
+ δ(λ, θ) = δ[1](λ, θ)ε + δ[2](λ, θ)ε2 + δ[3](λ, θ)ε3 + O(ε4).
284
+ (2.9e)
285
+ Inserting (2.9) in (2.7) and (2.8) implies Km = O(ε0) and
286
+ KL = O(ε). We make the requirements
287
+ Km(ε) = Km(−ε) , KL(ε) = −KL(−ε).
288
+ (2.10)
289
+ These
290
+ conditions
291
+ imply
292
+ that
293
+ under
294
+ the
295
+ change
296
+ ε → −ε the sense of rotation is reversed (recall that
297
+ K(∂φ) = −K(∂(−φ))).
298
+ From the metric (2.1), we see
299
+ that the 2-surfaces with u = u0 and λ = λ0, defined
300
+ such that R(u0, λ0, θ) =const have the induced metric
301
+ R2hABdxAdxB with area 4πR2(u0, λ0).
302
+ We assume
303
+ that the area of these 2-surfaces is invariant under
304
+ the change ε → −ε, which implies that R2 is an even
305
+ function of ε. Therefore R is either an even or an odd
306
+ function of ε.
307
+ However, if R were an odd function,
308
+ we had R(ε = 0) = 0, which is a non admissible solu-
309
+ tion.
310
+ In addition, ds2(∂φ, ∂φ) and ds2(∂θ, ∂θ) must be
311
+ independent of the sense of rotation implying that hφφ
312
+ and hθθ are even. However, due to the frame dragging
313
+ effect ds2(∂θ, ∂φ) must depend on the sense of rotation.
314
+ Therefore hθφ is an odd function of ε.
315
+ Using similar
316
+ arguments, because the Komar angular momentum KL
317
+ is an odd function of ε and taking into account (2.8)
318
+ and the parity behaviour of hAB and R2, we have that
319
+ W θ is even and W φ odd. Similarly, since Km must be a
320
+ even function of ε, W must be even in ε. Therefore,
321
+ R[2n+1] = W[2n+1] = 0,
322
+ (2.11a)
323
+ W θ
324
+ [2n+1] = 0,
325
+ (2.11b)
326
+ W φ
327
+ [2n] = 0,
328
+ (2.11c)
329
+ γ[2n+1] = δ[2n] = 0.
330
+ (2.11d)
331
+ To arrive at the last conditions (2.11d) we have taken
332
+ into account the odd parity of hθφ, which gives us
333
+ sinh(δ(ε)) = − sinh(δ(−ε)).
334
+ Hence, δ must be odd in
335
+ ε. Similarly, for hθθ and hφφ be even, γ(ε) must satisfies
336
+ e2γ(ε) = e2γ(−ε), which implies that γ is a even function
337
+ of ε.
338
+ We conclude
339
+ R = r + R[2]ε2 + +R[4]ε4 + O(ε6),
340
+ (2.12a)
341
+ W = V + W[2]ε2 + W[4]ε4 + O(ε6),
342
+ (2.12b)
343
+ W θ = W θ
344
+ [2]ε2 + W θ
345
+ [4]ε4 + O(ε4),
346
+ (2.12c)
347
+ W φ = W φ
348
+ [1]ε + W φ
349
+ [3]ε3 + O(ε5),
350
+ (2.12d)
351
+ γ = γ[2]ε2 + γ[4]ε4 + O(ε6),
352
+ (2.12e)
353
+ δ = δ[1]ε + δ[3]ε3 + O(ε5).
354
+ (2.12f)
355
+ A similar expansion was made by Hartle [30] in the
356
+ derivation of a metric for slowly rotating stars using a
357
+ a 3+1 decomposition of the metric. From (2.9) follows
358
+ that the Ricci tensor has the expansions
359
+ Rab = R[0]ab + R[1]abε + R[2]abε2 + R[3]abε3 + ... (2.13)
360
+ In fact, with the notation f[i] ∈ {γ[i], δ[i], R[i], W A
361
+ [i], W[i]},
362
+ it turns out for a perturbation at order n > 1 that
363
+ S[n]i = ˆSi(f[n]) + s[i](f[m<n])
364
+ (2.14)
365
+ H(γ)
366
+ K
367
+ = ˆH(γ)
368
+ K (f[n]) + h(γ)
369
+ K (f[m<n])
370
+ (2.15)
371
+ H(δ)
372
+ k
373
+ = ˆH(δ)
374
+ k (f[n]) + h(δ)
375
+ k (f[m<n])
376
+ (2.16)
377
+ where ˆSi, ˆH(γ)
378
+ K
379
+ and ˆH(δ)
380
+ k
381
+ are linear differential operators
382
+ of the indicated arguments. The functions s[i], h(γ)
383
+ K
384
+ and
385
+
386
+ 4
387
+ h(δ)
388
+ k
389
+ are nonlinear functions of the lower order perturba-
390
+ tions f[m] for m < n.
391
+ For the computations, it is useful to change the an-
392
+ gular coordinate according to y = − cosθ, introduce
393
+ s(y) =
394
+
395
+ 1 − y2 and transform W θ = s−1W y.
396
+ In ad-
397
+ dition, for a perturbation at order n it will shown useful
398
+ to make the following decomposition of the perturbation
399
+ f[n] in terms of associated Legendre polynomials, P m
400
+ ℓ (y),
401
+ R[n](λ, y) =R[n.ℓ](λ)P 0
402
+ ℓ (y)
403
+ (2.17)
404
+ W y
405
+ [n](λ, y) =W φ
406
+ [n.ℓ](λ)
407
+
408
+ s(y)P 1
409
+ ℓ (y)
410
+
411
+ (2.18)
412
+ W φ
413
+ [n](λ, y) =W φ
414
+ [n.ℓ](λ)
415
+ �P 1
416
+ ℓ (y)
417
+ s(y)
418
+
419
+ (2.19)
420
+ W[n](λ, y) =W[n,ℓ](λ)P 0
421
+ ℓ (y)
422
+ (2.20)
423
+ γ[n](λ, y) =γ[n.ℓ](λ)P 2
424
+ ℓ (y)
425
+ (2.21)
426
+ δ[n](λ, y) =δ[n.ℓ](λ)P 2
427
+ ℓ (y) .
428
+ (2.22)
429
+ We remark that this decomposition with respect to the
430
+ associated Legendre polynomials is in fact a decomposi-
431
+ tion in terms of axi-symmetric spin-weighed harmonics
432
+ (up to normalisation) obtained by setting m = 0 in the
433
+ standard sYℓm(y, φ).
434
+ III.
435
+ SOLUTION OF THE BACKGROUND AND
436
+ PERTURBATION EQUATIONS
437
+ A.
438
+ Solution background equations
439
+ The main equations for the background model are
440
+ 0 = r,λλ
441
+ r
442
+ (3.1a)
443
+ 0 = [(r2),λV − 2λ],λ ,
444
+ (3.1b)
445
+ From which we deduce the simple relations
446
+ r(λ) = r1(λ − λ0) + r0,
447
+ (3.2)
448
+ V (λ) = 2(λ − λ0) + 2r0r1V0 + A
449
+ 2rr1
450
+ ,
451
+ (3.3)
452
+ where A is a constant of integration. Since we have the
453
+ freedom of rescaling the affine parameter λ → αλ+β, we
454
+ can take without loss of generality
455
+ r(λ) = λ , V (λ) = 1 − A
456
+ 2λ.
457
+ (3.4)
458
+ The resulting spacetime is the Schwarzschild metric in
459
+ Eddington-Finkelstein coordinates, with a total Bondi
460
+ mass m0 related to the integration constant A by A =
461
+ 4m0.
462
+ Moreover, λ = A/2 corresponds to location the
463
+ horizon of the Schwarzschild horizon.
464
+ B.
465
+ Recurrent operators in the equations of the
466
+ perturbations
467
+ Assuming a formal expansion like
468
+ R(λ, y) ≈ λ + R[n](λ, y)εn
469
+ (3.5a)
470
+ W(λ, y) ≈ 1 − A
471
+ 2λ + W[n](λ, y)εn,
472
+ (3.5b)
473
+ W A(λ, y) ≈ W A
474
+ [n](λ, y)εn,
475
+ (3.5c)
476
+ γ(λ, y) ≈ γ[n](λ, y)εn,
477
+ (3.5d)
478
+ δ(λ, y) ≈ δ[n](λ, y)εn.
479
+ (3.5e)
480
+ Then
481
+ the
482
+ Ricci
483
+ tensor
484
+ also
485
+ has
486
+ the
487
+ expansion
488
+ Rab ≈ Rab[n]εn after dropping terms of higher order.
489
+ The O(εn) coefficients of the supplementary equations
490
+ gives
491
+ ˆS1(R[n], W[n], W y
492
+ [n]) =
493
+ 1
494
+ 2λ2
495
+
496
+ 1 − A
497
+
498
+ � �
499
+ λ2W[n],λ + AR[n]
500
+ λ
501
+
502
+
503
+ + (s2W[n],y),y
504
+ 2λ2
505
+ − A
506
+ 4λ2 W y
507
+ [n],y
508
+ (3.6a)
509
+ ˆS2(W[n], W y
510
+ [n]) =
511
+ 1
512
+ 2λ2
513
+
514
+ 1 − A
515
+
516
+ � (λ4W y
517
+ [n]),λ
518
+ s
519
+ + s
520
+ 2W[n],λy +
521
+ W y
522
+ [n]
523
+ s
524
+ (3.6b)
525
+ ˆS3(W φ
526
+ [n]) =
527
+ s2
528
+ 2λ2
529
+
530
+ 1 − A
531
+
532
+
533
+ (λ4W φ
534
+ [n],λ),λ + 1
535
+ 2
536
+
537
+ s4W φ
538
+ [n],y
539
+
540
+ ,y
541
+ (3.6c)
542
+
543
+ 5
544
+ Those for ˆH(γ)
545
+ K
546
+ are
547
+ ˆH(γ)
548
+ 1
549
+ (R[n]) = − 2
550
+ λR[n],λλ
551
+ (3.7a)
552
+ ˆH(γ)
553
+ 2
554
+ (R[n], γ[n], W y
555
+ [n]) =
556
+ 1
557
+ 2λ2
558
+ (λ4W y
559
+ [n],λ),λ
560
+ s
561
+ − s
562
+ �R[n],y
563
+ λ
564
+
565
+
566
+ + (γ[n],λs2),y
567
+ s
568
+ (3.7b)
569
+ ˆH(γ)
570
+ 3
571
+ (R[n], γ[n], W y
572
+ [n], W[n]) = −
573
+
574
+ λW[n]
575
+
576
+ ,λ −
577
+ ��
578
+ 1 − A
579
+
580
+
581
+ (λR[n]),λ
582
+
583
+
584
+
585
+
586
+ λ[n],ys2�
587
+ ,y
588
+ λ
589
+ +
590
+
591
+ λ4W y
592
+ [n]
593
+
594
+ ,λy
595
+ 2λ2
596
+ +(γ[n],ys4),y
597
+ s2
598
+ − 2γ[n]
599
+ (3.7c)
600
+ ˆH(γ)
601
+ 4
602
+ (γ[n], W y
603
+ [n]) = −
604
+
605
+ λ
606
+
607
+ λ − A
608
+ 2
609
+
610
+ γ[n],λ
611
+
612
+
613
+ + s2
614
+ 2
615
+
616
+ λ2W y
617
+ [n]
618
+ s2
619
+
620
+ ,λy
621
+ (3.7d)
622
+ and those for ˆH(δ)
623
+ k
624
+ ˆH(δ)
625
+ 1 (δ[n], W φ
626
+ [n]) =
627
+ s2
628
+ 2λ2 (λ4W φ
629
+ [n],λ),λ + (δ[n],λs2),y
630
+ (3.8a)
631
+ ˆH(δ)
632
+ 2 (δ[n], W φ
633
+ [n]) = −
634
+
635
+ λ
636
+
637
+ λ − A
638
+ 2
639
+
640
+ δ[n],λ
641
+
642
+
643
+ − s2
644
+ 2
645
+
646
+ λ2W φ
647
+ [n]
648
+
649
+ ,λy
650
+ (3.8b)
651
+ We observe that (3.7b) and (3.7d) as well as (3.8a) and
652
+ (3.8b) can be combined (see e.g. in [34]) to two fourth
653
+ order (master) equations
654
+ 0 =M(γ[n]) − s2R[n],λλyy
655
+ (3.9a)
656
+ 0 =M(δ[n])
657
+ (3.9b)
658
+ where
659
+ M(F) := 1
660
+ λ2 [λ4(λF),λλλ],λ + [(λF),λλys2],y
661
+ +
662
+ � A
663
+ 2λ + 2 − 4
664
+ s2
665
+
666
+ (λF),λλ − A
667
+ 2 [λ(λF),λλλ],λ
668
+ (3.10)
669
+ We emphasize that Eqs. (3.9a) and (3.9b) (similarly to
670
+ the Teukolsky master equations in 3+1 perturbation the-
671
+ ory) are the key equations to solve the system, because
672
+ they provide the initial data γ[n] or δ[n] needed to in-
673
+ tegrate the hypersurface equations of the characteristic
674
+ initial value problem.
675
+ C.
676
+ First order perturbations
677
+ Since γ[1], R[1], W y
678
+ [1] and W[1] are zero, we only have
679
+ to consider the equations
680
+ 0 = ˆS3(δ[1], W φ
681
+ [1])
682
+ (3.11)
683
+ 0 = ˆH(δ)
684
+ 1 (δ[1], W φ
685
+ [1])
686
+ (3.12)
687
+ 0 = ˆH(δ)
688
+ 2 (δ[1], W φ
689
+ [1])
690
+ (3.13)
691
+ whose explicit form can be read off from (3.6c), (3.8a)
692
+ and (3.8b). The corresponding master equation is
693
+ 0 = 1
694
+ λ2 [λ4(λδ[1]),λλλ],λ + [(λδ[1]),λλys2],y
695
+ +
696
+ � A
697
+ 2λ + 2 − 4
698
+ s2
699
+
700
+ (λδ[1]),λλ − A
701
+ 2
702
+
703
+ λ(λδ[1]),λλλ
704
+
705
+
706
+ (3.14)
707
+ which is in fact a second order equation for the variable
708
+ ψ[1] := (λδ[1]),λλ ,
709
+ (3.15)
710
+ namely
711
+ 0 =λ(2λ − A)ψ[1],λλ + (8λ − A)ψ[1],λ + Aψ[1]
712
+ λ
713
+ + 2
714
+
715
+ (s2ψ[1],y),y +
716
+ ��
717
+ 2 − 4
718
+ s2
719
+ ��
720
+ ψ[1]
721
+
722
+ ,
723
+ (3.16)
724
+ which admit a solution by separation of variables by set-
725
+ ting ψ[1](λ, y) = p[1](λ)S(y),
726
+ 0 =λ(2λ − A)p[1],λλ + (8λ − A)p[1],λ +
727
+ �A
728
+ λ + 2k
729
+
730
+ p[1],
731
+ (3.17)
732
+ 0 = d
733
+ dy
734
+
735
+ s2 dS
736
+ dy
737
+
738
+ +
739
+
740
+ 2 − k − 4
741
+ s2
742
+
743
+ S,
744
+ (3.18)
745
+ with k a constant. Identifying 2 − k = ℓ(ℓ + 1), we see
746
+ that (3.18) is an associated Legendre differential equation
747
+
748
+ 6
749
+ (A3), whose general solution is
750
+ Sℓ(y) = B0kP(ℓ, 2, y) + B1kQ(ℓ, 2, y),
751
+ (3.19)
752
+ where P(·) and Q(·) are the Legendre functions of first
753
+ kind and of second kind, respectively.
754
+ Requiring a regular solution at the poles y = ±1, im-
755
+ poses that ℓ must be a nonnegative integer and B1k = 0,
756
+ because P(ℓ, 2, −1) blows up at the pole y = −1 and
757
+ Q(ℓ, 2, ±1) blows up at the poles y = ±1.
758
+ Then the
759
+ remaining Legendre function P(·) is the associated Leg-
760
+ endre polynomial P 2
761
+ ℓ (y).
762
+ To find a solution for (3.15) and (3.16), we set
763
+ ψ[1](λ, y) = ψ[1.ℓ](λ)P 2
764
+ ℓ (y) , δ[1](λ, y) = δ[1.ℓ](λ)P 2
765
+ ℓ (y),
766
+ (3.20)
767
+ where a sum in ℓ is understood.
768
+ Note that P 2
769
+ 0 (y) =
770
+ P 2
771
+ 1 (y) = 0 , consequently, δ[1.0] = δ[1.1] = 0 without loss
772
+ of generality. Subsequent insertion into (3.16) while using
773
+ eq.(A3) gives us
774
+ 0 = − 1
775
+ 2λ(A − 2λ)d2ψ[1.ℓ]
776
+ dλ2
777
+ +
778
+
779
+ 4λ − A
780
+ 2
781
+ � dψ[1.ℓ]
782
+
783
+ +
784
+
785
+ 2 − ℓ(ℓ + 1) + A
786
+
787
+
788
+ ψ[1.ℓ] .
789
+ (3.21)
790
+ Using the parameter transformation x = 4λ
791
+ A − 1, similar
792
+ to [30], we find
793
+ 0 =(1 − x)d2ψ[1.ℓ]
794
+ dx2
795
+ − 4x + 2
796
+ x + 1
797
+ dψ[1.ℓ]
798
+ dx
799
+ + ℓ(ℓ + 1)(x + 1) − 2x − 4
800
+ (x + 1)2
801
+ ψ[1.ℓ],
802
+ (3.22)
803
+ which can also be written as
804
+ 0 = d
805
+ dy
806
+
807
+ (1 − x2) d
808
+ dx(1 − x)ψ[1.ℓ]
809
+
810
+ +
811
+
812
+ ℓ(ℓ + 1) −
813
+ 4
814
+ 1 − x2
815
+
816
+ (1 − x)ψ[1.ℓ].
817
+ (3.23)
818
+ Eq. (3.23) is an associated Legendre differential equation,
819
+ like (A3), with the general solution
820
+ ψ[1.ℓ](x) = B[1.ℓ]P 2
821
+ ℓ (x) + B[2.ℓ]Q2
822
+ ℓ(x)
823
+ 1 − x
824
+ (3.24)
825
+ Inverting the parameter transformation from x to λ yields
826
+ the general solution of (3.21) so that using (3.20)
827
+ ψ[1](λ, y) =
828
+ � AB[1.ℓ]
829
+ 2A − 4λ
830
+
831
+ P 2
832
+
833
+ �4λ
834
+ A − 1
835
+
836
+ P 2
837
+ ℓ (y)
838
+ +
839
+ � AB[2.ℓ]
840
+ 2A − 4λ
841
+
842
+ Q2
843
+
844
+ �4λ
845
+ A − 1
846
+
847
+ P 2
848
+ ℓ (y)
849
+ (3.25)
850
+ The field ψ[1] is related to the Weyl scalar Ψ0,
851
+ Ψ0 = −iψ[1]
852
+ λ ε + O(ε2) .
853
+ (3.26)
854
+ Inspection of (3.24) shows that Ψ becomes infinite for
855
+ λ → A/2 and for λ → ∞ if ℓ ≥ 2.
856
+ The first case
857
+ corresponds to the unperturbed location of the horizon
858
+ while the second one corresponds to the asymptotic re-
859
+ gion.
860
+ Consequently, also Ψ0 becomes infinite in these
861
+ cases. We require regularity of the scalar curvature Ψ0
862
+ at these locations, which implies B[1.ℓ] = B[2.ℓ] = 0. This
863
+ leaves us with the trivial solution ψ[1] = 0.
864
+ Integration of (3.15) with this trivial solution while
865
+ using (3.20) yields
866
+ δ[1.ℓ](λ) = Bδ
867
+ [0.ℓ] +
868
+
869
+ [1.ℓ]
870
+ λ
871
+ ,
872
+ (3.27)
873
+ where as aforementioned, since δ[1.0] = δ[1.1] = 0 we get
874
+ that Bδ
875
+ [0.0] = Bδ
876
+ [0.1] = Bδ
877
+ [1.0] = Bδ
878
+ [1.1] = 0. These modes
879
+ are physically irrelevant because δ[1] is expressed by the
880
+ angular base of P 2
881
+ ℓ −associated Legendre polynomials and
882
+ P 2
883
+ 0 = P 2
884
+ 1 = 0. Since δ[1] is now known, we are now in
885
+ position to integrate the hypersurface equation (3.12).
886
+ We insert (3.27) into (3.12), while using (A5), to find
887
+ (λ4W φ
888
+ [1],λ),λ = 2λ2
889
+ �dδ[1.ℓ]
890
+
891
+ � KℓP 1
892
+ ℓ (y)
893
+ s
894
+ .
895
+ (3.28)
896
+ where
897
+ Kℓ = 2 − ℓ(ℓ + 1) = (1 − ℓ)(2 + ℓ).
898
+ (3.29)
899
+ Then setting
900
+ W φ
901
+ [1](λ, y) =W φ
902
+ [1.ℓ](λ)P 1
903
+ ℓ (y)
904
+ s
905
+ ,
906
+ (3.30)
907
+ gives us
908
+ d
909
+
910
+
911
+ λ4 W φ
912
+ [1.ℓ]
913
+
914
+
915
+ = 2Bδ
916
+ [1.ℓ]Kℓ;
917
+ (3.31)
918
+ or after integration
919
+ W φ
920
+ [1.ℓ] =Bφ
921
+ [0.ℓ] −
922
+ KℓBδ
923
+ [1.ℓ]
924
+ λ2
925
+
926
+
927
+ [3.ℓ]
928
+ 3λ3 .
929
+ (3.32)
930
+ This give us for the first order axisymmetric perturba-
931
+ tions
932
+ δ[1](λ, y) =
933
+
934
+
935
+ [0.ℓ] +
936
+
937
+ [1.ℓ]
938
+ λ
939
+
940
+ P 2
941
+ ℓ (y)
942
+ (3.33)
943
+ W φ
944
+ [1](λ, y) =
945
+
946
+
947
+ [0.ℓ] −
948
+ KℓBδ
949
+ [1.ℓ]
950
+ λ2
951
+
952
+
953
+ [3.ℓ]
954
+ 3λ3
955
+
956
+ P 1
957
+ ℓ (y)
958
+ s
959
+ (3.34)
960
+ Again we set the unphysical modes Bφ
961
+ [0.0] = Bφ
962
+ [3.0] = 0,
963
+ because of behavior of the angular base functions of W φ
964
+ [1].
965
+ Inserting the obtained solutions into (3.13) yields while
966
+ using (A8)
967
+ 0 = A
968
+ 2
969
+
970
+ λδ[1.ℓ],λ
971
+
972
+ ,λ−
973
+
974
+ λ2δ[1.ℓ],λ
975
+
976
+ ,λ− 1
977
+ 2
978
+
979
+ λ2W φ
980
+ [1.ℓ]
981
+
982
+ ,λ (3.35)
983
+
984
+ 7
985
+ and together with (3.27) and (3.32) this gives us for any
986
+ ℓ ≥ 2
987
+ 0 = −Bφ
988
+ [0.ℓ]λ + 1
989
+ λ2
990
+
991
+ A
992
+ 2 Bδ
993
+ [0.ℓ] −
994
+
995
+ [3.ℓ]
996
+ 6
997
+
998
+ .
999
+ (3.36)
1000
+ Hence for any ℓ ≥ 2,
1001
+
1002
+ [0.ℓ] = 0 , Bδ
1003
+ [0.ℓ] =
1004
+
1005
+ [3.ℓ]
1006
+ 3A .
1007
+ (3.37)
1008
+ Moreover, inserting the obtained solution into the sup-
1009
+ plementary equation (3.11) while using (A9), we find
1010
+ ˆS3 =
1011
+
1012
+
1013
+ [0,ℓ]
1014
+ 2
1015
+
1016
+ ℓ(ℓ + 1)Bδ
1017
+ [1.ℓ]
1018
+ 2λ2
1019
+ +
1020
+ 3ABδ
1021
+ [1.ℓ] − Bφ
1022
+ [3.ℓ]
1023
+ 6λ3
1024
+
1025
+ × Kℓ × s(y)P 1
1026
+ ℓ (y) .
1027
+ (3.38)
1028
+ Considering (3.38) for the various modes of ℓ gives us:
1029
+ ℓ = 0 is trivial because P 1
1030
+ 0 = 0; the ℓ = 1 coefficient van-
1031
+ ishes since K1 = 0. Therefore the coefficients Bφ
1032
+ [0.1] and
1033
+
1034
+ [3.1] are unconstrained by the supplementary equation
1035
+ ˆS3. Finally considering ˆS3 = 0 for the ℓ > 1 coefficients
1036
+ while using (3.37) gives us
1037
+ 0 =ℓ(ℓ + 1)Bδ
1038
+ [1.ℓ]
1039
+ (3.39)
1040
+ which implies
1041
+
1042
+ [1.ℓ] = 0
1043
+ :
1044
+ ∀ℓ > 1 .
1045
+ (3.40)
1046
+ Furthermore, requiring an asymptotic Bondi frame (a
1047
+ non-rotating inertial observer at large distances), i.e.
1048
+ gabdxadxb → −du2 − dλdu + λ2qABdxAdxB
1049
+ (3.41)
1050
+ annuls the integration constants,
1051
+ W φ
1052
+ [0.1] = Bδ
1053
+ [0.ℓ] = 0.
1054
+ (3.42)
1055
+ From the above requirements, the final solution of the
1056
+ linear perturbations are
1057
+ δ[1](y, λ) = 0 ,
1058
+ W φ
1059
+ [1](y, λ) = − B
1060
+ 3λ3
1061
+ P 1
1062
+ ℓ (y)
1063
+ s
1064
+ = − B
1065
+ 3λ3
1066
+ y
1067
+ s,
1068
+ (3.43)
1069
+ where we redefined B := Bφ
1070
+ [3,1] for notational convenience
1071
+ because it is the only remaining integration constant.
1072
+ D.
1073
+ Quadratic perturbations
1074
+ Using the notation of Sec. III B, the relevant main
1075
+ equations (i.e. only those containing γ[2], R[2], W y
1076
+ [2] and
1077
+ W[2]) for the quadratic perturbations are found to be
1078
+ 0 = ˆS1(W[2], W y
1079
+ [2]) + B2s2
1080
+ 2λ6
1081
+
1082
+ 1 − A
1083
+
1084
+
1085
+ (3.44a)
1086
+ 0 = ˆS2(W[2], W y
1087
+ [2])
1088
+ (3.44b)
1089
+ 0 = ˆH(γ)
1090
+ 1
1091
+ (R[2])
1092
+ (3.44c)
1093
+ 0 = ˆH(γ)
1094
+ 2
1095
+ (R[2], γ[2], W y
1096
+ [2])
1097
+ (3.44d)
1098
+ 0 = ˆH(γ)
1099
+ 3
1100
+ (W[2], R[2], γ[2], W y
1101
+ [2]) − B2s2
1102
+ 4λ4
1103
+ (3.44e)
1104
+ 0 = ˆH(γ)
1105
+ 4
1106
+ (γ[2], W y
1107
+ [2]) + B2s2
1108
+ 4λ4
1109
+ (3.44f)
1110
+ The first hypersurface equation (3.44c) is readily inte-
1111
+ grated
1112
+ R[2] = CR20(y) + CR11(y)r.
1113
+ (3.45)
1114
+ Similarily to (3.9b), we can deduce a master equation for
1115
+ γ[2]
1116
+ 0 =M(γ[2]) − s2R2,λλyy − 5B2
1117
+ 2λ5 s2.
1118
+ (3.46)
1119
+ For finding a solution of the remaining fields γ[2], W y
1120
+ [2]
1121
+ and W[2], we need to solve the master equation (3.46).
1122
+ Defining
1123
+ ψ[2] = (λγ[2]),λλ
1124
+ (3.47)
1125
+ with Legendre decomposition
1126
+ ψ[2] = ψ[2.ℓ](λ)P 2
1127
+ ℓ (y)
1128
+ (3.48)
1129
+ while using (3.44c) gives us after insertion of (3.45),
1130
+ (3.47) and (3.48) into (3.46)
1131
+ 0 =
1132
+
1133
+ −1
1134
+ 2r(A − 2λ)d2ψ[2.ℓ]
1135
+ dλ2
1136
+ +
1137
+
1138
+ 4r − A
1139
+ 2
1140
+ � dψ[2.ℓ]
1141
+ dy
1142
+ +
1143
+
1144
+ 2 − ℓ(ℓ + 1) + A
1145
+
1146
+
1147
+ ψ[2.ℓ]
1148
+
1149
+ P 2
1150
+ ℓ − 5B2
1151
+ 2λ5 s2
1152
+ (3.49)
1153
+ To fully factor out the Legendre polynomials P 2
1154
+ ℓ , we re-
1155
+ call that P 2
1156
+ 2 (y) = 3s2. This allows us to write
1157
+ 0 =
1158
+ ��
1159
+ −1
1160
+ 2r(A − 2λ)d2ψ[2.ℓ]
1161
+ dλ2
1162
+ +
1163
+
1164
+ 4r − A
1165
+ 2
1166
+ � dψ[2.ℓ]
1167
+ dy
1168
+ +
1169
+
1170
+ 2 − ℓ(ℓ + 1) + A
1171
+
1172
+
1173
+ ψ[2.ℓ]
1174
+
1175
+ δℓ′
1176
+ ℓ − 5B2
1177
+ 6λ5 δℓ′
1178
+ 2
1179
+
1180
+ P 2
1181
+ ℓ′(y)
1182
+ (3.50)
1183
+ We can see that (3.50) resembles (3.21) if B = 0. It
1184
+ is in fact a inhomogeneous version of (3.21). We seek
1185
+ solutions of (3.50) as a superposition of a homogeneous
1186
+ solution, ψ(hom)
1187
+ [2.ℓ]
1188
+ for B = 0, and a particular solution
1189
+ ψ(part)
1190
+ [2.ℓ]
1191
+ for B ̸= 0, i.e.
1192
+ ψ[2.ℓ] = ψ(hom)
1193
+ [2.ℓ]
1194
+ + ψ(part)
1195
+ [2.ℓ]
1196
+ .
1197
+ (3.51)
1198
+
1199
+ 8
1200
+ The homogeneous solution ψ(hom)
1201
+ [2.ℓ]
1202
+ will be like (3.25).
1203
+ Also note that a particular solution needs to be found
1204
+ for the ℓ = 2 mode, only. We find ψ(part)
1205
+ [2.2]
1206
+ = −B2/(9Aλ4).
1207
+ Hence,
1208
+ ψ[2.ℓ](λ) =A
1209
+
1210
+ C[1.ℓ]P 2
1211
+
1212
+ � 4λ
1213
+ A − 1
1214
+
1215
+ + C[2.ℓ]Q2
1216
+
1217
+ � 4λ
1218
+ A − 1
1219
+
1220
+ 2A − 4λ
1221
+
1222
+ +
1223
+
1224
+ − B2
1225
+ 9Aλ4
1226
+
1227
+ δ2
1228
+ ℓ.
1229
+ (3.52)
1230
+ It follows by the same regularity arguments like in the
1231
+ discussion for (3.25) that in order the Weyl curvature
1232
+ scalar Ψ0 does not blow up at the horizon of the un-
1233
+ perturbed solution and towards null infinity we must set
1234
+ C[1.ℓ] = C[2.ℓ] = 0.
1235
+ Consequently a solution for the ψ[2.ℓ]–modes is
1236
+ ψ[2.ℓ](λ) =
1237
+
1238
+ − B2
1239
+ 9Aλ4
1240
+
1241
+ δ2
1242
+ ℓ .
1243
+ (3.53)
1244
+ Setting
1245
+ γ[2](λ, y) = γ[2.ℓ](λ)P 2
1246
+ ℓ (y),
1247
+ (3.54)
1248
+ we find after integration of (3.47)
1249
+ γ[2.ℓ](λ, y) = Cγ
1250
+ [0.ℓ] +
1251
+
1252
+ [1.ℓ]
1253
+ λ
1254
+
1255
+ B2
1256
+ 54Aλ3 δ2
1257
+
1258
+ (3.55)
1259
+ Insertion of (3.55) and (3.45) into (3.44d) gives us
1260
+ 0 =
1261
+ (λ4W y
1262
+ [2],r)
1263
+ 2sλ2
1264
+ + sCR20,y
1265
+ λ2
1266
+ +
1267
+ �dγ[2.ℓ]
1268
+
1269
+ � 1
1270
+ s
1271
+ d
1272
+ dy
1273
+
1274
+ s2P 2
1275
+
1276
+
1277
+ .
1278
+ (3.56)
1279
+ using (A5) we find
1280
+ 0 =
1281
+
1282
+ λ4 W y
1283
+ [2],r
1284
+ s
1285
+
1286
+
1287
+ + 2sCR20,y − 2λ2Kℓ
1288
+ �dγ[2.ℓ]
1289
+
1290
+
1291
+ P 1
1292
+ ℓ (y)
1293
+ (3.57)
1294
+ which indicates that the angular behaviour of W y
1295
+ [2]/s and
1296
+ sCR20,y are dictated by the associated Legendre polyno-
1297
+ mials P 1
1298
+ ℓ (y). As of (A7), we set (note Pℓ(y) = P 0
1299
+ ℓ (y))
1300
+ R[2](λ, y) =R[2.ℓ](λ)Pℓ(y) =
1301
+
1302
+ CR
1303
+ [20.ℓ] + CR
1304
+ [21.ℓ]λ
1305
+
1306
+ Pℓ(y),
1307
+ (3.58)
1308
+ W y
1309
+ [2](λ, y) =W y
1310
+ [2.ℓ](λ)s(y)P 1
1311
+ ℓ (y)
1312
+ (3.59)
1313
+ This gives us
1314
+ 0 = d
1315
+
1316
+
1317
+ λ4 d
1318
+ dλW y
1319
+ [2.ℓ]
1320
+
1321
+ − 2CR
1322
+ [20.ℓ] − 2λ2Kℓ
1323
+ �dγ[2.ℓ]
1324
+
1325
+
1326
+ ,
1327
+ (3.60)
1328
+ Integrating (3.60) yields
1329
+ W y
1330
+ [2.ℓ] =Cy
1331
+ [0.ℓ] +
1332
+ KℓCγ
1333
+ [1.ℓ] − CR
1334
+ [20.ℓ]
1335
+ λ2
1336
+
1337
+ Cy
1338
+ [3.ℓ]
1339
+ 3λ3 −
1340
+ B2
1341
+ 9Aλ4 δ2
1342
+
1343
+ (3.61)
1344
+ where we set the integration constants Cy
1345
+ [0.0] = Cy
1346
+ [3.0] = 0,
1347
+ because P 1
1348
+ 0 (y) = 0.
1349
+ Considering (3.7d) with (3.54),
1350
+ (3.59), (A8) and s2 = P 2
1351
+ ℓ (y)/3 gives us
1352
+
1353
+ λ2
1354
+
1355
+ 1 − A
1356
+
1357
+
1358
+ γ[2.ℓ],r
1359
+
1360
+
1361
+ = 1
1362
+ 2
1363
+
1364
+ λ2W y
1365
+ [2.ℓ]
1366
+
1367
+ ,λ + B2
1368
+ 12λ4 δ2
1369
+
1370
+ (3.62)
1371
+ so that after insertion of (3.55) and (3.61), we obtain
1372
+ λCy
1373
+ [0.ℓ] =
1374
+ A
1375
+ 2λ2
1376
+
1377
+
1378
+ [1.ℓ] −
1379
+ Cy
1380
+ [3.ℓ]
1381
+ 3A
1382
+
1383
+ +
1384
+ B2
1385
+ 9Aλ3
1386
+
1387
+ 1 + Kℓ
1388
+ 4
1389
+
1390
+ δ2
1391
+
1392
+ (3.63)
1393
+ implying for any ℓ ≥ 2
1394
+ Cy
1395
+ [0.ℓ] = 0 , Cy
1396
+ [3.ℓ] = 3ACγ
1397
+ [1.ℓ]
1398
+ (3.64)
1399
+ Next, proceed with the hypersurface equation (3.44e) for
1400
+ W[2]. Insertion of (3.54), (3.58) and (3.59) into (3.44e)
1401
+ gives us
1402
+ (λW[2]),λ =
1403
+
1404
+
1405
+ ��
1406
+ 1 − A
1407
+
1408
+
1409
+ (λR[2.ℓ]),λ
1410
+
1411
+
1412
+ + ℓ(ℓ + 1)
1413
+
1414
+ R[2.ℓ]
1415
+ λ
1416
+ +
1417
+ (λ4W y
1418
+ [2.ℓ]),λ
1419
+ 2λ2
1420
+ − Kℓγ[2,ℓ]
1421
+ ��
1422
+ P 0
1423
+ ℓ (y)
1424
+ − B2s2
1425
+ 4λ4
1426
+ (3.65)
1427
+ Using
1428
+ s2 = 1 − y2 = 2
1429
+ 3[P 0
1430
+ ℓ (y) − P 0
1431
+ 2 (y)]
1432
+ (3.66)
1433
+ as well as setting
1434
+ W[2](λ, y) = W[2.ℓ](λ)P 0
1435
+ ℓ (y)
1436
+ (3.67)
1437
+ yields
1438
+ (λW[2.ℓ]),λ = −
1439
+ ��
1440
+ 1 − A
1441
+
1442
+
1443
+ (λR[2.ℓ]),λ
1444
+
1445
+
1446
+ + ℓ(ℓ + 1)
1447
+
1448
+ R[2.ℓ]
1449
+ λ
1450
+ +
1451
+ (λ4W y
1452
+ [2.ℓ]),λ
1453
+ 2λ2
1454
+ − Kℓγ[2,ℓ]
1455
+
1456
+ − B2
1457
+ 6λ4 (δ0
1458
+ ℓ − δ2
1459
+ ℓ)
1460
+ (3.68)
1461
+ Since R[2.ℓ], W y
1462
+ [2.ℓ] and γ[2.ℓ] are known, we find after
1463
+ integration
1464
+ W[2.ℓ] = −KℓCR
1465
+ [21.ℓ] − ℓ(ℓ + 1)KℓCγ
1466
+ [0.ℓ] +
1467
+ CW
1468
+ [1.ℓ]
1469
+ λ
1470
+ +
1471
+ ACR
1472
+ [20.ℓ]
1473
+ 2λ2
1474
+ +
1475
+ ℓ(ℓ + 1)Cy
1476
+ [3.ℓ]
1477
+ 6λ2
1478
+ +
1479
+ � 2B2
1480
+ 9Aλ3 − B2
1481
+ 18λ4
1482
+
1483
+ δ2
1484
+ ℓ + B2
1485
+ 18λ4 δ0
1486
+
1487
+ (3.69)
1488
+
1489
+ 9
1490
+ where CW
1491
+ [1.ℓ] are integration constants.
1492
+ Calculation of (3.44a) and (3.44b) while using (3.58),
1493
+ (3.59), (3.66), (A1) and (A10) gives us
1494
+ 0 =
1495
+
1496
+ 1 − A
1497
+
1498
+ � �
1499
+ λ2W[2.ℓ],r + AR[2.ℓ]
1500
+ λ
1501
+
1502
+
1503
+ − ℓ(ℓ + 1)
1504
+
1505
+ W[2.ℓ] + A
1506
+ 2 W y
1507
+ [2,ℓ]
1508
+
1509
+ (3.70)
1510
+ 0 =
1511
+ 1
1512
+ 2λ2
1513
+
1514
+ 1 − A
1515
+
1516
+
1517
+ (λ4W y
1518
+ [2.ℓ]),λ − 1
1519
+ 2W[2.ℓ],r + W y
1520
+ [2.ℓ]
1521
+ (3.71)
1522
+ and insertion of the respective coefficient solutions
1523
+ (3.58),(3.61) and (3.69) yields
1524
+ 0 = ℓ(ℓ + 1)
1525
+
1526
+
1527
+ CW
1528
+ [1.ℓ]
1529
+ λ
1530
+ + Kℓ
1531
+
1532
+
1533
+ [1.ℓ] − CR
1534
+ [21.ℓ]
1535
+
1536
+ +
1537
+ Cy
1538
+ [3.ℓ] − 3ACγ
1539
+ [1.ℓ]
1540
+ 6λ2
1541
+
1542
+ (3.72)
1543
+ 0 =
1544
+ CW
1545
+ [1.ℓ]
1546
+ 2λ2 +
1547
+ (Cy
1548
+ [3.ℓ] − 3ACγ
1549
+ [1.ℓ])Kℓ
1550
+ 6λ3
1551
+ ,
1552
+ ∀ℓ ≥ 1
1553
+ (3.73)
1554
+ Therefore,
1555
+ CW
1556
+ [1.ℓ] =0 , ∀ℓ ≥ 1
1557
+ (3.74a)
1558
+ Cy
1559
+ [3.ℓ] =3ACγ
1560
+ [1.ℓ] , ∀ℓ ≥ 2
1561
+ (3.74b)
1562
+ CR
1563
+ [21.ℓ] =Cγ
1564
+ [1.ℓ] , ∀ℓ ≥ 2
1565
+ (3.74c)
1566
+ Note, (3.74b) is consistent with (3.64). The requirement
1567
+ of an asymptotic inertial observer leads to
1568
+
1569
+ [0.ℓ] = CR
1570
+ [21.ℓ] = 0
1571
+ (3.75)
1572
+ which gives with (3.74) that Cγ
1573
+ [1.ℓ] = Cy
1574
+ [3.ℓ] = 0. Thus,
1575
+ redefining C := CW
1576
+ [1.1], the quadratic perturbations are
1577
+ γ[2](λ, y) =
1578
+
1579
+
1580
+ B2
1581
+ 54Aλ3 δ2
1582
+
1583
+
1584
+ P 2
1585
+ ℓ (y)
1586
+ (3.76)
1587
+ R[2](λ, y) = 0
1588
+ (3.77)
1589
+ W y
1590
+ 2 (λ, y) =
1591
+
1592
+ − B2
1593
+ 9Aλ4 δ2
1594
+
1595
+
1596
+ s(y)P 1
1597
+ ℓ (y)
1598
+ (3.78)
1599
+ W[2](λ, y) = C
1600
+ λ + B2
1601
+ 18λ4 +
1602
+ � 2B2
1603
+ 9Aλ3 − B2
1604
+ 18λ4
1605
+
1606
+ P 0
1607
+ 2 (y) .
1608
+ (3.79)
1609
+ E.
1610
+ Third order perturbations
1611
+ Similarly, expressions for the higher order perturba-
1612
+ tions quantities f[i] can be obtained using the same pro-
1613
+ cedere as in the previous sections. In this and in the next
1614
+ subsection we show the fundamental results without re-
1615
+ peating intermediate steps.
1616
+ The relevant equations for the third perturbations are
1617
+ 0 = ˆS3(δ[3], W φ
1618
+ [3]) − B3s4
1619
+ 6Aλ6
1620
+ (3.80)
1621
+ 0 = ˆH(δ)
1622
+ 1 (δ[3], W φ
1623
+ [3]) − B3s4
1624
+ 6Aλ6
1625
+ (3.81)
1626
+ 0 = ˆH(δ)
1627
+ 2 (δ[3], W φ
1628
+ [3]) + 2B3ys2
1629
+ 3Aλ5
1630
+ (3.82)
1631
+ Similarily to (3.9b), we can deduce a master equation
1632
+ for δ[3]
1633
+ 0 =M(δ[3]) − 40B3
1634
+ 3Aλ6 s2y
1635
+ (3.83a)
1636
+ Using P 2
1637
+ 3 (y) = 15ys2 and following the steps of Sec. III C,
1638
+ we find
1639
+ δ[3](λ, y) =
1640
+
1641
+
1642
+ B3
1643
+ 162A2λ4
1644
+
1645
+ P 2
1646
+ 3 (y)
1647
+ (3.84)
1648
+ W[3](λ, y) =
1649
+
1650
+ − D
1651
+ 3λ3 −
1652
+ 2B3
1653
+ 135Aλ6
1654
+ � P 1
1655
+ 1 (y)
1656
+ s(y)
1657
+ +
1658
+
1659
+ B3
1660
+ 405Aλ6 −
1661
+ 4B3
1662
+ 81A2λ5
1663
+ � P 1
1664
+ 3 (y)
1665
+ s(y)
1666
+ (3.85)
1667
+ where D is the only free new remaining integration con-
1668
+ stant that appears at this order.
1669
+ F.
1670
+ Fourth order perturbations
1671
+ Here the relevant main equations are those containing
1672
+ γ[4], R[4], W y
1673
+ [4] and W[4] which are
1674
+ 0 = ˆS1(W[4], W y
1675
+ 4]) +
1676
+ � 14
1677
+ 9Aλ −
1678
+ 1
1679
+ 12λ2 − 35
1680
+ 6A2
1681
+ � B4s4
1682
+ 3λ8
1683
+ +
1684
+ ��
1685
+ 1 − A
1686
+
1687
+
1688
+ D − CB
1689
+ 2A +
1690
+ � 16
1691
+ Aλ2 −
1692
+ 7
1693
+ 3λ3
1694
+ � B3
1695
+ 9A
1696
+ � Bs2
1697
+ λ6
1698
+
1699
+ 8B4
1700
+ 27λ8A2 + CB2
1701
+ 3Aλ6
1702
+ (3.86a)
1703
+ 0 = ˆS2(W[4], W y
1704
+ [4]) +
1705
+ �(7A + 120λ)s2
1706
+ 12λ
1707
+ − 8
1708
+ � B4ys
1709
+ 9A2λ7
1710
+ + 2ysCB2
1711
+ 3Aλ5
1712
+ (3.86b)
1713
+ 0 = ˆH(γ)
1714
+ 1
1715
+ (R[4]) −
1716
+ B4s4
1717
+ 18A2λ8
1718
+ (3.86c)
1719
+ 0 = ˆH(γ)
1720
+ 2
1721
+ (R[2], γ[2], W y
1722
+ [2]) + B4ys3
1723
+ 27A2λ7
1724
+ (3.86d)
1725
+ 0 = ˆH(γ)
1726
+ 3
1727
+ (W[2], R[2], γ[2], W y
1728
+ [2]) + (A − 14r)B4s4
1729
+ 36A2λ7
1730
+ + 2B4s2
1731
+ 9A2λ6 + DBs4
1732
+ 2λ4
1733
+ (3.86e)
1734
+ 0 = ˆH(γ)
1735
+ 4
1736
+ (γ[2], W y
1737
+ [2]) +
1738
+
1739
+ 14 + A
1740
+ 2r
1741
+ � B4s4
1742
+ 9A2λ6
1743
+ +
1744
+ � B2C
1745
+ 2Aλ4 − BD
1746
+ 2λ4 −
1747
+ 38B2
1748
+ 27A2λ6
1749
+
1750
+ s2
1751
+ (3.86f)
1752
+
1753
+ 10
1754
+ The first hypersurface equation (3.86c) is readily inte-
1755
+ grated
1756
+ R[4](λ, y) =ER0(y) + ER1(y)λ −
1757
+ s4B4
1758
+ 1080A2λ5
1759
+ (3.87)
1760
+ or expressing in terms of the Legendre polynomials P 0
1761
+ ℓ (y)
1762
+ R[4](λ, y) =
1763
+
1764
+ ER
1765
+ [0.ℓ] + ER
1766
+ [1.ℓ]λ
1767
+
1768
+ P 0
1769
+ ℓ (y)
1770
+
1771
+ B4
1772
+ 135A2λ5
1773
+ �P 0
1774
+ 0 (y)
1775
+ 15
1776
+ − 2P 0
1777
+ 2 (y)
1778
+ 21
1779
+ + P 0
1780
+ 4 (y)
1781
+ 35
1782
+
1783
+ (3.88)
1784
+ Similarily to (3.9a) we can deduce a master equation
1785
+ for γ[4]
1786
+ 0 =M(γ[4]) − s2R[4],λλyy − 5B2Cs2
1787
+ Aλ5
1788
+ − 5BDs2
1789
+ λ5
1790
+ +
1791
+
1792
+ 358 − s2
1793
+
1794
+ 397 + A
1795
+ λ
1796
+ �� B4s2
1797
+ 9A2λ7
1798
+ (3.89a)
1799
+ Using the methods of Sec. III D together with the in-
1800
+ verted Legendre relations
1801
+ 1 =P 0
1802
+ 0 (y) = −P 1
1803
+ 1 (y)
1804
+ s
1805
+ (3.90a)
1806
+ y =P 0
1807
+ 1 (y) = −P 1
1808
+ 2 (y)
1809
+ 3s
1810
+ (3.90b)
1811
+ y2 =1
1812
+ 3 − 2
1813
+ 3P 0
1814
+ 2 (y) = 1 − 1
1815
+ 3P 2
1816
+ 2 (y)
1817
+ (3.90c)
1818
+ y3 = − 2P 1
1819
+ 4 (y)
1820
+ 35s
1821
+ + P 1
1822
+ 2 (y)
1823
+ 7s
1824
+ (3.90d)
1825
+ y4 =1
1826
+ 5 − 4P 0
1827
+ 2 (y)
1828
+ 7
1829
+ + 8P 0
1830
+ 4 (y)
1831
+ 35
1832
+ = 1 − 8
1833
+ 21P 2
1834
+ 2 (y) −
1835
+ 2
1836
+ 105P 2
1837
+ 4 (y)
1838
+ (3.90e)
1839
+ we deduce the following solution for the fourth order per-
1840
+ turbation
1841
+ R[4] = −
1842
+ B4
1843
+ 135A2λ5
1844
+ �P 0
1845
+ 0
1846
+ 15 − 2P 0
1847
+ 2
1848
+ 21 + P 0
1849
+ 4
1850
+ 35
1851
+
1852
+ (3.91a)
1853
+ γ[4] =
1854
+ � BD
1855
+ 27Aλ3 −
1856
+ B2C
1857
+ 27A2λ3 −
1858
+ B4
1859
+ 1134A2λ6
1860
+
1861
+ P 2
1862
+ 2
1863
+ +
1864
+
1865
+ B4
1866
+ 405A3λ5 +
1867
+ B4
1868
+ 17010A2λ6
1869
+
1870
+ P 2
1871
+ 4
1872
+ (3.91b)
1873
+ W y
1874
+ [4] =
1875
+ � 2BD
1876
+ 9Aλ4 − 2B2C
1877
+ 9A2λ4 −
1878
+ 2B4
1879
+ 2835A2λ7
1880
+
1881
+ P 1
1882
+ 2
1883
+ +
1884
+
1885
+ 2B4
1886
+ 81A3λ6 +
1887
+ B4
1888
+ 4725A2λ7
1889
+
1890
+ P 1
1891
+ 4
1892
+ (3.91c)
1893
+ W[4] =E
1894
+ λ − BD
1895
+ 9λ4 +
1896
+ 4B4
1897
+ 405A2λ6 −
1898
+ B4
1899
+ 675Aλ7
1900
+ +
1901
+
1902
+ 2B4
1903
+ 945Aλ7 −
1904
+ 2B4
1905
+ 81A2λ6 + 4B2C
1906
+ 9A2λ3 − 4BD
1907
+ 9Aλ3 + BD
1908
+ 9λ4
1909
+
1910
+ P 0
1911
+ 2
1912
+ +
1913
+
1914
+
1915
+ 8B4
1916
+ 81A3λ5 +
1917
+ 2B4
1918
+ 135A2λ6 −
1919
+ B4
1920
+ 1575Aλ7
1921
+
1922
+ P 0
1923
+ 4
1924
+ (3.91d)
1925
+ Note that E is the only remaining new integration con-
1926
+ stant, all other vanish because of the reasons mentioned
1927
+ in Sec. III D.
1928
+ G.
1929
+ Perturbations in terms of Komar quantities
1930
+ The solution of the perturbation involve the free inte-
1931
+ gration constants A, B, C, D and E. These free constants
1932
+ determine the Komar mass, Km, and the Komar angular
1933
+ momentum, KL, which can be found by calculation of
1934
+ (2.7) and (2.8)
1935
+ m := Km = A
1936
+ 4 − C
1937
+ 2 ε2 − E
1938
+ 2 ε4 + O(ε5)
1939
+ (3.92)
1940
+ L := KL = −B
1941
+ 6 ε + D
1942
+ 6 ε3 + O(ε5)
1943
+ (3.93)
1944
+ If ε = 0, Km = A/4 corresponds to the mass m0 of the
1945
+ unperturbed system. Furthermore, we can see that L =
1946
+ O(ε). This allows us relate ε with the angular momentum
1947
+ L of the system. To do that we have to solve the cubic
1948
+ equation
1949
+ 0 = D
1950
+ 6 ε3 − B
1951
+ 6 ε + L
1952
+ (3.94)
1953
+ for ε. This equation also shows that in order to make the
1954
+ substitution of ε by L, we seek the solution ε(L) = O(L).
1955
+ The root of (3.94) which fulfils this requirement is
1956
+ ε = − 6
1957
+ B L − 216D
1958
+ B4 L3 + O(L5)
1959
+ (3.95)
1960
+ Subsequent insertion of this expansion into (3.92) and
1961
+ solving for A gives us
1962
+ A = 4m + 72C
1963
+ B2 L2 + 2592EB − 2CD
1964
+ B5
1965
+ L4 + O(L6) (3.96)
1966
+ The relations (3.95) and (3.96) allow us to substitute
1967
+ A and ε, by the physical quantities m and L. Insertion
1968
+ of (3.95) and (3.96) into the solution of the perturba-
1969
+ tions and subsequent expansion up to O(L4) allows us to
1970
+ eliminate the integration constants C, D and E from the
1971
+ perturbations. This gives us
1972
+
1973
+ 11
1974
+ R(λ, y) = λ −
1975
+ �P 0
1976
+ 0
1977
+ 5 − 2P 0
1978
+ 2
1979
+ 7
1980
+ + 3P 0
1981
+ 4
1982
+ 35
1983
+
1984
+ L4
1985
+ 5m2λ5 + O(L6)
1986
+ (3.97a)
1987
+ W(λ, y) = 1 − 2m
1988
+ λ +
1989
+ � 2
1990
+ λ4 +
1991
+
1992
+ 2
1993
+ mλ3 − 2
1994
+ λ4
1995
+
1996
+ P 0
1997
+ 2
1998
+
1999
+ L2 +
2000
+
2001
+ 4
2002
+ 5m2λ6 −
2003
+ 12
2004
+ 25λ7 +
2005
+
2006
+ 24
2007
+ 35mλ7 − 2L4
2008
+ m2λ6
2009
+
2010
+ P 0
2011
+ 2
2012
+ +
2013
+
2014
+
2015
+ 2
2016
+ m3λ5 +
2017
+ 6
2018
+ 5m2λ6 −
2019
+ 36
2020
+ 175mλ7
2021
+
2022
+ P 0
2023
+ 4
2024
+
2025
+ L4 + O(L6)
2026
+ (3.97b)
2027
+ W y(λ, y) =
2028
+
2029
+
2030
+ 1
2031
+ mλ4 (sP 1
2032
+ 2 )
2033
+
2034
+ L2 +
2035
+
2036
+
2037
+ 2
2038
+ 35m2λ7 (sP 1
2039
+ 2 ) +
2040
+
2041
+ 1
2042
+ 2m3λ6 +
2043
+ 3
2044
+ 175m2λ7
2045
+
2046
+ (sP 1
2047
+ 4 )
2048
+
2049
+ L4 + O(L6)
2050
+ (3.97c)
2051
+ W φ(λ, y) =
2052
+
2053
+ − 2
2054
+ λ3
2055
+ P 1
2056
+ 1
2057
+ s
2058
+
2059
+ L +
2060
+
2061
+ 4
2062
+ 5mλ6
2063
+ P 1
2064
+ 1
2065
+ s +
2066
+
2067
+ 2
2068
+ 3m2λ5 −
2069
+ 2
2070
+ 15mλ6
2071
+ � P 1
2072
+ 3
2073
+ s
2074
+
2075
+ L3 + O(L5)
2076
+ (3.97d)
2077
+ γ(λ, y) =
2078
+
2079
+
2080
+ 1
2081
+ 6mλ3 P 2
2082
+ 2
2083
+
2084
+ L2 +
2085
+
2086
+
2087
+ 1
2088
+ 14m2λ6 P 2
2089
+ 2 +
2090
+
2091
+ 1
2092
+ 20m3λ5 +
2093
+ 1
2094
+ 210m2λ6
2095
+
2096
+ P 2
2097
+ 4
2098
+
2099
+ L4 + O(L6)
2100
+ (3.97e)
2101
+ δ(λ, y) =
2102
+
2103
+ 1
2104
+ 12m2λ4 P 2
2105
+ 3
2106
+
2107
+ L3 + O(L5)
2108
+ (3.97f)
2109
+ We see in (3.97) that the perturbations are determined by the mass and angular momentum, i.e. the solution has two
2110
+ hairs. To show that this solutions represents the Kerr solution in affine-null coordinates, we introduce the specific
2111
+ angular momentum, a := L/m. In terms of a, (3.97) read after changing to the angular coordinate θ
2112
+ R(λ, θ) = λ − 3m2 sin4 θ
2113
+ 40λ5
2114
+ a4 + O(a6)
2115
+ (3.98a)
2116
+ W(λ, θ) = 1 − 2m
2117
+ λ +
2118
+ �2m
2119
+ λ3 +
2120
+
2121
+ −3m
2122
+ λ3 + 3m2
2123
+ λ4
2124
+
2125
+ sin2 θ
2126
+
2127
+ a2
2128
+ +
2129
+
2130
+ −2m
2131
+ λ5 +
2132
+ �10m
2133
+ λ5
2134
+ − 3m2
2135
+ λ6
2136
+
2137
+ sin2 θ +
2138
+
2139
+ −35m
2140
+ 4λ5 + 21m2
2141
+ 4λ6 − 9m3
2142
+ 10λ7
2143
+
2144
+ sin4 θ
2145
+
2146
+ a4 + O(a6)
2147
+ (3.98b)
2148
+ W θ(λ, θ) =
2149
+
2150
+ −3m
2151
+ λ4 a2 +
2152
+ �5m
2153
+ λ6 −
2154
+ �35m
2155
+ 4λ6 + 3m2
2156
+ 10λ7
2157
+
2158
+ sin2 θ
2159
+
2160
+ a4
2161
+
2162
+ sin θ cos θ + O(a6)
2163
+ (3.98c)
2164
+ W φ(λ, θ) = 2m
2165
+ λ3 a +
2166
+
2167
+ −4m
2168
+ λ5 +
2169
+ �5m
2170
+ λ5 − m2
2171
+ λ6
2172
+
2173
+ sin2 θ
2174
+
2175
+ a3 + O(a5)
2176
+ (3.98d)
2177
+ γ(λ, θ) =
2178
+
2179
+ −m sin2 θ
2180
+ 2λ3
2181
+
2182
+ a2 +
2183
+ �9m sin2 θ
2184
+ 4λ5
2185
+ +
2186
+
2187
+ −21m
2188
+ 8λ5 − m2
2189
+ 4λ6
2190
+
2191
+ sin4 θ
2192
+
2193
+ a4 + O(a6)
2194
+ (3.98e)
2195
+ δ(λ, θ) = −5m cosθ sin2 θ
2196
+ 4λ4
2197
+ a3 + O(a5)
2198
+ (3.98f)
2199
+ Comparing with [28], we find agreement for R which corresponds to their areal coordinate r. Calculation of the metric
2200
+
2201
+ 12
2202
+ components gab using (3.98) gives us
2203
+ guu(λ, θ) = −1 + 2m
2204
+ λ +
2205
+ ��3m
2206
+ λ3 + m2
2207
+ λ4
2208
+
2209
+ sin2 θ − 2m
2210
+ λ3
2211
+
2212
+ a2
2213
+ +
2214
+ �2m
2215
+ λ5 −
2216
+ �10m
2217
+ λ5
2218
+ + 4m2
2219
+ λ6
2220
+
2221
+ sin2 θ +
2222
+ �35m
2223
+ 4λ5 + 23m2
2224
+ 4λ6 + 9m3
2225
+ 10λ7
2226
+
2227
+ sin4 θ
2228
+
2229
+ a4 + O(a6)
2230
+ (3.99a)
2231
+ guλ(λ, θ) = −1
2232
+ (3.99b)
2233
+ guθ(λ, θ) =
2234
+ ��3m
2235
+ λ2
2236
+
2237
+ a2 +
2238
+
2239
+ −5m
2240
+ λ4 +
2241
+ �35m
2242
+ λ4
2243
+ + 23m2
2244
+ 10λ5
2245
+
2246
+ sin2 θ
2247
+
2248
+ a4
2249
+
2250
+ sin θ cos θ + O(a6)
2251
+ (3.99c)
2252
+ guφ(λ, θ) =
2253
+ ��
2254
+ −2m
2255
+ λ
2256
+
2257
+ a +
2258
+ �4m
2259
+ λ3 −
2260
+ �5m
2261
+ λ3 + m2
2262
+ λ4
2263
+
2264
+ sin2 θ
2265
+
2266
+ a3
2267
+
2268
+ sin2 θ + O(a5)
2269
+ (3.99d)
2270
+ gθθ(λ, θ) = λ2 +
2271
+
2272
+ −m sin2 θ
2273
+ λ
2274
+
2275
+ a2 +
2276
+ � 9m
2277
+ 2λ3 sin2 θ −
2278
+ �21m
2279
+ 4λ3 + 3m2
2280
+ 20λ4
2281
+
2282
+ sin4 θ
2283
+
2284
+ a4 + O(a6)
2285
+ (3.99e)
2286
+ gθφ(λ, θ) =
2287
+
2288
+ −5m sin3 θ cos θ
2289
+ 2λ2
2290
+
2291
+ a3 + O(a5)
2292
+ (3.99f)
2293
+ gφφ(λ, θ) =
2294
+
2295
+ λ2 +
2296
+ �m sin2 θ
2297
+ λ
2298
+
2299
+ a2 +
2300
+
2301
+ − 9m
2302
+ 2λ3 sin2 θ +
2303
+ �21m
2304
+ 4λ3 + 17m2
2305
+ 20λ4
2306
+
2307
+ sin4 θ
2308
+
2309
+ a4
2310
+
2311
+ sin2 θ + O(a6)
2312
+ (3.99g)
2313
+ Eqs.(3.99) constitute our final expression for the slowly
2314
+ rotating stationary and axially symmetric (Kerr) metric
2315
+ adapted to null coordinates which asymptotically match
2316
+ an inertial Bondi frame. At difference of all previous ap-
2317
+ proaches, it was obtained as an explicit solution of the
2318
+ Einstein equations. After comparison with [28], we find
2319
+ agreement up to a typo in their equation for gθφ. We also
2320
+ note care should be taken when comparing [28]’s expres-
2321
+ sions with ours. First, [28] present a Bondi-Sachs form of
2322
+ the metric, while we have an affine-null metric approach-
2323
+ ing a Bondi frame, the difference is in the choice of radial
2324
+ coordinate, and the two agree only up to O(λ−4) with one
2325
+ another. Second, [28] make a large λ expansion while we
2326
+ make a small a expansion, this results in powers of λ−k
2327
+ absorbed by order symbols in [28]. A slowly rotating ver-
2328
+ sion of the Kerr metric in null affine coordinates at second
2329
+ order in a was also obtained by Dozmorov who made a
2330
+ null tetrad rotations starting with the Kerr metric as ex-
2331
+ pressed in Boyer-Lindquist coordinates [35]. In the next
2332
+ section we show an alternative procedure to recover the
2333
+ slowly rotating Kerr metric components as expressed in
2334
+ (3.99) by doing appropriate coordinates transformations.
2335
+ IV.
2336
+ APPROXIMATED AFFINE-NULL METRIC
2337
+ DERIVED FROM THE KERR-METRIC
2338
+ Here, starting with the Kerr metric expressed in Boyer-
2339
+ Lindquist coordinates (BL) {ˆt, ˆr, ˆθ, ˆφ}, we present an
2340
+ explicit transformation to affine-null coordinates up to
2341
+ fourth order in a. The Kerr metric in BL coordinates
2342
+ reads:
2343
+ ds2 = gˆtˆtdˆt2+gˆtˆφdˆtdˆφ+gˆrˆrdˆr2+gˆrˆrdˆr2+gˆθˆθdˆθ2+g ˆφˆφdˆφ2;
2344
+ (4.1)
2345
+ with
2346
+ gˆtˆt = −
2347
+
2348
+ 1 − 2mˆr
2349
+ Σ
2350
+
2351
+ ,
2352
+ (4.2)
2353
+ gˆt ˆφ = −2maˆr sin2 ˆθ
2354
+ Σ
2355
+ ,
2356
+ (4.3)
2357
+ gˆrˆr = Σ
2358
+ ∆,
2359
+ (4.4)
2360
+ gˆθˆθ = Σ,
2361
+ (4.5)
2362
+ g ˆφˆφ =
2363
+
2364
+ ˆr2 + a2 + 2ma2ˆr sin2 ˆθ
2365
+ Σ
2366
+
2367
+ sin2 ˆθ,
2368
+ (4.6)
2369
+ with ∆ = ˆr2 − 2mˆr + a2 and Σ = ˆr2 + a2 cos2 ˆθ. The u
2370
+ null coordinate must satisfy the eikonal equation,
2371
+ gab∇au∇bu = 0,
2372
+ (4.7)
2373
+ Inspired by [28], we propose the following expansion for
2374
+ u,
2375
+ u = ˆt − ˆr − 2m ln
2376
+ � ˆr − 2m
2377
+ 2m
2378
+
2379
+ +
2380
+
2381
+
2382
+ i=1
2383
+ fi(ˆr, ˆθ)ai.
2384
+ (4.8)
2385
+ Note that for a = 0 this expression reduces to the stan-
2386
+ dard outgoing Schwarzschild null coordinate. By replac-
2387
+ ing (4.8) into (4.7), we obtain a set of differential equa-
2388
+ tions for fi(ˆr, ˆθ) that can be solved iteratively. Conserv-
2389
+ ing terms up to fourth order in a we find that only the
2390
+ even coefficients f2n(ˆr, ˆθ) are non–vanishing with:
2391
+
2392
+ 13
2393
+ f2(ˆr, ˆθ) =
2394
+ 5ˆr − 2m
2395
+ 4ˆr(2m − ˆr) + cos 2ˆθ
2396
+ 4ˆr
2397
+ − ln(1 − 2m
2398
+ ˆr )
2399
+ 2m
2400
+ ,(4.9)
2401
+ f4(ˆr, ˆθ) = (2ˆr + m)
2402
+ 16ˆr4
2403
+ sin4(2ˆθ) − 3 ln(1 − 2m
2404
+ ˆr )
2405
+ 8m3
2406
+ −4m2 − 9mˆr + 3ˆr2
2407
+ 4m2ˆr(ˆr − 2m)2 ,
2408
+ (4.10)
2409
+ Similarly, affine-null coordinates {λ, θ, φ} can be ob-
2410
+ tained from the requirements
2411
+ gab∇au∇bλ = −1,
2412
+ (4.11a)
2413
+ gab∇au∇bθ = gab∇au∇bφ = 0,
2414
+ (4.11b)
2415
+ by assuming relations of the form:
2416
+ λ = ˆr +
2417
+
2418
+
2419
+ i=1
2420
+ ˆΛi(ˆθ, ˆr)ai,
2421
+ (4.12)
2422
+ θ = ˆθ +
2423
+
2424
+
2425
+ i=1
2426
+ ˆΘi(ˆθ, ˆr)ai,
2427
+ (4.13)
2428
+ φ = ˆφ +
2429
+
2430
+
2431
+ i=1
2432
+ ˆΦi(ˆθ, ˆr)ai,
2433
+ (4.14)
2434
+ and replacing into the set (4.11), the coefficients functions
2435
+ ˆΛi, ˆΘi, ˆΦi can be obtained in the same way as u. After
2436
+ that, the resulting relations can be inverted in order to
2437
+ express the BL coordinates in terms of the affine-null
2438
+ coordinates. Following these steps up to fourth order,
2439
+ the final transformation coordinates reads:
2440
+ ˆt = u + λ + 2m ln( λ
2441
+ 2m − 1) +
2442
+ �ln(1 − 2m
2443
+ λ )
2444
+ 2m
2445
+ + 3 m cos(2 θ) + 4 λ − 3 m
2446
+ (4 λ − 8 m) λ
2447
+
2448
+ a2
2449
+ +
2450
+
2451
+ −m
2452
+
2453
+ 175 λ2 − 224 mλ − 72 m2�
2454
+ (cos (2 θ))2
2455
+ 320 (λ − 2 m)2 λ4
2456
+ − m
2457
+
2458
+ 25 λ2 + 64 mλ + 72 m2�
2459
+ cos (2 θ)
2460
+ 160 (λ − 2 m)2 λ4
2461
+ +3 ln
2462
+
2463
+ 1 − 2 m
2464
+ λ
2465
+
2466
+ 8 m3
2467
+ + 240 λ5 − 720 λ4m + 320 λ3m2 + 225 λ2m3 − 96 λ m4 + 72 m5
2468
+ 320 m2λ4 (λ − 2 m)2
2469
+
2470
+ a4 + O(a6)
2471
+ (4.15)
2472
+ ˆr = λ − (λ + m) sin2 θ
2473
+ 2λ2
2474
+ a2 +
2475
+ �sin2 θ(5 cos 2θ + 3))
2476
+ 16λ3
2477
+ + m sin2 θ(7 cos 2θ + 1)
2478
+ 16λ4
2479
+ − m2 sin4 θ
2480
+ 5λ5
2481
+
2482
+ a4 + O(a6)
2483
+ (4.16)
2484
+ ˆθ = θ − sin (2 θ)
2485
+ 4 λ2
2486
+ a2 + sin (2 θ) (3 λ cos (2 θ) + m cos(2 θ) − m)
2487
+ 16 λ5
2488
+ a4 + O(a6)
2489
+ (4.17)
2490
+ ˆφ = φ +
2491
+ � 1
2492
+ λ + ln(1 − 2m
2493
+ λ )
2494
+ 2m
2495
+
2496
+ a +
2497
+ �ln(1 − 2m
2498
+ λ )
2499
+ 4m
2500
+ + m(2m + 5λ) cos 2θ
2501
+ 8(λ − 2m)λ4
2502
+ −6m4 − m3λ + 8m2λ2 + 12mλ3 − 12λ4
2503
+ 24m2(λ − 2m)λ4
2504
+
2505
+ a3 + O(a5)
2506
+ (4.18)
2507
+ Finally, with these transformations in hand, we obtain
2508
+ the same metric components in affine-null coordinates
2509
+ up to fourth order in a as given in (3.99) in the previous
2510
+ Section.
2511
+ V.
2512
+ LOCALIZING THE EVENT HORIZON AND
2513
+ ERGOSPHERE IN AFFINE-NULL CORDINATES
2514
+ In this Section we show that the affine-null coordinates
2515
+ for the slowly rotating Kerr metric cover the ergosphere
2516
+ and the (past) event horizon r+. In order to find them in
2517
+ a consistent way, they must be localized at O(a4). Recall
2518
+ that in BL coordinates the Kerr metric has the external
2519
+
2520
+ 14
2521
+ ergosphere placed at
2522
+ ˆrerg =m +
2523
+
2524
+ m2 − a2 cos2 ˆθ
2525
+ =2m − a2 cos2 ˆθ
2526
+ m
2527
+ − a4 cos4 ˆθ
2528
+ 8m3
2529
+ + O(a6),
2530
+ (5.1)
2531
+ and the event horizon at
2532
+ r+ = m +
2533
+
2534
+ m2 − a2 = 2m − a2
2535
+ 2m − a4
2536
+ 8m3 + O(a6). (5.2)
2537
+ The boundary of the external ergosphere is obtained by
2538
+ looking for the timelike surface Γ where the stationary
2539
+ Killing vector field ∂u becomes a null vector field that is
2540
+ where
2541
+ guu|Γ = 0.
2542
+ (5.3)
2543
+ Taking into account the expression for guu as found in the
2544
+ first equation of (3.99), the ergosphere will be located at
2545
+ a given λ = λerg(θ), with
2546
+ λerg(θ) =
2547
+ 2
2548
+
2549
+ i=0
2550
+ λerg[2i](θ)a2i + O(a6).
2551
+ (5.4)
2552
+ where the even expansion is a consequence of the symme-
2553
+ try assumption of Sec. II. Inserting (5.4) into (5.3), and
2554
+ after re-expanding in powers of a we find
2555
+ λerg(θ) =2 m −
2556
+
2557
+ 7 cos2 θ − 3
2558
+
2559
+ 8m
2560
+ a2
2561
+
2562
+
2563
+ 51 cos4 θ − 2 cos2 θ + 31
2564
+
2565
+ 640 m3
2566
+ a4 + O(a6),
2567
+ (5.5)
2568
+ which gives the location of the (external) ergosphere in
2569
+ affine-null coordinates.
2570
+ By replacing (5.5) into (4.16),
2571
+ and after a re-expansion in powers of a it can be checked
2572
+ that the standard fourth order expression for the BL ex-
2573
+ pression of the ergosphere as given by (5.1) is recovered.
2574
+ Similarly, for the (Killing) event horizon we search
2575
+ a null surface Σ described in affine-null coordinates by
2576
+ Σ(λ, θ) = λ − λH(θ) = 0.
2577
+ Hence, its normal vector
2578
+ Na = ∇aΣ must satisfy N aNa = 0 which implies the
2579
+ following differential equation for λH(θ) = 0,
2580
+ gabNaNb = W + 2W θ ∂λH(θ)
2581
+ ∂θ
2582
+ + hφφ
2583
+ R2
2584
+ �∂λH(θ)
2585
+ ∂θ
2586
+ �2
2587
+ = 0.
2588
+ (5.6)
2589
+ Let us assume an expansion for λH(θ) similar to (5.4),
2590
+ i.e.
2591
+ λH(θ) =
2592
+ 2
2593
+
2594
+ i=0
2595
+ λH[2i](θ)a2i + O(a6);
2596
+ (5.7)
2597
+ with λH[0] = 2m (the Schwarzschild value for the loca-
2598
+ tion of the horizon).
2599
+ Introducing (5.7) into (5.6); re-
2600
+ expanding again in powers of a, we find (omitting the
2601
+ O(a6) term)
2602
+ 0 =
2603
+ �λH[2]
2604
+ 2m + 3 cos2 θ + 1
2605
+ 16m2
2606
+
2607
+ a2
2608
+ +
2609
+
2610
+ (λH[2],θ)2
2611
+ 4m2
2612
+ − 3 sin θ cos θ
2613
+ 8m3
2614
+ λH[2],θ −
2615
+ λ2
2616
+ H[2]
2617
+ 4m2 − 3(cos2 θ + 1)
2618
+ 16m3
2619
+ λH[2] − 127 cos4 θ − 320m3λH[4] − 84 cos2 θ − 3
2620
+ 640m4
2621
+
2622
+ a4.
2623
+ (5.8)
2624
+ So that solving for the coefficient λH[2] and λH[4] gives
2625
+ us
2626
+ λH(θ) =2m − (1 + 3 cos2 θ)
2627
+ 8m
2628
+ a2
2629
+ + (29 cos4 θ − 78 cos2 θ − 31)
2630
+ 640m3
2631
+ a4 + O(a6),
2632
+ (5.9)
2633
+ which gives the location of the (past) event horizon in
2634
+ affine-null coordinates. By replacing into (4.16) and af-
2635
+ ter a reexpansion in a up to fourth order, the well known
2636
+ value (5.2) for the BL radial coordinate of the event hori-
2637
+ zon is recovered. At this location, the resulting compo-
2638
+ nents of the metric are regular.
2639
+ VI.
2640
+ SUMMARY
2641
+ We have derived high-order slow rotation approxima-
2642
+ tion of the Kerr metric in affine-null coordinates.
2643
+ To
2644
+ achieve this aim a metric in affine-null coordinates was
2645
+ expanded off a spherically symmetric background metric
2646
+ that corresponds to a Schwarzschild metric in outgoing
2647
+ Eddington Finkelstein coordinates. This quasi-spherical
2648
+ expansion was done with respect to a general smallness
2649
+ parameter ε. Subject to stationarity and axial symmetry
2650
+
2651
+ 15
2652
+ the perturbations did not depend on the u and φ coor-
2653
+ dinate.
2654
+ Moreover, requiring even parity of the Komar
2655
+ integral of stationary (giving the mass of the system)
2656
+ and odd parity of the Komar integral of axial symme-
2657
+ try (giving the angular momentum of the system), we
2658
+ argued that on the one hand the metric functions γ, R,
2659
+ W θ and W have only even perturbations in ε while on
2660
+ the other hand the metric fields δ and W φ have only odd
2661
+ perturbations in ε. This fact significantly simplifies the
2662
+ integration of the perturbation equations resulting form
2663
+ the quasi-spherical expansion of the Ricci tensor. In ad-
2664
+ dition, we find that the integration of the perturbation
2665
+ equations follows an alternative hierarchical structure.
2666
+ Meaning with the spherically symmetric background so-
2667
+ lution at hand, the linear perturbations only involve the
2668
+ functions δ and W φ and its integration provides (after
2669
+ application of the boundary condition of an asymptotic
2670
+ inertial observer) one free integration constant B.
2671
+ At
2672
+ next order, the quadratic perturbations turn out to be
2673
+ a linear combination of the derivatives of functions γ,
2674
+ R, W θ and W together with nonlinear terms containing
2675
+ the integration constants A of the background model and
2676
+ the free integration constant B of the linear perturbation.
2677
+ Their integration also yields a free integration constant,
2678
+ C. Following up the next order, there only differential
2679
+ equations involving the cubic perturbations of δ and W φ
2680
+ as well as the integration constants A, B and C charac-
2681
+ terizing the lower order perturbations. This alternative
2682
+ scheme between the perturbations of (δ, W φ) and those of
2683
+ (γ, R, W θ, W) continues up to any order and is in fact a
2684
+ result of the symmetry assumptions. A common feature
2685
+ in solving for the even and odd-parity modes of ε, is that
2686
+ at any order there is a fourth order master equation for ei-
2687
+ ther the perturbation in γ or the perturbation in δ. With
2688
+ the solution of this master equation, the remaining per-
2689
+ turbations can be solve by mere integration. After having
2690
+ obtained the perturbed solution and calculation of the
2691
+ Komar mass and Komar angular momentum, the arising
2692
+ free integration constants A, B, C, ... can expressed by the
2693
+ Komar mass and Komar angular momentum or by mass
2694
+ and specific angular momentum. Hence, the solution de-
2695
+ pends only on two free physical parameters. Since the
2696
+ Komar angular momentum is O(ε), it turns out that the
2697
+ formal expansion parameter ε relates to the specific angu-
2698
+ lar momentum and the previously made quasi-spherical
2699
+ approximation is in fact a slow rotation approximation,
2700
+ like those of Hartle and Thorne [30, 36]. By successively
2701
+ solving Einstein equations, we thus derived a slow rota-
2702
+ tion approximation of the Kerr-metric up to fourth order
2703
+ in the specific angular momentum. This solution is fur-
2704
+ ther verified for correctness using a ’standard’ approach
2705
+ by obtaining a different representation of a given metric
2706
+ in another coordinate chart via a coordinate transforma-
2707
+ tion. The slowly rotating Kerr metric presented here also
2708
+ obeys the peeling property, which can be seen considering
2709
+ the Weyl scalars in (6.1)
2710
+ Ψ0 =
2711
+ �3ma2
2712
+ λ5
2713
+ + i15ma3
2714
+ λ6
2715
+ cos θ
2716
+
2717
+ sin2 θ + O(λ−7)
2718
+ (6.1a)
2719
+ Ψ1 = i3
2720
+
2721
+ 2ma
2722
+ 2λ4
2723
+ sin θ + O(λ−5)
2724
+ (6.1b)
2725
+ Ψ2 = m
2726
+ λ3 + i3ma cosθ
2727
+ λ4
2728
+ + O(λ−5)
2729
+ (6.1c)
2730
+ Ψ3 = i3
2731
+
2732
+ 2ma
2733
+ 4λ4
2734
+ sin θ + O(λ−5)
2735
+ (6.1d)
2736
+ Ψ4 = 3
2737
+
2738
+ 2ma2
2739
+ 4λ5
2740
+ sin2 θ + O(λ−6)
2741
+ (6.1e)
2742
+ Moreover it is easily checked that the (only) conserved
2743
+ Newman Penrose constant [21] vanishes [28].
2744
+ What is is interesting to remark is that up to the con-
2745
+ sidered order of approximation of our work and those of
2746
+ [28], the small a expansion and the large λ expansion
2747
+ coincide. It would be interesting to see up until which
2748
+ order this is the case. Such analysis might give insight
2749
+ on the validity and universality of general small param-
2750
+ eter expansions of the Kerr spacetime in relation to null
2751
+ coordinates.
2752
+ It may also give insight if a closed form
2753
+ solution of the Kerr metric with a surface forming null
2754
+ coordinate can be obtained at all. The method presented
2755
+ here offers the possibility to calculate any type of approx-
2756
+ imate rotating null-metric solution that is stationary, ax-
2757
+ ially symmetric and has a known spherically symmetric
2758
+ background, like e.g.
2759
+ those to describe compact mat-
2760
+ ter systems or with a cosmological constant. Indeed, the
2761
+ study presented here (solving the characteristic equations
2762
+ in this affine-null, metric formulation for vacuum space-
2763
+ times) is the natural starting point for further studying
2764
+ matter system under the given symmetry assumptions.
2765
+ Some of such questions we are currently investigating.
2766
+ Acknowledgements
2767
+ The authors thanks J. Winicour, L. Lehner, N. Ster-
2768
+ gioulas, E. M¨uller and G. Dotti for discussions at (early)
2769
+ stages of the project. T.M acknowledges financial sup-
2770
+ port from the FONDECYT de iniciaci´on 2019 (Project
2771
+ No. 11190854) of the ”Agencia Nacional de Investigaci´on
2772
+ y Desarrollo” in Chile. E.G gratefully acknowledges the
2773
+ hospitality extended to him during his stay at the Fac-
2774
+ ultad de Ingenier´ıa, Universidad Diego Portales and the
2775
+ financial support from CONICET and SeCyT-UNC.
2776
+ Appendix A: Useful Relations between Legendre
2777
+ Polynomials
2778
+ For completeness, we list some properties of the Legen-
2779
+ dre differential equations and relations between the Leg-
2780
+ endre polynomials.
2781
+
2782
+ 16
2783
+ The Legendre differential equation for the Legendre
2784
+ polynomials Pℓ(y) is
2785
+ d
2786
+ dy
2787
+
2788
+ (1 − y2)dPℓ
2789
+ dy
2790
+
2791
+ + ℓ(ℓ + 1)Pℓ = 0
2792
+ (A1)
2793
+ where the Legendre Polynomials Pℓ(y) are defined
2794
+ Pℓ(y) =
2795
+ 1
2796
+ 2ℓℓ!
2797
+ dℓ
2798
+ dyℓ (y2 − 1)ℓ
2799
+ (A2)
2800
+ The associated Legendre differential equation is
2801
+ d
2802
+ dy
2803
+
2804
+ (1 − y2)dP m
2805
+
2806
+ dy
2807
+
2808
+ +
2809
+
2810
+ ℓ(ℓ + 1) −
2811
+ m2
2812
+ 1 − y2
2813
+
2814
+ P m
2815
+
2816
+ = 0 (A3)
2817
+ where P m
2818
+ ℓ (y) are the associated Legendre polynomials,
2819
+ defined via
2820
+ P m
2821
+ ℓ (y) =(−)m(1 − y2)m/2 dmPℓ(y)
2822
+ dxm
2823
+ =(−)m
2824
+ 2ℓℓ! (1 − y2)m/2 dℓ+m
2825
+ dyℓ+m (y2 − 1)ℓ
2826
+ (A4)
2827
+ which also shows P 0
2828
+ ℓ (y) = Pℓ(y). From these definitions,
2829
+ some useful identities can be derived
2830
+ d
2831
+ dy
2832
+
2833
+ (1 − y2)P 2
2834
+ ℓ (y)
2835
+
2836
+ = [ℓ(ℓ+1)−2](1−y2)1/2P 1
2837
+ ℓ (y) (A5)
2838
+ d
2839
+ dy
2840
+
2841
+ (1 − y2)2 dP 2
2842
+
2843
+ dy
2844
+
2845
+ 1 − y2
2846
+ − 2P 2
2847
+
2848
+ = ℓ(ℓ + 1)(ℓ + 2)(ℓ − 1)Pℓ(y)
2849
+ (A6)
2850
+ P 1
2851
+
2852
+ = −(1 − y2)1/2 dP 0
2853
+
2854
+ dy
2855
+ (A7)
2856
+ P 2
2857
+
2858
+ = (1 − y2)d2P 0
2859
+
2860
+ dy2
2861
+ (A8)
2862
+ d
2863
+ dy
2864
+
2865
+ (1 − y2)2 d
2866
+ dy
2867
+ P 1
2868
+
2869
+ (1 − y2)1/2
2870
+
2871
+ = [2−ℓ(ℓ+1)](1−y2)1/2P 1
2872
+
2873
+ (A9)
2874
+ d
2875
+ dy(1 − y2)1/2P 1
2876
+ ℓ = ℓ(ℓ + 1)P 0
2877
+
2878
+ (A10)
2879
+ Appendix B: Komar charges
2880
+ Depending on the Killing vector Xa ∈ {∂u, ∂φ}, we
2881
+ take the Komar charges to be
2882
+ KX = −kX
2883
+
2884
+
2885
+ ∇[aXb]dΣab
2886
+ (B1)
2887
+ with kX = 1, −1/2 for a timelike ( e.g.
2888
+ ∂u) or rota-
2889
+ tional Killing vector (e.g. ∂φ), respectively. Consider the
2890
+ general null metric with the nonzero contravariant com-
2891
+ ponents g01, g11, g1A and gAB. The corresponding line
2892
+ element is
2893
+ gabdxadxb = (g11 + gABg1Ag1B)
2894
+ �dx0
2895
+ g01
2896
+ �2
2897
+ + 2
2898
+ �dx0
2899
+ g01
2900
+
2901
+ dx1
2902
+ − 2gABg1AdxB
2903
+ �dx0
2904
+ g01
2905
+
2906
+ + gABdxAdxB
2907
+ (B2)
2908
+ where gACgCB = δB
2909
+ A. Defining the null vectors
2910
+ l = −dx0 = −g01∂1
2911
+ (B3)
2912
+ and
2913
+ n = −1
2914
+ 2
2915
+ g11
2916
+ (g01)2 dx0 + dx1
2917
+ g01 = ∂0 + 1
2918
+ 2
2919
+ g11
2920
+ g01 ∂1 + g1A
2921
+ g01 ∂A (B4)
2922
+ which obey lana + 1 = lala = nana = 0, the surface
2923
+ element follows as
2924
+ dΣab = 2k[anb]
2925
+
2926
+ det(gAB)dx2dx3
2927
+ (B5)
2928
+ with xA = (x2, x3) being any angular coordinates for the
2929
+ units sphere. Setting gAB = R2hAB with hAB having
2930
+ the determinant of the unit sphere metric qAB, q(xC) :=
2931
+ det(hAB). The corresponding volume element is defined
2932
+ as d2q := √qdx2dx3 and we have � d2q = 4π. Hence,
2933
+ dΣab = 2l[anb]R2d2q .
2934
+ This allows us to write the Komar integal as
2935
+ K(X) = −kX
2936
+
2937
+
2938
+ (2lanb∂[aXb])R2d2q ,
2939
+ (B6)
2940
+ Since
2941
+ 2lanb∂[aXb] = 2l[anb]Xb,a
2942
+ (B7)
2943
+ = (lanb − lbna)Xb,a
2944
+ (B8)
2945
+ = l1(nbXb,1) − l1(nbX1,b)
2946
+ (B9)
2947
+ = −g01[(nbXb,1) − (nbX1,b)] , (B10)
2948
+ we have
2949
+ K(X) = kX
2950
+
2951
+ � �
2952
+ (nbXb,1) − (nbX1,b)
2953
+
2954
+ g01R2d2q, (B11)
2955
+ Taking the Killing vector to be X = Xa∂a and specifica-
2956
+ tion to an affine null metric
2957
+ g01 = ǫ , g1A = ǫW A , g11 = W ,
2958
+ g0A = −R2hABW B , gAB = R2hAB
2959
+ (B12)
2960
+ and ǫ2 = 1 gives us
2961
+ 2lanb∂[aXb] = −
2962
+
2963
+ W,1 − R2hABW AW B
2964
+ ,1
2965
+
2966
+ X0
2967
+ + R2�
2968
+ hABW B
2969
+ ,1 XA − 2hABW BXA
2970
+ ,1
2971
+
2972
+ + ǫ(X1
2973
+ ,1 − X0
2974
+ ,0) − WX0
2975
+ ,1 − W AX0
2976
+ ,A
2977
+ (B13)
2978
+
2979
+ 17
2980
+ Assuming the timelike Killing vector X = ∂0 gives us
2981
+ 2lanb∂[aXb] = −
2982
+
2983
+ W,1 − R2hABW AW B
2984
+ ,1 .
2985
+
2986
+ Thus for the above form of the Killing vector we have the
2987
+ related Komar charge using kX = 1
2988
+ K(∂0) = 1
2989
+
2990
+ � �
2991
+ − ǫ
2992
+
2993
+ W,1 − R2hABW AW B
2994
+ ,1
2995
+ ��
2996
+ R2d2q.
2997
+ (B14)
2998
+ With the rotational Killing X = ∂3, we have
2999
+ 2lanb∂[aXb] = R2h3BW B
3000
+ ,1
3001
+ so that the Komar charge is with kX = − 1
3002
+ 2
3003
+ K(∂3) = − ǫ
3004
+ 16π
3005
+ � �
3006
+ R4h3BW B
3007
+ ,1
3008
+
3009
+ d2q.
3010
+ (B15)
3011
+ [1] H. Bondi, Nature (London) 186, 535 (1960).
3012
+ [2] H. Bondi, M. G. J. van der Burg, and A. W. K. Metzner,
3013
+ Proceedings of the Royal Society of London Series A 269,
3014
+ 21 (1962).
3015
+ [3] R. K. Sachs, Proceedings of the Royal Society of London
3016
+ Series A 270, 103 (1962).
3017
+ [4] R. P. Kerr, Phys. Rev. Lett. 11, 237 (1963).
3018
+ [5] E. T. Newman, E. Couch, K. Chinnapared, A. Exton,
3019
+ A. Prakash, and R. Torrence, Journal of Mathematical
3020
+ Physics 6, 918 (1965).
3021
+ [6] P. Jordan, J. Ehlers, and R. K. Sachs, General Relativity
3022
+ and Gravitation 45, 2691 (2013).
3023
+ [7] R. G´omez, L. Lehner, R. L. Marsa, and J. Winicour,
3024
+ Phys. Rev. D 57, 4778 (1998), gr-qc/9710138.
3025
+ [8] J. Winicour, Living Reviews in Relativity 15, 2 (2012).
3026
+ [9] T. M¨adler and J. Winicour, Scholarpedia 11, 33528
3027
+ (2016), 1609.01731.
3028
+ [10] G. Barnich and C. Troessaert, Journal of High Energy
3029
+ Physics 2010, 62 (2010), 1001.1541.
3030
+ [11] S. Pasterski, A. Strominger, and A. Zhiboedov, Journal
3031
+ of High Energy Physics 2016, 53 (2016), 1502.06120.
3032
+ [12] T. M¨adler and J. Winicour, Classical and Quantum
3033
+ Gravity 33, 175006 (2016), 1605.01273.
3034
+ [13] D. A. Nichols,
3035
+ Phys. Rev. D
3036
+ 95,
3037
+ 084048 (2017),
3038
+ 1702.03300.
3039
+ [14] T. M¨adler and J. Winicour, Classical and Quantum
3040
+ Gravity 35, 035009 (2018), 1708.08774.
3041
+ [15] T. M¨adler and J. Winicour, Classical and Quantum
3042
+ Gravity 36, 095009 (2019), 1811.04711.
3043
+ [16] J. Winicour, Phys. Rev. D 87, 124027 (2013), 1303.6969.
3044
+ [17] T. M¨adler, Phys. Rev. D 99, 104048 (2019), 1810.04743.
3045
+ [18] E. Gallo, C. Kozameh, T. M¨adler, O. M. Moreschi, and
3046
+ A. Perez, Phys. Rev. D 104, 084048 (2021), 2107.10120.
3047
+ [19] J. A. Crespo, H. P. de Oliveira, and J. Winicour, Phys.
3048
+ Rev. D 100, 104017 (2019), 1910.03439.
3049
+ [20] O. Baake and T. M¨adler, in prep. (2023).
3050
+ [21] N. T. Bishop and L. R. Venter, Phys. Rev. D 73, 084023
3051
+ (2006), gr-qc/0506077.
3052
+ [22] M. A. Arga˜naraz and O. M. Moreschi, Phys. Rev. D 104,
3053
+ 024049 (2021).
3054
+ [23] S. J. Fletcher and A. W. C. Lun, Classical and Quantum
3055
+ Gravity 20, 4153 (2003).
3056
+ [24] S. Jahanur Hoque and A. Virmani,
3057
+ arXiv e-prints
3058
+ arXiv:2108.01098 (2021), 2108.01098.
3059
+ [25] S. A. Hayward, Phys. Rev. Lett. 92, 191101 (2004), gr-
3060
+ qc/0401111.
3061
+ [26] M. A. Arga˜naraz and O. M. Moreschi, Phys. Rev. D 105,
3062
+ 084012 (2022).
3063
+ [27] E. Gallo and O. M. Moreschi, Phys. Rev. D 89, 084009
3064
+ (2014), 1404.2475.
3065
+ [28] S. Bai, Z. Cao, X. Gong, Y. Shang, X. Wu, and Y. K.
3066
+ Lau, Phys. Rev. D 75, 044003 (2007), gr-qc/0701171.
3067
+ [29] X. Gong, Y. Shang, S. Bai, Z. Cao, Z. Luo, and Y. K.
3068
+ Lau, Phys. Rev. D 76, 107501 (2007).
3069
+ [30] J. B. Hartle, Astrophys. J. 150, 1005 (1967).
3070
+ [31] J.
3071
+ Tafel,
3072
+ Class.
3073
+ Quant.
3074
+ Grav.
3075
+ 39,
3076
+ 115013
3077
+ (2022),
3078
+ 2105.09372.
3079
+ [32] R. G´omez, S. Husa, and J. Winicour, Phys. Rev. D 64,
3080
+ 024010 (2001), gr-qc/0009092.
3081
+ [33] M. G. J. van der Burg, Proceedings of the Royal Society
3082
+ of London Series A 294, 112 (1966).
3083
+ [34] T. M¨adler, Phys. Rev. D 87, 104016 (2013), 1212.3316.
3084
+ [35] I. M. Dozmorov, Fizika 18, 95 (1975).
3085
+ [36] J. B. Hartle and K. S. Thorne, Astrophys. J. 153, 807
3086
+ (1968).
3087
+
9NE4T4oBgHgl3EQfdgy1/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
AdFLT4oBgHgl3EQfEy_H/content/tmp_files/2301.11985v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
AdFLT4oBgHgl3EQfEy_H/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
BNE5T4oBgHgl3EQfTA98/content/tmp_files/2301.05533v1.pdf.txt ADDED
@@ -0,0 +1,421 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+
3
+
4
+
5
+
6
+
7
+ Investigation of radiation hardness of silicon semiconductor
8
+ detectors under irradiation with fission products of 252Cf
9
+ nuclide.
10
+ N V Bazlov1,2, A V Derbin1, I S Drachnev1, I M Kotina1, O I Konkov1,3, I S
11
+ Lomskaya1, M S Mikulich1, V N Muratova1, D A Semenov1, M V Trushin1 and
12
+ E V Unzhakov1
13
+ 1 NRC "Kurchatov Institute" - PNPI, Gatchina, Russia
14
+ 2 Saint-Petersburg State University, Universitetskaya nab. 7/9, St. Petersburg, Russia
15
+ 3 Ioffe Physical-Technical Institute of the Russian Academy of Sciences, St.
16
+ Petersburg, Russia
17
+
18
+ e-mail: [email protected]
19
+ Abstract. Influence of the prolonged irradiation by fission products of 252Cf radionuclide on
20
+ the operational parameters of silicon-lithium Si(Li) p-i-n detectors, Si surface barrier detectors
21
+ and Si planar p+n detector was investigated. The obtained results revealed a linear shift of the
22
+ fission fragment peaks positions towards the lower energies with increase of the irradiation
23
+ dose for all investigated detectors. The rate of the peaks shift was found to depend strongly on
24
+ the detector type and the strength of the electric field in the detector’s active region, but not on
25
+ the temperature of irradiation (room or liquid nitrogen temperature). Based on the obtained
26
+ results, the possibility of integration of the investigated types of Si semiconductor detectors in
27
+ a radionuclide neutron calibration source is considered.
28
+ 1. Introduction
29
+ Heavy nuclides subjected to spontaneous fission decay accompanied by emission of several fast
30
+ neutrons can be utilized as a compact neutron calibration source. The most common spontaneous
31
+ fission source is 252Cf which undergoes α-decay and spontaneous fission with a branching ratio of
32
+ 97:3, whereas each spontaneous fission event liberates 3.8 neutrons and 9.7 gamma-ray photons on
33
+ average [1]. The timing of the moment of neutron production can be fixed by detecting the fission
34
+ fragments signal with a semiconductor detector.
35
+ Semiconductor detectors possess sufficiently high energy resolution for detection of the high-
36
+ energy heavy ions. The main obstacle for the integration of such detectors in the neutron calibration
37
+ source could be their limited lifetime under the influence of the nuclide radiation [2]. Degradation of
38
+ the detector’s operational parameters effectively proceeds just in case of irradiation by alpha particles
39
+ and fission fragments (FF), which are capable of transferring a significant fraction of their energy to
40
+ the atoms of the detector lattice. Therefore, the degradation of the semiconductor detector will limit
41
+ the maximum neutron source activity and/or the source expiration period.
42
+ This article is devoted to the investigations of degradation of the operational parameters of several
43
+ types of silicon semiconductor detectors under prolonged irradiation with fission products of 252Cf (-
44
+
45
+
46
+
47
+
48
+
49
+
50
+
51
+ particles and fission fragments). The main issue was to study the rate of degradation of different
52
+ detector types under irradiation by 252Cf fission products at various irradiation conditions. Irradiation
53
+ was performed at room and liquid nitrogen temperatures as well as with different detector’s
54
+ operational biases, i.e. with different electric field strength in the detectors active regions. Results of
55
+ the preceding investigations were presented in previous articles [3-5].
56
+ 2. Detectors and experimental setup
57
+ Three types of silicon semiconductor detectors were under investigations. Detectors of the first type
58
+ are SiLi p-i-n detectors produced from p-type silicon ingot with resistivity of 2.5 kΩ×cm and carrier
59
+ lifetime of 1000 µs. Two similar detectors with a sensitive region of 20 mm in diameter and 4 mm
60
+ thick were produced using standard Li drift technology [6]. The thickness of the undrifted p-type layer
61
+ in these detectors (i.e. the entrance window thickness) usually amounts to 300-500 nm [7], which is
62
+ kept to suppress the excessive growth of the leakage current at high operation reverse voltage [8].
63
+ Detectors of the second type were two surface-barrier (SB) detectors fabricated from p-type boron-
64
+ doped silicon wafer of (111) orientation and 10 mm in diameter. The resistivity and the carrier lifetime
65
+ were 1 kΩ×cm and 1000 µs, respectively. The front side of the wafers was covered by a thin layer of
66
+ amorphous silicon which served as a passivation coating [9]. The ohmic contact was made by
67
+ sputtering of Pd layer on the whole rear side of the wafer, whereas the rectifying one – by evaporation
68
+ of Al dot with diameter of 7 mm in the center of the wafer’s front side. Detector of the third type was
69
+ p+n planar detector with the thickness of 300 m produced in Ioffe Physical-Technical Institute
70
+ (entrance window thickness was about 50 nm and the voltage of full depletion – nearly 150 V).
71
+ Irradiation by a 252Cf source was performed in vacuum cryostat typically during 10-20 days. The
72
+ source representing a stainless steel substrate covered by an active layer under the thin protective
73
+ coating was mounted 1 cm above the detector front surface that was collimated in order to exclude
74
+ side surface effects of incomplete charge collection. The spectra of the fission products of 252Cf were
75
+ recorded continuously during the whole irradiation period in short 1-hour series, what allowed us to
76
+ observe the spectra evolution directly. Detector reverse current was also monitored during the whole
77
+ irradiation period on 5-second basis with the following averaging on 1-hour measurement series.
78
+ Details of the measurement setup were presented in [3-5].
79
+
80
+
81
+
82
+
83
+ Figure 1. (a) The first and the last spectra measured by SB2 detector in the beginning and at the end of
84
+ the prolonged irradiation period. The following peaks are marked: constant amplitude generator peak
85
+ (g), peak of -particles at 6.118 MeV (), peak at doubled energy of -particles (2) and the peaks
86
+ due to FFs of light (LF) and heavy (HF) groups. (b) Dependence of the light and heavy FF peaks
87
+ visible energies on exposure by FFs.
88
+
89
+ a)
90
+ 106
91
+ α
92
+ first spectrum
93
+ last spectrum
94
+ g
95
+ 105
96
+ Counts per hour
97
+ 104
98
+ 103
99
+
100
+ HF
101
+ 102
102
+ 101
103
+ 100
104
+ 0
105
+ 20
106
+ 40
107
+ 60
108
+ 80
109
+ 100
110
+ Energy, MeVb)
111
+ Heavy Fragments
112
+ 80
113
+ Light Fragments
114
+ Peak Position, MeV
115
+ 75
116
+ 70
117
+ 65
118
+ 60
119
+ 55
120
+ 0
121
+ 1
122
+ 2
123
+ 3
124
+ 4
125
+ Exposure, FF*107
126
+
127
+
128
+
129
+
130
+
131
+ 3. Experimental results
132
+ In order to study the influence of temperature of irradiation on the degradation of the detector’s
133
+ parameters, the irradiation of identical SiLi detectors was performed at room (SiLi1 detector) and
134
+ liquid nitrogen (SiLi2 detector) temperature, respectively. To study the influence of external electric
135
+ field strength on the detector’s parameters degradation, two identical SB detectors were subjected to
136
+ the irradiation with different applied reverse biases, i.e. with different electric field strengths in their
137
+ active regions. The operating biases applied to the respective detector during the irradiation period, the
138
+ corresponding surface electric field strengths and the total exposures are collected in Table 1.
139
+ For all investigated detectors the similar signs of operational parameters degradation as a result of
140
+ the prolonged irradiation by 252Cf fission products were revealed. As an example, Figure 1a represents
141
+ the spectra recorded by SB2 detector at the beginning and at the end of the prolonged irradiation
142
+ period. The peak at 6.1 MeV corresponding to α-particles and another peak at doubled energy of the α-
143
+ particles caused by their accidental coincidences were used as reference points for the calibration of
144
+ the energy scale. Two broad unresolved peaks appearing at higher energies correspond to fission
145
+ fragments of light and heavy groups, respectively.
146
+ The main effect of the detector degradation is a gradual shift of fission fragments visible energy
147
+ towards the lower values, see Figure 1a. The positions of the peaks corresponding to heavy (HF) and
148
+ light (LF) fission fragments were approximated using the Gaussian function for each 1-hour series.
149
+ The dependences of the peaks positions with exposure by fission fragments can be well described by
150
+ linear functions (Figure 1b) for any masses of fission fragments and for all investigated detectors. The
151
+ obtained slope coefficients are summarized in Table 1. It is interesting to note, that the obtained
152
+ coefficients for the peaks of light and heavy fission fragments groups differ approximately by the
153
+ factor of 2 – this holds for all types of investigated detectors and for all irradiation conditions. In more
154
+ details this fact will be discussed separately in the next paper. A similar approximation of the positions
155
+ of α-peaks didn’t reveal any measurable shift with the irradiation dose for all studied detectors [4-5].
156
+ Another sign of the detector’s operational parameters degradation under irradiation is the rapid
157
+ increase of the leakage current which proceeds linear with the number of absorbed fission products
158
+ [3]. The obtained slope coefficients of the leakage current growth are also collected in Table 1.
159
+ 4. Discussion
160
+ It could be noted in Figure 1 that the peak energies of light and heavy groups of fission fragments are
161
+ below the predicted values of 104 MeV and 79 MeV [10], respectively, even on the spectrum
162
+ measured by non-irradiated detectors. The same is true for all other investigated detectors. This effect
163
+ is known as pulse-height defect (PHD) in heavy charged particles spectroscopy by semiconductor
164
+ detectors implying that the measured pulse height amplitude for heavy charged particles is somewhat
165
+ lower than that for -particles of the same energy [1]. It is generally considered that PHD is caused by
166
+ a combination of energy losses (i) in the detector dead layer/entrance window, (ii) due to the atomic
167
+ collisions and (iii) due to recombination of the electron-hole pairs created by the incident heavy
168
+ particle. Whereas energy losses by (i) and (ii) mechanisms are well understood, the full understanding
169
+ of the charge losses due to recombination is still missing. Two models were suggested supposing that
170
+ enhanced carrier recombination proceeds either in the bulk region on the radiation-induced defects
171
+ created by incident FFs [11], or at the surface states of the semiconductor [12]. The later model is
172
+ consistent with the TRIM [13] simulation results (Figure 2) showing that the density of electron-hole
173
+ pairs generated by fission fragments reaches the maximum in the near-surface region of the detector
174
+ and then gradually drops down towards the bulk, suggesting therefore that decisive influence on PHD
175
+ would have the carrier recombination at the surface states.
176
+ Previously, the PHD of about 7-10 MeV was reported for 252Cf fission fragments detection by
177
+ semiconductor detectors not subjected to the prolonged irradiation [10]. These PHD values are close to
178
+ those ones obtained for the investigated planar and SB1 detectors operated at high reverse bias – see
179
+ Table 1. We believe, that higher PHD values in non-irradiated SiLi are related with rather thick
180
+ entrance window in these detectors. Whereas the increase of PHD for SB2 detector operated at lower
181
+
182
+
183
+
184
+
185
+
186
+
187
+
188
+ Table 1. Irradiation conditions and the degradation of the operational parameters of the investigated
189
+ detectors: Ub – applied bias during irradiation; Fs – surface electric field strength; PHDLF/ PHDHF –
190
+ pulse-height defects for light and heavy fragments peaks registered by non-irradiated detectors; NFF
191
+ and N – exposure by fission fragments and -particles, respectively; ∆EHF/∆NFF – slope coefficient
192
+ describing the linear shift of heavy fission fragment maximum; ∆ELF/∆NFF – slope coefficient
193
+ describing the linear shift of light fission fragment maximum; ∆I/∆N – rate of the reverse current
194
+ increase relative to the total number of the registered fission products (wasn’t measured for SiLi2
195
+ detector); NFFmax – maximal permissible exposure by fission fragments; t – expected active operation
196
+ period of the detector in a neutron source.
197
+
198
+
199
+ p+n planar
200
+ SB1
201
+ SB2
202
+ SiLi1
203
+ SiLi2
204
+ Ub, V
205
+ 150
206
+ 200
207
+ 30
208
+ 400
209
+ 400
210
+ Fs, kV/cm
211
+ 8.5
212
+ 40
213
+ 17
214
+ 1.5
215
+ 1.5
216
+ PHDLF/PHDHF, MeV
217
+ 8/10
218
+ 9/11
219
+ 18/19
220
+ 28/29
221
+ 35/37
222
+ NFF *108
223
+ 1.1
224
+ 0.45
225
+ 0.43
226
+ 3.4
227
+ 1
228
+ N*
229
+ 0.5
230
+ 0.20
231
+ 0.19
232
+ 1.5
233
+ 0.44
234
+ ∆EHF/∆NFF*10-5, keV/FF
235
+ -0.9
236
+ -1.8
237
+ -8.9
238
+ -3.6
239
+ -5.7
240
+ ∆ELF/∆NFF*10-5, keV/FF
241
+ -1.9
242
+ -3.9
243
+ -20
244
+ -6.2
245
+ -12
246
+ ∆I/∆N*10-16, A/ion
247
+ 8.9
248
+ 14
249
+ 8.0
250
+ 4.4
251
+ -
252
+ NFFmax *108
253
+ 22
254
+ 12
255
+ 2.2
256
+ 6.9
257
+ 4.7
258
+ t, years
259
+ 11.6
260
+ 6.3
261
+ 1.2
262
+ 3.6
263
+ 2.5
264
+
265
+
266
+ electric field (Table 1) reflects the influence of the electric field strength on the charge carrier
267
+ collection efficiency, i.e. on the recombination of the generated electron-hole pairs (note that the
268
+ active layer thickness in SB2 detector exceeds the projection range of incident FFs even at 30V).
269
+ As a result of the prolonged irradiation by 252Cf fission products, the linear shift of FF peaks
270
+ positions, i.e. the linear increase of PHD for fission fragments peaks, was revealed. Since the task of
271
+ semiconductor detector operating as a part of neutron calibration source is the reliable detection of
272
+ fission fragments signal, the irradiated detector could be considered to be "degraded" when the
273
+ spectrum of the heavy fission fragment overlaps with much more intense signal at double energy of α-
274
+ peak, what prevents us from discrimination between them [3]. The values of maximal “permissible”
275
+ exposure by fission fragments NFFmax corresponding to the beginning of the peaks overlap at three
276
+ standard deviations from their maxima were estimated for each detector using the corresponding slope
277
+ coefficients derived for HF peak and the results are presented in Table 1.
278
+
279
+
280
+
281
+ Figure 2. TRIM simulated vacancies
282
+ distribution profiles (solid lines) and
283
+ linear densities of electron-hole pairs
284
+ (dashed lines) generated by light and
285
+ heavy FFs with mean energies and
286
+ masses of 104 MeV and 79 MeV, 106
287
+ amu and 142 amu, respectively.
288
+
289
+ Heavy Fragments
290
+ 300
291
+ 2
292
+ Light Fragments
293
+ 250
294
+ 200
295
+ 150
296
+ 100
297
+ 50
298
+ 0
299
+ 0
300
+ 0
301
+ 5
302
+ 10
303
+ 15
304
+ 20
305
+ Depth, μm
306
+
307
+
308
+
309
+
310
+
311
+ Permissible exposure values NFFmax for the investigated detectors appeared to vary approximately
312
+ by one order of magnitude. The highest NFFmax values were found for planar and SB1 detector operated
313
+ at 200V. Reduction of the operating bias and thus the electric field strength in case of SB2 detector has
314
+ led to considerable decrease of the expected permissible exposure value. Therefore, the electric field
315
+ strength affects not only the PHD on non-irradiated detector, but also the value of the expected
316
+ maximal exposure. However the NFFmax exposure values for SiLi detectors – which operated with
317
+ lowest electric field as compared with other detectors – are significantly higher than that for SB2
318
+ detector. Thus the expected maximal exposure appeared to be more sensitive to the electric field
319
+ strength in the surface barrier detectors and less sensitive in SiLi and planar detectors. It follows then
320
+ that not only the electric field strength, but also a detector’s internal structure defines the PHD growth
321
+ under irradiation and the maximal permissible exposure.
322
+ According to TRIM simulations, irradiation of Si detectors by fission fragments will lead to the
323
+ creation of vacancy-interstitial pairs and therefore to the formation of high density of radiation-
324
+ induced defects in the region from detector surface till the depth of 17 μm with the maxima at 14-16
325
+ μm (Figure 2). Additionally TRIM indicates, that the energy of FFs is high enough to damage the
326
+ detector surface by sputtering. Therefore, prolonged irradiation with fission fragments will lead to an
327
+ increase of the carrier recombination rate both in Si bulk and on the surface of the semiconductor, thus
328
+ contributing to the PHD growth.
329
+ The transition region in the detectors produced by planar and by SiLi technology (p+n and p-i
330
+ transition regions, respectively) is located inside the crystalline matrix at the typical depths of 50-500
331
+ nm from the surface. Apparently, the contribution of the surface recombination to the charge carrier
332
+ losses will be more significant for surface-barrier detectors than for SiLi and planar ones, whereas the
333
+ contribution of bulk defects – approximately similar in all detectors, what may be the reason for
334
+ different sensitivity of NFFmax exposure to the electric field strength in these detectors. Additional
335
+ investigations are needed to determine the dominant charge loss channel.
336
+ Suggested neutron calibration source should operate also at cryogenic temperatures (liquid nitrogen
337
+ or slightly above). Performed irradiation of SiLi2 detector at liquid nitrogen temperature has shown,
338
+ that in contrast to the electric field, temperature of irradiation seems to have no or only minor
339
+ influence on the expected value of maximal exposure as it could be concluded from the comparable
340
+ NFFmax values obtained for SiLi detectors irradiated at different temperatures. Somewhat smaller NFFmax
341
+ exposure obtained for SiLi2 detector is probably related with thicker entrance window in this detector.
342
+ Knowing the maximal expected exposure values NFFmax, it is possible to estimate the duration of
343
+ active “lifetime” of neutron calibration source. For the operation of neutron calibration source the
344
+ reasonable neutron activity would be the around 20 neutrons/s and taking into account that each
345
+ spontaneous fission releases in average 3.7 fast neutrons, the activity of 20 neutrons/s would
346
+ correspond to ~6 spontaneous fissions per second. Therefore, considering the maximal exposure value
347
+ from Table 1, the duration of active “lifetime” of such neutron calibration source will be 1.2-11.6
348
+ years (without taking into account the decay of the radiation source).
349
+ During this operation period, a significant increase of leakage current up to ~100 μA can be
350
+ expected at room temperature, as can be calculated from the obtained coefficients of leakage current
351
+ growth (Table 1). Such high reverse current is unacceptable and therefore the detector cooling in order
352
+ to reduce the reverse current during the neutron source operation will be required. The coefficients of
353
+ current growth upon irradiation by fission products of 252Cf appeared to be an order of magnitude
354
+ higher than the corresponding coefficients of 7-17×10–17 A/α determined by us earlier for the identical
355
+ detectors subjected to long-term irradiation by -particles [4]. This fact confirms that prolonged
356
+ irradiation by FFs leads to the creation of the effective recombination-generation defect centers
357
+ participating in charge carrier recombination and the reverse current growth.
358
+ 5. Conclusions
359
+ Prolonged irradiation of three different types of Si semiconductor detectors by fission products of 252Cf
360
+ nuclide has led to a gradual increase of pulse-height defect for the fission fragments peaks in all
361
+
362
+
363
+
364
+
365
+
366
+
367
+
368
+ investigated detectors. This will eventually lead to the overlap with more intense -peak and therefore
369
+ to the impossibility of further reliable detection of fission fragments by the semiconductor detector and
370
+ thus to the limitation of the operation period of neutron calibration source. Obtained experimental
371
+ results suggest, that in order to assure the longest operation period of the neutron calibration source it
372
+ is worth to use the semiconductor detectors with lowest surface recombination rate and with highest
373
+ possible electric field strength in their active region. Among the investigated detectors, the planar one
374
+ most fully meets these requirements, whereas in relatively thick SiLi detectors it is difficult to achieve
375
+ the high electric field strength and surface-barrier detectors may suffer from high surface
376
+ recombination. With properly chosen semiconductor detector the expected active operation period of
377
+ 252Cf-based neutron calibration source may reach up to 12 years.
378
+
379
+ Acknowledgements
380
+ The reported study was funded by RFBR, project number 20-02-00571
381
+ References
382
+ [1]
383
+ Knoll G F 2000 Radiation Detection and Measurement, 3rd ed. (New York: John Wiley and
384
+ Sons) ISBN 978-0-471-07338-3, 978-0-471-07338-3
385
+ [2]
386
+ Moll M 2018 IEEE Transactions on Nuclear Science 65 1561–1582
387
+ [3]
388
+ Bakhlanov S V, Derbin A V, Drachnev I S, Konkov O I, Kotina I M, Kuzmichev A M,
389
+ Lomskaya I S, Mikulich M S, Muratova N V, Niyazova N V, Semenov D A, Trushin M V
390
+ and Unzhakov E V 2021 Journal of Physics: Conference Series 2103 012138
391
+ [4]
392
+ Bakhlanov S V, Bazlov N V, Chernobrovkin I D, Derbin A V, Drachnev I S, Kotina I M,
393
+ Konkov O I, Kuzmichev A M, Mikulich M S, Muratova N V, Trushin M V and Unzhakov E
394
+ V 2021 Journal of Physics: Conference Series 2103 012139
395
+ [5]
396
+ Bazlov N V, Bakhlanov S V, Derbin A V, Drachnev I S, Eremin V K, Kotina I M, Muratova V
397
+ N, Pilipenko N V, Semenov D A, Unzhakov E V and Chmel E A 2018 Instruments and
398
+ Experimental Techniques 61 323
399
+ [6]
400
+ Bazlov N V, Bakhlanov S V, Derbin A V, Drachnev I S, Izegov G A, Kotina I M, Muratova V
401
+ N, Niyazova N V, Semenov D A, Trushin M V, Unzhakov E V, Chmel E A 2020 Instrum.
402
+ Exp. Tech. 63(1) 25
403
+ [7]
404
+ Alekseev I E, Bakhlanov S V, Derbin A V, Drachnev I S, Kotina I M, Lomskaya I S, Muratova
405
+ V N, Niyazova N V, Semenov D A, Trushin M V, Unzhakov E V 2020 Physical Review C
406
+ 102 064329
407
+ [8]
408
+ Kozai M, Fuke H, Yamada M, Perez K, Erjavec T, Hailey C J, Madden N, Rogers F, Saffold N,
409
+ Seyler D, Shimizu Y, Tokuda K, Xiao M 2019 Nuclear Inst. and Methods in Physics
410
+ Research A 947 162695
411
+ [9]
412
+ Kotina I M, Danishevskii A M, Konkov O I, Terukov E I, Tuhkonen L M 2014 Semiconductors
413
+ 48(9) 1167
414
+ [10] Paasch K, Krause H, Scobel W 1984 Nuclear Inst. and Methods in Physics Research 221 558
415
+ [11] Eremin V K, Il’yashenko I N, Strokan N B, Shmidt B 1995 Fiz. Tekh. Poluprovodn. 29(1) 79
416
+ [in Russian]
417
+ [12] Tsyganov Y S 2013 Physics of Particles and Nuclei 44(1) 92
418
+ [13] Ziegler J F, Biersack J P, Ziegler M D SRIM – Stopping and Range of Ions in Matter
419
+ www.srim.org (accessed: May 2022)
420
+
421
+
CdFJT4oBgHgl3EQftC3b/content/tmp_files/2301.11616v1.pdf.txt ADDED
@@ -0,0 +1,1537 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ A Systematic Mapping Study on Responsible AI
2
+ Risk Assessment
3
+ Boming Xia∗†, Qinghua Lu∗†, Harsha Perera∗, Liming Zhu∗†, Zhenchang Xing∗, Yue Liu∗†, Jon Whittle∗
4
+ ∗CSIRO’s Data61, Sydney, Australia
5
+ †University of New South Wales, Sydney, Australia
6
+ Abstract—The rapid development of artificial intelligence (AI)
7
+ has led to increasing concerns about the capability of AI systems
8
+ to make decisions and behave responsibly. Responsible AI (RAI)
9
+ refers to the development and use of AI systems that benefit
10
+ humans, society, and the environment while minimising the risk
11
+ of negative consequences. To ensure responsible AI, the risks as-
12
+ sociated with AI systems’ development and use must be identified,
13
+ assessed and mitigated. Various AI risk assessment frameworks
14
+ have been released recently by governments, organisations, and
15
+ companies. However, it can be challenging for AI stakeholders to
16
+ have a clear picture of the available frameworks and determine
17
+ the most suitable ones for a specific context. Additionally, there
18
+ is a need to identify areas that require further research or
19
+ development of new frameworks. To fill the gap, we present a
20
+ mapping study of 16 existing RAI risk assessment frameworks
21
+ from the industry, governments, and non-government organiza-
22
+ tions (NGOs). We identify key characteristics of each framework
23
+ and analyse them in terms of RAI principles, stakeholders,
24
+ system lifecycle stages, geographical locations, targeted domains,
25
+ and assessment methods. Our study provides a comprehensive
26
+ analysis of the current state of the frameworks and highlights
27
+ areas of convergence and divergence among them. We also
28
+ identify the deficiencies in existing frameworks and outlines the
29
+ essential characteristics a concrete framework should possess.
30
+ Our findings and insights can help relevant stakeholders choose
31
+ suitable RAI risk assessment frameworks and guide the design
32
+ of future frameworks towards concreteness.
33
+ Index Terms—artificial intelligence, machine learning, risk
34
+ assessment, impact assessment, responsible AI, risk mitigation,
35
+ pattern
36
+ I. INTRODUCTION
37
+ The adoption of artificial intelligence (AI) in various appli-
38
+ cation domains has led to numerous advantages, such as im-
39
+ proved efficiency and reduced cost in manufacturing. However,
40
+ the risks associated with AI systems have also attracted signif-
41
+ icant attention from both industry and academia [1]–[3]. For
42
+ example, an AI system may make biased decisions that lead
43
+ to unintended discrimination [4]–[6]. Also, the AI system’s
44
+ dataset may contain sensitive information, risking violation of
45
+ laws such as EU General Data Protection Regulation (GDPR)1
46
+ and EU AI Act (proposed)2. The AI incident database3 has
47
+ collected over 2200 (as of January 2023) reported real-world
48
+ incidents caused by AI systems.
49
+ Responsible AI (RAI) is developing and applying AI sys-
50
+ tems that benefit humans, society, and the environment (HSE)
51
+ 1https://gdprinfo.eu/
52
+ 2https://artificialintelligenceact.eu/
53
+ 3https://incidentdatabase.ai/
54
+ while minimizing the associated risks. A number of RAI
55
+ principle frameworks that AI systems and stakeholders should
56
+ adhere to have been released recently, such as Australia’s
57
+ AI Ethics Principles4 and European Commission’s Ethics
58
+ guidelines for trustworthy AI5. Many organizations have de-
59
+ veloped principle-driven RAI risk assessment frameworks to
60
+ implement RAI based on the RAI principles, m (e.g., US NIST
61
+ AI risk management framework [7], EU Assessment List for
62
+ Trust AI framework [8]). These frameworks are designed to
63
+ help organizations and individuals systematically assess and
64
+ mitigate potential risks associated with AI systems. Despite
65
+ the availability of these RAI risk assessment frameworks, AI
66
+ system stakeholders need to gain a holistic view of the existing
67
+ frameworks to choose the most appropriate one for their
68
+ context. Also, it is unclear how effective these frameworks
69
+ are at assessing and mitigating RAI risks.
70
+ To bridge the gaps, we have performed a systematic map-
71
+ ping study on the existing RAI risk assessment frameworks.
72
+ The main objectives of this study are: 1) to provide a summary
73
+ of the current available higher-quality AI risk frameworks to
74
+ which researchers and practitioners can refer; 2) to investigate
75
+ the capabilities and limitations of the RAI risk assessment
76
+ frameworks; and 3) to provide insights for future research and
77
+ development on concrete AI risk assessment frameworks.
78
+ The main contributions of this study are:
79
+ • We present a comprehensive qualitative and quantitative
80
+ analysis and synthesis of 16 state-of-practice RAI risk
81
+ assessment frameworks selected from the grey literature.
82
+ • We provide empirical findings and insights on the capa-
83
+ bilities and limitations of the existing frameworks and
84
+ highlight the essentials for developing concrete RAI risk
85
+ assessment frameworks.
86
+ The remainder of this paper is organized as follows: Sec-
87
+ tion II presents the methodology and research questions (RQ)
88
+ followed by the results and findings for each RQ in section
89
+ III. Section IV discusses RAI risk assessment framework
90
+ “concreteness” and threats to validity. Then, section V lists
91
+ related work while section VI concludes the paper with a
92
+ summary and the future work of this study.
93
+ 4https://www.industry.gov.au/publications/australias-artificial-intelligence-
94
+ ethics-framework/australias-ai-ethics-principles
95
+ 5https://ec.europa.eu/digital-single-market/en/news/ethics-guidelines-
96
+ trustworthy-ai
97
+ arXiv:2301.11616v1 [cs.SE] 27 Jan 2023
98
+
99
+ Fig. 1: Methodology overview.
100
+ II. METHODOLOGY
101
+ We perform the systematic mapping study following
102
+ Kitchenham’s guideline [9] on conducting literature reviews
103
+ in software engineering. The overall methodology is presented
104
+ in Figure 1. To investigate the capabilities of the existing AI
105
+ risk assessment frameworks, we derived the following RQs:
106
+ • RQ1: Who have published RAI risk assessment frame-
107
+ works?
108
+ • RQ2: What are the characteristics of the existing RAI
109
+ risk assessment frameworks?
110
+ – RQ2.1 What RAI principles are addressed?
111
+ – RQ2.2 Who are the stakeholders?
112
+ ∗ RQ2.2.1 Who conducts the assessment?
113
+ ∗ RQ2.2.2 Whose activities are assessed?
114
+ – RQ2.3 What is the scope of the frameworks?
115
+ ∗ RQ2.3.1 Which development stages are covered
116
+ by the frameworks?
117
+ ∗ RQ2.3.2 Where can the frameworks be applied?
118
+ ∗ RQ2.3.3 Which domains/sectors are the frame-
119
+ works designed for?
120
+ • RQ3: How are the RAI risks assessed?
121
+ – RQ3.1: What are the inputs?
122
+ – RQ3.2: What is the assessment process?
123
+ – RQ3.3: What are the outputs?
124
+ The data sources include ACM, IEEE, Science Direct,
125
+ Springer, Google scholar for academic papers and Google
126
+ Search for industrial frameworks. The search was conducted
127
+ in November 2022 and the search term is (“artificial in-
128
+ telligence” OR “machine learning” OR AI OR ML) AND
129
+ (impact OR risk) AND (assess OR assessment OR assessing
130
+ OR evaluate OR evaluation OR evaluating OR measure OR
131
+ measurement OR measuring OR mitigate OR mitigation OR
132
+ mitigating OR manage OR managing OR management). The
133
+ study only includes frameworks with relatively concrete AI
134
+ risk assessment solutions and excludes frameworks discussing
135
+ high-level AI risks.
136
+ Although the literature search included academic databases,
137
+ academic papers and frameworks are excluded considering the
138
+ following reasons: a) one of our inclusion criteria is to include
139
+ frameworks that are currently being used in practice (e.g.,
140
+ governmental/industrial/international organizations consulting
141
+ extensively with practitioners and extracting proven uses); b)
142
+ based on the inclusion and exclusion criteria, the academic pa-
143
+ pers we collected are either discussing the identified industrial
144
+ frameworks, or lack of details on concrete AI risk assessment
145
+ solutions. In the end, we selected 16 industrial frameworks to
146
+ be investigated. The complete research protocol is available
147
+ online6. We adopt Australia’s AI ethics principles in this
148
+ study: Human, societal and environmental (HSE) wellbeing,
149
+ Human-centered values, Fairness, Privacy protection and se-
150
+ curity, Reliability and safety, Transparency and explainability,
151
+ Contestability, and Accountability.
152
+ III. RESEARCH RESULTS
153
+ This section presents the results and findings of each RQ.
154
+ A. RQ1: Who have published RAI risk assessment frame-
155
+ works?
156
+ As illustrated in Table I and Fig. 2, we identified 16 frame-
157
+ works to be included in this study. These frameworks have
158
+ been published by organizations based in the United States
159
+ (US), the United Kingdom (UK), the European Union (EU),
160
+ Canada (CA), Australia (AU), Singapore (SA), the Netherlands
161
+ (NL), and Germany (DE). Additionally, one framework has
162
+ been released by the World Economic Forum, an international
163
+ (INT) organization. Although we did not set a specific time
164
+ limit for the literature search, as shown in Fig. 2b, the majority
165
+ of the frameworks (10 out of 16, 62.5%) were published
166
+ or last updated (some frameworks tend to be updated over
167
+ time) in 2022, followed by 1 framework (6.25%) published in
168
+ 2021, 3 frameworks (18.75%) published in 2020, 0 framework
169
+ published in 2019, and 2 (12.5%) frameworks published in
170
+ 2018. Also, a significant proportion of the frameworks (9
171
+ out of 16, 56.25%) were published by government agencies
172
+ worldwide, with 6 of them last updated in 2022. This suggests
173
+ that the issue of RAI risks has been gaining significant
174
+ attention worldwide, particularly among government agencies.
175
+ In terms of the number of published frameworks (Fig. 2a
176
+ and Fig. 2c), the US leads in research and development on RAI
177
+ risk assessment and published 6 related frameworks, including
178
+ 3 frameworks from US government agencies (i.e., I1 by
179
+ National Institute of Standards and Technology (NIST), and I8
180
+ 6https://docs.google.com/document/d/1F sAmRI7zvJBYyiF96cn5oNtG3O8
181
+ psyB/edit?usp=sharing&ouid=111846093034327217492&rtpof=true&sd=true
182
+
183
+ Retrieved
184
+ Academic: 10988
185
+ Grey: 3160
186
+ Tentative
187
+ Academic: 104
188
+ Grey: 198
189
+ Tentative
190
+ Academic: 104+39=143
191
+ Grey: 198+24=222
192
+ Selected
193
+ Research protocol
194
+ Academic: 0
195
+ Grey: 16
196
+ ReportingTABLE I: Industrial AI risk assessment frameworks (collected in November 2022).
197
+ Demographics (RQ1)
198
+ Characteristics (RQ2)
199
+ Processes (RQ3)
200
+ No.
201
+ Frameworks
202
+ Region
203
+ Affiliation
204
+ Affiliation type First release Last update
205
+ RAI Princples
206
+ Stakeholders
207
+ Stages
208
+ Region
209
+ Sector
210
+ Type
211
+ Mitigation
212
+ *Risk factors
213
+ I1
214
+ AI risk management framework
215
+ (AI RMF) [7]
216
+ US
217
+ National Institute of Standards
218
+ and technology (NIST)
219
+ Government
220
+ 2022.05
221
+ 2022.08
222
+ All principles
223
+ Specified
224
+ All stages
225
+ Region-
226
+ agnostic
227
+ Sector-
228
+ agnostic
229
+ Descriptive
230
+ *Yes
231
+ Hazard,
232
+ exposure,
233
+ vulnerability
234
+ I2
235
+ Assessment list for trustworthy AI
236
+ (ALTAI) [8]
237
+ EU
238
+ European Commission High-
239
+ Level Expert Group on AI
240
+ Government
241
+ 2019.06
242
+ 2020.07
243
+ All principles
244
+ Specified
245
+ All stages
246
+ Region-
247
+ agnostic
248
+ Sector-
249
+ agnostic
250
+ Procedural
251
+ Yes
252
+ Hazard,
253
+ exposure,
254
+ vulnerability
255
+ I3
256
+ Algorithm Impact Assessment
257
+ tool (AIA) [10]
258
+ CA
259
+ Government of Canada
260
+ Government
261
+ 2019
262
+ 2022.11
263
+ Not specified
264
+ Not specified
265
+ Planning &
266
+ requirements
267
+ analysis,
268
+ design, testing
269
+ Region-
270
+ agnostic
271
+ Sector-
272
+ agnostic
273
+ Procedural
274
+ Yes
275
+ Hazard,
276
+ exposure,
277
+ vulnerability
278
+ I4
279
+ Fundamental rights and algorithm
280
+ impact assessment (FRAIA) [11]
281
+ NL
282
+ Ministry of the Interior and
283
+ Kingdom Relations (BZK)
284
+ Government
285
+ 2022.03
286
+ N/A
287
+ Not specified
288
+ Specified
289
+ All stages
290
+ Region-
291
+ agnostic
292
+ Public
293
+ sectors
294
+ Procedural
295
+ *Yes
296
+ Hazard,
297
+ exposure,
298
+ vulnerability
299
+ I5
300
+ AI and data protection risk toolkit
301
+ [12]
302
+ UK
303
+ Information Commissioner’s
304
+ Office (ICO)
305
+ Government
306
+ 2021
307
+ 2022.05
308
+ HSE wellbeing, human-
309
+ centered values, fairness,
310
+ privacy protection &
311
+ security, reliability &
312
+ safety, transparency &
313
+ explainability,
314
+ accountability
315
+ Specified
316
+ All stages
317
+ Reusable
318
+ anywhere
319
+ with
320
+ adjustments
321
+ Sector-
322
+ agnostic
323
+ Procedural
324
+ No
325
+ Hazard,
326
+ exposure,
327
+ vulnerability
328
+ I6
329
+ Model AI governance framework
330
+ [13]
331
+ SA
332
+ Personal Data Protection
333
+ Commission (PDPC)
334
+ Government
335
+ 2019.01
336
+ 2020.01
337
+ HSE wellbeing, human-
338
+ centered values, fairness,
339
+ transparency &
340
+ explainability, reliability
341
+ & safety
342
+ Not specified
343
+ Not specified
344
+ Region-
345
+ agnostic
346
+ Sector-
347
+ agnostic
348
+ Descriptive
349
+ Yes
350
+ Hazard,
351
+ exposure,
352
+ vulnerability
353
+ I7
354
+ NSW artificial intelligence
355
+ assurance framework [14]
356
+ AU
357
+ NSW Government
358
+ Government
359
+ 2022.03
360
+ N/A
361
+ HSE wellbeing, human-
362
+ centered values, fairness,
363
+ privacy protection &
364
+ security, reliability &
365
+ safety, transparency &
366
+ explainability,
367
+ accountability
368
+ Specified
369
+ All stages
370
+ Reusable
371
+ anywhere
372
+ with
373
+ adjustments
374
+ Sector-
375
+ agnostic
376
+ Procedural
377
+ *Yes
378
+ Hazard,
379
+ exposure,
380
+ vulnerability,
381
+ mitigation risk
382
+ I8
383
+ Ethics & algorithms toolkit [15]
384
+ US
385
+ GovEX, the City and County
386
+ of San Francisco, Harvard
387
+ DataSmart, and Data
388
+ Community DC
389
+ Government
390
+ involved
391
+ 2018
392
+ N/A
393
+ HSE wellbeing, human-
394
+ centered values, fairness,
395
+ privacy protection
396
+ & security, reliability &
397
+ safety, transparency &
398
+ explainability,
399
+ accountability
400
+ Specified
401
+ Not specified
402
+ Region-
403
+ agnostic
404
+ Sector-
405
+ agnostic
406
+ Procedural
407
+ No
408
+ Hazard,
409
+ exposure,
410
+ vulnerability
411
+ I9
412
+ RFD-BUS012A artificial
413
+ intelligence assessment tool [16]
414
+ US
415
+ Pennsylvania Office of
416
+ Administration
417
+ Government
418
+ 2018.09
419
+ 2022.08
420
+ Not specified
421
+ Not specified
422
+ Planning &
423
+ requirements
424
+ analysis, design
425
+ Region-
426
+ agnostic
427
+ Sector-
428
+ agnostic
429
+ Procedural
430
+ No
431
+ Vulnerability
432
+ I10
433
+ Model rules on impact assessment
434
+ of algorithmic decision-making
435
+ systems used by public
436
+ administration [17]
437
+ EU
438
+ European Law Institute
439
+ (ELI)
440
+ NGO
441
+ 2022.01
442
+ N/A
443
+ Not specified
444
+ Not specified
445
+ Not specified
446
+ Region-
447
+ agnostic
448
+ Public
449
+ sectors
450
+ Procedural
451
+ Yes
452
+ Hazard,
453
+ exposure,
454
+ vulnerability
455
+ I11 Artificial intelligence for children
456
+ toolkit [18]
457
+ INT
458
+ World Economic Forum
459
+ (WEF)
460
+ NGO
461
+ 2022.03
462
+ N/A
463
+ HSE wellbeing, human-
464
+ centered values, fairness,
465
+ privacy protection &
466
+ security, reliability &
467
+ safety, transparency &
468
+ explainability,
469
+ contestability
470
+ Specified
471
+ Not specified
472
+ Region-
473
+ agnostic
474
+ Children
475
+ & youth
476
+ Procedural
477
+ Yes
478
+ Hazard,
479
+ exposure,
480
+ vulnerability
481
+ I12
482
+ Recommended practices for
483
+ assessing the impact of
484
+ autonomous and intelligent
485
+ systems on human well-being [19]
486
+ US
487
+ IEEE
488
+ NGO
489
+ 2020.05
490
+ N/A
491
+ HSE wellbeing, human-
492
+ centered values
493
+ Specified
494
+ All stages
495
+ Region-
496
+ agnostic
497
+ Sector-
498
+ agnostic
499
+ Descriptive
500
+ *Yes
501
+ Hazard,
502
+ exposure,
503
+ vulnerability
504
+ I13 Responsible AI impact assessment
505
+ template [20]
506
+ US
507
+ Microsoft
508
+ Industry
509
+ 2022.06
510
+ N/A
511
+ All principles
512
+ Not specified
513
+ Not specified
514
+ Region-
515
+ agnostic
516
+ Sector-
517
+ agnostic
518
+ Procedural
519
+ No
520
+ Hazard,
521
+ exposure,
522
+ vulnerability
523
+ I14 Algorithmic impact assessments
524
+ (AIAs) in healthcare [21]
525
+ UK
526
+ Ada Lovelace Institute
527
+ NGO
528
+ 2022.01
529
+ N/A
530
+ HSE wellbeing, human
531
+ centered values, fairness,
532
+ privacy protection &
533
+ security, reliability &
534
+ safety, transparency &
535
+ explainability
536
+ Specified
537
+ Planning &
538
+ requirements
539
+ analysis,
540
+ design, tetsing,
541
+ deployment,
542
+ monitoring
543
+ UK only
544
+ UK
545
+ healthcare Procedural
546
+ *Yes
547
+ Hazard,
548
+ exposure,
549
+ vulnerability
550
+ I15 Artificial intelligence impact
551
+ assessment [22]
552
+ NL
553
+ ECP, Platform for the
554
+ Information Society
555
+ NGO
556
+ 2018
557
+ N/A
558
+ Not specified
559
+ Specified
560
+ Not specified
561
+ Region-
562
+ agnostic
563
+ Sector-
564
+ agnostic
565
+ Procedural
566
+ *Yes
567
+ Hazard,
568
+ exposure,
569
+ vulnerability,
570
+ mitigation risk
571
+ I16
572
+ Automated decision-making
573
+ systems in the public sector: an
574
+ impact assessment
575
+ tool for public authorities [23]
576
+ DE
577
+ Algorithm Watch
578
+ NGO
579
+ 2021.06
580
+ N/A
581
+ All principles
582
+ Not specified
583
+ Not specified
584
+ Region-
585
+ agnostic
586
+ Public
587
+ sectors
588
+ Procedural
589
+ *Yes
590
+ Hazard,
591
+ exposure,
592
+ vulnerability
593
+
594
+ (a) Percentage by region.
595
+ (b) Number by year.
596
+ (c) Number by year and region.
597
+ Fig. 2: Demographics of collected frameworks.
598
+ and I9 by two state/city government agencies) and 2 from US-
599
+ based organizations (i.e., I12 by IEEE and I13 by Microsoft).
600
+ The UK, EU, and Netherlands rank second with 2 frameworks
601
+ developed. In the UK, one framework was developed by a
602
+ government agency (Information Commissioner’s Office, ICO)
603
+ and Ada Lovelace Institute published the other specifically
604
+ for the proposed National Medical Imaging Platform of the
605
+ National Health Service (NHS). The EU published 2 frame-
606
+ works on RAI risk assessment in 2020 (last updated) and
607
+ 2022, respectively. Notably, the EU has drafted its AI Act,
608
+ which marks a significant step towards operationalizing RAI
609
+ by legislation. The Netherlands published 2 frameworks, one
610
+ in 2018 by a non-government organization (ECP) and the other
611
+ in 2022 by its Ministry of the Interior and Kingdom Relations
612
+ (BZK). Australia’s New South Wales government published
613
+ the nation’s first AI Assurance Framework in 2022. Singapore
614
+ had its AI governance framework published in 2019, while
615
+ it launched AI Verify in May 2022 to objectively assess AI
616
+ systems in a verifiable way. The Canadian government released
617
+ its Algorithm impact assessment tool in 2021. The World
618
+ Economic Forum (WEF) published a toolkit for managing
619
+ RAI risks to children. Lastly, one framework published by
620
+ a German-based organization, Algorithm Watch, is identified
621
+ in this study.
622
+ Finding to RQ1: The growing number of RAI risk
623
+ assessment frameworks worldwide indicates increasing
624
+ global concern about the risks associated with the
625
+ development and use of AI systems and a growing
626
+ recognition of RAI approaches to assess and mitigate
627
+ RAI risks.
628
+ B. RQ2: What are the characteristics of the existing AI risk
629
+ assessment frameworks?
630
+ This subsection discusses the characteristics of the collected
631
+ frameworks based on the following aspects: RAI principles,
632
+ stakeholders, software development lifecycle stages, geograph-
633
+ ical locations, and targeted sectors.
634
+ To improve presentation, we first classify the frameworks
635
+ based on whether they have clear specifications on differ-
636
+ ent characteristics (e.g., whether RAI principles/stakeholder-
637
+ s/stages are specified). Then, we further categorize them to see
638
+ whether they have formulated the assessment and mitigation
639
+ based on different sub-categories of those characteristics (e.g.,
640
+ different RAI principles).
641
+ 1) RQ2.1 What RAI principles are addressed?: This sub-
642
+ RQ aims to investigate the RAI principles (i.e., the correspond-
643
+ ing risk category) addressed by the identified frameworks. We
644
+ have mapped the various principles from different frameworks
645
+ to Australia’s AI ethics principles (see Table I).
646
+ As illustrated in Fig. 3a, among the 16 identified frame-
647
+ works, 11 frameworks (I1, I2, I5-I8, I11-I14, I16) have speci-
648
+ fied their guiding principles. 5 frameworks (I3, I4, I9, I10, I15)
649
+ do not explicitly state their corresponding principles, although
650
+ they may implicitly encompass these principles through their
651
+ framework description and introduction (e.g., I3) or references
652
+ to other existing frameworks, standards, and guidelines (e.g.,
653
+ I4). Among the 11 frameworks with specified guiding princi-
654
+ ples, only 5 frameworks (i.e., I1, I2, I11, I13, I16) organize
655
+ their sets of RAI risk assessment questions or checklists based
656
+ on different RAI principles.
657
+ All the 11 frameworks that explicitly specify the guid-
658
+ ing principles or targeted risks consider HSE wellbeing and
659
+ human-centred values. Out of these 11 frameworks, 10 frame-
660
+ works cover fairness, reliability & safety, transparency &
661
+ explainability. The only exception is framework I12, which
662
+ focuses mainly on HSE wellbeing. Privacy protection & se-
663
+ curity is covered by 9 (I1, I2, I5, I7, I8, I11, I13, I14, I16)
664
+ and accountability is covered by 8 frameworks (I1, I2, I5, I7,
665
+ I8, I11, I13, I16). Only 5 frameworks (I1, I2, I11, I13, I16)
666
+ include contestability (Fig. 3b).
667
+ 2) RQ2.2: Who are the stakeholders?: This subsection
668
+ examines the stakeholders involved in the frameworks from
669
+ two perspectives: the framework user(s) who are responsible
670
+ for conducting the risk assessment (i.e., assessor), and those
671
+ whose activities are being assessed (i.e., assessee). The stake-
672
+ holders classification is based on [24], where the stakeholders
673
+ are categorized into three levels: industry-level, organization-
674
+ level, and team-level (see Fig. 4).
675
+
676
+ INT, 1
677
+ DE,1
678
+ SA, 1
679
+ US, 5
680
+ CA, 1
681
+ AU,1
682
+ NL, 2
683
+ EU,2
684
+ UK, 21
685
+ 3
686
+ 6
687
+ 1
688
+ 1
689
+ 2
690
+ 11
691
+ 1
692
+ 1
693
+ 1
694
+ 2
695
+ 1
696
+ 1
697
+ 1
698
+ 1
699
+ 3
700
+ 1
701
+ 1
702
+ 1(a) Principle overview.
703
+ (b) Principle coverage.
704
+ Fig. 3: Ethical principles covered by the identified frameworks.
705
+ Fig. 4: Stakeholders classification [24].
706
+ Fig. 5 shows that 10 of the collected frameworks (I1, I2,
707
+ I4, I5, I7, I8, I11, I12, I14, I15) have mentioned their targeted
708
+ stakeholders. For example, NIST’s AI RMF (I1) specifies
709
+ the framework is intended for “AI actors” defined by the
710
+ Organisation for Economic Co-operation and Development
711
+ (OECD), while EU’s ALTAI (I2) has listed the example
712
+ stakeholders in its guide on “How to complete ALTAI7”.
713
+ However, only the Netherlands BZK’s FRAIA (I4) has clearly
714
+ specified the different stakeholders associated with different
715
+ assessment stages to answer stage-specific questions.
716
+ The data synthesis results show that all 16 frameworks
717
+ involve the participation of RAI governors as the assessors and
718
+ development teams (e.g., data scientists, system developers) as
719
+ the assessee. RAI governors are those who set and enforce RAI
720
+ 7https://altai.insight-centre.org/Home/HowToComplete
721
+ Fig. 5: Stakeholders of collected frameworks.
722
+ policies within an organization or community, and they can be
723
+ internal or external.
724
+ One issue identified through the data synthesis process is
725
+ the lack of consideration of more diverse and inclusive (i.e.,
726
+ comprehensive) roles of stakeholders from different levels. For
727
+ example, industry-level procurers are largely neglected, with
728
+ only I1, I2 and I7 considering this aspect. Team-level speaking,
729
+ all 10 frameworks with identified stakeholders require input
730
+ from AI system development teams (i.e., assessees) on infor-
731
+ mation such as intended use, data source, data privacy, and
732
+ algorithm transparency. The assessees typically include prod-
733
+ uct managers, project managers, team leaders, data scientists,
734
+ and system developers. However, 9 out of 10 frameworks fail
735
+ to consider more diverse roles of assessees (e.g., architects,
736
+ UI/UX designers [7], [24]). The US NIST’s AI RMF (I1) is
737
+ distinguished by its inclusion of a broader range of stake-
738
+ holders involved in various stages of AI system development
739
+ and post-development, such as procurement, deployment, and
740
+ operations. However, I1 does not explicitly present categorized
741
+ assessments and mitigations based on different stakeholders.
742
+ Finding to RQ2.1 & RQ2.2: The current RAI risk as-
743
+ sessment frameworks are developed with ad-hoc scope
744
+ and focus, making them difficult for organizations to
745
+ use effectively in practice. This can be seen in the lack
746
+ of consideration for certain key stakeholders, lifecycle
747
+ stages, or ethical principles in their assessment and
748
+ mitigation, failing to identify and mitigate important
749
+ risks.
750
+ 3) RQ2.3 What is the scope of the frameworks?: With the
751
+ RQ, we aim to explore the scope of the existing AI risk
752
+ assessment frameworks.
753
+ a) RQ2.3.1: Which development stages are covered by the
754
+ frameworks?
755
+ This RQ aims to investigate the stages covered by the
756
+ collected frameworks in the AI system development lifecycle
757
+ (AI-SDLC). By referencing several existing sources with AI-
758
+ SDLC [1], [7], [12], [25], we first summarized the typical
759
+ stages included in (AI) SDLC (i.e., planning & requirement
760
+ analysis, design, implementation, testing, deployment, opera-
761
+ tion & monitoring). Then, we adapted the tasks in each stage
762
+
763
+ uestions
764
+ tlassified,
765
+ Principles
766
+ Principles
767
+ 5
768
+ not
769
+ specified, 11
770
+ Questions
771
+ specified,
772
+ not
773
+ 5
774
+ classified
775
+ 611
776
+ 11
777
+ 10
778
+ 9
779
+ 10
780
+ 10
781
+ 5
782
+ 8Industry-level stakeholders
783
+ AI technology producers/procurers
784
+ Al impacted subjects
785
+ AI solution producers/procurers
786
+ RAI governors
787
+ AI users/consumers
788
+ RAI tool producers/procurers
789
+ Organization-level stakeholders
790
+ Employees
791
+ Board members ·Executives
792
+ Managers·
793
+ Team-level stakeholders
794
+ Product managers · Project managers · Team leaders
795
+ Business analysts · Architects
796
+ ·UX/UI designers
797
+ Data scientists · Developers · Testers · OperatorsQuestions
798
+ not
799
+ Stakeholder
800
+ classified, 9
801
+ Stakeholder
802
+ specified, 10
803
+ not
804
+ specified, 6with additional AI-specific context and derived an AI-SDLC
805
+ (Fig. 6). The detailed results of the AI-SDLC stages covered
806
+ by the collected frameworks are presented in Table I.
807
+ Fig. 7 shows that 7 of the collected frameworks do not
808
+ specify AI system lifecycle stages. Although the other 9
809
+ frameworks (I1-I5, I7, I9, I12, I14) have clarified when they
810
+ can be applied during the AI system lifecycle, The UK ICO’s
811
+ AI and Data Protection Risk toolkit (I5) is the only one that has
812
+ categorized AI risk assessment and evaluation processes based
813
+ on different stages of the AI system lifecycle. Netherlands
814
+ BZK’s FRAIA (I4) is similarly structured in a more coarse-
815
+ grained way in that the assessment is conducted based on three
816
+ stages: input, throughput, and output.
817
+ 6 (I1, I2, I4, I5, I7, I12) out of 9 frameworks with specified
818
+ AI system lifecycle stages can be used to evaluate potential
819
+ risks throughout the entire AI system lifecycle. The other 3
820
+ frameworks (I3, I9, I14) focus on the initial stages of ideation
821
+ (i.e., planning & requirements analysis, design). In addition,
822
+ I3 covers the testing stage, while I14 covers the testing,
823
+ deployment and follow-up monitoring stages.
824
+ b) RQ2.3.2: Where can the frameworks be applied?
825
+ With the RQ, we aim to explore whether there are geograph-
826
+ ical constraints to applying the existing AI risk assessment
827
+ frameworks.
828
+ The government-developed frameworks (i.e., I1-I9) can be
829
+ applied anywhere, although some of them may require adjust-
830
+ ments considering region-specific elements in the frameworks.
831
+ For example, The UK ICO’s AI and Data Protection Risk
832
+ toolkit (I5) is aligned with the UK’s General Data Protection
833
+ Regulation (GDPR), and the AU NSW’s AI Assurance Frame-
834
+ work (I7) references relevant policies in the Australian state
835
+ of New South Wales. 6 out of 7 frameworks developed by
836
+ NGOs and industrial companies (I10-I13, I15-I16) are region-
837
+ agnostic. At the same time, I14 is specially designed for UK
838
+ NHS’s planned National Medical Imaging platform.
839
+ c) RQ2.3.2 Which domains/sectors are the frameworks
840
+ designed for?
841
+ This RQ intends to investigate the domains and sectors
842
+ where the frameworks can be applied.
843
+ Most of the collected frameworks are generally designed
844
+ across various domains. However, 5 frameworks have been
845
+ designed for specific purposes. FRAIA (I4), Model Rules (I10)
846
+ and Impact Assessment Tool for Public Authorities (I16) are
847
+ intended for evaluating the development and deployment of
848
+ AI systems in the public sector. The AI for Children toolkit
849
+ (I11) is specifically designed for AI systems that may impact
850
+ children and youth as potential users. AIAs in Healthcare
851
+ (I15) is intended to assess risks associated with designing and
852
+ developing AI systems that require access to the UK National
853
+ Medical Imaging Platform.
854
+ Finding to RQ2.3: The current RAI risk assessment
855
+ frameworks generally consider the entire lifecycle of
856
+ AI systems rather than focusing only on the AI model
857
+ pipeline. However, these frameworks do not provide
858
+ clear guidance on extending or adapting them to fit
859
+ diverse contexts. This limitation restricts the effective-
860
+ ness of RAI risk assessment frameworks as the risks
861
+ and mitigation may vary depending on the context (e.g.
862
+ different organizations, sectors, or regions) in which AI
863
+ systems are used.
864
+ C. RQ3: How are risks assessed?
865
+ This section presents the assessment processes of the col-
866
+ lection frameworks.
867
+ The frameworks are categorized into two types: procedural
868
+ and descriptive. The descriptive frameworks are less concrete
869
+ by providing general non-prescriptive assessment and mitiga-
870
+ tion and not referring to more specific and concrete solutions.
871
+ In contrast, procedural frameworks are more structured and in-
872
+ clude more detailed steps (e.g., inputs, processes, outputs) for
873
+ conducting AI risk assessments. The procedural frameworks
874
+ can also contain suggested mitigation solutions, assessment
875
+ templates, or checklists.
876
+ The collected frameworks examine underlying risks and/or
877
+ corresponding mitigation plans. To better present the results,
878
+ we summarize the different types of risks (i.e., risk factors)
879
+ the frameworks take into account. We adapted the risk cate-
880
+ gorization from a traditional risk management framework [26]
881
+ and added AI-specific context. The adapted risk factors are
882
+ categorized as follows:
883
+ • Hazard: A hazard refers to any dangerous situation or
884
+ condition arising from AI systems or related activities/ar-
885
+ tifacts that can result in harm to HSE wellbeing. Hazards
886
+ are sources of harm or exploit external to AI systems.
887
+ • Exposure: Exposure refers to individuals, property, sys-
888
+ tems, or other elements located within zones affected by
889
+ AI-related hazards that are therefore at risk of potential
890
+ losses.
891
+ • Vulnerability: Vulnerability pertains to the characteris-
892
+ tics and circumstances of an AI system or related artifacts
893
+ that make it susceptible to the detrimental effects of a
894
+ hazard. Compared to hazards, vulnerabilities are internal
895
+ weaknesses/issues of AI systems.
896
+ • Risks by/after mitigation (Mitigation risk): Mitigation
897
+ risks refer to the potential newly introduced risks brought
898
+ about by the implementation of specific mitigation, re-
899
+ silience, or control measures, or residual risks that persist
900
+ even after the implementation of mitigation measures.
901
+ For each of the collected frameworks, we summarized their
902
+ types (i.e., descriptive or procedural) and examined mitigation
903
+ measures and risk factors in Table I.
904
+ We only articulate the answers to RQ3.1 (framework inputs)
905
+ and RQ3.3 (framework outputs) for the procedural framework,
906
+ as the descriptive frameworks do not have direct inputs or
907
+ outputs. RQ3.2 (assessment processes) fits all frameworks, and
908
+ the answer is thus presented based on all frameworks collected.
909
+
910
+ Fig. 6: AI system lifecycle (adapted from [1], [7], [12], [25]).
911
+ Fig. 7: Stages covered by collected frameworks.
912
+ 1) RQ3.1: What are the inputs?: This RQ investigate the
913
+ inputs and the forms of inputs of the procedural frameworks.
914
+ The procedural frameworks are all based on certain forms of
915
+ questionnaires (e.g., self-assessment template, checklist etc.).
916
+ The inputs to these frameworks are answers to predefined
917
+ questions provided by relevant stakeholders (e.g., development
918
+ teams including system developers, data scientists etc.).
919
+ 2 frameworks (EU ALTAI, I2 and CA AIA, I3) are de-
920
+ signed as interactive online tools. Users can input the required
921
+ information about their AI systems and get instant feedback
922
+ based on their inputs. Similarly, I5 and I9 are based on
923
+ excel sheets where users can fill in system details or check
924
+ if the recommended practices for minimizing potentials are
925
+ met. I13 and I14 provide self-assessment templates where
926
+ predefined questions regarding the AI system (e.g., intended
927
+ use, stakeholders, benefits/harms) need to be answered. The
928
+ other seven procedural frameworks (I4, I7, I8, I10, I11, I15,
929
+ I16) are available as published reports, where more detailed
930
+ descriptions of the contexts are given. In these reports, AI
931
+ risk/impact assessment questionnaires/checklists are given. It
932
+ is important to note that in Q&A-style assessments, both the
933
+ quality of the answers and the underpinning methodology used
934
+ to generate them are crucial factors, rather than relying solely
935
+ on subjective inputs from the assessors.
936
+ Finding to RQ3.1: The current RAI risk assessment
937
+ frameworks primarily rely on subjective evaluation
938
+ from the assessors via a series of questions or check-
939
+ lists without the support of more objective tools and
940
+ techniques, leading to potentially biased results.
941
+ 2) RQ3.2: What are the processes?: This section discusses
942
+ how risk assessments are conducted in both descriptive and
943
+ procedural frameworks.
944
+ The descriptive industrial frameworks include AI RMF (I1)
945
+ by US NIST, Model AI governance framework (I6) by Sin-
946
+ gapore, and recommended practices for assessing the impact
947
+ of autonomous and intelligent systems on human well-being
948
+ (I12) by IEEE.
949
+ AI RMF (I1) is a framework with four components (map,
950
+ measure, manage, and govern) that gives organizations rec-
951
+ ommendations to adopt and adapt to their specific needs. AI
952
+ RMF (I1) is a non-prescriptive framework that aims to identify,
953
+ assess, and manage context-related risks by presenting desired
954
+ outcomes and general approaches for risk management. It pro-
955
+ motes the development of a culture of active risk management
956
+ through its recommendations and non-exhaustive solutions
957
+ presented in its companion playbook8. AI RMF (I1) is a
958
+ non-prescriptive framework that aims to identify, assess, and
959
+ manage context-related risks by presenting desired outcomes
960
+ and general approaches for risk management. It promotes the
961
+ development of a culture of active risk management through
962
+ its recommendations and non-exhaustive solutions presented
963
+ in its companion playbook. Similarly, Singapore’s Model
964
+ Framework (I6) and IEEE’s standard on AI impact assessment
965
+ (I12) are designed to be flexible by providing higher-level
966
+ guidance on the assessment processes.
967
+ 8https://pages.nist.gov/AIRMF/
968
+
969
+ Questions
970
+ not
971
+ Stage not
972
+ Stage
973
+ classified, 8
974
+ specified, 7
975
+ specified, 9Planning &
976
+ Business and ethical requirements analysis: identification of the system's concept and objectives, stakeholders (and possible impacts to
977
+ Requirements analysis
978
+ stakeholders), intended uses (application domain) etc. Ethical considerations (ethics application) etc
979
+ Architectural/structural design (e.g., software architecture design, AI/ML paradigm design (e.g., centralized/distributed/decentralized),
980
+ Design
981
+ detailed design of desired behavior of AI/non-AI components (e.g., UI design, data source identification, model/algorithm selection)
982
+ System construction of both AI (e.g., data collection and processing, existing/new model/algorithm creation/selection) and non-Al
983
+ Implementation
984
+ components, including unit testing and integration testing.
985
+ Implemented system tested against a finite set of test cases. AI/ML model(s) verified & validated on test data, model output calibrated
986
+ Testing
987
+ and interpreted
988
+ Deployment
989
+ Deploying (e.g., canary/blue-green/shadow deployment) the tested system, and verifying regulatory/ethical compliance
990
+ Operation &
991
+ Operating, continuously monitoring (assess both intended and unintended system/model output and impacts), feedback gathering, and
992
+ Monitoring
993
+ maintenance of the deployed systemAs for the procedural frameworks, the assessment pro-
994
+ cesses are based on the input answers, where potential risks
995
+ are identified through the Q&A processes. The assessment
996
+ and evaluation processes of various procedural frameworks
997
+ can be grouped into four categories: risk/principle-based (I2,
998
+ I5, I7, I8, I11, I16), system development process-based (I4,
999
+ I5), essential system component-based (I3, I9), and sys-
1000
+ tem description- and requirements-based (I10, I13, I14, I15).
1001
+ The risk/principle-based assessments include questions de-
1002
+ signed for each of the different risks/principles. The process-
1003
+ based assessments include questions throughout different AI-
1004
+ SDLC stages, from planning to monitoring & operations.
1005
+ The component-based assessments are formulated based on
1006
+ essential components (e.g., algorithms, data). The system
1007
+ description- and requirements-based solutions offer mecha-
1008
+ nisms for the assessee to provide information about their AI
1009
+ systems and reflect on compliance with specific requirements.
1010
+ For the more developed tools and frameworks, such as
1011
+ I2 and I3, the risk scores and potential risks are calculated
1012
+ automatically based on the selections/inputs. As for other
1013
+ procedural frameworks, such as report- or excel-based ones,
1014
+ they identify and assess risks by the assessment conductors
1015
+ through a more manual process. The assessors evaluate the
1016
+ system’s details, such as intended and unintended uses, stake-
1017
+ holders, data integrity, algorithmic explainability, and consult
1018
+ with external or internal stakeholders if necessary. This process
1019
+ enables a seemingly systematic analysis of an AI system to
1020
+ evaluate its impact and risks.
1021
+ A valuable part of the AI risk assessment is the mitigation
1022
+ plans suggested by some frameworks. In Table I, we summa-
1023
+ rize whether clear mitigation considerations are included in the
1024
+ frameworks by examining the questions/recommendations in-
1025
+ cluded in each of the 16 frameworks (Yes: Mitigation specified.
1026
+ *Yes: Mitigation included but not specified. No: Mitigation not
1027
+ included). Only 5 (I2, I3, I6, I10, I11) out of 16 frameworks
1028
+ have specified mitigation-related aspects. 7 frameworks (I1,
1029
+ I4, I7, I12, I14-I16) have more or less included risk mitigation
1030
+ measures without clearly specifying them. 4 frameworks (I5,
1031
+ I8, I9, I13) do not cover mitigation.
1032
+ As for the risk factors, none of the frameworks specified
1033
+ the different risk factors they considered. However, given the
1034
+ potential value of such categorization in helping organizations
1035
+ better triage and prioritize risks, we examined the frameworks
1036
+ and their questions/recommendations and extracted the risk
1037
+ factors each framework takes into account (see Table I and Fig.
1038
+ 8). Despite their respective focus (e.g., I14 focuses on hazards
1039
+ while touching vulnerability and exposure), all 16 frameworks
1040
+ consider potential vulnerability, and 15 frameworks cover haz-
1041
+ ard and exposure. However, mitigation risks are significantly
1042
+ underemphasized and only covered by AU NSW AI Assurance
1043
+ Framework (I7) and ECP’s AI impact assessment framework
1044
+ (I15). Even for these two frameworks that consider mitigation
1045
+ risks, they do not provide a comprehensive assessment rather
1046
+ briefly mention such risks. For example, in I7: “Are there any
1047
+ residual risks?”, and in I15: “Considering planned mitigations,
1048
+ could the AI system cause significant or irreversible harms?”.
1049
+ Fig. 8: Risk factors considered by collected frameworks.
1050
+ Finding 1 to RQ3.2: RAI risk assessment frameworks
1051
+ need to distinguish among risk factors (i.e., hazard,
1052
+ exposure, vulnerability, and mitigation risk). Although
1053
+ collected frameworks categorically encompass these
1054
+ factors to some degree, they may focus on particular
1055
+ factors while briefly touching on others. Further, mit-
1056
+ igation risks are significantly neglected. This can lead
1057
+ to potential failure to identify and mitigate crucial RAI
1058
+ risks.
1059
+ Finding 2 to RQ3.2: Existing RAI risk assessment
1060
+ frameworks provide some information on assessment
1061
+ procedures but fail to clearly specify inputs/outputs,
1062
+ stakeholders, and resources needed at each step.
1063
+ Finding 3 to RQ3.2: Many RAI risk assessment frame-
1064
+ works plainly list assessment measures (e.g., questions,
1065
+ checklists) without considering their interconnections
1066
+ or dependencies, leading to an inefficient assessment
1067
+ process.
1068
+ 3) RQ3.3: What are the outputs?: This section discuss the
1069
+ outputs of the procedural frameworks.
1070
+ Whether the output of a documented report is specified or
1071
+ not, the outputs of the procedural frameworks are, or at least
1072
+ should be, risk/impact assessment reports. Some frameworks,
1073
+ such as I2, which creates a visualization of the risk level
1074
+ correlated to the RAI principles, and I3, which calculates
1075
+ risk and mitigation scores for various risk areas and gener-
1076
+ ates the level of impact, generate reports automatically. For
1077
+ I10, assessors must manually generate a report based on the
1078
+ questionnaire and their answers to the questions. The other
1079
+ procedural frameworks (I4, I5, I7-I9, I11, I13-I16) serve as
1080
+ (self-)assessment tools to guide assessors in identifying risks
1081
+ and do not require the preparation of a report. However, since
1082
+ assessors should clearly document all the answers and the
1083
+ related questions when using the procedural frameworks, the
1084
+ processes result in documented assessment reports.
1085
+
1086
+ 15
1087
+ 15
1088
+ 16
1089
+ 2Finding to RQ3: While organizations may rely on RAI
1090
+ risk assessment frameworks for potential mitigation
1091
+ solutions, current frameworks either fail to provide
1092
+ concrete mitigation solutions, or lack a structured way
1093
+ to present the solutions. This makes it challenging for
1094
+ organizations to address identified risks effectively.
1095
+ IV. DISCUSSION
1096
+ A. On the concreteness of RAI risk assessment frameworks
1097
+ 1) Relative concreteness: A risk assessment framework
1098
+ may appear concrete at one level but too abstract for the
1099
+ next. For example, management teams may consider certain
1100
+ assessment questions concrete, while development teams may
1101
+ find them not doable. Additionally, even seemingly concrete
1102
+ checklists or templates for RAI risk assessment may only
1103
+ be effective if assessors have a standardized and trustworthy
1104
+ approach to completing each item. Therefore, it is essential
1105
+ to have well-structured, concrete, and reusable solutions (e.g.,
1106
+ design patterns [24], [27]) in the lower level that align/connect
1107
+ with higher-level practices such as governance guidelines to
1108
+ ensure a comprehensive and effective risk assessment.
1109
+ 2) Trivialized concreteness: Our mapping study reveals that
1110
+ many frameworks trivialize the concept of “concreteness” by:
1111
+ • Applying existing assessment concepts to new AI-specific
1112
+ artifacts/processes without further specifying potential
1113
+ solutions. Examples include acknowledging the existence
1114
+ of RAI risks and broadly mentioning that they need to
1115
+ be identified, documented, and mitigated.
1116
+ • Identifying new concepts in AI and providing some
1117
+ sub-categorization without providing potential solutions.
1118
+ Examples include acknowledging bias as a common issue
1119
+ in AI systems and listing different sources of bias (e.g.,
1120
+ data, algorithm), but not providing specified solutions to
1121
+ different biases.
1122
+ • Identifying important RAI risks and referring to poten-
1123
+ tially stale non-AI frameworks, which may not be suitable
1124
+ for addressing RAI risks.
1125
+ It’s important to note that while higher-level frameworks
1126
+ may seem abstract, they do not always trivialize concreteness.
1127
+ These frameworks are generally more abstract because they
1128
+ need to be widely applicable and less prone to obsolescence.
1129
+ They can be helpful, particularly for management teams,
1130
+ as they point out areas where organizations can uplift their
1131
+ practices. The key criteria for determining if a higher-level
1132
+ framework is concrete or not include:
1133
+ • Whether high-level abstractions of potential assessment
1134
+ and/or mitigation measures are underpinned (specified or
1135
+ reasonably inferable) by lower-level concrete assessment
1136
+ techniques.
1137
+ • Whether there is a clear understanding among higher-
1138
+ level stakeholders (e.g., management) about the inputs,
1139
+ processes, outputs, as well as required personnel and
1140
+ resources to complete the assessment. This understanding
1141
+ may not necessarily require technical expertise but rather
1142
+ an understanding of the trust placed in the lower-level
1143
+ concrete assessment techniques utilized.
1144
+ B. Essentials to “Concreteness”
1145
+ We summarize the essential qualities, elements, and pro-
1146
+ cesses that a concrete RAI risk assessment framework should
1147
+ process in Fig. 9.
1148
+ A concrete RAI risk assessment framework should have the
1149
+ following characteristics: 1) The assessment and/or mitigation
1150
+ (e.g., questions/checklists/recommendations) proposed at one
1151
+ level/stage are reasonably underpinned/aligned/connected to
1152
+ other level/stage even if the measure itself is narrow-scoped
1153
+ and not directly covering different levels/stages. 2) The or-
1154
+ ganization of assessment and mitigation should be layered,
1155
+ considering their dependencies on each other. This allows
1156
+ for a clearer assessment logic and more efficient assessment
1157
+ processes. Currently, only EU’s ALTAI (I2) achieves a certain
1158
+ level of interconnectivity by providing an interactive online
1159
+ assessment tool where the following questions may vary
1160
+ depending on the previous answers. 3) The framework should
1161
+ be extensible, dynamic, and adaptive in that it can be adapted
1162
+ and extended to more specific contexts. All the qualities above
1163
+ result in enhanced assessment efficiency.
1164
+ The elements to be covered by a comprehensive RAI
1165
+ risk assessment framework should include different dimen-
1166
+ sions, contexts, measurements, and mitigations. Assessment
1167
+ and mitigation should be organized based on different RAI
1168
+ principles, RAI stakeholders, and AI-SDLC stages. Existing
1169
+ organizational governance structures and measures should also
1170
+ be considered. Even if a framework focuses on a specific
1171
+ aspect (e.g., stage/principle-specific, designed for assessment
1172
+ instead of mitigation), it needs to be well connected (i.e., in-
1173
+ terconnected) with other aspects. Additionally, the framework
1174
+ should consider and specify contextual elements such as appli-
1175
+ cable regions, sectors, and compatibility with an organization’s
1176
+ existing risk management processes and structures. Moreover,
1177
+ different risk factors should be considered, and corresponding
1178
+ assessment and mitigation be presented. Especially, mitiga-
1179
+ tion risks are significantly neglected by existing frameworks.
1180
+ Reusable mitigation plans should be suggested in a structured
1181
+ way, along with their pros and cons considered.
1182
+ Specifications on the procedures required to conduct the
1183
+ RAI risk assessment can help assessors and assessees from
1184
+ different levels better understand the inputs/processes/outputs
1185
+ and required stakeholders and resources (e.g., data, tools,
1186
+ funds) for each step. However, only half (I1, I3, I4, I7, I10, I12,
1187
+ I14, I15) of the 16 frameworks provided such specifications
1188
+ to a certain extent. Furthermore, 7 out of the 8 frameworks
1189
+ merely stated the steps needed to conduct the assessment, with
1190
+ only I4 specifying stakeholders involved in each step. None
1191
+ of the frameworks provides details on the resources required
1192
+ to complete each step.
1193
+ C. Threats to validity
1194
+ External. The term “AI risk assessment” along with a set
1195
+ of other terms such as “AI risk management” and “AI impact
1196
+
1197
+ Fig. 9: Essentials to building a concrete RAI risk assessment framework.
1198
+ assessment” has been used to mean largely the same topic:
1199
+ identification, assessment/measurement, and mitigation of RAI
1200
+ risks. We extended our search terms with a set of supportive
1201
+ terms that are being used interchangeably in the search string
1202
+ to ensure that all the relevant work were covered to mitigate
1203
+ this issue. Another issue is that we only include the publicly
1204
+ accessible RAI risk assessment frameworks, although some
1205
+ organizations have their own frameworks for internal use.
1206
+ Internal. To mitigate the threat of not finding all rele-
1207
+ vant studies, we conducted a rigorous search using defined
1208
+ keywords with support terms and conducted snowballing to
1209
+ recover the missing studies from the literature. To address
1210
+ the bias from data collection and synthesis, one researcher
1211
+ performed the tasks and the other researcher reviewed and
1212
+ double-checked the results. The two researchers discussed the
1213
+ inconsistency and reached a common ground.
1214
+ V. RELATED WORK
1215
+ The pressing need to manage RAI risks has attracted
1216
+ significant attention in both industry and academia. Many
1217
+ studies on RAI risks have been published in recent years.
1218
+ However, they heavily focus on AI risk conceptualization and
1219
+ taxonomy (e.g., [28]–[31]) and provide no concrete solutions
1220
+ (i.e., assessment/mitigation techniques) to RAI risks. With the
1221
+ increasing interest in managing RAI risks, more actionable
1222
+ solutions to managing RAI risks have been proposed. Zhang
1223
+ et al. [32] proposed to evaluate model risks by inspecting their
1224
+ behavior on counterfactuals. Schwee et al. [33] introduced a
1225
+ toolchain for assessing privacy risks. The toolchain takes in
1226
+ a model trained from the dataset to be shared and creates a
1227
+ privacy risk report. Yajima et al. [34] showcased their work in
1228
+ progress on assessing machine learning security risks. Failure
1229
+ mode and effect analysis (FMEA) has been adopted/extended
1230
+ for assessing RAI risks in [35]–[37].
1231
+ Notably, EY and Trilateral Research published a survey of
1232
+ AI risk assessment methodologies in January 2022 [38]. With
1233
+ the objective of providing RAI governors with noteworthy
1234
+ practices and regulations in the field, this survey presents
1235
+ a high-level overview of the global landscape of AI risk
1236
+ assessment. The report discusses: 1) regulations and legislation
1237
+ worldwide containing AI risk assessment related elements;
1238
+ 2) solutions to RAI risk assessment by several international
1239
+ organizations; 3) standards related to AI risk management
1240
+ and governance; 4) a brief overview of part of the proposed
1241
+ approaches in industry and academia. While this report is cate-
1242
+ gorically comprehensive, it mainly aims to help RAI governors
1243
+ grasp the worldwide outline of the field. Furthermore, it lacks
1244
+ detailed and systematic analysis of the existing frameworks.
1245
+ In contrast, the objective of this study is to provide RAI
1246
+ practitioners with a systematic summary of the existing RAI
1247
+ risk assessment frameworks, and shed light on the future
1248
+
1249
+ Interconnected
1250
+ Layered
1251
+ Industry-level
1252
+ Efficient
1253
+ Organization-level
1254
+ Different levels
1255
+ Qualities
1256
+ Extensible
1257
+ Team-level
1258
+ Dynamic
1259
+ RAI principles
1260
+ Assessors
1261
+ Adaptive
1262
+ RAI Stakeholders
1263
+ Different Roles
1264
+ Assessees
1265
+ Dimension
1266
+ AI software
1267
+ lifecycle stages
1268
+ Existing (other) risk
1269
+ assessment frameworks
1270
+ Region
1271
+ Governance structure
1272
+ Essentials to
1273
+ Elements
1274
+ Context
1275
+ concreteness
1276
+ Sector
1277
+ Hazard
1278
+ Organization
1279
+ Exposure
1280
+ Measurement
1281
+ Vulnerability
1282
+ Risk factors
1283
+ Tools and techniques
1284
+ Mitigation risk
1285
+ for assessment
1286
+ Control
1287
+ Mitigation
1288
+ Reusable solutions
1289
+ Steps
1290
+ Stakeholders
1291
+ Processes
1292
+ Inputs & outputs
1293
+ Resourcesdevelopment of concrete RAI risk assessment frameworks.
1294
+ VI. CONCLUSION AND FUTURE WORK
1295
+ This paper conducts a systematic mapping study to evaluate
1296
+ the capabilities and limitations of existing RAI risk assessment
1297
+ frameworks. We examine key characteristics of a concrete
1298
+ framework, including specified RAI principles, stakeholders,
1299
+ AI system lifecycle stages, applicable regions and sectors, risk
1300
+ factors, and reusable mitigations. We provide insights to help
1301
+ facilitating the development of concrete RAI risk assessment
1302
+ frameworks. These includes presenting the assessment and
1303
+ mitigation measures in an interconnected and layered way and
1304
+ specifying the assessment procedures as well as associated
1305
+ inputs/outputs, stakeholders, and resources. For future work,
1306
+ we are developing a question bank with questions clearly
1307
+ labelled with respect to different characteristics, mitigations,
1308
+ and risk factors etc. Based on the question bank, we plan to
1309
+ develop a concrete RAI risk assessment framework.
1310
+ REFERENCES
1311
+ [1] Q. Lu, L. Zhu, X. Xu, J. Whittle, and Z. Xing, “Towards a
1312
+ roadmap on software engineering for responsible ai,” in Proceedings
1313
+ of the 1st International Conference on AI Engineering: Software
1314
+ Engineering for AI, ser. CAIN ’22.
1315
+ New York, NY, USA: Association
1316
+ for Computing Machinery, 2022, p. 101–112. [Online]. Available:
1317
+ https://doi.org/10.1145/3522664.3528607
1318
+ [2] N. Sambasivan, S. Kapania, H. Highfill, D. Akrong, P. Paritosh, and
1319
+ L. M. Aroyo, ““everyone wants to do the model work, not the data
1320
+ work”: Data cascades in high-stakes ai,” in proceedings of the 2021
1321
+ CHI Conference on Human Factors in Computing Systems, 2021, pp.
1322
+ 1–15.
1323
+ [3] J. Gao, J. Yao, and Y. Shao, “Towards reliable learning for high stakes
1324
+ applications,” in Proceedings of the AAAI Conference on Artificial
1325
+ Intelligence, vol. 33, no. 01, 2019, pp. 3614–3621.
1326
+ [4] X. Ferrer, T. van Nuenen, J. M. Such, M. Cot´e, and N. Criado, “Bias and
1327
+ discrimination in ai: a cross-disciplinary perspective,” IEEE Technology
1328
+ and Society Magazine, vol. 40, no. 2, pp. 72–80, 2021.
1329
+ [5] S. F. Nasim, M. R. Ali, and U. Kulsoom, “Artificial intelligence incidents
1330
+ & ethics a narrative review,” International Journal of Technology,
1331
+ Innovation and Management (IJTIM), vol. 2, no. 2, 2022.
1332
+ [6] P. R. Nagbøl, O. M¨uller, and O. Krancher, “Designing a risk assessment
1333
+ tool for artificial intelligence systems,” in International Conference
1334
+ on Design Science Research in Information Systems and Technology.
1335
+ Springer, 2021, pp. 328–339.
1336
+ [7] US National Institute of Standards and Technology, NIST, “AI
1337
+ Risk Management Framework: Second Draft.” [Online]. Available:
1338
+ https://https://www.nist.gov/system/files/documents/2022/08/18/AI
1339
+ RMF 2nd draft.pdf
1340
+ [8] European
1341
+ Commission
1342
+ High-Level
1343
+ Expert
1344
+ Group
1345
+ on
1346
+ Artificial
1347
+ Intelligence,
1348
+ “The
1349
+ Assessment
1350
+ List
1351
+ for
1352
+ Trustworthy
1353
+ Artificial
1354
+ Intelligence.” [Online]. Available: https://altai.insight-centre.org/
1355
+ [9] B. Kitchenham, O. P. Brereton, D. Budgen, M. Turner, J. Bailey, and
1356
+ S. Linkman, “Systematic literature reviews in software engineering–
1357
+ a systematic literature review,” Information and software technology,
1358
+ vol. 51, no. 1, pp. 7–15, 2009.
1359
+ [10] Government of Canada, “Algorithmic Impact Assessment tool .”
1360
+ [Online].
1361
+ Available:
1362
+ https://www.canada.ca/en/government/system/
1363
+ digital-government/digital-government-innovations/responsible-use-ai/
1364
+ algorithmic-impact-assessment.html
1365
+ [11] Government
1366
+ of
1367
+ the
1368
+ Netherlands,
1369
+ Ministry
1370
+ of
1371
+ the
1372
+ Interior
1373
+ and
1374
+ Kingdom Relations, “Fundamental Rights and Algorithms Impact
1375
+ Assessment (FRAIA) .” [Online]. Available: https://www.government.nl/
1376
+ documents/reports/2021/07/31/impact-assessment-fundamental-rights-
1377
+ and- algorithms#: ∼:text=The%20Fundamental%20Rights%20and%
1378
+ 20Algorithm,deployment%20of%20an%20algorithmic%20system.
1379
+ [12] UK Information Commissioner’s Office, “AI and data protection risk
1380
+ toolkit.” [Online]. Available: https://ico.org.uk/for-organisations/guide-
1381
+ to-data-protection/key-dp-themes/guidance-on-ai-and-data-protection/
1382
+ ai-and-data-protection-risk-toolkit/
1383
+ [13] Personal
1384
+ Data
1385
+ Protection
1386
+ Commission
1387
+ Singapore,
1388
+ “Model
1389
+ AI
1390
+ Governance Framework.” [Online]. Available: https://www.pdpc.gov.sg/
1391
+ help-and-resources/2020/01/model-ai-governance-framework
1392
+ [14] Australia
1393
+ NSW
1394
+ Government,
1395
+ “NSW
1396
+ Artificial
1397
+ Intelli-
1398
+ gence
1399
+ Assurance
1400
+ Framework.”
1401
+ [Online].
1402
+ Available:
1403
+ https:
1404
+ //www.digital.nsw.gov.au/policy/artificial- intelligence/nsw- artificial-
1405
+ intelligence-assurance-framework
1406
+ [15] GovEx, the City and County of San Francisco, Harvard DataSmart, Data
1407
+ Community DC, “Ethics & Algorithms Toolkit.” [Online]. Available:
1408
+ http://ethicstoolkit.ai/
1409
+ [16] US
1410
+ Pennsylvania
1411
+ Office
1412
+ of
1413
+ Administration,
1414
+ “RFD BUS012A
1415
+ Artificial
1416
+ Intelligence
1417
+ Assessment
1418
+ Too.”
1419
+ [Online].
1420
+ Available:
1421
+ https://www.oa.pa.gov/Policies/Documents/rfd bus012a.xlsx
1422
+ [17] European Law Institute, “Model Rules on Impact Assessment of Algo-
1423
+ rithmic Decision-Making Systems Used by Public Administration.” [On-
1424
+ line]. Available: https://www.europeanlawinstitute.eu/fileadmin/user
1425
+ upload/p eli/Publications/ELI Model Rules on Impact Assessment
1426
+ of ADMSs Used by Public Administration.pdf
1427
+ [18] World Economic Forum, “Artificial Intelligence for Children Toolkit.”
1428
+ [Online]. Available: https://www3.weforum.org/docs/WEF Artificial
1429
+ Intelligence for Children 2022.pdf
1430
+ [19] “Ieee recommended practice for assessing the impact of autonomous
1431
+ and intelligent systems on human well-being,” IEEE Std 7010-2020, pp.
1432
+ 1–96, 2020.
1433
+ [20] Microsoft, “Microsoft Responsible AI Impact Assessment Template.”
1434
+ [Online]. Available: https://blogs.microsoft.com/wp-content/uploads/
1435
+ prod/sites/5/2022/06/Microsoft-RAI-Impact-Assessment-Template.pdf
1436
+ [21] Ada
1437
+ Lovelace
1438
+ Institute,
1439
+ “Algorithmic
1440
+ impact
1441
+ assessment
1442
+ in
1443
+ healthcare.” [Online]. Available: https://www.adalovelaceinstitute.org/
1444
+ project/algorithmic-impact-assessment-healthcare/
1445
+ [22] ECP, Platform for the Information Society, “Artificial Intelligence Impact
1446
+ Assessment (English version).” [Online]. Available: https://ecp.nl/
1447
+ publicatie/artificial-intelligence-impact-assessment-english-version/
1448
+ [23] Algorithm Watch, “Automated Decision-Making Systems in the Public
1449
+ Sector: An Impact Assessment Tool for Public Authorities.” [Online].
1450
+ Available:
1451
+ https://uploads- ssl.webflow.com/5f57d40eb1c2ef22d8a8
1452
+ ca7e/61a8efeb70b677a50f9c24cb 2021 AW Decision Public Sector
1453
+ EN v5.pdf
1454
+ [24] Q. Lu, L. Zhu, X. Xu, J. Whittle, D. Zowghi, and A. Jacquet, “Respon-
1455
+ sible ai pattern catalogue: a multivocal literature review,” arXiv preprint
1456
+ arXiv:2209.04963, 2022.
1457
+ [25] S. Amershi, A. Begel, C. Bird, R. DeLine, H. Gall, E. Kamar, N. Na-
1458
+ gappan, B. Nushi, and T. Zimmermann, “Software engineering for
1459
+ machine learning: A case study,” in 2019 IEEE/ACM 41st International
1460
+ Conference on Software Engineering: Software Engineering in Practice
1461
+ (ICSE-SEIP).
1462
+ IEEE, 2019, pp. 291–300.
1463
+ [26] Emergency Management Victoria, “Colac Otway Shire Municipal
1464
+ Emergency Management Plan: 4.3 Hazard, Exposure, Vulnerability and
1465
+ Resilience.” [Online]. Available: http://memp.colacotway.vic.gov.au/
1466
+ ch01s04s03.php
1467
+ [27] Q. Lu, L. Zhu, X. Xu, and J. Whittle, “Responsible-ai-by-design: a
1468
+ pattern collection for designing responsible ai systems,” IEEE Software,
1469
+ 2023.
1470
+ [28] B. Perry and R. Uuk, “Ai governance and the policymaking process: key
1471
+ considerations for reducing ai risk,” Big data and cognitive computing,
1472
+ vol. 3, no. 2, p. 26, 2019.
1473
+ [29] K. Sotala, “Disjunctive scenarios of catastrophic ai risk,” in Artificial
1474
+ Intelligence Safety and Security.
1475
+ Chapman and Hall/CRC, 2018, pp.
1476
+ 315–337.
1477
+ [30] M. Ciupa and K. Abney, “Conceptualizing ai risk,” in Proceedings of
1478
+ the 8th International Conference on Computer Science, Engineering and
1479
+ Applications (CCSEA 2018), 2018, pp. 73–81.
1480
+ [31] C. Garvey, “Ai risk mitigation through democratic governance: Intro-
1481
+ ducing the 7-dimensional ai risk horizon,” in Proceedings of the 2018
1482
+ AAAI/ACM Conference on AI, Ethics, and Society, 2018, pp. 366–367.
1483
+ [32] Z. Zhang, V. Setty, and A. Anand, “Sparcassist: A model risk assessment
1484
+ assistant based on sparse generated counterfactuals,” in Proceedings
1485
+ of the 45th International ACM SIGIR Conference on Research and
1486
+ Development in Information Retrieval, ser. SIGIR ’22.
1487
+ New York,
1488
+ NY, USA: Association for Computing Machinery, 2022, p. 3219–3223.
1489
+
1490
+ [33] J. H. Schwee, F. C. Sangogboye, F. D. Salim, and M. B. Kjærgaard,
1491
+ “Tool-chain for supporting privacy risk assessments,” in Proceedings of
1492
+ the 7th ACM International Conference on Systems for Energy-Efficient
1493
+ Buildings, Cities, and Transportation, 2020, pp. 140–149.
1494
+ [34] J. Yajima, M. Inui, T. Oikawa, F. Kasahara, I. Morikawa, and
1495
+ N. Yoshioka, “A new approach for machine learning security risk
1496
+ assessment: Work in progress,” in Proceedings of the 1st International
1497
+ Conference on AI Engineering: Software Engineering for AI, ser. CAIN
1498
+ ’22. New York, NY, USA: Association for Computing Machinery, 2022,
1499
+ p. 52–53. [Online]. Available: https://doi.org/10.1145/3522664.3528613
1500
+ [35] G. A. Dominguez, K. Kawaai, and H. Maruyama, “Fails: a tool for
1501
+ assessing risk in ml systems,” in 2021 28th Asia-Pacific Software
1502
+ Engineering Conference Workshops (APSEC Workshops).
1503
+ IEEE, 2021,
1504
+ pp. 1–4.
1505
+ [36] S. Rismani and A. Moon, “How do ai systems fail socially?: an engi-
1506
+ neering risk analysis approach,” in 2021 IEEE International Symposium
1507
+ on Ethics in Engineering, Science and Technology (ETHICS).
1508
+ IEEE,
1509
+ 2021, pp. 1–8.
1510
+ [37] J. Li and M. Chignell, “Fmea-ai: Ai fairness impact assessment using
1511
+ failure mode and effects analysis,” AI and Ethics, pp. 1–14, 2022.
1512
+ [38] EY,
1513
+ Trilateral
1514
+ Research,
1515
+ “A
1516
+ survey
1517
+ of
1518
+ artificial
1519
+ intelligence
1520
+ risk
1521
+ assessment
1522
+ methodologies:
1523
+ The
1524
+ global
1525
+ state
1526
+ of
1527
+ play
1528
+ and
1529
+ leading
1530
+ practices
1531
+ identified.”
1532
+ [Online].
1533
+ Available:
1534
+ https:
1535
+ //www.trilateralresearch.com/wp-content/uploads/2022/01/A-survey-of-
1536
+ AI-Risk-Assessment-Methodologies-full-report.pdf
1537
+
CdFJT4oBgHgl3EQftC3b/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
GNFLT4oBgHgl3EQfGi_B/content/tmp_files/2301.11993v1.pdf.txt ADDED
@@ -0,0 +1,3951 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Quantum Langevin theory for two coupled phase-conjugated electromagnetic waves
2
+ Yue Jiang,1, 2, ∗ Yefeng Mei,3, † and Shengwang Du4, ‡
3
+ 1JILA, National Institute of Standards and Technology and the University of Colorado, Boulder, Colorado 80309, USA
4
+ 2Department of Physics, University of Colorado, Boulder, Colorado 80309, USA
5
+ 3Department of Physics, University of Michigan, Ann Arbor, Michigan 48109, USA
6
+ 4Department of Physics, The University of Texas at Dallas, Richardson, Texas 75080, USA
7
+ (Dated: January 31, 2023)
8
+ While loss-gain-induced Langevin noises have been intensively studied in quantum optics, the ef-
9
+ fect of a complex-valued nonlinear coupling coefficient on the noises of two coupled phase-conjugated
10
+ optical fields has never been questioned before. Here, we provide a general macroscopic phenomeno-
11
+ logical formula of quantum Langevin equations for two coupled phase-conjugated fields with linear
12
+ loss (gain) and complex nonlinear coupling coefficient. The macroscopic phenomenological formula
13
+ is obtained from the coupling matrix to preserve the field commutation relations and correlations,
14
+ which does not require knowing the microscopic details of light-matter interaction and internal
15
+ atomic structures. To validate this phenomenological formula, we take spontaneous four-wave mix-
16
+ ing in a double-Λ four-level atomic system as an example to numerically confirm that our macroscopic
17
+ phenomenological result is consistent with that obtained from the microscopic Heisenberg-Langevin
18
+ theory. Finally, we apply the quantum Langevin equations to study the effects of linear gain and
19
+ loss, complex phase mismatching, as well as complex nonlinear coupling coefficient in entangled
20
+ photon pair (biphoton) generation, particularly to their temporal quantum correlations.
21
+ I.
22
+ INTRODUCTION
23
+ Quantum Langevin equations is a common approach
24
+ to studying an open quantum system involving loss or
25
+ gain, where the stochastic coupling between the system
26
+ and its environment is molded as a set of Langevin noise
27
+ operators [1–5]. For example, in the parametric down-
28
+ conversion (PDC) process, a pump laser beam passes
29
+ through a χ(2) nonlinear crystal and is down-converted
30
+ into a pair of phase-conjugated electromagnetic (EM)
31
+ waves.
32
+ In the simplest case with the perfect phase-
33
+ matching condition and an undepleted pump beam, with-
34
+ out linear loss or gain, the two phase-conjugated single-
35
+ mode fields are governed by the following coupled equa-
36
+ tions [6]
37
+
38
+ ∂z
39
+ �ˆa1
40
+ ˆa†
41
+ 2
42
+
43
+ = M
44
+ �ˆa1
45
+ ˆa†
46
+ 2
47
+
48
+ =
49
+
50
+ 0
51
+
52
+ −iκ
53
+ 0
54
+ � �ˆa1
55
+ ˆa†
56
+ 2
57
+
58
+ ,
59
+ (1)
60
+ where ˆam and ˆa†
61
+ m (m = 1, 2) are the field annihilation
62
+ and creation operators, M is the 2 × 2 coupling matrix,
63
+ and κ is the (real) nonlinear coupling coefficient. Here
64
+ we consider only the forward-wave case with both fields
65
+ propagating along the same +z direction. If losses are
66
+ presented during the propagation of the two fields, the
67
+ coupling matrix is
68
+ M =
69
+
70
+ −α1
71
+
72
+ −iκ −α2
73
+
74
+ ,
75
+ (2)
76
77
78
79
+ and their coupled equations become [3, 7]
80
+
81
+ ∂z
82
+ �ˆa1
83
+ ˆa†
84
+ 2
85
+
86
+ =
87
+
88
+ −α1
89
+
90
+ −iκ −α2
91
+ � �ˆa1
92
+ ˆa†
93
+ 2
94
+
95
+ +
96
+ �√2α1 ˆf1
97
+ √2α2 ˆf †
98
+ 2
99
+
100
+ ,
101
+ (3)
102
+ where αm > 0 are the loss (absorption) coefficients,
103
+ and ˆfm are the associated Langevin noise operators sat-
104
+ isfying [ ˆfm(ω, z), ˆf †
105
+ n(ω′, z′)] = δmnδ(ω − ω′)δ(z − z′).
106
+ If there is linear gain instead of loss, for example in
107
+ channel 1, i.e., α1 < 0, equation (3) can be modi-
108
+ fied by taking √2α1 ˆf1 → √−2α1 ˆf †
109
+ 1.
110
+ One can show
111
+ that these Langevin noise operators are necessary to pre-
112
+ serve the commutation relations during propagation, i.e.
113
+ [ˆam(ω, z), ˆa†
114
+ n(ω′, z)] = [ˆam(ω, 0), ˆa†
115
+ n(ω′, 0)] = δmnδ(ω −
116
+ ω′).
117
+ Equation (3) has been widely applied for PDC pro-
118
+ cesses where the nonlinear coupling coefficient κ is real
119
+ [3, 7–9].
120
+ However, in a more general case of cou-
121
+ pled phase-conjugated fields, such as four-wave mixing
122
+ (FWM) near atomic resonances [10–12], the nonlinear
123
+ coupling coefficient κ can take a complex value involving
124
+ complicated atomic transitions.
125
+ In this case, equation
126
+ (3) is not valid and its solution does not preserve com-
127
+ mutation relations of the fields. What are the general
128
+ quantum Langevin coupled equations accounting for the
129
+ complex nonlinear coupling coefficient?
130
+ To answer the question, the common approach is to
131
+ derive quantum Langevin equations by solving the light-
132
+ matter coupled Heisenberg equations, which requires
133
+ knowing microscopic details of light-matter interaction
134
+ such as atomic populations and transitions [11–13]. The
135
+ complexity of this approach increases dramatically as
136
+ more atomic transitions are involved and it is extremely
137
+ difficult for experimentalists to follow, particularly in
138
+ some situations where it is impossible to obtain full mi-
139
+ croscopic details. Then our reduced question becomes:
140
+ arXiv:2301.11993v1 [quant-ph] 27 Jan 2023
141
+
142
+ 2
143
+ Is it possible to obtain self-consistent quantum Langevin
144
+ coupled equations from the general expression of the cou-
145
+ pling matrix? We call this the macroscopic phenomeno-
146
+ logical approach. To our best knowledge, there has been
147
+ no published work in investigating Langevin noises in-
148
+ duced by a complex nonlinear coupling coefficient κ.
149
+ In this article, for the first time, we provide a gen-
150
+ eral macroscopic phenomenological formula of quantum
151
+ Langevin equations for two coupled phase-conjugated
152
+ fields with linear loss (gain) and complex nonlinear cou-
153
+ pling coefficient, in both forward- and backward-wave
154
+ configurations. The macroscopic phenomenological for-
155
+ mula is obtained from the coupling matrix by preserv-
156
+ ing commutation relations and correlations of the fields,
157
+ which does not require knowing the microscopic details of
158
+ light-matter interaction and internal atomic structures.
159
+ We aim to make it readable and accessible for experi-
160
+ mental researchers in the quantum optics community.
161
+ This article is structured as follows. In Sec. II, to ful-
162
+ fill the requirement of preserving commutation relations,
163
+ we formulate the general macroscopic phenomenologi-
164
+ cal quantum Langevin coupled equations and their solu-
165
+ tions from the coupling matrix taking into account linear
166
+ loss (gain) and complex nonlinear coupling coefficient,
167
+ in both forward- and backward-wave configurations. In
168
+ Sec. III, taking spontaneous four-wave mixing (SFWM)
169
+ in a double-Λ four-level atomic system as an example,
170
+ we derive the coupled Langevin equations from micro-
171
+ scopic light-atom Heisenberg interaction for this special
172
+ case. We numerically confirm that the macroscopic phe-
173
+ nomenological solution in Sec. II agrees well with the
174
+ microscopic approach. In Sec. IV, we apply the quan-
175
+ tum Langevin theory to study effects of linear gain and
176
+ loss, complex phase mismatching, and complex nonlinear
177
+ coupling coefficient in entangled photon pair (biphoton)
178
+ generation, particularly to their temporal quantum cor-
179
+ relations. We conclude in the last section V.
180
+ II.
181
+ QUANTUM LANGEVIN EQUATIONS
182
+ Here we consider the two coupled single-mode phase-
183
+ conjugated fields in either forward-wave or backward-
184
+ wave configuration, as illustrated in Fig. 1.
185
+ In the
186
+ forward-wave configuration [Fig. 1(a)], both fields prop-
187
+ agate along +z direction through a nonlinear medium
188
+ with a length L.
189
+ In the backward-wave configuration
190
+ [Fig. 1(b)], the two fields propagate in opposing direc-
191
+ tions. The field annihilation operators ˆam(t, z) can be
192
+ expressed as
193
+ ˆa1(t, z) =
194
+ 1
195
+
196
+
197
+
198
+ dωˆa1(ω, z)ei( ω
199
+ c z−ωt),
200
+ ˆa2(t, z) =
201
+ 1
202
+
203
+
204
+
205
+ dωˆa2(ω, z)ei(± ω
206
+ c z−ωt),
207
+ (4)
208
+ where ± represents that field 2 propagates along +z or
209
+ −z direction, for the forward-wave or backward-wave
210
+ configuration, respectively.
211
+ The filed operators satisfy
212
+ the following commutation relations
213
+
214
+ ˆam (t, z) , ˆa†
215
+ n (t′, z)
216
+
217
+ = δmnδ(t − t′),
218
+
219
+ ˆam (ω, z) , ˆa†
220
+ n (ω′, z)
221
+
222
+ = δmnδ(ω − ω′).
223
+ (5)
224
+ In the forward-wave configuration, both fields are input
225
+ at z = 0, or ˆa1(0) and ˆa2(0) are the “initial” boundary
226
+ conditions. The general coupling matrix is [14]
227
+ MF =
228
+
229
+ −α1 + i ∆k
230
+ 2
231
+
232
+ −iκ
233
+ −α∗
234
+ 2 − i ∆k
235
+ 2
236
+
237
+ ,
238
+ (6)
239
+ where αm = −i ωm
240
+ 2c χm with χm being linear suscepti-
241
+ bility, and ∆k (real) is the phase mismatching in vac-
242
+ uum. In general, αm is complex valued, whose real part
243
+ Re{αm} > 0 represents loss (or gain for Re{αm} < 0)
244
+ and imaginary part represents phase velocity dispersion.
245
+ The nonlinear coupling coefficient κ can also be complex-
246
+ valued. In the backward-wave configuration, the general
247
+ coupling matrix becomes [12, 15]
248
+ MB =
249
+
250
+ −α1 + i ∆k
251
+ 2
252
+
253
+
254
+ α∗
255
+ 2 − i ∆k
256
+ 2
257
+
258
+ ,
259
+ (7)
260
+ and the “initial” boundary conditions are ˆa1(0) and
261
+ ˆa2(L): field 1 is input at z = 0 and field 2 is input at
262
+ z = L.
263
+ One can show that, under the following unitary gauge
264
+ transformation
265
+ �ˆa1
266
+ ˆa†
267
+ 2
268
+
269
+ =
270
+
271
+ eiθ/2
272
+ 0
273
+ 0
274
+ e−iθ/2
275
+ � �ˆa1
276
+ ˆa†
277
+ 2
278
+
279
+ = U
280
+ �ˆa1
281
+ ˆa†
282
+ 2
283
+
284
+ =
285
+ � ˆa1eiθ/2
286
+ ˆa†
287
+ 2e−iθ/2
288
+
289
+ ,
290
+ (8)
291
+ the corresponding coupling matrix become
292
+ MF(θ) = UMFU† =
293
+
294
+ −α1 + i ∆k
295
+ 2
296
+ iκeiθ
297
+ −iκe−iθ
298
+ −α∗
299
+ 2 − i ∆k
300
+ 2
301
+
302
+ ,
303
+ (9)
304
+ and
305
+ MB(θ) = UMBU† =
306
+
307
+ −α1 + i ∆k
308
+ 2
309
+ iκeiθ
310
+ iκe−iθ
311
+ α∗
312
+ 2 − i ∆k
313
+ 2
314
+
315
+ .
316
+ (10)
317
+ As physics is preserved and unchanged under the above
318
+ gauge transformation, we take θ = 0 throughout this
319
+ article for convenience and simplification.
320
+ In presence of linear loss or gain, i.e., Re{αm} ̸= 0, or
321
+ complex nonlinear coupling coefficient, κ ̸= κ∗, the two-
322
+ mode coupled equations must include Langevin noise op-
323
+ erators to preserve the commutation relations of the field
324
+ operators in Eq. (5). The noise operators should only
325
+ be related to Re{αm} and Im{κ}. As κ is real, the cou-
326
+ pled equations in forward-wave configuration should be
327
+ reduced to the known Eq. (3). For both forward- and
328
+ backward-wave configurations in the same nonlinear ma-
329
+ terial, the noise origin is the same except field 2 prop-
330
+ agates along ±z direction for different configurations.
331
+ With these guidelines, we provide quantum Langevin
332
+ equations for the two phase-conjugated fields from their
333
+ coupling matrix in the following subsections.
334
+
335
+ 3
336
+ 𝑧
337
+ 𝑎��
338
+ 𝑎��
339
+
340
+ Medium
341
+ 0
342
+ 𝐿
343
+ 𝜅
344
+ 𝑧
345
+ 𝑎��
346
+ 𝑎��
347
+
348
+ Medium
349
+ 0
350
+ 𝐿
351
+ 𝜅
352
+ (b)
353
+ (a)
354
+ Figure 1.
355
+ Schematics of two coupled phase-conjugated electromagnetic waves:
356
+ (a) forward-wave configuration, and (b)
357
+ backward-wave configuration. κ is the nonlinear coupling coefficient between the two modes.
358
+ A.
359
+ Forward-Wave Configuration
360
+ In
361
+ the
362
+ forward-wave
363
+ configuration
364
+ as
365
+ shown
366
+ in
367
+ Fig. 1(a), we find that its quantum Langevin coupled
368
+ equations can be expressed in the following general form
369
+
370
+ ∂z
371
+ �ˆa1
372
+ ˆa†
373
+ 2
374
+
375
+ = MF
376
+ �ˆa1
377
+ ˆa†
378
+ 2
379
+
380
+ + NFR
381
+ � ˆf1
382
+ ˆf †
383
+ 2
384
+
385
+ + NFI
386
+ � ˆf †
387
+ 1ˆf2
388
+
389
+ (11)
390
+ with the “initial” condition at z = 0:
391
+
392
+ ˆam(ω, 0), ˆa†
393
+ n(ω′, 0)
394
+
395
+ = δmnδ(ω − ω′).
396
+ (12)
397
+ The Langevin noise operators satisfy
398
+
399
+ ˆfm(ω, z), ˆf †
400
+ n(ω′, z′)
401
+
402
+ = δmnδ(ω − ω′)δ(z − z′)
403
+ (13)
404
+ and have the following correlations
405
+
406
+ ˆf †
407
+ m(ω, z) ˆfn(ω′, z′)
408
+
409
+ = 0,
410
+
411
+ ˆfm(ω, z) ˆf †
412
+ n(ω′, z′)
413
+
414
+ = δmnδ(ω − ω′)δ(z − z′),
415
+
416
+ ˆfm(ω, z) ˆfn(ω′, z′)
417
+
418
+ =
419
+
420
+ ˆf †
421
+ m(ω, z) ˆf †
422
+ n(ω′, z′)
423
+
424
+ = 0.
425
+ (14)
426
+ The Langevin noise matrix is given by
427
+ NF ≡
428
+
429
+ −(MF + MF
430
+ ∗) = NFR + iNFI,
431
+ (15)
432
+ where NFR and NFI are the real and imaginary parts of
433
+ the matrix NF (i.e., NFmn = NFRmn + iNFImn), respec-
434
+ tively.
435
+ We obtain the solution of Eq. (11) at the output sur-
436
+ face z = L as the following
437
+ �ˆa1 (L)
438
+ ˆa†
439
+ 2 (L)
440
+
441
+ = eMFL
442
+ �ˆa1 (0)
443
+ ˆa†
444
+ 2 (0)
445
+
446
+ +
447
+ � L
448
+ 0
449
+ eMF(L−z)
450
+
451
+ NFR
452
+ � ˆf1 (z)
453
+ ˆf †
454
+ 2 (z)
455
+
456
+ + NFI
457
+ � ˆf †
458
+ 1 (z)
459
+ ˆf2 (z)
460
+ ��
461
+ dz.
462
+ (16)
463
+ Defining
464
+ eMFL ≡
465
+
466
+ A B
467
+ C D
468
+
469
+ ,
470
+ (17)
471
+ eMF(L−z) ≡
472
+
473
+ A1 (z) B1 (z)
474
+ C1 (z) D1 (z)
475
+
476
+ ,
477
+ (18)
478
+ we rewrite Eq. (16) as
479
+ �ˆa1 (L)
480
+ ˆa†
481
+ 2 (L)
482
+
483
+ =
484
+
485
+ A B
486
+ C D
487
+ � �ˆa1 (0)
488
+ ˆa†
489
+ 2 (0)
490
+
491
+ +
492
+ � L
493
+ 0
494
+
495
+ A1 (z) B1 (z)
496
+ C1 (z) D1 (z)
497
+ � �
498
+ NFR
499
+ � ˆf1 (z)
500
+ ˆf †
501
+ 2 (z)
502
+
503
+ + NFI
504
+ � ˆf †
505
+ 1 (z)
506
+ ˆf2 (z)
507
+ ��
508
+ dz.
509
+ (19)
510
+ We numerically confirm that the solution preserves the
511
+ commutation relations
512
+
513
+ ˆam(ω, L), ˆa†
514
+ n(ω′, L)
515
+
516
+ =
517
+
518
+ ˆam(ω, 0), ˆa†
519
+ n(ω′, 0)
520
+
521
+ = δmnδ(ω − ω′).
522
+ (20)
523
+ Now we examine some special cases.
524
+ Case 1: We first consider the coupling matrix MF in Eq.
525
+ (6) where the nonlinear coupling coefficient κ is real and
526
+ both modes have losses (Re{αm} ≥ 0) . This works for
527
+ most PDC processes [3, 7]. Under such a condition, we
528
+ have the following diagonalized noise matrix
529
+ NF = NFR =
530
+ ��
531
+ 2Re{α1}
532
+ 0
533
+ 0
534
+
535
+ 2Re{α2}
536
+
537
+ ,
538
+ (21)
539
+ and the coupled Langevin equations
540
+
541
+ ∂z
542
+ �ˆa1
543
+ ˆa†
544
+ 2
545
+
546
+ = MF
547
+ �ˆa1
548
+ ˆa†
549
+ 2
550
+
551
+ +
552
+ ��
553
+ 2Re{α1} ˆf1
554
+
555
+ 2Re{α2} ˆf †
556
+ 2
557
+
558
+ ,
559
+ (22)
560
+ which is the well-known result in literature [3, 7].
561
+ Case 2: κ is real, the mode 1 has linear loss (Re{α1} =
562
+ α ≥ 0), and the mode 2 has linear gain (Re{α2} = −g ≤
563
+ 0). The noise matrix becomes
564
+ NF =
565
+ �√
566
+
567
+ 0
568
+ 0
569
+ i√2g
570
+
571
+ .
572
+ (23)
573
+ We have the following coupled Langevin equations
574
+
575
+ ∂z
576
+ �ˆa1
577
+ ˆa†
578
+ 2
579
+
580
+ = MF
581
+ �ˆa1
582
+ ˆa†
583
+ 2
584
+
585
+ +
586
+ �√
587
+ 2α ˆf1
588
+ √2g ˆf2
589
+
590
+ .
591
+ (24)
592
+
593
+ 4
594
+ Case 3: The two modes are perfectly phase-matched
595
+ without linear gain or loss: ∆k = 0, α1 = α2 = 0, but
596
+ the nonlinear coupling coefficient is complex-valued κ =
597
+ η + iζ. In this case, the coupled matrix is
598
+ MF =
599
+
600
+ 0
601
+ −ζ + iη
602
+ ζ − iη
603
+ 0
604
+
605
+ .
606
+ (25)
607
+ The noise matrix becomes
608
+ NF = Θ(ζ)
609
+
610
+ ζ
611
+
612
+ 1
613
+ 1
614
+ −1 1
615
+
616
+ + iΘ(−ζ)
617
+
618
+ −ζ
619
+
620
+ 1
621
+ 1
622
+ −1 1
623
+
624
+ ,
625
+ (26)
626
+ where Θ(ζ) is Heaviside step function, Θ(ζ) = 1 if ζ > 0,
627
+ Θ(ζ) = 0 if ζ ≤ 0. The Langevin coupled equations are
628
+
629
+ ∂z
630
+ �ˆa1
631
+ ˆa†
632
+ 2
633
+
634
+ =MF
635
+ �ˆa1
636
+ ˆa†
637
+ 2
638
+
639
+ + Θ(ζ)
640
+
641
+ ζ
642
+
643
+ 1
644
+ 1
645
+ −1 1
646
+ � � ˆf1
647
+ ˆf †
648
+ 2
649
+
650
+ + Θ(−ζ)
651
+
652
+ −ζ
653
+
654
+ 1
655
+ 1
656
+ −1 1
657
+ � � ˆf †
658
+ 1ˆf2
659
+
660
+ .
661
+ (27)
662
+ Eq. (27) shows that a complex-valued nonlinear coupling
663
+ coefficient also leads to Langevin noises even when there
664
+ is no linear gain or loss. This is revealed by this article
665
+ for the first time.
666
+ Case 4: As κ is real and there is no linear loss or gain
667
+ (α1 = α2 = 0), the coupled equations can be written as
668
+ i ∂
669
+ ∂z
670
+ �ˆa1
671
+ ˆa†
672
+ 2
673
+
674
+ =
675
+
676
+ − ∆k
677
+ 2
678
+ −κ
679
+ κ
680
+ ∆k
681
+ 2
682
+ � �ˆa1
683
+ ˆa†
684
+ 2
685
+
686
+ = ˆH
687
+ �ˆa1
688
+ ˆa†
689
+ 2
690
+
691
+ .
692
+ (28)
693
+ The effective Hamiltonian ˆH has anti-parity-time (APT)
694
+ symmetry, which has been demonstrated in FWM in cold
695
+ atoms [14, 16].
696
+ B.
697
+ Backward-Wave Configuration
698
+ In the back-wave configuration as shown in Fig. 1(b),
699
+ the quantum Langevin coupled equations can be ex-
700
+ pressed in the following general form
701
+
702
+ ∂z
703
+ �ˆa1
704
+ ˆa†
705
+ 2
706
+
707
+ = MB
708
+ �ˆa1
709
+ ˆa†
710
+ 2
711
+
712
+ + NBR
713
+ � ˆf1
714
+ ˆf †
715
+ 2
716
+
717
+ + NBI
718
+ � ˆf †
719
+ 1ˆf2
720
+
721
+ .
722
+ (29)
723
+ Different
724
+ from
725
+ the
726
+ forward-wave
727
+ configuration,
728
+ the
729
+ “boundary” condition is
730
+
731
+ ˆa1(ω, 0), ˆa†
732
+ 1(ω′, 0)
733
+
734
+ =
735
+
736
+ ˆa2(ω, L), ˆa†
737
+ 2(ω′, L)
738
+
739
+ = δ(ω − ω′).
740
+ (30)
741
+ The Langevin noise operators satisfy the same commu-
742
+ tation relations and correlations in Eqs. (13) and (14).
743
+ The Langevin noise matrix is given by
744
+ NB ≡
745
+
746
+ 1
747
+ 0
748
+ 0 −1
749
+ � ��
750
+ −MB11 −MB12
751
+ MB21
752
+ MB22
753
+
754
+ +
755
+
756
+ −MB11 −MB12
757
+ MB21
758
+ MB22
759
+ �∗
760
+ = NBR + iNBI,
761
+ (31)
762
+ where NBR and NBI are the real and imaginary parts of
763
+ the matrix NB, respectively. One can show that the noise
764
+ matrix defined in Eq. (31) has the same origin as that
765
+ in the forward-wave configuration in the same nonlinear
766
+ material:
767
+ NB =
768
+
769
+ 1
770
+ 0
771
+ 0 −1
772
+
773
+ NF.
774
+ (32)
775
+ We note that the choice of noise matrix is not unique.
776
+ For example, transformation ˆf1 → − ˆf1 or/and ˆf2 → − ˆf2
777
+ does not affect computing any physical observable. We
778
+ elaborate on this more in Appendix A.
779
+ We obtain the solution of Eq. (29) at z = L as follow-
780
+ ing
781
+ �ˆa1 (L)
782
+ ˆa†
783
+ 2 (L)
784
+
785
+ = eMBL
786
+ �ˆa1 (0)
787
+ ˆa†
788
+ 2 (0)
789
+
790
+ +
791
+ � L
792
+ 0
793
+ eMB(L−z)
794
+
795
+ NBR
796
+ � ˆf1 (z)
797
+ ˆf †
798
+ 2 (z)
799
+
800
+ + NBI
801
+ � ˆf †
802
+ 1 (z)
803
+ ˆf2 (z)
804
+ ��
805
+ dz.
806
+ (33)
807
+ We define
808
+ eMBL ≡
809
+ � ¯A
810
+ ¯B
811
+ ¯C
812
+ ¯D
813
+
814
+ ,
815
+ (34)
816
+ eMB(L−z) ≡
817
+ � ¯A1 (z)
818
+ ¯B1 (z)
819
+ ¯C1 (z)
820
+ ¯D1 (z)
821
+
822
+ .
823
+ (35)
824
+ Different from the forward-wave case, in the backward-
825
+ wave configuration, the mode 1 input is at z = 0 and the
826
+ mode 2 input is at z = L. With known ˆa1(0) and ˆa2(L),
827
+ we rearrange Eq. (33) and obtain solutions for ˆa1(L) and
828
+ ˆa2(0):
829
+ �ˆa1 (L)
830
+ ˆa†
831
+ 2 (0)
832
+
833
+ =
834
+
835
+ A B
836
+ C D
837
+ � �ˆa1 (0)
838
+ ˆa†
839
+ 2 (L)
840
+
841
+ +
842
+
843
+ 1 −B
844
+ 0 −D
845
+ � � L
846
+ 0
847
+ � ¯A1 (z)
848
+ ¯B1 (z)
849
+ ¯C1 (z)
850
+ ¯D1 (z)
851
+ � �
852
+ NBR
853
+ � ˆf1 (z)
854
+ ˆf †
855
+ 2 (z)
856
+
857
+ + NBI
858
+ � ˆf †
859
+ 1 (z)
860
+ ˆf2 (z)
861
+ ��
862
+ dz,
863
+ (36)
864
+
865
+ 5
866
+ where
867
+ A = ¯A −
868
+ ¯B ¯C
869
+ ¯D ,
870
+ B =
871
+ ¯B
872
+ ¯D,
873
+ C = −
874
+ ¯C
875
+ ¯D,
876
+ D = 1
877
+ ¯D.
878
+ (37)
879
+ We numerically confirm that Eq. (36) preserves the com-
880
+ mutation relations
881
+
882
+ ˆa1(ω, L), ˆa†
883
+ 1(ω′, L)
884
+
885
+ =
886
+
887
+ ˆa1(ω, 0), ˆa†
888
+ 1(ω′, 0)
889
+
890
+ ,
891
+
892
+ ˆa2(ω, 0), ˆa†
893
+ 2(ω′, 0)
894
+
895
+ =
896
+
897
+ ˆa2(ω, L), ˆa†
898
+ 2(ω′, L)
899
+
900
+ .
901
+ (38)
902
+ Similarly to the forward-wave configuration, we exam-
903
+ ine the following four special cases.
904
+ Case 1: We assume the nonlinear coupling coefficient κ
905
+ is real and both modes have losses (Re{αm} ≥ 0). Under
906
+ such a condition, we have the following diagonalized noise
907
+ matrix
908
+ NB =
909
+ ��
910
+ 2Re{α1}
911
+ 0
912
+ 0
913
+
914
+
915
+ 2Re{α2}
916
+
917
+ ,
918
+ (39)
919
+ and the coupled Langevin equations
920
+
921
+ ∂z
922
+ �ˆa1
923
+ ˆa†
924
+ 2
925
+
926
+ = MB
927
+ �ˆa1
928
+ ˆa†
929
+ 2
930
+
931
+ +
932
+ � �
933
+ 2Re{α1} ˆf1
934
+
935
+
936
+ 2Re{α2} ˆf †
937
+ 2
938
+
939
+ .
940
+ (40)
941
+ Case 2: κ is real, mode 1 has linear loss (Re{α1} = α ≥
942
+ 0), and mode 2 has linear gain (Re{α2} = −g ≤ 0). The
943
+ noise matrix becomes
944
+ NF =
945
+ �√
946
+
947
+ 0
948
+ 0
949
+ −i√2g
950
+
951
+ .
952
+ (41)
953
+ We have the following coupled Langevin equations
954
+
955
+ ∂z
956
+ �ˆa1
957
+ ˆa†
958
+ 2
959
+
960
+ = MB
961
+ �ˆa1
962
+ ˆa†
963
+ 2
964
+
965
+ +
966
+ � √
967
+ 2α ˆf1
968
+ −√2g ˆf2
969
+
970
+ .
971
+ (42)
972
+ Case 3: The two modes are perfectly phase-matched
973
+ without linear gain and loss: ∆k = 0, α1 = α2 = 0,
974
+ but the nonlinear coupling coefficient is complex-valued
975
+ κ = η + iζ. In this case, the coupled matrix is
976
+ MB =
977
+
978
+ 0
979
+ −ζ + iη
980
+ −ζ + iη
981
+ 0
982
+
983
+ .
984
+ (43)
985
+ The noise matrix becomes
986
+ NB = Θ(ζ)
987
+
988
+ ζ
989
+
990
+ 1
991
+ 1
992
+ 1 −1
993
+
994
+ + iΘ(−ζ)
995
+
996
+ −ζ
997
+
998
+ 1
999
+ 1
1000
+ 1 −1
1001
+
1002
+ .
1003
+ (44)
1004
+ The Langevin coupled equations are
1005
+
1006
+ ∂z
1007
+ �ˆa1
1008
+ ˆa†
1009
+ 2
1010
+
1011
+ =MB
1012
+ �ˆa1
1013
+ ˆa†
1014
+ 2
1015
+
1016
+ + Θ(ζ)
1017
+
1018
+ ζ
1019
+
1020
+ 1
1021
+ 1
1022
+ 1 −1
1023
+ � � ˆf1
1024
+ ˆf †
1025
+ 2
1026
+
1027
+ + Θ(−ζ)
1028
+
1029
+ −ζ
1030
+
1031
+ 1
1032
+ 1
1033
+ 1 −1
1034
+ � � ˆf †
1035
+ 1ˆf2
1036
+
1037
+ .
1038
+ (45)
1039
+ Eq. (45) shows that in the backward-wave configuration,
1040
+ a complex-valued nonlinear coupling coefficient also leads
1041
+ to Langevin noises even though there is no linear gain or
1042
+ loss.
1043
+ Case 4: As κ is real and there are equal losses in both
1044
+ modes (α1 = α2 = α > 0) with perfect phase matching
1045
+ (∆k = 0), the coupled equations can be written as
1046
+ i ∂
1047
+ ∂z
1048
+ �ˆa1
1049
+ ˆa†
1050
+ 2
1051
+
1052
+ =
1053
+
1054
+ −iα −κ
1055
+ −κ
1056
+
1057
+ � �ˆa1
1058
+ ˆa†
1059
+ 2
1060
+
1061
+ = ˆH
1062
+ �ˆa1
1063
+ ˆa†
1064
+ 2
1065
+
1066
+ .
1067
+ (46)
1068
+ Interestingly, the effective Hamiltonian ˆH here follows
1069
+ parity-time (PT) symmetry [17, 18].
1070
+ III.
1071
+ MICROSCOPIC ORIGIN OF LANGEVIN
1072
+ NOISES: SFWM
1073
+ One could validate the above phenomenological ap-
1074
+ proach of quantum Langevin coupled equations by con-
1075
+ firming the microscopic origin of the Langevin noises.
1076
+ However,
1077
+ for two systems with the same quantum
1078
+ Langevin equations, their microscopic structures may be
1079
+ quite different. Therefore it is impossible to sort all mi-
1080
+ croscopic systems. In this section, we focus on SFWM in
1081
+ a double-Λ four-level atomic system [10–12, 19, 20] with
1082
+ electromagnetically induced transparency (EIT) [21, 22],
1083
+ and show that the phenomenological approach in the
1084
+ above section agrees with the numerical results from the
1085
+ microscopic quantum theory of light-atom interaction.
1086
+ We start from a single-atom picture, considering an
1087
+ EM wave couples the atomic transition |j⟩ and |k⟩. The
1088
+ induced single atom polarization ˆpjk ∝ µjkˆσjk, where
1089
+ µjk is the electric dipole moment matrix element, ˆσjk =
1090
+ |j ⟩⟨ k| is single atom transition operator from state |k⟩ to
1091
+ |j⟩. In the Heisenberg-Langevin picture, the single-atom
1092
+ transition operator can be expressed as
1093
+ ˆσjk = ˆσ(0)
1094
+ jk +
1095
+
1096
+ µν
1097
+ βµν ˆf (σ)
1098
+ µν ,
1099
+ (47)
1100
+ where ˆσ(0)
1101
+ jk = ⟨ˆσjk⟩ is the zeroth-order steady state so-
1102
+ lution. The single atom noise operator between atomic
1103
+ transition |ν⟩ → |µ⟩ is represented by ˆf (σ)
1104
+ µν , which satisfies
1105
+ the following correlations:
1106
+ ⟨ ˆf (σ)
1107
+ µν (ω) ˆf (σ)†
1108
+ µ′ν′ (ω′)⟩ = ⟨ ˆf (σ)
1109
+ µν (ω) ˆf (σ)
1110
+ ν′µ′(ω′)⟩
1111
+ = Dµν,ν′µ′δ(ω − ω′),
1112
+ ⟨ ˆf (σ)†
1113
+ µν (ω) ˆf (σ)
1114
+ µ′ν′(ω′)⟩ = ⟨ ˆf (σ)
1115
+ νµ (ω) ˆf (σ)
1116
+ µ′ν′(ω′)⟩
1117
+ = Dνµ,µ′ν′δ(ω − ω′),
1118
+ (48)
1119
+ where Dµν,ν′µ′ and Dνµ,µ′ν′ are diffusion coefficients.
1120
+ In a continuous medium with atomic number density
1121
+ n, the noises from different atoms are uncorrelated. We
1122
+ have the spatially averaged atomic operator
1123
+ ˆ¯σjk ≡ ˆσ(0)
1124
+ jk +
1125
+ 1
1126
+
1127
+ nA
1128
+
1129
+ µν
1130
+ βµν ˆ¯f (σ)
1131
+ µν ,
1132
+ (49)
1133
+
1134
+ 6
1135
+ |1⟩
1136
+ |2⟩
1137
+ Δ�
1138
+ 𝑎���
1139
+ 𝑧
1140
+ 0
1141
+ 𝐿
1142
+ (a)
1143
+ (b)
1144
+ |3⟩
1145
+ 𝜔��
1146
+ 𝜔�
1147
+ 𝐸�
1148
+ 𝐸�
1149
+ 𝑎��
1150
+ |4⟩
1151
+ 𝜔�
1152
+ 𝜔�
1153
+ 𝜛
1154
+ 𝜛
1155
+ Figure 2. Spontaneous four-wave mixing (SFWM) in a double-Λ four-level cold atomic medium. (a) Backward-wave geometry
1156
+ of SFWM optical configuration. Driven by counter-propagating pump (Ep) and coupling (Ec) beams, phase-matched backward
1157
+ Stokes (ˆas) and anti-Stokes (ˆaas) are spontaneously generated from a laser-cooled atomic medium. (b) Atomic energy-level
1158
+ diagram. The pump (ωp) laser is detuned with ∆p from transition |1⟩ → |4⟩, and the coupling (ωc) laser is on-resonant with
1159
+ transition |2⟩ → |3⟩. Stokes (ωs) photons are spontaneously generated from transition |4⟩ → |2⟩, and anti-Stokes (ωas) photons
1160
+ from transition |3⟩ → |1⟩. ϖ = ωas − ω13 is the anti-Stokes photon frequency detuning from transition |1⟩ → |3⟩.
1161
+ where A is the single-mode cross-section area, and the
1162
+ spatially averaged atomic noise operators ˆ¯f (σ)
1163
+ µν satisfy the
1164
+ following modified correlations
1165
+ ⟨ ˆ¯f (σ)
1166
+ µν (ω, z) ˆ¯f (σ)†
1167
+ µ′ν′ (ω′, z′)⟩ = ⟨ ˆ¯f (σ)
1168
+ µν (ω, z) ˆ¯f (σ)
1169
+ ν′µ′(ω′, z′)⟩
1170
+ = Dµν,ν′µ′δ(ω − ω′)δ(z − z′),
1171
+ ⟨ ˆ¯f (σ)†
1172
+ µν (ω, z) ˆ¯f (σ)
1173
+ µ′ν′(ω′, z′)⟩ = ⟨ ˆ¯f (σ)
1174
+ νµ (ω, z) ˆ¯f (σ)
1175
+ µ′ν′(ω′, z′)⟩
1176
+ = Dνµ,µ′ν′δ(ω − ω′)δ(z − z′),
1177
+ (50)
1178
+ where the diffusion coefficients are the same as those from
1179
+ the single-atom picture.
1180
+ The electric field and polarization are described as
1181
+ ˆE(t, z) = 1
1182
+ 2
1183
+
1184
+ ˆE(+)(t, z) + ˆE(−)(t, z)
1185
+
1186
+ ,
1187
+ ˆP(t, z) = 1
1188
+ 2
1189
+
1190
+ ˆP (+)(t, z) + ˆP (−)(t, z)
1191
+
1192
+ ,
1193
+ (51)
1194
+ Where ˆE(+), ˆP (+) and ˆE(−), ˆP (−) are positive and nega-
1195
+ tive frequency parts. We take the following Fourier trans-
1196
+ form
1197
+ ˆE(+)(t, z) =
1198
+ 1
1199
+
1200
+
1201
+
1202
+ dω ˆE(ω, z)ei(± ω
1203
+ c z−ωt),
1204
+ ˆP (+)(t, z) =
1205
+ 1
1206
+
1207
+
1208
+
1209
+ dω ˆP(ω, z)ei(± ω
1210
+ c z−ωt),
1211
+ (52)
1212
+ where ˆE(ω, z), ˆP(ω, z) are complex amplitudes in fre-
1213
+ quency domain.
1214
+ The Maxwell equation under slowly
1215
+ varying envelope approximation (SVEA) can be written
1216
+ as
1217
+ ±∂ ˆE(ω, z)
1218
+ ∂z
1219
+ = i
1220
+ 2ωη ˆP(ω, z),
1221
+ (53)
1222
+ where ± represents for propagation direction along ±z,
1223
+ and free space impedance η = 1/(cε0) = 377 Ohm, with
1224
+ c being the speed of light in vacuum, and ε0 the vacuum
1225
+ permittivity. With quantized electric field
1226
+ ˆE(ω, z) =
1227
+
1228
+ 2ℏω
1229
+ cε0Aˆa(ω, z),
1230
+ (54)
1231
+ and
1232
+ ˆP(ω, z) = 2nµjkˆ¯σjk(ω, z),
1233
+ (55)
1234
+ we obtain the Langevin equation for the EM field in the
1235
+ atomic medium
1236
+ ±∂ˆa(ω, z)
1237
+ ∂z
1238
+ = i nAgjkˆ¯σjk(ω, z)
1239
+ = i nAgjkˆσ(0)
1240
+ jk (ω, z) + ˆ¯F(ω, z),
1241
+ (56)
1242
+ where
1243
+ gjk = µjk
1244
+
1245
+ ωjk
1246
+ 2cε0ℏA,
1247
+ ˆ¯F(ω, z) = i
1248
+
1249
+ nAgjk
1250
+
1251
+ µν
1252
+ βµν ˆ¯f (σ)
1253
+ µν (ω, z)
1254
+ = iµjk
1255
+ � nωjk
1256
+ 2cε0ℏ
1257
+
1258
+ µν
1259
+ βµν ˆ¯f (σ)
1260
+ µν (ω, z).
1261
+ (57)
1262
+ Here gjk = g∗
1263
+ kj is single photon-atom coupling strength.
1264
+ Now we turn to the backward-wave SFWM in a double-
1265
+ Λ four-level atomic system as illustrated in Fig. 2. In
1266
+ presence of counter-propagating pump (Ep, ωp) and cou-
1267
+ pling (Ec, ωc) laser beams, phase-matched Stokes (ωs)
1268
+ and anti-Stokes (ωas) are spontaneously generated and
1269
+ propagate through the medium in opposing directions.
1270
+ In the rotating reference frame, the interaction Hamilto-
1271
+ nian for a single atom is
1272
+ ˆV = − ℏ
1273
+
1274
+ g31ˆaasˆσ31 + g13ˆa†
1275
+ asˆσ13
1276
+
1277
+ − ℏ
1278
+
1279
+ g42ˆasˆσ42 + g24ˆa†
1280
+ sˆσ24
1281
+
1282
+ − 1
1283
+ 2ℏ (Ωcˆσ32 + Ω∗
1284
+ c ˆσ23) − 1
1285
+ 2ℏ
1286
+
1287
+ Ωpˆσ41 + Ω∗
1288
+ pˆσ14
1289
+
1290
+ − ℏ∆pˆσ44 − ℏϖˆσ33 − ℏϖˆσ22,
1291
+ (58)
1292
+ where Ωc = µ32Ec/ℏ is coupling Rabi frequency. The
1293
+ coupling laser is on-resonant with transition |2⟩ → |3⟩.
1294
+ Ωp = µ41Ep/ℏ is pump Rabi frequency.
1295
+ The pump
1296
+ laser is far detuned from the transition |1⟩ → |4⟩ with
1297
+ ∆p = ωp − ω14 so that the atomic population mainly oc-
1298
+ cupies the ground state |1⟩. We take this ground-state
1299
+
1300
+ 7
1301
+ approximation through this section.
1302
+ With continuous-
1303
+ wave pump and coupling driving fields, the energy con-
1304
+ servation leads to ωas+ωs = ωc+ωp. Here ϖ = ωas−ω13
1305
+ is the anti-Stokes frequency detuning and thus the Stokes
1306
+ frequency detuning is ωs − ωs0 = −ϖ.
1307
+ The atomic evolution is governed by the following
1308
+ Heisenberg-Langevin equation [11]
1309
+
1310
+ ∂t ˆσjk = i
1311
+ ℏ[ ˆV , ˆσjk] − γjkˆσjk + rA
1312
+ jk + ˆf (σ)
1313
+ jk ,
1314
+ (59)
1315
+ where γjk = γkj (nonzero only as j ̸= k) are dephasing
1316
+ rates, rA
1317
+ jk (nonzero only as j = k) are the population
1318
+ transfer resulting from spontaneous emission decay. The
1319
+ full equation of motion can be found in Appendix B. The
1320
+ diffusion coefficients Djk,j′k′ can be obtained through the
1321
+ Einstein relation
1322
+ Djk,j′k′ = ∂
1323
+ ∂t ⟨ˆσjkˆσj′k′⟩
1324
+
1325
+
1326
+ ˆAjkˆσj′k′
1327
+
1328
+
1329
+
1330
+ ˆσjk ˆAj′k′
1331
+
1332
+ ,
1333
+ (60)
1334
+ where ˆAjk =
1335
+
1336
+ ∂t ˆσjk − ˆf (σ)
1337
+ jk . For the SFWM governed by
1338
+ Eq. (59), we have [11, 12]
1339
+
1340
+ ��
1341
+ D12,21 D12,24
1342
+ D42,21 D42,24
1343
+ D12,31 D12,34
1344
+ D42,31 D42,34
1345
+ D13,21 D13,24
1346
+ D43,21 D43,24
1347
+ D13,31 D13,34
1348
+ D43,31 D43,34
1349
+
1350
+ ��
1351
+ =
1352
+
1353
+ ��
1354
+ 2γ12 ⟨ˆσ11⟩ + Γ31 ⟨ˆσ33⟩ + Γ41 ⟨ˆσ44⟩ γ12 ⟨ˆσ14⟩
1355
+ 0
1356
+ 0
1357
+ γ12 ⟨ˆσ41⟩
1358
+ 0
1359
+ 0
1360
+ 0
1361
+ 0
1362
+ 0
1363
+ Γ3 ⟨ˆσ11⟩ + Γ31 ⟨ˆσ33⟩ + Γ41 ⟨ˆσ44⟩ Γ3 ⟨ˆσ14⟩
1364
+ 0
1365
+ 0
1366
+ Γ3 ⟨ˆσ41⟩
1367
+ Γ3 ⟨ˆσ44⟩
1368
+
1369
+ �� ,
1370
+ (61)
1371
+
1372
+ ��
1373
+ D21,12 D21,42
1374
+ D24,12 D24,42
1375
+ D21,13 D21,43
1376
+ D24,13 D24,43
1377
+ D31,12 D31,42
1378
+ D34,12 D34,42
1379
+ D31,13 D31,43
1380
+ D34,13 D34,43
1381
+
1382
+ ��
1383
+ =
1384
+
1385
+ ��
1386
+ 2γ12 ⟨ˆσ22⟩ + Γ32 ⟨ˆσ33⟩ + Γ42 ⟨ˆσ44⟩
1387
+ 0
1388
+ γ12 ⟨ˆσ23⟩
1389
+ 0
1390
+ 0
1391
+ Γ4 ⟨ˆσ22⟩ + Γ32 ⟨ˆσ33⟩ + Γ42 ⟨ˆσ44⟩
1392
+ 0
1393
+ Γ4 ⟨ˆσ23⟩
1394
+ γ12 ⟨ˆσ32⟩
1395
+ 0
1396
+ 0
1397
+ 0
1398
+ 0
1399
+ Γ4 ⟨ˆσ32⟩
1400
+ 0
1401
+ Γ4 ⟨ˆσ33⟩
1402
+
1403
+ �� .
1404
+ (62)
1405
+ Solving Eq. (59) under the ground-state approximation
1406
+ ⟨ˆσ11⟩ ∼= 1 with weak pump excitation ∆p ≫ {Ωp, Γ4}, we
1407
+ get the single-atom steady-state solutions (with µν =
1408
+ 12, 13, 42, 43)
1409
+ ˆσ13 = ˆσ(0)
1410
+ 13 +
1411
+
1412
+ µν
1413
+ βas
1414
+ µν ˆf (σ)
1415
+ µν ,
1416
+ ˆσ42 = ˆσ(0)
1417
+ 42 +
1418
+
1419
+ µν
1420
+ βs
1421
+ µν ˆf (σ)
1422
+ µν ,
1423
+ (63)
1424
+ where
1425
+ ˆσ(0)
1426
+ 13 =4 (ϖ + iγ12)
1427
+ T (ϖ)
1428
+ g31ˆaas
1429
+ +
1430
+ ΩcΩp
1431
+ T (ϖ) (∆p + iγ14)g24ˆa†
1432
+ s,
1433
+ ˆσ(0)
1434
+ 42 =(ϖ + iγ13)
1435
+ T (ϖ)
1436
+ |Ωp|2
1437
+ (∆p − iγ24)
1438
+ 1
1439
+ (∆p + iγ14)g24ˆa†
1440
+ s
1441
+ +
1442
+ Ω∗
1443
+ pΩ∗
1444
+ c
1445
+ T (ϖ) (∆p − iγ24)g31ˆaas,
1446
+ (64)
1447
+ βas
1448
+ 12 = i2Ωc
1449
+ T (ϖ),
1450
+ βas
1451
+ 13 = −i4 (ϖ + iγ12)
1452
+ T (ϖ)
1453
+ ,
1454
+ βas
1455
+ 42 = −
1456
+ iΩcΩp
1457
+ T (ϖ) (∆p − iγ24),
1458
+ βas
1459
+ 43 =
1460
+ i2Ωp (ϖ + iγ12)
1461
+ T (ϖ) (∆p − iγ34),
1462
+ βs
1463
+ 12 = i2 (ϖ + iγ13)
1464
+ T (ϖ)
1465
+ Ω∗
1466
+ p
1467
+ (∆p − iγ24),
1468
+ βs
1469
+ 13 = −
1470
+ iΩ∗
1471
+ pΩ∗
1472
+ c
1473
+ T (ϖ) (∆p − iγ24),
1474
+ βs
1475
+ 42 = −
1476
+ i
1477
+ (∆p − iγ24),
1478
+ βs
1479
+ 43 = −
1480
+ iΩ∗
1481
+ c
1482
+ 2 (∆p − iγ24) (∆p − iγ34),
1483
+ (65)
1484
+ where T(ϖ) ≡ |Ωc|2 − 4 (ϖ + iγ13) (ϖ + iγ12). We then
1485
+ obtain the ensemble spatially averaged atomic operators
1486
+
1487
+ 8
1488
+ -50
1489
+ 0
1490
+ 50
1491
+ 0
1492
+ 0.2
1493
+ 0.4
1494
+ 0.6
1495
+ 0.8
1496
+ 1
1497
+ -50
1498
+ 0
1499
+ 50
1500
+ -10
1501
+ -8
1502
+ -6
1503
+ -4
1504
+ -2
1505
+ 0
1506
+ 10-7
1507
+ -50
1508
+ 0
1509
+ 50
1510
+ 0
1511
+ 0.2
1512
+ 0.4
1513
+ 0.6
1514
+ 0.8
1515
+ 1
1516
+ Macro
1517
+ Micro
1518
+ NLN
1519
+ -50
1520
+ 0
1521
+ 50
1522
+ -2
1523
+ -1
1524
+ 0
1525
+ 1
1526
+ 2
1527
+ 3
1528
+ 10-8
1529
+ (
1530
+ -
1531
+ )
1532
+ (
1533
+ -
1534
+ )
1535
+ (
1536
+ -
1537
+ )
1538
+ (
1539
+ -
1540
+ )
1541
+ (a)
1542
+ (b)
1543
+ (c)
1544
+ (d)
1545
+ Figure 3. Comparison of commutation relations between the macroscopic (“Macro”, blue solid lines) and microscopic (“Micro”,
1546
+ red dashed lines) approaches in the group delay regime: (a) [ˆaas(L), ˆa†
1547
+ as(L)], (b) [ˆaas(L), ˆa†
1548
+ as(L)] − δ(ϖ − ϖ′), (c) [ˆas(0), ˆa†
1549
+ s(0)],
1550
+ and (d)[ˆas(0), ˆa†
1551
+ s(0)] − δ(ϖ − ϖ′). The results with no Langevin noise operators (“NLN”) are shown as black dotted lines in
1552
+ (a) and (c).
1553
+ for generating anti-Stokes and Stokes fields from Eq. (49)
1554
+ ˆ¯σ13 = ˆσ(0)
1555
+ 13 +
1556
+ 1
1557
+
1558
+ nA
1559
+
1560
+ µν
1561
+ βas
1562
+ µν ˆ¯f
1563
+ (σ)
1564
+ µν ,
1565
+ ˆ¯σ42 = ˆσ(0)
1566
+ 42 +
1567
+ 1
1568
+
1569
+ nA
1570
+
1571
+ µν
1572
+ βs
1573
+ µν ˆ¯f
1574
+ (σ)
1575
+ µν .
1576
+ (66)
1577
+ Following the procedures in Eqs. (56) and (57),
1578
+ ∂ˆaas(ω, z)
1579
+ ∂z
1580
+ = i nAg13ˆ¯σ13(ω, z),
1581
+ ∂ˆa†
1582
+ s(ω, z)
1583
+ ∂z
1584
+ = i nAg42ˆ¯σ42(ω, z),
1585
+ (67)
1586
+ we get coupled equations for counter-propagating anti-
1587
+ Stokes (propagating along +z) and Stokes (propagating
1588
+ along −z) fields in the backward-wave configuration
1589
+
1590
+ ∂z
1591
+ �ˆaas
1592
+ ˆa†
1593
+ s
1594
+
1595
+ =
1596
+
1597
+ −αas + i ∆k
1598
+ 2
1599
+ iκas
1600
+ iκs
1601
+ α∗
1602
+ s − i ∆k
1603
+ 2
1604
+ � �ˆaas
1605
+ ˆa†
1606
+ s
1607
+
1608
+ +
1609
+ � ˆ¯Fas
1610
+ − ˆ¯F †
1611
+ s
1612
+
1613
+ ,
1614
+ (68)
1615
+ where
1616
+ ˆ¯Fas = ig13
1617
+
1618
+ nA
1619
+
1620
+ βas
1621
+ 12 ˆ¯f (σ)
1622
+ 12 + βas
1623
+ 13 ˆ¯f (σ)
1624
+ 13 + βas
1625
+ 42 ˆ¯f (σ)
1626
+ 42 + βas
1627
+ 43 ˆ¯f (σ)
1628
+ 43
1629
+
1630
+ ,
1631
+ ˆ¯F †
1632
+ s = −ig42
1633
+
1634
+ nA
1635
+
1636
+ βs
1637
+ 12 ˆ¯f (σ)
1638
+ 12 + βs
1639
+ 13 ˆ¯f (σ)
1640
+ 13 + βs
1641
+ 42 ˆ¯f (σ)
1642
+ 42 + βs
1643
+ 43 ˆ¯f (σ)
1644
+ 43
1645
+
1646
+ ,
1647
+ (69)
1648
+ and
1649
+ αas = −iωas
1650
+ 2c χas,
1651
+ αs = −iωs
1652
+ 2c χs,
1653
+ κas =
1654
+ √ωasωs
1655
+ 2c
1656
+ χ(3)
1657
+ as EpEc,
1658
+ κs =
1659
+ √ωsωas
1660
+ 2c
1661
+ χ(3)∗
1662
+ s
1663
+ E∗
1664
+ pE∗
1665
+ c ,
1666
+ χas = 4n |µ13|2
1667
+ ε0ℏ
1668
+ (ϖ + iγ12)
1669
+ T (ϖ)
1670
+ ,
1671
+ χs = n |µ24|2
1672
+ ε0ℏ
1673
+ (ϖ − iγ13)
1674
+ T ∗ (ϖ)
1675
+ |Ωp|2
1676
+ ∆2p + γ2
1677
+ 14
1678
+ ,
1679
+ χ(3)
1680
+ as = nµ13µ32µ24µ41
1681
+ ε0ℏ3
1682
+ 1
1683
+ T (ϖ)
1684
+ 1
1685
+ (∆p + iγ14),
1686
+ χ(3)
1687
+ s
1688
+ = nµ13µ32µ24µ41
1689
+ ε0ℏ3
1690
+ 1
1691
+ T ∗ (ϖ)
1692
+ 1
1693
+ (∆p + iγ14),
1694
+ (70)
1695
+
1696
+ 9
1697
+ -10
1698
+ -5
1699
+ 0
1700
+ 5
1701
+ 10
1702
+ 1
1703
+ 1.0001
1704
+ 1.0002
1705
+ 1.0003
1706
+ 1.0004
1707
+ Macro
1708
+ Micro
1709
+ -10
1710
+ -5
1711
+ 0
1712
+ 5
1713
+ 10
1714
+ 0
1715
+ 1
1716
+ 2
1717
+ 3
1718
+ 4
1719
+ 10-4
1720
+ -10
1721
+ -5
1722
+ 0
1723
+ 5
1724
+ 10
1725
+ 1
1726
+ 1.0001
1727
+ 1.0002
1728
+ 1.0003
1729
+ 1.0004
1730
+ -10
1731
+ -5
1732
+ 0
1733
+ 5
1734
+ 10
1735
+ 0
1736
+ 1
1737
+ 2
1738
+ 3
1739
+ 4
1740
+ 10-4
1741
+ (
1742
+ -
1743
+ )
1744
+ (
1745
+ -
1746
+ )
1747
+ (
1748
+ -
1749
+ )
1750
+ (
1751
+ -
1752
+ )
1753
+ (a)
1754
+ (b)
1755
+ (c)
1756
+ (d)
1757
+ Figure 4.
1758
+ Four real correlations of Stokes and anti-Stokes fields in the group delay regime:
1759
+ (a) ⟨ˆaas(L)ˆa†
1760
+ as(L)⟩, (b)
1761
+ ⟨ˆa†
1762
+ as(L)ˆaas(L)⟩, (c) ⟨ˆas(0)ˆa†
1763
+ s(0)⟩, and (d) ⟨ˆa†
1764
+ s(0)ˆas(0)⟩.
1765
+ The macroscopic (“Macro”) and microscopic (“Micro”) approaches
1766
+ are shown as blue solid and red dashed lines, respectively.
1767
+ The expressions for βas
1768
+ µν and βs
1769
+ µν are listed in Eqs. (65).
1770
+ ∆k = (ωas−ωs)/c−(⃗kc+⃗kp)· ˆz is the phase mismatching
1771
+ in vacuum.
1772
+ Here the complex αas represents the EIT
1773
+ loss and phase dispersion.
1774
+ α∗
1775
+ s is the Raman gain and
1776
+ dispersion along −z propagation direction. One can show
1777
+ that the nonlinear coupling coefficients can be expressed
1778
+ as κas = κeiθ and κs = κe−iθ, where
1779
+ κ =
1780
+ √ωasωs
1781
+ 2c
1782
+ nµ13µ24
1783
+ ε0ℏ
1784
+ ����
1785
+ ΩpΩc
1786
+ ∆p + iγ14
1787
+ ����
1788
+ 1
1789
+ T(ϖ),
1790
+ (71)
1791
+ and θ is the phase of ΩpΩc/(∆p + iγ14). As a result, κas
1792
+ and κs fulfill the gauge transformation discussed in Sec.
1793
+ II. Therefore, to be consistent with the treatment in Sec.
1794
+ II, we rewrite Eq. (68) to
1795
+
1796
+ ∂z
1797
+ �ˆaas
1798
+ ˆa†
1799
+ s
1800
+
1801
+ = MB
1802
+ �ˆaas
1803
+ ˆa†
1804
+ s
1805
+
1806
+ +
1807
+ � ˆFas
1808
+ − ˆF †
1809
+ s
1810
+
1811
+ ,
1812
+ (72)
1813
+ where
1814
+ MB =
1815
+
1816
+ −αas + i ∆k
1817
+ 2
1818
+
1819
+
1820
+ α∗
1821
+ s − i ∆k
1822
+ 2
1823
+
1824
+ ,
1825
+ ˆFas = ˆ¯Fase−iθ/2,
1826
+ ˆF †
1827
+ s = ˆ¯F †
1828
+ s eiθ/2.
1829
+ (73)
1830
+ Similarly, we rewrite the SFWM quantum Langevin
1831
+ equations in the forward-wave configuration in Ap-
1832
+ pendix C.
1833
+ We now turn to compare Eq.
1834
+ (72) with Eq.
1835
+ (29)
1836
+ from the phenomenological approach in Sec. II, where
1837
+ we take mode 1 as anti-Stokes and mode 2 as Stokes in
1838
+ the backward-wave configuration.
1839
+ From Eq.
1840
+ (29), we
1841
+ have
1842
+ ˆFas = NBR11 ˆf1 + NBI11 ˆf †
1843
+ 1 + NBI12 ˆf2 + NBR12 ˆf †
1844
+ 2,
1845
+ ˆF †
1846
+ s = −NBR21 ˆf1 − NBI21 ˆf †
1847
+ 1 − NBI22 ˆf2 − NBR22 ˆf †
1848
+ 2.
1849
+ (74)
1850
+ Therefore, we obtain ˆFas and ˆF †
1851
+ s from two different ap-
1852
+ proaches: Eq. (69) from the microscopic photon-atom
1853
+ interaction, and Eq. (74) from the macroscopic phe-
1854
+ nomenological approach.
1855
+ Although we remark that
1856
+ the atomic noise operators ˆ¯f (σ)
1857
+ µν
1858
+ are different from the
1859
+ field noise operators ˆfm, the correlations of ˆFas and ˆFs
1860
+ uniquely determine the system performance. While we
1861
+ find it difficult to analytically prove the two approaches
1862
+ are equivalent, we could numerically compute and com-
1863
+ pare the commutation relations and correlations of ˆaas,
1864
+ ˆa†
1865
+ as, ˆas, and ˆa†
1866
+ s.
1867
+ We consider here the backward-wave SFWM in laser-
1868
+ cooled
1869
+ 85Rb atoms with relevant atomic energy lev-
1870
+ els being |1⟩ =
1871
+ ��52S1/2, F = 2
1872
+
1873
+ , |2⟩ =
1874
+ ��52S1/2, F = 3
1875
+
1876
+ ,
1877
+
1878
+ 10
1879
+ -2
1880
+ -1
1881
+ 0
1882
+ 1
1883
+ 2
1884
+ Macro
1885
+ Micro
1886
+ -2
1887
+ -1
1888
+ 0
1889
+ 1
1890
+ 2
1891
+ -2
1892
+ -1
1893
+ 0
1894
+ 1
1895
+ 2
1896
+ -10
1897
+ -5
1898
+ 0
1899
+ 5
1900
+ 10
1901
+ -2
1902
+ -1
1903
+ 0
1904
+ 1
1905
+ 2
1906
+ -10
1907
+ 0
1908
+ 10
1909
+ -2
1910
+ -1
1911
+ 0
1912
+ 1
1913
+ 2
1914
+ -10
1915
+ -5
1916
+ 0
1917
+ 5
1918
+ 10
1919
+ -2
1920
+ -1
1921
+ 0
1922
+ 1
1923
+ 2
1924
+ -2
1925
+ -1
1926
+ 0
1927
+ 1
1928
+ 2
1929
+ -2
1930
+ -1
1931
+ 0
1932
+ 1
1933
+ 2
1934
+ -2
1935
+ -1
1936
+ 0
1937
+ 1
1938
+ 2
1939
+ -10
1940
+ -5
1941
+ 0
1942
+ 5
1943
+ 10
1944
+ -2
1945
+ -1
1946
+ 0
1947
+ 1
1948
+ 2
1949
+ -10
1950
+ 0
1951
+ 10
1952
+ -2
1953
+ -1
1954
+ 0
1955
+ 1
1956
+ 2
1957
+ -10
1958
+ -5
1959
+ 0
1960
+ 5
1961
+ 10
1962
+ -2
1963
+ -1
1964
+ 0
1965
+ 1
1966
+ 2
1967
+ 10-2 (
1968
+ -
1969
+ )
1970
+ 10-2 (
1971
+ -
1972
+ )
1973
+ 10-2 (
1974
+ -
1975
+ )
1976
+ 10-2 (
1977
+ -
1978
+ )
1979
+ 10-2 (
1980
+ -
1981
+ )
1982
+ 10-2 (
1983
+ -
1984
+ )
1985
+ (a)
1986
+ (b)
1987
+ (c)
1988
+ (d)
1989
+ (e)
1990
+ (f)
1991
+ Figure 5.
1992
+ Twelve complex correlations of Stokes and anti-Stokes fields in the group delay regime: (a) ⟨ˆaas(L)ˆaas(L)⟩ =
1993
+ ⟨ˆa†
1994
+ as(L)ˆa†
1995
+ as(L)⟩∗, (b) ⟨ˆaas(L)ˆas(0)⟩ = ⟨ˆa†
1996
+ s(0)ˆa†
1997
+ as(L)⟩∗, (c) ⟨ˆaas(L)ˆa†
1998
+ s(0)⟩ = ⟨ˆas(0)ˆa†
1999
+ as(L)⟩∗, (d) ⟨ˆa†
2000
+ as(L)ˆas(0)⟩ = ⟨ˆa†
2001
+ s(0)ˆaas(L)⟩∗,
2002
+ (e) ⟨ˆas(0)ˆaas(L)⟩ = ⟨ˆa†
2003
+ as(L)ˆa†
2004
+ s(0)⟩∗, and (f) ⟨ˆas(0)ˆas(0)⟩ = ⟨ˆa†
2005
+ s(0)ˆa†
2006
+ s(0)⟩∗. The macroscopic (“Macro”) and microscopic (“Mi-
2007
+ cro”) approaches are shown as blue solid and red dashed lines, respectively.
2008
+ |3⟩ =
2009
+ ��52P1/2, F = 3
2010
+
2011
+ , |4⟩ =
2012
+ ��52P3/2, F = 3
2013
+
2014
+ . The decay
2015
+ and dephasing rates for corresponding energy levels are
2016
+ Γ3 = Γ4 = 2π × 6 MHz, Γ31 = 5
2017
+ 9Γ3, Γ32 = 4
2018
+ 9Γ3, Γ41 =
2019
+ 4
2020
+ 9Γ4, Γ42 = 5
2021
+ 9Γ4, γ13 = γ23 = γ14 = γ24 = 2π × 3 MHz,
2022
+ and γ12 = 2π × 0.03 MHz. With vacuum inputs in both
2023
+ Stokes (z = L) and anti-Stokes (z = 0) modes, we have
2024
+ ⟨ˆaas(ϖ, 0)ˆa†
2025
+ as(ϖ′, 0)⟩ = ⟨ˆas(ϖ, L)ˆa†
2026
+ s(ϖ′, L)⟩ = δ(ϖ − ϖ′)
2027
+ and ⟨ˆa†
2028
+ as(ϖ, 0)ˆaas(ϖ′, 0)⟩ = ⟨ˆa†
2029
+ s(ϖ, L)ˆas(ϖ′, L)⟩ = 0.
2030
+ There is also no correlation between Stokes and anti-
2031
+ Stokes fields at their inputs.
2032
+ We numerically compute SFWM in two different
2033
+ regimes to confirm the consistency between the macro-
2034
+ scopic and microscopic theories. i) The first is the group
2035
+ delay regime, where the SFWM spectrum bandwidth is
2036
+ determined by the EIT slow-light induced phase mis-
2037
+ matching [10].
2038
+ The working parameters are:
2039
+ Ωp =
2040
+ 2π × 1.2 MHz, Ωc = 2π × 12 MHz, ∆p = 2π × 500 MHz.
2041
+ The cold atomic medium with length L = 2 cm has den-
2042
+ sity n = 5.1 × 1016 m−3, corresponding to an atomic
2043
+ optical depth OD = 80 on the anti-Stokes resonance
2044
+ transition. ii) The second is the Rabi oscillation regime,
2045
+ where biphoton correlation reveals single-atom dynamics
2046
+ [10]. The working parameters are: Ωp = 2π × 1.2 MHz,
2047
+ Ωc = 2π ×24 MHz, ∆p = ωp −ω14 = 2π ×500 MHz. The
2048
+ cold atomic medium with length L = 0.2 cm has density
2049
+ n = 6.4×1014 m−3, corresponding to OD = 0.1. In both
2050
+ cases, we take ∆k = 127 rad/m.
2051
+ The numerical results in the group delay regime are
2052
+
2053
+ 11
2054
+ -50
2055
+ 0
2056
+ 50
2057
+ 0
2058
+ 0.2
2059
+ 0.4
2060
+ 0.6
2061
+ 0.8
2062
+ 1
2063
+ -50
2064
+ 0
2065
+ 50
2066
+ -10
2067
+ -5
2068
+ 0
2069
+ 10-9
2070
+ -50
2071
+ 0
2072
+ 50
2073
+ 0
2074
+ 0.2
2075
+ 0.4
2076
+ 0.6
2077
+ 0.8
2078
+ 1
2079
+ Macro
2080
+ Micro
2081
+ NLN
2082
+ -50
2083
+ 0
2084
+ 50
2085
+ -1
2086
+ 0
2087
+ 1
2088
+ 2
2089
+ 10-11
2090
+ (
2091
+ -
2092
+ )
2093
+ (
2094
+ -
2095
+ )
2096
+ (
2097
+ -
2098
+ )
2099
+ (
2100
+ -
2101
+ )
2102
+ (a)
2103
+ (b)
2104
+ (c)
2105
+ (d)
2106
+ Figure 6. Comparison of commutation relations between the macroscopic (“Macro”, blue solid lines) and microscopic (“Micro”,
2107
+ red dashed lines) approaches in the damped Rabi oscillation regime: (a) [ˆaas(L), ˆa†
2108
+ as(L)], (b) [ˆaas(L), ˆa†
2109
+ as(L)] − δ(ϖ − ϖ′), (c)
2110
+ [ˆas(0), ˆa†
2111
+ s(0)], and (d)[ˆas(0), ˆa†
2112
+ s(0)] − δ(ϖ − ϖ′). The results with no Langevin noise operators (“NLN”) are shown as black
2113
+ dotted lines in (a) and (c).
2114
+ plotted in Figs. 3, 4, and 5.
2115
+ The commutation re-
2116
+ lations [ˆaas(L), ˆa†
2117
+ as(L)] and [ˆas(0), ˆa†
2118
+ s(0)] are shown in
2119
+ Fig. 3.
2120
+ Both macroscopic and microscopic approaches
2121
+ agree well with each other [Figs. 3(a) and (c)], with neg-
2122
+ ligible relative small difference < 1.0 × 10−6 [Figs. 3(b)
2123
+ and (d)]. As expected, the macroscopic phenomenologi-
2124
+ cal results give perfect flat lines at [ˆaas(L,ϖ),ˆa†
2125
+ as(L,ϖ′)]
2126
+ δ(ϖ−ϖ′)
2127
+ =
2128
+ [ˆas(0,ϖ),ˆa†
2129
+ s(0,ϖ′)]
2130
+ δ(ϖ−ϖ′)
2131
+ = 1 which is the starting point of Sec.
2132
+ II. The microscopic results of field commutations are
2133
+ consistent with the macroscopic approach, but with <
2134
+ 1.0 × 10−6 deviation at some spectra points.
2135
+ As we
2136
+ understand, these small spectra discrepancies may be
2137
+ caused by the ground-state and zeroth-order approxi-
2138
+ mations we take for solving the microscopic Heisenberg-
2139
+ Langevin equations (59). If the Langevin noise operators
2140
+ are not taken into account, as shown in the black dotted
2141
+ curves in Figs. 3(a) and (c), the anti-Stokes commuta-
2142
+ tion relation is not preserved and displays EIT transmis-
2143
+ sion spectrum, while Stokes commutation relation still
2144
+ approximately holds due to the negligible gain or loss in
2145
+ Stokes channel under the ground-state approximation.
2146
+ Figure 4 displays four real-valued correlations of
2147
+ Stokes and anti-Stokes fields:
2148
+ (a )⟨ˆaas(L)ˆa†
2149
+ as(L)⟩, (b)
2150
+ ⟨ˆa†
2151
+ as(L)ˆaas(L)⟩, (c) ⟨ˆas(0)ˆa†
2152
+ s(0)⟩, and (d) ⟨ˆa†
2153
+ s(0)ˆas(0)⟩.
2154
+ Figure 5 shows the twelve (six pairs) complex-valued
2155
+ correlations
2156
+ of
2157
+ Stokes
2158
+ and
2159
+ anti-Stokes
2160
+ fields:
2161
+ (a)
2162
+ ⟨ˆaas(L)ˆaas(L)⟩ = ⟨ˆa†
2163
+ as(L)ˆa†
2164
+ as(L)⟩∗, (b) ⟨ˆaas(L)ˆas(0)⟩ =
2165
+ ⟨ˆa†
2166
+ s(0)ˆa†
2167
+ as(L)⟩∗, (c) ⟨ˆaas(L)ˆa†
2168
+ s(0)⟩ = ⟨ˆas(0)ˆa†
2169
+ as(L)⟩∗, (d)
2170
+ ⟨ˆa†
2171
+ as(L)ˆas(0)⟩ = ⟨ˆa†
2172
+ s(0)ˆaas(L)⟩∗, (e) ⟨ˆas(0)ˆaas(L)⟩ =
2173
+ ⟨ˆa†
2174
+ as(L)ˆa†
2175
+ s(0)⟩∗, and (f) ⟨ˆas(0)ˆas(0)⟩ = ⟨ˆa†
2176
+ s(0)ˆa†
2177
+ s(0)⟩∗.
2178
+ The macroscopic solutions agree well with those obtained
2179
+ from the microscopic approach.
2180
+ The numerical results in the Rabi oscillation regime are
2181
+ plotted in Figs. 6, 7, and 8. The macroscopic phenomeno-
2182
+ logical results also agree remarkably well with those from
2183
+ the microscopic theory.
2184
+ IV.
2185
+ BIPHOTON GENERATION
2186
+ We now turn to apply the quantum Langevin the-
2187
+ ory to study time-frequency entangled photon pair
2188
+ (biphoton) generation through spontaneous four-wave
2189
+ mixing
2190
+ process,
2191
+ especially
2192
+ in
2193
+ a
2194
+ variety
2195
+ of
2196
+ situa-
2197
+ tions involving gain, loss, and/or complex nonlinear
2198
+ coupling
2199
+ coefficient.
2200
+ We
2201
+ consider
2202
+ continuous-wave
2203
+ pumping
2204
+ whose
2205
+ time
2206
+ translation
2207
+ symmetry
2208
+ leads
2209
+ to
2210
+ frequency
2211
+ anti-correlation
2212
+ ω1 + ω2
2213
+ =constant
2214
+ between the paired photons.
2215
+ In the spontaneous
2216
+
2217
+ 12
2218
+ -50
2219
+ 0
2220
+ 50
2221
+ 1
2222
+ 1+0.5E-7
2223
+ 1+1.0E-7
2224
+ 1+1.5E-7
2225
+ Macro
2226
+ Micro
2227
+ -50
2228
+ 0
2229
+ 50
2230
+ 0
2231
+ 5
2232
+ 10
2233
+ 15
2234
+ 10-8
2235
+ -50
2236
+ 0
2237
+ 50
2238
+ 1
2239
+ 1+0.2E-7
2240
+ 1+0.4E-7
2241
+ 1+0.6E-7
2242
+ 1+0.8E-7
2243
+ 1+1.0E-7
2244
+ -50
2245
+ 0
2246
+ 50
2247
+ 0
2248
+ 2
2249
+ 4
2250
+ 6
2251
+ 8
2252
+ 10
2253
+ 10-8
2254
+ (
2255
+ -
2256
+ )
2257
+ (
2258
+ -
2259
+ )
2260
+ (
2261
+ -
2262
+ )
2263
+ (
2264
+ -
2265
+ )
2266
+ (a)
2267
+ (b)
2268
+ (c)
2269
+ (d)
2270
+ Figure 7. Four real correlations of Stokes and anti-Stokes fields in the damped Rabi oscillation regime: (a) ⟨ˆaas(L)ˆa†
2271
+ as(L)⟩, (b)
2272
+ ⟨ˆa†
2273
+ as(L)ˆaas(L)⟩, (c) ⟨ˆas(0)ˆa†
2274
+ s(0)⟩, and (d) ⟨ˆa†
2275
+ s(0)ˆas(0)⟩. The macroscopic (“Macro”) and microscopic (“Micro”) approaches are
2276
+ shown as blue solid and red dashed lines, respectively.
2277
+ four-wave mixing process, both input states are vac-
2278
+ uum:
2279
+ ⟨ˆa†
2280
+ 1(ϖ, 0)ˆa1(ϖ′, 0)⟩ = ⟨ˆa†
2281
+ 2(ϖ, 0)ˆa2(ϖ′, 0)⟩ = 0,
2282
+ ⟨ˆa1(ϖ′, 0)ˆa†
2283
+ 1(ϖ, 0)⟩
2284
+ =
2285
+ ⟨ˆa2(ϖ′, 0)ˆa†
2286
+ 2(ϖ, 0)⟩
2287
+ =
2288
+ δ(ϖ
2289
+ − ϖ′)
2290
+ for
2291
+ the
2292
+ forward-wave
2293
+ configuration,
2294
+ and ⟨ˆa†
2295
+ 1(ϖ, 0)ˆa1(ϖ′, 0)⟩
2296
+ =
2297
+ ⟨ˆa†
2298
+ 2(ϖ, L)ˆa2(ϖ′, L)⟩
2299
+ =
2300
+ 0,
2301
+ ⟨ˆa1(ϖ, 0)ˆa†
2302
+ 1(ϖ′, 0)⟩ = ⟨ˆa2(ϖ, L)ˆa†
2303
+ 2(ϖ′, L)⟩ = δ(ϖ − ϖ′)
2304
+ for the backward-wave configuration. From Eq. (4), with
2305
+ ω1 = ω10 + ϖ and ω2 = ω20 − ϖ, we have
2306
+ ˆa1(t, z1) = eiω10( z1
2307
+ c −t)
2308
+
2309
+
2310
+
2311
+ dϖˆa1(ϖ, z1)eiϖ( z1
2312
+ c −t)e−i ∆k
2313
+ 2 z1,
2314
+ ˆa2(t, z2) = eiω20(± z2
2315
+ c −t)
2316
+
2317
+
2318
+
2319
+ dϖˆa2(ϖ, z2)eiϖ(± z2
2320
+ c −t)e−i ∆k
2321
+ 2 z2,
2322
+ (75)
2323
+ where ± represents the forward-wave (+) or backward-
2324
+ wave (−) configuration, z = z1 and z = z2 are the
2325
+ output positions of channels 1 and 2, respectively. For
2326
+ the forward-wave configuration, z1 = z2 = L. For the
2327
+ backward-wave configuration, z1 = L and z2 = 0. The
2328
+ phase mismatching in vacuum ∆k = (ωas ±ωs)/c−(⃗kc +
2329
+ ⃗kp)· ˆz ≃ (ωas0 ±ωs0)/c−(⃗kc +⃗kp)· ˆz is nearly a constant.
2330
+ The vacuum time delay zi/c constants are usually very
2331
+ small in usual experimental conditions, from now on we
2332
+ ignore these constants for simplification and rewrite the
2333
+ above equations to (otherwise one just needs to make a
2334
+ time translation t → t − zi/c)
2335
+ ˆa1(t, z1) = e−iω10t
2336
+
2337
+
2338
+
2339
+ dϖˆa1(ϖ, z1)e−iϖt,
2340
+ ˆa2(t, z2) = e−iω20t
2341
+
2342
+
2343
+
2344
+ dϖˆa2(ϖ, z2)eiϖt.
2345
+ (76)
2346
+ The photon rate in channel m can be computed from
2347
+ Rm ≡
2348
+
2349
+ ˆa†
2350
+ m (t, zm) ˆam (t, zm)
2351
+
2352
+ = 1
2353
+
2354
+ �� ∞
2355
+ −∞
2356
+ dϖdϖ′e−iϖteiϖ′t �
2357
+ ˆa†
2358
+ m (ϖ′, zm) ˆam (ϖ, zm)
2359
+
2360
+ .
2361
+ (77)
2362
+ Here we are particularly interested in the two-photon
2363
+ Glauber correlation in the time domain, which can be
2364
+ computed from the following two different orders
2365
+ G(2)
2366
+ 2,1 (t2, t1)
2367
+ ≡⟨ˆa†
2368
+ 1 (t1, z1) ˆa†
2369
+ 2 (t2, z2) ˆa2 (t2, z2) ˆa1 (t1, z1)⟩
2370
+ =|⟨ˆa2 (t2, z2) ˆa1 (t1, z1)⟩|2
2371
+ + |⟨ˆa†
2372
+ 2 (t2, z2) ˆa1 (t1, z1)⟩|2 + R1R2,
2373
+ (78)
2374
+
2375
+ 13
2376
+ -5
2377
+ 0
2378
+ 5
2379
+ Macro
2380
+ Micro
2381
+ -5
2382
+ 0
2383
+ 5
2384
+ -5
2385
+ 0
2386
+ 5
2387
+ -50
2388
+ 0
2389
+ 50
2390
+ -5
2391
+ 0
2392
+ 5
2393
+ -50
2394
+ 0
2395
+ 50
2396
+ -5
2397
+ 0
2398
+ 5
2399
+ -50
2400
+ 0
2401
+ 50
2402
+ -5
2403
+ 0
2404
+ 5
2405
+ -5
2406
+ 0
2407
+ 5
2408
+ -5
2409
+ 0
2410
+ 5
2411
+ -5
2412
+ 0
2413
+ 5
2414
+ -50
2415
+ 0
2416
+ 50
2417
+ -5
2418
+ 0
2419
+ 5
2420
+ -50
2421
+ 0
2422
+ 50
2423
+ -5
2424
+ 0
2425
+ 5
2426
+ -50
2427
+ 0
2428
+ 50
2429
+ -5
2430
+ 0
2431
+ 5
2432
+ 10-5 (
2433
+ -
2434
+ )
2435
+ 10-5 (
2436
+ -
2437
+ )
2438
+ 10-5 (
2439
+ -
2440
+ )
2441
+ 10-5 (
2442
+ -
2443
+ )
2444
+ 10-5 (
2445
+ -
2446
+ )
2447
+ 10-5 (
2448
+ -
2449
+ )
2450
+ (a)
2451
+ (b)
2452
+ (c)
2453
+ (d)
2454
+ (e)
2455
+ (f)
2456
+ Figure 8.
2457
+ Twelve complex correlations of Stokes and anti-Stokes fields in the damped Rabi oscillation regime:
2458
+ (a)
2459
+ ⟨ˆaas(L)ˆaas(L)⟩ = ⟨ˆa†
2460
+ as(L)ˆa†
2461
+ as(L)⟩∗, (b) ⟨ˆaas(L)ˆas(0)⟩ = ⟨ˆa†
2462
+ s(0)ˆa†
2463
+ as(L)⟩∗, (c) ⟨ˆaas(L)ˆa†
2464
+ s(0)⟩ = ⟨ˆas(0)ˆa†
2465
+ as(L)⟩∗, (d) ⟨ˆa†
2466
+ as(L)ˆas(0)⟩ =
2467
+ ⟨ˆa†
2468
+ s(0)ˆaas(L)⟩∗, (e) ⟨ˆas(0)ˆaas(L)⟩ = ⟨ˆa†
2469
+ as(L)ˆa†
2470
+ s(0)⟩∗, and (f) ⟨ˆas(0)ˆas(0)⟩ = ⟨ˆa†
2471
+ s(0)ˆa†
2472
+ s(0)⟩∗. The macroscopic (“Macro”) and mi-
2473
+ croscopic (“Micro”) approaches are shown as blue solid and red dashed lines, respectively.
2474
+ G(2)
2475
+ 1,2 (t1, t2)
2476
+ ≡⟨ˆa†
2477
+ 2 (t2, z2) ˆa†
2478
+ 1 (t1, z1) ˆa1 (t1, z1) ˆa2 (t2, z2)⟩
2479
+ =|⟨ˆa1 (t1, z1) ˆa2 (t2, z2)⟩|2
2480
+ + |⟨ˆa†
2481
+ 2 (t2, z2) ˆa1 (t1, z1)⟩|2 + R1R2,
2482
+ (79)
2483
+ where we have applied the Gaussian moment theorem
2484
+ [23, 24] to decompose the fourth-order field correlations
2485
+ to the sum of the products of second-order field corre-
2486
+ lations.
2487
+ The first term in Eqs.
2488
+ (78) and (79) can be
2489
+ expressed as |Ψ2,1(t2, t1)|2 and |Ψ1,2(t1, t2)|2, where
2490
+ Ψ2,1(t2, t1) = ⟨ˆa2 (t2, z2) ˆa1 (t1, z1)⟩
2491
+ = e−iω20t2e−iω10t1ψ2,1(t1 − t2),
2492
+ (80)
2493
+ Ψ1,2(t1, t2) = ⟨ˆa1 (t1, z1) ˆa2 (t2, z2)⟩
2494
+ = e−iω20t2e−iω10t1ψ1,2(t1 − t2),
2495
+ (81)
2496
+ are the two-photon wavefunctions with the relative parts
2497
+ ψ2,1(t1 − t2)
2498
+ = 1
2499
+
2500
+ ��
2501
+ dϖdϖ′⟨ˆa2(ϖ′, z2)ˆa1(ϖ, z1)⟩e−iϖ(t1−t2). (82)
2502
+ ψ1,2(t1 − t2)
2503
+ = 1
2504
+
2505
+ ��
2506
+ dϖdϖ′⟨ˆa1(ϖ, z1)ˆa2(ϖ′, z2)⟩e−iϖ(t1−t2). (83)
2507
+ One can show that the second term in Eqs. (78) and (79)
2508
+ is zero if the nonlinear coupling coefficient is real-valued,
2509
+
2510
+ 14
2511
+ and it is usually very small as compared to other terms.
2512
+ The third term in Eqs. (78) and (79) is the accidental
2513
+ coincidence counts. The two-photon wavefunction and
2514
+ Glauber correlation satisfy the following exchange sym-
2515
+ metry
2516
+ ψ21(t1 − t2) = ψ2,1(t1 − t2) = ψ1,2(t1 − t2),
2517
+ Ψ21(t2, t1) = Ψ2,1(t2, t1) = Ψ1,2(t1, t2),
2518
+ G(2)
2519
+ 21 (t2, t1) = G(2)
2520
+ 2,1 (t2, t1) = G(2)
2521
+ 1,2 (t1, t2) .
2522
+ (84)
2523
+ The normalized two-photon correlation is defined as
2524
+ g(2)
2525
+ 21 (t2, t1) ≡ G(2)
2526
+ 21 (t2, t1)
2527
+ R1R2
2528
+ .
2529
+ (85)
2530
+ As the system has time translation symmetry with
2531
+ continuous-wave pumping, G(2)
2532
+ 21 (t2, t1) = G(2)
2533
+ 21 (t1 − t2)
2534
+ depends only on the relative time t1 − t2.
2535
+ A.
2536
+ Loss and Gain
2537
+ To simplify and unify the descriptions for account-
2538
+ ing both forward- and backward-wave cases, we define
2539
+ “input-output” fields:
2540
+ ˆa1,in ≡ ˆa1(0), ˆa2,in ≡ ˆa2(0),
2541
+ ˆa1,out ≡ ˆa1(L), and ˆa2,out ≡ ˆa2(L) for the forward-wave
2542
+ case; ˆa1,in ≡ ˆa1(0), ˆa2,in ≡ ˆa2(L), ˆa1,out ≡ ˆa1(L), and
2543
+ ˆa2,out ≡ ˆa2(0) for the backward-wave case. In this sub-
2544
+ section, we aim to investigate the roles of loss and gain
2545
+ in biphoton generation, considering linear loss in mode 1
2546
+ (Re{α1} = α ≥ 0) and linear gain (Re{α2} = −g ≤ 0)
2547
+ in mode 2. We also assume κ is real, or its contribution
2548
+ to Langevin noises is much smaller than the linear gain
2549
+ and loss, i.e., Im{κ} ≪ {α, g}. In this case, for forward-
2550
+ and backward-wave configurations, the noise matrix is
2551
+ reduced to
2552
+ NF,B =
2553
+ �√
2554
+
2555
+ 0
2556
+ 0
2557
+ ±i√2g
2558
+
2559
+ .
2560
+ (86)
2561
+ Hence, the output fields in Eqs. (19) and (36) can be
2562
+ rewritten as
2563
+ �ˆa1,out
2564
+ ˆa†
2565
+ 2,out
2566
+
2567
+ =
2568
+
2569
+ A B
2570
+ C D
2571
+ � �ˆa1,in
2572
+ ˆa†
2573
+ 2,in
2574
+
2575
+ +
2576
+ � L
2577
+ 0
2578
+
2579
+ X11 X12
2580
+ X21 X22
2581
+ � � ˆf1 (z)
2582
+ ˆf2 (z)
2583
+
2584
+ dz.
2585
+ (87)
2586
+ where Xmn are combined coefficients. We further rewrite
2587
+ Eq. (87) as
2588
+ ˆa1,out = Aˆa1,in + Bˆa†
2589
+ 2,in +
2590
+ � L
2591
+ 0
2592
+
2593
+ X11 ˆf1(z) + X12 ˆf2(z)
2594
+
2595
+ ,
2596
+ ˆa2,out = C∗ˆa†
2597
+ 1,in + D∗ˆa2,in +
2598
+ � L
2599
+ 0
2600
+
2601
+ X∗
2602
+ 21 ˆf †
2603
+ 1(z) + X∗
2604
+ 22 ˆf †
2605
+ 2(z)
2606
+
2607
+ .
2608
+ (88)
2609
+ As shown in Eq.
2610
+ (84), there are two different orders
2611
+ [⟨: ˆa2ˆa1 :⟩ or ⟨: ˆa1ˆa2 :⟩] to compute the two-photon wave-
2612
+ function and Galuber correlation. Although these two
2613
+ orders are equivalent, the numerical computation com-
2614
+ plexity may be significantly different. Computing bipho-
2615
+ ton wavefunction in Eq. (83) in the order ⟨: ˆa1ˆa2 :⟩ in-
2616
+ volves nonzero noise field correlations ⟨ ˆfm ˆf †
2617
+ m⟩, while in
2618
+ the order ⟨: ˆa2ˆa1 :⟩ [Eq.
2619
+ (82)] these noise field corre-
2620
+ lations disappear because of ⟨ ˆf †
2621
+ m ˆfm⟩ = 0.
2622
+ These field
2623
+ correlations in the frequency domain can be expressed as
2624
+ ⟨ˆa2out (ϖ′) ˆa1out (ϖ)⟩ = δ(ϖ − ϖ′) [BD∗] ,
2625
+ (89)
2626
+ ⟨ˆa1out (ϖ) ˆa2out (ϖ′)⟩
2627
+ = δ(ϖ − ϖ′)
2628
+
2629
+ AC∗ +
2630
+ � L
2631
+ 0
2632
+ dz (X11X∗
2633
+ 21 + X12X∗
2634
+ 22)
2635
+
2636
+ .
2637
+ (90)
2638
+ Therefore, we obtain the biphoton wavefunction follow-
2639
+ ing the order ⟨: ˆa2ˆa1 :⟩
2640
+ ψ21(τ) =
2641
+ ��
2642
+ dϖdϖ′⟨ˆa2,out(ϖ′)ˆa1,out(ϖ)⟩e−iϖτ
2643
+ =
2644
+
2645
+ dϖBD∗e−iϖτ.
2646
+ (91)
2647
+ where τ = t1 − t2. If following the order ⟨: ˆa1ˆa2 :⟩, we
2648
+ have
2649
+ ψ12(τ) =
2650
+ ��
2651
+ dϖdϖ′⟨ˆa1,out(ϖ)ˆa2,out(ϖ′)⟩e−iϖτ
2652
+ =
2653
+
2654
+
2655
+
2656
+ AC∗ +
2657
+ � L
2658
+ 0
2659
+ dz (X11X∗
2660
+ 21 + X12X∗
2661
+ 22)
2662
+
2663
+ e−iϖτ.
2664
+ (92)
2665
+ One can show that the second term in Eqs. (78) and (79)
2666
+ is zero in this loss-gain configuration. The single-channel
2667
+ photon rates can be obtained as
2668
+ R1 = 1
2669
+
2670
+
2671
+ |B|2dϖ,
2672
+ R2 = 1
2673
+
2674
+ � �
2675
+ |C|2 +
2676
+ � L
2677
+ 0
2678
+ dz
2679
+
2680
+ |X21|2 + |X22|2�
2681
+
2682
+ dϖ.
2683
+ (93)
2684
+ It is interesting to remark that, in the loss-gain config-
2685
+ uration, the biphoton field correlation following the order
2686
+ ⟨: ˆagainˆaloss :⟩ does not involve noise field correlations as
2687
+ shown in Eqs. (89) and (91), which dramatically reduces
2688
+ the computation complexity. On the other side, taking
2689
+ the order ⟨: ˆalossˆagain :⟩ must include noise field corre-
2690
+ lations as shown in Eqs. (90) and (92). This may be
2691
+ understood in the heralded photon picture [25]: When
2692
+ a photon in a lossy channel is detected (annihilated) by
2693
+ a detector, we can always ensure there is its partner (or
2694
+ paired) photon in another channel; On the other side,
2695
+ when a photon is detected in a gain channel which pro-
2696
+ duces multiple photons, we can not always ensure it has
2697
+ a partner photon in another channel. The exchange sym-
2698
+ metry can only be preserved by taking into account the
2699
+ Langevin noises.
2700
+
2701
+ 15
2702
+ 0
2703
+ 1
2704
+ 2
2705
+ 109
2706
+ Macro
2707
+ Micro
2708
+ NLN
2709
+ -0.5
2710
+ 0
2711
+ 0.5
2712
+ 1
2713
+ 0
2714
+ 1
2715
+ 2
2716
+ (a)
2717
+ (b)
2718
+ Figure 9. Two-photon Glauber correlation in time domain in
2719
+ the group delay regime: (a) G(2)
2720
+ s,as(τ) and (b) G(2)
2721
+ as,s(τ). The
2722
+ simulation conditions are the same as that in Figs. 3, 4, and
2723
+ 5. NLN: no Langevin noise included.
2724
+ In the SFWM described in Sec. III, the anti-Stokes
2725
+ photons experience finite EIT loss due to the ground
2726
+ state dephasing (γ12 ̸= 0), and the Stokes photons prop-
2727
+ agate with negligible but small Raman gain.
2728
+ Figure
2729
+ 9 displays the two-photon Glauber correlation in the
2730
+ group delay regime with the same parameters as those
2731
+ in Figs. 3, 4 and 5. As shown in Fig. 9(a) and (b), both
2732
+ macroscopic and microscopic approaches with Langevin
2733
+ noises give consistent results. As expected, the compu-
2734
+ tation of G(2)
2735
+ s,as(τ) (following the order ⟨: ˆasˆaas :⟩) with-
2736
+ out Langevin noise operators (black dotted line: NLN)
2737
+ agrees with the exact results obtained from both macro-
2738
+ scopic (blue solid line) and microscopic (red dashed line)
2739
+ approaches, shown in Fig. 9(a).
2740
+ On the contrary, the
2741
+ computation of G(2)
2742
+ as,s(τ) (following the order ⟨: ˆaasˆas :⟩)
2743
+ without Langevin noise operators deviates significantly
2744
+ from the exact results, as shown in Fig. 9(b).
2745
+ B.
2746
+ Complex Phase Mismatching
2747
+ Different from the Heisenberg picture where the evo-
2748
+ lution of field operators is governed by their Langevin
2749
+ coupled equations, reference [10] provides a perturbation
2750
+ theory to describe biphoton state in the interaction pic-
2751
+ ture. The solution from Heisenberg-Langevin theory may
2752
+ contain correlations of more than two photons, while the
2753
+ perturbation theory focuses only on the two-photon state
2754
+ by ignoring higher-order terms. These two treatments are
2755
+ expected to give the same results in the limit of small pa-
2756
+ rameter gain. Although the perturbation theory in the
2757
+ interaction picture provides a much clear physics picture
2758
+ of two-photon state, treating loss and gain requires a
2759
+ proper justification. In the perturbation theory, linear
2760
+ loss and gain are included in the complex phase mis-
2761
+ matching ∆˜k(ϖ) [10]. For the SFWM described in Sec.
2762
+ III, Ref. [10] derives the biphoton relative wavefunction
2763
+ with perturbation theory as
2764
+ ψ(τ) = iL
2765
+
2766
+
2767
+ dϖκ(ϖ)Φ(ϖ)e−iϖτ,
2768
+ (94)
2769
+ where the longitudinal detuning function is
2770
+ Φ(ϖ) = sinc
2771
+
2772
+ ∆˜kL
2773
+ 2
2774
+
2775
+ ei(kas+ks)L,
2776
+ (95)
2777
+ There is a statement in Ref. [10]: “It is found that to be
2778
+ consistent with the Heisenberg–Langevin theory in the
2779
+ low-gain limit, the argument in Φ should be replaced by
2780
+ ∆˜k =
2781
+
2782
+ ⃗kas + ⃗k∗
2783
+ s − ⃗kc − ⃗kp
2784
+
2785
+ · ˆz, where ⃗k∗
2786
+ s is the conjugate
2787
+ of ⃗ks.” For the SFWM in the double-Λ four-level atomic
2788
+ system, there is small Raman gain in the Stokes chan-
2789
+ nel. What happens if there is loss in the Stokes channel?
2790
+ Should we take ⃗k∗
2791
+ s or ⃗ks in the complex phase mismatch-
2792
+ ing ∆˜k(ϖ)? Although Ref. [10] takes ⃗k∗
2793
+ s for Stokes pho-
2794
+ tons with gain, it is not clear whether it still holds for the
2795
+ case with loss. In this subsection, we do not only provide
2796
+ a justification for the above statement in Ref. [10] from
2797
+ the quantum Langevin theory by taking small parametric
2798
+ gain approximation, but also extend the complex phase
2799
+ mismatching to the case with loss in the Stokes channel.
2800
+ We take the same backward-wave configuration in
2801
+ Ref. [10]. We assume anti-Stokes photons in mode 1 are
2802
+ lossless with EIT and there is gain (or loss) in Stokes
2803
+ mode 2. The small parametric gain fulfills |κ| ≪ {α, g}.
2804
+ In the backward-wave configuration, using Eq.
2805
+ (7),
2806
+ (34), and (37), we obtain analytical expressions of
2807
+ A, B, C, and D as
2808
+ A =
2809
+
2810
+ q2 − 4κ2e−(α1−α∗
2811
+ 2)L/2
2812
+ qsinh
2813
+
2814
+ L
2815
+ 2
2816
+
2817
+ q2 − 4κ2
2818
+
2819
+ +
2820
+
2821
+ q2 − 4κ2cosh
2822
+
2823
+ L
2824
+ 2
2825
+
2826
+ q2 − 4κ2
2827
+ �,
2828
+ B =
2829
+ 2iκ
2830
+ q +
2831
+
2832
+ q2 − 4κ2coth( L
2833
+ 2
2834
+
2835
+ q2 − 4κ2)
2836
+ ,
2837
+ C =
2838
+ −2iκ
2839
+ q +
2840
+
2841
+ q2 − 4κ2coth( L
2842
+ 2
2843
+
2844
+ q2 − 4κ2)
2845
+ ,
2846
+ D =
2847
+
2848
+ q2 − 4κ2e(α1−α∗
2849
+ 2)L/2
2850
+ qsinh
2851
+
2852
+ L
2853
+ 2
2854
+
2855
+ q2 − 4κ2
2856
+
2857
+ +
2858
+
2859
+ q2 − 4κ2cosh
2860
+
2861
+ L
2862
+ 2
2863
+
2864
+ q2 − 4κ2
2865
+ �,
2866
+ (96)
2867
+ where q ≡ α1 + α∗
2868
+ 2 − i∆k. In the small parametric gain
2869
+ approximation, we have
2870
+
2871
+ q2 − 4κ2 ≈ q
2872
+ = α1 + α∗
2873
+ 2 − i∆k = −i(∆k1 − ∆k2
2874
+ ∗ + ∆k),
2875
+ (97)
2876
+ and
2877
+ α1 − α∗
2878
+ 2 = −i(∆k1 + ∆k2
2879
+ ∗).
2880
+ (98)
2881
+
2882
+ 16
2883
+ where ∆km = ωm
2884
+ 2c χm is the wavenumber difference from
2885
+ that in vacuum. Hence, we simplify A, B, C, and D to
2886
+ A =exp [i∆k1L] exp
2887
+ �i∆kL
2888
+ 2
2889
+
2890
+ ,
2891
+ B =iκLsinc
2892
+ �(∆k1 − ∆k∗
2893
+ 2 + ∆k)L
2894
+ 2
2895
+
2896
+ × exp
2897
+ �i(∆k1 − ∆k∗
2898
+ 2 + ∆k)L
2899
+ 2
2900
+
2901
+ ,
2902
+ C = − iκLsinc
2903
+ �(∆k1 − ∆k∗
2904
+ 2 + ∆k)L
2905
+ 2
2906
+
2907
+ × exp
2908
+ �i(∆k1 − ∆k∗
2909
+ 2 + ∆k)L
2910
+ 2
2911
+
2912
+ ,
2913
+ D =exp [−i∆k∗
2914
+ 2L] exp
2915
+ �i∆kL
2916
+ 2
2917
+
2918
+ .
2919
+ (99)
2920
+ We first look at the case with gain in the Stokes (mode
2921
+ 2). As discussed in Sec. IV A, we take the order ⟨: ˆa2ˆa1 :⟩
2922
+ ψ21(τ) =
2923
+ ��
2924
+ dϖdϖ′⟨ˆa2,out(ϖ′)ˆa1,out(ϖ)⟩e−iϖτ
2925
+ =
2926
+
2927
+ dϖBD∗e−iϖτ,
2928
+ (100)
2929
+ where
2930
+ BD∗ = iκLsinc
2931
+ �(∆k1 − ∆k∗
2932
+ 2 + ∆k)L
2933
+ 2
2934
+
2935
+ × exp
2936
+ �i(∆k1 − ∆k∗
2937
+ 2 + 2∆k2)L
2938
+ 2
2939
+
2940
+ .
2941
+ (101)
2942
+ Comparing Eqs. (100) and (101) with Eqs. (94) and (95),
2943
+ particularly for the argument in the sinc function, we
2944
+ have ∆˜k = ∆k1 − ∆k∗
2945
+ 2 + ∆k = k1 − k∗
2946
+ 2 − kc + kp =
2947
+ kas − k∗
2948
+ s − kc + kp which is consistent with the statement
2949
+ in Ref. [10].
2950
+ We now look at the case with loss in the Stokes (mode
2951
+ 2). We take the order ⟨: ˆa1ˆa2 :⟩ and have
2952
+ ψ12(τ) =
2953
+ ��
2954
+ dϖdϖ′⟨ˆa1,out(ϖ)ˆa2,out(ϖ′)⟩e−iϖτ
2955
+ =
2956
+
2957
+ dϖAC∗e−iϖτ,
2958
+ (102)
2959
+ where
2960
+ AC∗ = iκ∗Lsinc
2961
+ �(∆k∗
2962
+ 1 − ∆k2 + ∆k)L
2963
+ 2
2964
+
2965
+ exp
2966
+ �i(2∆k1 − ∆k∗
2967
+ 1 + ∆k2)L
2968
+ 2
2969
+
2970
+ .
2971
+ (103)
2972
+ Comparing Eqs. (102) and (103) with Eqs. (94) and (95),
2973
+ we have ∆˜k = ∆k∗
2974
+ 1 − ∆k2 + ∆k = k1 − k2 − kc + kp =
2975
+ kas − ks − kc + kp, which is different from the case with
2976
+ gain. Here we have taken k1 ≃ k∗
2977
+ 1 for lossless mode 1.
2978
+ Although our discussion is based on the backward-
2979
+ wave configuration, the conclusion can be extended to
2980
+ the forward-wave configuration, which is derived in de-
2981
+ tail in Appendix D. Therefore, in the case with gain in
2982
+ the Stokes mode 2, the complex phase mismatching is
2983
+ ∆˜k =
2984
+
2985
+ ⃗kas + ⃗k∗
2986
+ s − ⃗kc − ⃗kp
2987
+
2988
+ · ˆz. In the case with loss in
2989
+ the Stokes mode 2, the complex phase mismatching be-
2990
+ comes ∆˜k =
2991
+
2992
+ ⃗kas + ⃗ks − ⃗kc − ⃗kp
2993
+
2994
+ · ˆz.
2995
+ C.
2996
+ Complex Nonlinear Coupling Coefficient and
2997
+ Rabi Oscillation
2998
+ As illustrated in Fig. 2, we can understand the SFWM
2999
+ process in the following picture. After a Stoke and anti-
3000
+ Stokes photon pair is born from a single atom follow-
3001
+ ing the atomic transitions [Fig. 2(b)], the paired pho-
3002
+ tons then propagate through the medium [Fig. 2(a)]. As
3003
+ the photon pair can be generated at any atom inside the
3004
+ medium, the overall two-photon wavefunction (or prob-
3005
+ ability amplitude) is a superposition of all possible such
3006
+ generation-propagation two-photon Feynman paths. Fol-
3007
+ lowing this picture, when the propagation effect can be
3008
+ ignored, the biphoton state reveals the single atom dy-
3009
+ namics, which is connected to the nonlinear coupling co-
3010
+ efficient. In the following, we consider SFWM in the limit
3011
+ of small optical depth (OD) where the linear propaga-
3012
+ tion effect is small and show how the complex spectrum
3013
+ of nonlinear coupling coefficient reveals single-atom Rabi
3014
+ oscillation.
3015
+ We
3016
+ rewrite
3017
+ the
3018
+ nonlinear
3019
+ coupling
3020
+ coefficient
3021
+ in
3022
+ Eq. (71) as:
3023
+ κ(ϖ) = J
3024
+
3025
+ 1
3026
+ (ϖ − Ωe/2 + iγe) −
3027
+ 1
3028
+ (ϖ + Ωe/2 + iγe)
3029
+
3030
+ ,
3031
+ (104)
3032
+ where
3033
+ J = −
3034
+ √ωasωsnµ13µ24
3035
+ 8cε0ℏΩe
3036
+ ����
3037
+ ΩpΩc
3038
+ ∆p + iγ14
3039
+ ���� .
3040
+ (105)
3041
+ Here Ωe =
3042
+
3043
+ |Ωc|2 − (γ13 − γ12)2 is the effective coupling
3044
+ Rabi frequency, and γe = (γ12 + γ13)/2 is the effective
3045
+ dephasing rate. Obviously, the nonlinear coupling coeffi-
3046
+ cient κ(ϖ) has a complex spectrum, with two resonances
3047
+ separated by the effective coupling Rabi frequency Ωe. In
3048
+ the ground-state approximation with major atomic pop-
3049
+ ulation in state |1⟩, the undepleted pump laser beam is
3050
+ far detuned from the transition |1⟩ → |4⟩ and its exci-
3051
+ tation is weak such that we can take χs ≃ 0. On the
3052
+ other side, from Eq.
3053
+ (70) we have the complex linear
3054
+ susceptibility for anti-Stokes photons
3055
+ χas(ϖ) = −n |µ13|2
3056
+ ε0ℏ
3057
+ (ϖ + iγ12)
3058
+ (ϖ − Ωe/2 + iγe)(ϖ + Ωe/2 + iγe)
3059
+ (106)
3060
+ Although the anti-Stokes photon absorption at ϖ = 0
3061
+ is suppressed by the EIT effect, there are two absorp-
3062
+ tion resonances appearing at ϖ = ±Ωe/2 which coin-
3063
+ cide with the two resonances of nonlinear coupling coef-
3064
+ ficient in Eq. (104). We take the pump laser with weak
3065
+
3066
+ 17
3067
+ intensity (∝ |Ωp|2) and large detuning (∆p) such that
3068
+ Re{αas(ϖ = ±Ωe/2)}>Im{κ(ϖ = ±Ωe/2)}, which are
3069
+ usually satisfied in the ground state condition. As the
3070
+ propagation effect is small and the phase matching is not
3071
+ important, the paired photons are mostly generated from
3072
+ the two resonances (ϖ = ±Ωe/2) of the nonlinear cou-
3073
+ pling coefficient.
3074
+ In the forward-wave configuration, with the coupling
3075
+ matrix
3076
+ MF =
3077
+
3078
+ −αas + i ∆k
3079
+ 2
3080
+
3081
+ −iκ
3082
+ −i ∆k
3083
+ 2
3084
+
3085
+ ,
3086
+ (107)
3087
+ and short medium length L satisfying |MFL| ≪ 1, we
3088
+ have approximately
3089
+
3090
+ A B
3091
+ C D
3092
+
3093
+ = eMFL ∼= 1 + MFL
3094
+ =
3095
+
3096
+ 1 − αasL + i ∆k
3097
+ 2 L
3098
+ iκL
3099
+ −iκL
3100
+ 1 − i ∆k
3101
+ 2 L
3102
+
3103
+ .
3104
+ (108)
3105
+ As discussed in Sec. IV A, the biphoton field correlation
3106
+ following the order ⟨: ˆasˆaas :⟩ does not need count the
3107
+ Langevin noise operators:
3108
+ ⟨ˆas(ϖ′, L)ˆaas(ϖ, L)⟩ = BD∗δ(ϖ − ϖ′)
3109
+ = iκL(1 + i∆k
3110
+ 2 L)δ(ϖ − ϖ′)
3111
+ ∼= iκ(ϖ)Lδ(ϖ − ϖ′),
3112
+ (109)
3113
+ where we have neglected higher order terms O(L2). From
3114
+ Eq. (82), we have the relative biphoton wavefunction
3115
+ ψs−as(τ) = iL
3116
+
3117
+
3118
+ dϖκ(ϖ)e−iϖτ,
3119
+ (110)
3120
+ which is the Fourier transform of the nonlinear coupling
3121
+ coefficient with τ = tas − ts. Substituting Eq. (104) into
3122
+ Eq. (110) we obtain
3123
+ ψs−as(τ) = LJe−γeτ[e−iΩeτ/2 − eiΩeτ/2]Θ(τ)
3124
+ = −2iLJe−γeτ sin
3125
+ �Ωeτ
3126
+ 2
3127
+
3128
+ Θ(τ),
3129
+ (111)
3130
+ where Θ(τ) is the Heaviside function.
3131
+ Equation (111)
3132
+ shows a damped Rabi oscillation, resulting from the beat-
3133
+ ing between biphotons generated from the two resonances
3134
+ at ϖ = ±Ωe/2. The Heaviside function shows the anti-
3135
+ Stokes photon is always generated after its paired Stokes
3136
+ photon following the time order of atomic transitions
3137
+ |1⟩ → |4⟩ → |2⟩ → |3⟩ → |1⟩ in an SFWM cycle shown in
3138
+ Fig. 2(b).
3139
+ In the backward-wave configuration, the coupling ma-
3140
+ trix becomes
3141
+ MB =
3142
+
3143
+ −αas + i ∆k
3144
+ 2
3145
+
3146
+
3147
+ −i ∆k
3148
+ 2
3149
+
3150
+ .
3151
+ (112)
3152
+ 0
3153
+ 2
3154
+ 4
3155
+ 6
3156
+ 105
3157
+ Macro
3158
+ Micro
3159
+ NLN
3160
+ 0
3161
+ 2
3162
+ 4
3163
+ 6
3164
+ -0.4
3165
+ -0.2
3166
+ 0
3167
+ 0.2
3168
+ 0.4
3169
+ 0
3170
+ 2
3171
+ 4
3172
+ 6
3173
+ |
3174
+ s-as|2
3175
+ (a)
3176
+ (b)
3177
+ (c)
3178
+ Figure 10. Two-photon Glauber correlation in time domain
3179
+ in the damped Rabi oscillation regime: (a) G(2)
3180
+ s,as(τ) and (b)
3181
+ G(2)
3182
+ as,s(τ). The simulation conditions are the same as that in
3183
+ Figs. 6, 7, and 8. (c) shows the analytic solution for the bipho-
3184
+ ton waveform |ψs−as(τ)|2. NLN: no Langevin noise included.
3185
+ With |MBL| ≪ 1 we have
3186
+ � ¯A
3187
+ ¯B
3188
+ ¯C
3189
+ ¯D
3190
+
3191
+ = eMBL ∼= 1 + MBL
3192
+ =
3193
+
3194
+ 1 − αasL + i ∆k
3195
+ 2 L
3196
+ iκL
3197
+ iκL
3198
+ 1 − i ∆k
3199
+ 2 L
3200
+
3201
+ ,
3202
+ (113)
3203
+ and
3204
+
3205
+ A B
3206
+ C D
3207
+
3208
+ =
3209
+
3210
+ 1 − αasL + i ∆k
3211
+ 2 L
3212
+ iκL
3213
+ −iκL
3214
+ 1 + i ∆k
3215
+ 2 L
3216
+
3217
+ ,
3218
+ (114)
3219
+ where we have neglect higher order terms O(L2). Simi-
3220
+ larly, we have
3221
+ ⟨ˆas(ϖ′, 0)ˆaas(ϖ, L)⟩ ∼= iκ(ϖ)Lδ(ϖ − ϖ′),
3222
+ (115)
3223
+ which is the same as Eq. (109) of the forward-wave con-
3224
+ figuration. Therefore, we obtain Rabi oscillations in both
3225
+ forward- and backward-wave configurations.
3226
+ Equation
3227
+ (111) is identical to the result derived from the pertur-
3228
+ bation theory in the interaction picture [10].
3229
+
3230
+ 18
3231
+ Figure 10 displays the two-photon Glauber correlation
3232
+ in the damped Rabi oscillation regime with the same pa-
3233
+ rameters as those in Figs.
3234
+ 6, 7 and 8.
3235
+ As illustrated
3236
+ in Fig. 10(a) and (b), both macroscopic and microscopic
3237
+ approaches with Langevin noises give consistent results.
3238
+ As expected, the computation of G(2)
3239
+ s,as(τ) (following the
3240
+ order ⟨: ˆasˆaas :⟩) without Langevin noise operators (dot
3241
+ points) agrees with the exact results obtained from both
3242
+ microscopic (red dashed line) and macroscopic (blue solid
3243
+ line) approaches, shown in Fig. 10(a).
3244
+ On the con-
3245
+ trary, the computation of G(2)
3246
+ as,s(τ) (following the order
3247
+ ⟨: ˆaasˆas :⟩) without Langevin noise operators (dot points:
3248
+ NLN) deviates significantly from the exact results and vi-
3249
+ olates the causality, as shown in Fig. 10(b). Fig. 10(c)
3250
+ shows the result from the analytic solution in Eq. (111)
3251
+ which agree well with the exact results in Figs. 10(a) and
3252
+ (b).
3253
+ It is interesting to examine a system without gain and
3254
+ loss whose Langevin noises are purely contributed by the
3255
+ complex nonlinear coupling coefficient. In this case, the
3256
+ above approximation and conclusion do not hold. Let’s
3257
+ now consider the case 3 with the forward-wave config-
3258
+ uration in Sec. II A, where α1 = α2 = ∆k = 0, and
3259
+ κ = η + iζ. As shown in Sec. II A, the noise matrix is
3260
+ different as ζ is positive or negative. We first consider
3261
+ ζ > 0, the Langevin coupled equations (27) becomes
3262
+
3263
+ ∂z
3264
+ �ˆa1
3265
+ ˆa†
3266
+ 2
3267
+
3268
+ =
3269
+
3270
+ 0
3271
+
3272
+ −iκ
3273
+ 0
3274
+ � �ˆa1
3275
+ ˆa†
3276
+ 2
3277
+
3278
+ +
3279
+
3280
+ ζ
3281
+
3282
+ 1
3283
+ 1
3284
+ −1 1
3285
+ � � ˆf1
3286
+ ˆf †
3287
+ 2
3288
+
3289
+ .
3290
+ (116)
3291
+ Under the condition |MFL| ≪ 1, we solve Eq.
3292
+ (116) to
3293
+ the first order of L and have
3294
+ ˆa1(L) ∼= ˆa1(0) + iκLˆa†
3295
+ 2(0) +
3296
+
3297
+ ζ
3298
+ � L
3299
+ 0
3300
+ dz
3301
+
3302
+ ˆf1 + ˆf †
3303
+ 2
3304
+
3305
+ ,
3306
+ ˆa2(L) ∼= ˆa2(0) + iκ∗Lˆa†
3307
+ 1(0) +
3308
+
3309
+ ζ
3310
+ � L
3311
+ 0
3312
+ dz
3313
+
3314
+ − ˆf †
3315
+ 1 + ˆf2
3316
+
3317
+ .
3318
+ (117)
3319
+ The two-photon field correlations are
3320
+ ⟨ˆa1(L)ˆa2(L)⟩ = ⟨ˆa2(L)ˆa1(L)⟩ ∼= i
3321
+ 2(κ + κ∗)Lδ(ϖ − ϖ′).
3322
+ (118)
3323
+ As ζ < 0, the Langevin coupled equations (27) becomes
3324
+
3325
+ ∂z
3326
+ �ˆa1
3327
+ ˆa†
3328
+ 2
3329
+
3330
+ =
3331
+
3332
+ 0
3333
+
3334
+ −iκ
3335
+ 0
3336
+ � �ˆa1
3337
+ ˆa†
3338
+ 2
3339
+
3340
+ +
3341
+
3342
+ −ζ
3343
+
3344
+ 1
3345
+ 1
3346
+ −1 1
3347
+ � � ˆf †
3348
+ 1ˆf2
3349
+
3350
+ . (119)
3351
+ Under the condition |MFL| ≪ 1, we solve Eq.
3352
+ (119) to
3353
+ the first order of L and have
3354
+ ˆa1(L) ∼= ˆa1(0) + iκLˆa†
3355
+ 2(0) +
3356
+
3357
+ −ζ
3358
+ � L
3359
+ 0
3360
+ dz
3361
+
3362
+ ˆf †
3363
+ 1 + ˆf2
3364
+
3365
+ ,
3366
+ ˆa2(L) ∼= ˆa2(0) + iκ∗Lˆa†
3367
+ 1(0) +
3368
+
3369
+ −ζ
3370
+ � L
3371
+ 0
3372
+ dz
3373
+
3374
+ − ˆf1 + ˆf †
3375
+ 2
3376
+
3377
+ .
3378
+ (120)
3379
+ The two-photon field correlations are
3380
+ ⟨ˆa1(L)ˆa2(L)⟩ = ⟨ˆa2(L)ˆa1(L)⟩ ∼= i
3381
+ 2(k + k∗)Lδ(ϖ − ϖ′),
3382
+ (121)
3383
+ which is the same as Eq. (118). The biphoton relative
3384
+ wavefunction is
3385
+ ψ21(τ) = ψ∗
3386
+ 21(−τ) = iL
3387
+
3388
+
3389
+ dϖ1
3390
+ 2(k + k∗)e−iϖτ.
3391
+ (122)
3392
+ One can prove that under the same limit |MBL| ≪ 1, the
3393
+ backward-wave configuration gives the same two-photon
3394
+ field correlation [Eqs.
3395
+ (118) and (121)] and temporal
3396
+ wavefunction [Eq. (122)]. Equation (122) suggests the
3397
+ biphoton temporal wavefunction has time reversal sym-
3398
+ metry when there is no linear gain and loss.
3399
+ V.
3400
+ CONCLUSION
3401
+ In summary, we provide a macroscopic phenomenolog-
3402
+ ical formula of quantum Langevin equations for two cou-
3403
+ pled phase-conjugated fields with linear loss (gain) and
3404
+ complex nonlinear coupling coefficient, in both forward-
3405
+ and backward-wave configurations.
3406
+ The macroscopic
3407
+ phenomenological formula, obtained from the coupling
3408
+ matrix and the requirement of preserving commutation
3409
+ relations of field operators during propagation, does not
3410
+ require knowing microscopic details of light-matter inter-
3411
+ action and internal atomic structures. To validate this
3412
+ phenomenological formula, we take SFWM in a double-
3413
+ Λ four-level atomic system as an example to numeri-
3414
+ cally confirm that our macroscopic phenomenological re-
3415
+ sult is consistent with that obtained from microscopic
3416
+ Heisenberg-Langevin theory. As compared to the com-
3417
+ plicated microscopic theory which varies from system
3418
+ to system, the macroscopic coupled equations are much
3419
+ more friendly to experimentalists. We apply the quantum
3420
+ Langevin equations to study the effects of gain and/or
3421
+ loss as well as complex nonlinear coupling coefficient in
3422
+ biphoton generation, particularly to the temporal quan-
3423
+ tum correlations. We show that the computation com-
3424
+ plexity can be dramatically reduced by taking a proper
3425
+ order of field operators based on loss and gain. Making
3426
+ a comparison between the quantum Langevin theory (in
3427
+ the Heisenberg picture) and the perturbation theory (in
3428
+ the interaction picture [10]), we extend the expression of
3429
+ complex phase mismatching to account for loss and gain.
3430
+ At last, we reveal Rabi oscillation in SFWM biphoton
3431
+ temporal correlation when the propagation effect is small.
3432
+ Although in this article we focus on biphoton generation
3433
+ from the spontaneous parametric process, the quantum
3434
+ Langevin coupled equations can also be used to study
3435
+ two-mode squeezing, parametric oscillation, and other
3436
+ quantum light state generation.
3437
+ ACKNOWLEDGMENTS
3438
+ S.D.
3439
+ acknowledges
3440
+ support
3441
+ from
3442
+ DOE
3443
+ (DE-
3444
+ SC0022069),
3445
+ AFOSR
3446
+ (FA9550-22-1-0043)
3447
+ and
3448
+ NSF
3449
+ (CNS-2114076, 2228725).
3450
+
3451
+ 19
3452
+ Appendix A: Noise Matrix in Backward-Wave
3453
+ Configuration
3454
+ In the macroscopic quantum Langevin equations, the
3455
+ requirement of preserving commutation relations allows
3456
+ multiple choices of the noise matrix. For example, ˆf1 →
3457
+ − ˆf1 or/and ˆf2 → − ˆf2 do not affect any computation re-
3458
+ sults of physical observables involving pairs of Langevin
3459
+ noise operators. As an example, here we provide several
3460
+ equivalent noise matrices for backward-wave configura-
3461
+ tion:
3462
+ NB1 ≡
3463
+
3464
+ 1
3465
+ 0
3466
+ 0 −1
3467
+ � ��
3468
+ −MB11 −MB12
3469
+ MB21
3470
+ MB22
3471
+
3472
+ +
3473
+
3474
+ −MB11 −MB12
3475
+ MB21
3476
+ MB22
3477
+ �∗
3478
+ =
3479
+
3480
+ 1
3481
+ 0
3482
+ 0 −1
3483
+
3484
+ NF,
3485
+ NB2 ≡ NB1
3486
+
3487
+ 1
3488
+ 0
3489
+ 0 −1
3490
+
3491
+ =
3492
+ ��
3493
+ −MB11 MB12
3494
+ −MB21 MB22
3495
+
3496
+ +
3497
+
3498
+ −MB11 MB12
3499
+ −MB21 MB22
3500
+ �∗
3501
+ ,
3502
+ NB3 ≡ NB1
3503
+
3504
+ −1 0
3505
+ 0
3506
+ 1
3507
+
3508
+ ,
3509
+ NB4 ≡ NB1
3510
+
3511
+ −1
3512
+ 0
3513
+ 0
3514
+ −1
3515
+
3516
+ = −NB1.
3517
+ (A1)
3518
+ We take the first choice NB1 in the main text [see Eq. (31)
3519
+ in Sec. II B] so that it is consistent with the microscopic
3520
+ treatment in Sec. III.
3521
+ Appendix B: Heisenberg-Langevin Equations of
3522
+ SFWM
3523
+ The full Heisenberg equation of motion can be written
3524
+ as
3525
+ ˙ˆS = i( ˆO ˆS − ˆS ˆO) + ˆG + ˆF,
3526
+ (B1)
3527
+ where
3528
+ ˆS =
3529
+
3530
+ ��
3531
+ ˆσ11 ˆσ12 ˆσ13 ˆσ14
3532
+ ˆσ21 ˆσ22 ˆσ23 ˆσ24
3533
+ ˆσ31 ˆσ32 ˆσ33 ˆσ34
3534
+ ˆσ41 ˆσ42 ˆσ43 ˆσ44
3535
+
3536
+ �� ,
3537
+ (B2)
3538
+ ˆO = −
3539
+
3540
+ ��
3541
+ 0
3542
+ 0
3543
+ g31ˆaas Ωp/2
3544
+ 0
3545
+ ϖ
3546
+ Ωc/2
3547
+ g42ˆas
3548
+ g13ˆa∗
3549
+ as Ω∗
3550
+ c/2
3551
+ ϖ
3552
+ 0
3553
+ Ω∗
3554
+ p/2
3555
+ g24ˆa∗
3556
+ s
3557
+ 0
3558
+ ∆p
3559
+
3560
+ �� ,
3561
+ (B3)
3562
+ ˆG =
3563
+
3564
+ ��
3565
+ Γ31ˆσ33 + Γ41ˆσ44
3566
+ −γ12ˆσ12
3567
+ −γ13ˆσ13 −γ14ˆσ14
3568
+ −γ12ˆσ21
3569
+ Γ32ˆσ33 + Γ42ˆσ44 −γ23ˆσ23 −γ24ˆσ24
3570
+ −γ13ˆσ31
3571
+ −γ23ˆσ32
3572
+ −Γ3ˆσ33 −γ34ˆσ34
3573
+ −γ14ˆσ41
3574
+ −γ24ˆσ42
3575
+ −γ34ˆσ43 −Γ4ˆσ44
3576
+
3577
+ �� ,
3578
+ (B4)
3579
+ ˆF =
3580
+
3581
+ ����
3582
+ ˆf (σ)
3583
+ 11
3584
+ ˆf (σ)
3585
+ 12
3586
+ ˆf (σ)
3587
+ 13
3588
+ ˆf (σ)
3589
+ 14
3590
+ ˆf (σ)
3591
+ 21
3592
+ ˆf (σ)
3593
+ 22
3594
+ ˆf (σ)
3595
+ 23
3596
+ ˆf (σ)
3597
+ 24
3598
+ ˆf (σ)
3599
+ 31
3600
+ ˆf (σ)
3601
+ 32
3602
+ ˆf (σ)
3603
+ 33
3604
+ ˆf (σ)
3605
+ 34
3606
+ ˆf (σ)
3607
+ 41
3608
+ ˆf (σ)
3609
+ 42
3610
+ ˆf (σ)
3611
+ 43
3612
+ ˆf (σ)
3613
+ 44
3614
+
3615
+ ���� .
3616
+ (B5)
3617
+ Γm = Γm1 + Γm2 is the total spontaneous decay rate of
3618
+ excited state |m⟩, where m = 3, or 4, and Γmj is the decay
3619
+ rate from state |m⟩ to |j⟩. For the two hyperfine ground
3620
+ states, there are Γ1 = Γ2 = 0.
3621
+ For cold atoms with
3622
+ only spontaneous emisson decay, the dephasing rates γjk
3623
+ (j ̸= k) between states |k⟩ and |j⟩ are γ13 = γ23 = Γ3/2,
3624
+ γ14 = γ24 = Γ4/2, γ34 = (Γ3+Γ4)/2. γ12 is the dephasing
3625
+ rate between two hyperfine ground states |1⟩ and |2⟩.
3626
+ Appendix C: Microscopic SFWM Quantum
3627
+ Langevin Equations in Forward-Wave Configuration
3628
+ Although Sec. III focuses on numerical confirmation
3629
+ of backward-wave SFWM, we remark that it may be
3630
+ helpful for general readers to write the SFWM quantum
3631
+ Langevin equations in the forward-wave configuration as
3632
+ well.
3633
+ In the forward-wave configuration with both Stokes
3634
+ and anti-Stokes fields propagating along +z direction,
3635
+ the coupled Langevin equations become
3636
+
3637
+ ∂z
3638
+ �ˆaas
3639
+ ˆa†
3640
+ s
3641
+
3642
+ = MF
3643
+ �ˆaas
3644
+ ˆa†
3645
+ s
3646
+
3647
+ +
3648
+ � ˆFas
3649
+ ˆF †
3650
+ s
3651
+
3652
+ ,
3653
+ (C1)
3654
+ where
3655
+ MF =
3656
+
3657
+ −αas + i ∆k
3658
+ 2
3659
+
3660
+ −iκ
3661
+ −α∗
3662
+ s − i ∆k
3663
+ 2
3664
+
3665
+ ,
3666
+ (C2)
3667
+ with ∆k = (ωas+ωs)/c−(⃗kc+⃗kp)·ˆz. The noise operators
3668
+ ˆFas and ˆF †
3669
+ s , defined in Eq. (69), originate from micro-
3670
+ scopic atom-light interaction. To compare Eq. (C1) with
3671
+ Eq. (11) from the phenomenological approach in Sec. II,
3672
+ we take mode 1 as anti-Stokes and mode 2 as Stokes in
3673
+ the forward-wave configuration. From Eq. (11), we can
3674
+ also obtain ˆFas and ˆF †
3675
+ s from the noise matrix:
3676
+ ˆFas = NFR11 ˆf1 + NFI11 ˆf †
3677
+ 1 + NFI12 ˆf2 + NFR12 ˆf †
3678
+ 2,
3679
+ ˆF †
3680
+ s = NFR21 ˆf1 + NFI21 ˆf †
3681
+ 1 + NFI22 ˆf2 + NFR22 ˆf †
3682
+ 2.
3683
+ (C3)
3684
+ Appendix D: Complex Phase Mismatching in
3685
+ Forward-Wave Configuration
3686
+ In the forward-wave configuration, similar to the
3687
+ backward-wave configuration in Sec. IV B, we assume
3688
+ anti-Stokes photons in mode 1 are lossless with EIT and
3689
+ there is gain (or loss) in Stokes mode 2. The small para-
3690
+ metric gain fulfills |κ| ≪ {α, g}. Using Eq. (6) and (17),
3691
+
3692
+ 20
3693
+ we obtain analytical expressions of A, B, C, and D as
3694
+ A =
3695
+
3696
+ q2 + 4κ2cosh
3697
+
3698
+ L
3699
+ 2
3700
+
3701
+ q2 + 4κ2
3702
+
3703
+ − qsinh
3704
+
3705
+ L
3706
+ 2
3707
+
3708
+ q2 + 4κ2
3709
+
3710
+
3711
+ q2 + 4κ2e(α1+α∗
3712
+ 2)L/2
3713
+ ,
3714
+ B =
3715
+ 2iκsinh
3716
+
3717
+ L
3718
+ 2
3719
+
3720
+ q2 + 4κ2
3721
+
3722
+
3723
+ q2 + 4κ2e(α1+α∗
3724
+ 2)L/2 ,
3725
+ C =
3726
+ −2iκsinh
3727
+
3728
+ L
3729
+ 2
3730
+
3731
+ q2 + 4κ2
3732
+
3733
+
3734
+ q2 + 4κ2e(α1+α∗
3735
+ 2)L/2 ,
3736
+ D =
3737
+
3738
+ q2 + 4κ2cosh
3739
+
3740
+ L
3741
+ 2
3742
+
3743
+ q2 + 4κ2
3744
+
3745
+ + qsinh
3746
+
3747
+ L
3748
+ 2
3749
+
3750
+ q2 + 4κ2
3751
+
3752
+
3753
+ q2 + 4κ2e(α1+α∗
3754
+ 2)L/2
3755
+ ,
3756
+ (D1)
3757
+ where q ≡ α1 − α∗
3758
+ 2 − i∆k. In the small parametric gain
3759
+ approximation, we have
3760
+
3761
+ q2 − 4κ2 ≈ q
3762
+ = α1 − α∗
3763
+ 2 − i∆k = −i(∆k1 + ∆k∗
3764
+ 2 + ∆k),
3765
+ (D2)
3766
+ and
3767
+ α1 + α∗
3768
+ 2 = −i(∆k1 − ∆k∗
3769
+ 2),
3770
+ (D3)
3771
+ where ∆km = ωm
3772
+ 2c χm is the wavenumber difference from
3773
+ that in vacuum. Hence, we simplify A, B, C, and D to
3774
+ A =exp [i∆k1L] exp
3775
+ �i∆kL
3776
+ 2
3777
+
3778
+ ,
3779
+ B =iκLsinc
3780
+ �(∆k1 + ∆k∗
3781
+ 2 + ∆k)L
3782
+ 2
3783
+
3784
+ × exp
3785
+ �i(∆k1 − ∆k∗
3786
+ 2)L
3787
+ 2
3788
+
3789
+ ,
3790
+ C = − iκLsinc
3791
+ �(∆k1 + ∆k∗
3792
+ 2 + ∆k)L
3793
+ 2
3794
+
3795
+ × exp
3796
+ �i(∆k1 − ∆k∗
3797
+ 2)L
3798
+ 2
3799
+
3800
+ ,
3801
+ D =exp [−i∆k∗
3802
+ 2L] exp
3803
+ �−i∆kL
3804
+ 2
3805
+
3806
+ .
3807
+ (D4)
3808
+ We first look at the case with gain in the Stokes (mode
3809
+ 2). As discussed in Sec. IV A, we take the order ⟨: ˆa2ˆa1 :⟩
3810
+ ψ21(τ) =
3811
+ ��
3812
+ dϖdϖ′⟨ˆa2,out(ϖ′)ˆa1,out(ϖ)⟩e−iϖτ
3813
+ =
3814
+
3815
+ dϖBD∗e−iϖτ,
3816
+ (D5)
3817
+ where
3818
+ BD∗ = iκLsinc
3819
+ �(∆k1 + ∆k∗
3820
+ 2 + ∆k)L
3821
+ 2
3822
+
3823
+ × exp
3824
+ �i(∆k1 − ∆k∗
3825
+ 2 + 2∆k2 + ∆k)L
3826
+ 2
3827
+
3828
+ .
3829
+ (D6)
3830
+ Comparing Eqs. (D5) and (D6) with Eqs. (94) and (95),
3831
+ particularly for the argument in the sinc function, we
3832
+ have ∆˜k = ∆k1 + ∆k∗
3833
+ 2 + ∆k = k1 + k∗
3834
+ 2 − kc − kp =
3835
+ kas + k∗
3836
+ s − kc − kp which is consistent with the statement
3837
+ in Ref. [10].
3838
+ We now look at the case with loss in the Stokes (mode
3839
+ 2). We take the order ⟨: ˆa1ˆa2 :⟩ and have
3840
+ ψ12(τ) =
3841
+ ��
3842
+ dϖdϖ′⟨ˆa1,out(ϖ)ˆa2,out(ϖ′)⟩e−iϖτ
3843
+ =
3844
+
3845
+ dϖAC∗e−iϖτ,
3846
+ (D7)
3847
+ where
3848
+ AC∗ = iκ∗Lsinc
3849
+ �(∆k∗
3850
+ 1 + ∆k2 + ∆k)L
3851
+ 2
3852
+
3853
+ × exp
3854
+ �i(2∆k1 − ∆k∗
3855
+ 1 + ∆k2 + ∆k)L
3856
+ 2
3857
+
3858
+ .
3859
+ (D8)
3860
+ Comparing Eqs. (D7) and (D8) with Eqs. (94) and (95),
3861
+ we have ∆˜k = ∆k∗
3862
+ 1 + ∆k2 + ∆k = k1 + k2 − kc +
3863
+ kp = kas + ks − kc − kp, which is different from the
3864
+ case with gain.
3865
+ Here we have taken k1 ≃ k∗
3866
+ 1 for loss-
3867
+ less mode 1.
3868
+ Therefore, in the case with loss in the
3869
+ Stokes mode 2, the complex phase mismatching becomes
3870
+ ∆˜k =
3871
+
3872
+ ⃗kas + ⃗ks − ⃗kc − ⃗kp
3873
+
3874
+ · ˆz.
3875
+ [1] C. W. Gardiner and M. J. Collett, Input and output in
3876
+ damped quantum systems: Quantum stochastic differen-
3877
+ tial equations and the master equation, Phys. Rev. A 31,
3878
+ 3761 (1985).
3879
+ [2] M. O. Scully and M. S. Zubairy, Quantum optics (1999).
3880
+ [3] Y. Yamamoto and A. Imamoglu, Mesoscopic quantum
3881
+ optics, Mesoscopic Quantum Optics (1999).
3882
+ [4] C. Gardiner, P. Zoller, and P. Zoller, Quantum noise:
3883
+ a handbook of Markovian and non-Markovian quantum
3884
+ stochastic methods with applications to quantum optics
3885
+ (Springer Science & Business Media, 2004).
3886
+ [5] R. Benguria and M. Kac, Quantum langevin equation,
3887
+ Phys. Rev. Lett. 46, 1 (1981).
3888
+ [6] R. W. Boyd, Nonlinear optics (Academic press, 2020).
3889
+ [7] S. Shwartz, R. N. Coffee, J. M. Feldkamp, Y. Feng, J. B.
3890
+ Hastings, G. Y. Yin, and S. E. Harris, X-ray parametric
3891
+ down-conversion in the langevin regime, Phys. Rev. Lett.
3892
+ 109, 013602 (2012).
3893
+ [8] U. A. Javid and Q. Lin, Quantum correlations from
3894
+ dynamically modulated optical nonlinear interactions,
3895
+ Phys. Rev. A 100, 043811 (2019).
3896
+ [9] G. Shafiee, D. V. Strekalov, A. Otterpohl, F. Sedlmeir,
3897
+
3898
+ 21
3899
+ G. Schunk, U. Vogl, H. G. L. Schwefel, G. Leuchs, and
3900
+ C. Marquardt, Nonlinear power dependence of the spec-
3901
+ tral properties of an optical parametric oscillator below
3902
+ threshold in the quantum regime, New Journal of Physics
3903
+ 22, 073045 (2020).
3904
+ [10] S. Du, J. Wen, and M. H. Rubin, Narrowband biphoton
3905
+ generation near atomic resonance, J. Opt. Soc. Am. B
3906
+ 25, C98 (2008).
3907
+ [11] P. Kolchin, Electromagnetically-induced-transparency-
3908
+ based paired photon generation, Phys. Rev. A 75, 033814
3909
+ (2007).
3910
+ [12] L. Zhao, Y. Su, and S. Du, Narrowband biphoton gener-
3911
+ ation in the group delay regime, Phys. Rev. A 93, 033815
3912
+ (2016).
3913
+ [13] C. H. Raymond Ooi, Q. Sun, M. S. Zubairy, and M. O.
3914
+ Scully, Correlation of photon pairs from the double ra-
3915
+ man amplifier: Generalized analytical quantum langevin
3916
+ theory, Phys. Rev. A 75, 013820 (2007).
3917
+ [14] Y. Jiang, Y. Mei, Y. Zuo, Y. Zhai, J. Li, J. Wen, and
3918
+ S. Du, Anti-parity-time symmetric optical four-wave mix-
3919
+ ing in cold atoms, Phys. Rev. Lett. 123, 193604 (2019).
3920
+ [15] Y. Mei, X. Guo, L. Zhao, and S. Du, Mirrorless opti-
3921
+ cal parametric oscillation with tunable threshold in cold
3922
+ atoms, Phys. Rev. Lett. 119, 150406 (2017).
3923
+ [16] X.-W. Luo, C. Zhang, and S. Du, Quantum squeez-
3924
+ ing and sensing with pseudo-anti-parity-time symmetry,
3925
+ Phys. Rev. Lett. 128, 173602 (2022).
3926
+ [17] C. M. Bender and S. Boettcher, Real spectra in non-
3927
+ Hermitian Hamiltonians having PT symmetry, Phys.
3928
+ Rev. Lett. 80, 5243 (1998).
3929
+ [18] M.-A. Miri and A. Al`u, Exceptional points in optics and
3930
+ photonics, Science 363, 10.1126/science.aar7709 (2019).
3931
+ [19] D. A. Braje, V. Bali´c, S. Goda, G. Y. Yin, and S. E. Har-
3932
+ ris, Frequency mixing using electromagnetically induced
3933
+ transparency in cold atoms, Phys. Rev. Lett. 93, 183601
3934
+ (2004).
3935
+ [20] V. Bali´c, D. A. Braje, P. Kolchin, G. Y. Yin, and S. E.
3936
+ Harris, Generation of paired photons with controllable
3937
+ waveforms, Phys. Rev. Lett. 94, 183601 (2005).
3938
+ [21] S. E. Harris, Electromagnetically induced transparency,
3939
+ Phys. Today 50, 36 (1997).
3940
+ [22] M. Fleischhauer, A. Imamoglu, and J. P. Marangos, Elec-
3941
+ tromagnetically induced transparency: Optics in coher-
3942
+ ent media, Rev. Mod. Phys. 77, 633 (2005).
3943
+ [23] L. Mandel and E. Wolf, Optical Coherence and Quantum
3944
+ Optics (Cambridge University Press, Cambridge Eng-
3945
+ land, New York, 1994).
3946
+ [24] W. H. Louisell, Optical Coherence and Quantum Optics
3947
+ (Wiley, New York, 1973).
3948
+ [25] S. Du, Quantum-state purity of heralded single photons
3949
+ produced from frequency-anticorrelated biphotons, Phys.
3950
+ Rev. A 92, 043836 (2015).
3951
+
GNFLT4oBgHgl3EQfGi_B/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
I9AyT4oBgHgl3EQfTfcL/content/tmp_files/2301.00104v1.pdf.txt ADDED
@@ -0,0 +1,1865 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.00104v1 [cs.CR] 31 Dec 2022
2
+ Separating Computational and Statistical Differential Privacy
3
+ (Under Plausible Assumptions)
4
+ Badih Ghazi∗
5
+ Rahul Ilango†
6
+ Pritish Kamath‡
7
+ Ravi Kumar§
8
+ Pasin Manurangsi¶
9
+ Abstract
10
+ Computational differential privacy (CDP) is a natural relaxation of the standard notion of
11
+ (statistical) differential privacy (SDP) proposed by Beimel, Nissim, and Omri (CRYPTO 2008)
12
+ and Mironov, Pandey, Reingold, and Vadhan (CRYPTO 2009). In contrast to SDP, CDP only
13
+ requires privacy guarantees to hold against computationally-bounded adversaries rather than
14
+ computationally-unbounded statistical adversaries. Despite the question being raised explicitly
15
+ in several works (e.g., Bun, Chen, and Vadhan, TCC 2016), it has remained tantalizingly open
16
+ whether there is any task achievable with the CDP notion but not the SDP notion. Even a
17
+ candidate such task is unknown. Indeed, it is even unclear what the truth could be!
18
+ In this work, we give the first construction of a task achievable with the CDP notion but not
19
+ the SDP notion. More specifically, under strong but plausible cryptographic assumptions, we
20
+ construct a task for which there exists an ε-CDP mechanism with ε = O(1) achieving 1 − o(1)
21
+ utility, but any (ε, δ)-SDP mechanism, including computationally unbounded ones, that achieves
22
+ a constant utility must use either a super-constant ε or a non-negligible δ. To prove this, we
23
+ introduce a new approach for showing that a mechanism satisfies CDP: first we show that a
24
+ mechanism is “private” against a certain class of decision tree adversaries, and then we use
25
+ cryptographic constructions to “lift” this into privacy against computational adversaries. We
26
+ believe this approach could be useful to devise further tasks separating CDP from SDP.
27
+ ∗Google Research, Mountain View. [email protected].
28
+ †MIT. Part of this work was done during an internship at Google Research. [email protected].
29
+ ‡Google Research, Mountain View. [email protected].
30
+ §Google Research, Mountain View. [email protected].
31
+ ¶Google Research, Thailand. [email protected].
32
+
33
+ Contents
34
+ 1
35
+ Introduction
36
+ 1
37
+ 2
38
+ Overview of the Results
39
+ 3
40
+ 2.1
41
+ The d-Distance Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
+ 3
43
+ 2.2
44
+ SDP Lower Bound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
+ 4
46
+ 2.3
47
+ A CDP Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
+ 4
49
+ 2.4
50
+ Final Steps
51
+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
+ 6
53
+ 2.5
54
+ On the Plausiblility of the Cryptographic Assumptions . . . . . . . . . . . . . . . . .
55
+ 6
56
+ 3
57
+ Preliminaries
58
+ 7
59
+ 3.1
60
+ Dataset and Adjacency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
61
+ 8
62
+ 3.2
63
+ Mechanism, Utility Function, and Usefulness
64
+ . . . . . . . . . . . . . . . . . . . . . .
65
+ 8
66
+ 3.3
67
+ Notions of Differential Privacy
68
+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
+ 8
70
+ 4
71
+ Low Diameter Set Problem and Nearby Point Problem
72
+ 9
73
+ 4.1
74
+ Simplification of Input Representation . . . . . . . . . . . . . . . . . . . . . . . . . .
75
+ 10
76
+ 4.2
77
+ Nearby Point Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
+ 10
79
+ 4.3
80
+ Verifiable Low Diameter Set Problem . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
+ 10
82
+ 4.4
83
+ From Low Diameter Set Problem to Nearby Point Problem
84
+ . . . . . . . . . . . . . .
85
+ 11
86
+ 5
87
+ CDP Mechanism for Verifiable Low Diameter Set Problem
88
+ 11
89
+ 5.1
90
+ CDP Mechanism without Verifiability
91
+ . . . . . . . . . . . . . . . . . . . . . . . . . .
92
+ 12
93
+ 5.1.1
94
+ Additional Preliminaries: Cryptographic Primitives . . . . . . . . . . . . . . .
95
+ 12
96
+ 5.1.2
97
+ Public-Coin Differing-Inputs Circuits from CRKHFs . . . . . . . . . . . . . .
98
+ 13
99
+ 5.1.3
100
+ From Differing-Inputs Circuits to CDP . . . . . . . . . . . . . . . . . . . . . .
101
+ 15
102
+ 5.2
103
+ CDP Mechanism for VLDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
104
+ 17
105
+ 5.2.1
106
+ Witness-Indistinguishable Proofs . . . . . . . . . . . . . . . . . . . . . . . . .
107
+ 17
108
+ 5.2.2
109
+ Making Utility Function Efficient Using Witness-Indistinguishable Proofs
110
+ . .
111
+ 17
112
+ 6
113
+ SDP Lower Bounds for the Nearby Point Problem
114
+ 19
115
+ 6.1
116
+ Additional Preliminaries: Tools from Differential Privacy
117
+ . . . . . . . . . . . . . . .
118
+ 20
119
+ 6.2
120
+ Weak Hardness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
121
+ 21
122
+ 6.3
123
+ Boosting the Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
124
+ 22
125
+ 6.4
126
+ Boosting the Failure Probability
127
+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
128
+ 23
129
+ 6.5
130
+ Putting Things Together: Proof of Theorem 30 . . . . . . . . . . . . . . . . . . . . .
131
+ 24
132
+ 7
133
+ Putting Things Together: Proof of Theorem 5
134
+ 24
135
+ 8
136
+ Conclusion and Discussion
137
+ 24
138
+ A Comparison of various diO assumptions
139
+ 29
140
+
141
+ 1
142
+ Introduction
143
+ The framework of differential privacy (DP) [DMNS06, DKM+06] gives formal privacy guarantees
144
+ on the outputs of randomized algorithms.
145
+ It has been the subject of a significant body of re-
146
+ search, leading to numerous practical deployments including the US census [Abo18], and industrial
147
+ applications [EPK14, Sha14, Gre16, App17, DKY17, KT18, RSP+21].
148
+ The definition of DP requires privacy against computationally unbounded, i.e., statistical, adver-
149
+ saries. A natural modification is to instead only require privacy against computationally bounded
150
+ adversaries. In cryptography, considering computationally bounded adversaries instead of statisti-
151
+ cal ones enables a vast array of applications, like public-key cryptography. Could the same be true
152
+ for DP? A good survey of the area can be found in Vadhan’s monograph [Vad17, Section 10]. De-
153
+ spite Beimel, Nissim, and Omri [BNO08] defining computational differential privacy (CDP) in 2008
154
+ (definitions that were further extended by Mironov, Pandey, Reingold, and Vadhan [MPRV09]),
155
+ the central question of separating it from statistical differential privacy (SDP)1, in the standard
156
+ client-server model, remains open:
157
+ Question 1. [Vad17, Open Problem 10.7]
158
+ Is there a computational task solvable by a single cura-
159
+ tor with computational differential privacy but is impossible to achieve with information-theoretic
160
+ differential privacy?2
161
+ There have been several positive and negative results towards resolving this question. In the
162
+ positive direction, it is known that in the multi-party setting, CDP is stronger than SDP [MMP+10,
163
+ MPRV09]. Roughly speaking, this is because secure multi-party computation enables many data cu-
164
+ rators to simulate acting as a single central curator, without compromising privacy. Still, the multi-
165
+ party setting seems very different than the single-curator (aka central) setting. Indeed, [MMP+10]
166
+ remark3 that their “strong separation between (information-theoretic) differential privacy and com-
167
+ putational differential privacy ... stands in sharp contrast with the client-server setting where ...
168
+ there are not even candidates for a separation.”
169
+ In the central setting, Bun, Chen, and Vadhan [BCV16] show there is a task for which there
170
+ is a CDP mechanism, but any SDP mechanism for this task must be inefficient (modulo certain
171
+ cryptographic assumptions). We stress that the task they consider does have an inefficient SDP
172
+ mechanism (with parameters that match their CDP mechanism), so it does not resolve Question 1.
173
+ While this may seem like a minor technical point, we emphasize that it is of crucial importance.
174
+ Perhaps the main practical motivation behind studying CDP is the hope that there are CDP mech-
175
+ anisms for natural tasks with parameters that beat the lower bounds against SDP mechanisms. But
176
+ if, as in the case of the result in [BCV16], there exists (even an inefficient) SDP mechanism match-
177
+ ing the parameters of the CDP mechanism, then clearly there is no hope of the CDP mechanism’s
178
+ parameters beating SDP lower bounds.
179
+ In the negative direction, Mironov, Pandey, Reingold, and Vadhan [MPRV09] (building on
180
+ Green and Tao [GT08], Tao and Ziegler [TZ08], and Reingold, Trevisan, Tulsiani, and Vad-
181
+ han [RTTV08]) show a “dense model theorem” for pairs of random variables with “pseudodensity”
182
+ with each other. [MPRV09] note that (roughly speaking) extending this dense model theorem to
183
+ handle multiple pairs of random variables would prove that any CDP mechanism could be converted
184
+ into an SDP mechanism; such an extension is still open [Vad17, Open Problem 10.8].
185
+ 1For the formal definitions of CDP and SDP, we refer the reader to Section 3.
186
+ 2We state this verbatim from [Vad17].
187
+ 3This remark is also quoted by Groce, Katz, and Yerukhimovich [GKY11].
188
+ 1
189
+
190
+ Groce, Katz, and Yerukhimovich [GKY11] show that CDP mechanisms for certain tasks where
191
+ the output is low-dimensional actually do imply SDP mechanisms. Many natural statistical tasks
192
+ fall into this category, and consequently, such tasks cannot separate CDP from SDP. This result
193
+ was further strengthened by [BCV16]. Furthermore, [GKY11] show that CDP mechanisms con-
194
+ structed in a black-box way from a variety of cryptographic objects, such as one-way functions,
195
+ random oracles, trapdoor permutations, and cryptographic hash functions, cannot separate CDP
196
+ from SDP.
197
+ In summary, there are at least two barriers to separate CDP from SDP:
198
+ 1. High-dimensionality: One needs to consider (perhaps non-natural) tasks with high dimen-
199
+ sional outputs;
200
+ 2. Exotic cryptography: One needs to use cryptography somewhat specially (perhaps either
201
+ an exotic primitive or in a non-black-box manner).
202
+ In light of these both positive and negative results as well as the lack of a candidate separation, it
203
+ was not even clear what the truth could be: is there any task for which there is a CDP mechanism
204
+ but not an SDP one?
205
+ Our Contributions.
206
+ We show, under plausible cryptographic hypotheses, that there are indeed
207
+ tasks for which there exist CDP mechanisms but no SDP mechanisms. This not only positively
208
+ answers Question 1 but also negatively answers the dense model extension question [Vad17, Open
209
+ Problem 10.8]. We state this result now informally and formalize it later in Section 2. We also delay
210
+ discussing our precise cryptographic assumptions to Section 2.5, where we discuss their plausibility
211
+ in detail.
212
+ Theorem 2. [Informal version of Theorem 5] Under cryptographic assumptions, there exists a task
213
+ for which there is a CDP mechanism but no SDP mechanism.
214
+ Let us take a step back to discuss the implications of Theorem 2. Although (as we will see in
215
+ a moment) our task is specifically constructed for the purpose of separating CDP and SDP, the
216
+ fact that we can separate them at all opens up a possibility that such a separation even holds for
217
+ some “natural” tasks. Indeed, some of the current lower bound techniques for SDP—such as the
218
+ ubiquitous “packing lower bounds”4 (see [HT10])—do not necessarily rule out CDP mechanisms.
219
+ It seems prudent to carefully reexamine the current lower bound techniques to see whether they
220
+ also apply to CDP. The ultimate hope for this program would be to employ CDP to overcome the
221
+ known SDP lower bounds for some more “natural” tasks. (Of course, such tasks would also give a
222
+ more “natural” separation of CDP and SDP.)
223
+ In fact, the technical approach we use in our construction already suggests a general approach
224
+ for constructing non-trivial CDP mechanisms that could apply to more tasks. We discuss this in
225
+ more detail in Section 2, but the idea is as follows. In order to show a task has a CDP mechanism,
226
+ first show there is a mechanism for that task that is “private” against a certain class of decision
227
+ tree adversaries. Then, second, use cryptographic assumptions to “lift” this into privacy against
228
+ computational adversaries.
229
+ 4Specifically, when the packing lower bound requires the use of super-polynomially many datasets, the correspond-
230
+ ing adversary does not necessarily run in polynomial time.
231
+ 2
232
+
233
+ Organization.
234
+ The rest of the paper is organized as follows.
235
+ Section 2 provides a high-level
236
+ overview of our techniques as well as a discussion of our cryptographic assumptions and their plau-
237
+ sibility. Section 3 contains the background material and Section 4 formally defines the problems.
238
+ We provide our CDP mechanism in Section 5, and prove lower bounds against SDP mechanisms in
239
+ Section 6. These two components are put together to prove the main result in Section 7. Finally,
240
+ we discuss the open problems and future directions in Section 8.
241
+ 2
242
+ Overview of the Results
243
+ We will next discuss the high-level overview of our techniques.
244
+ We will sometimes have to be
245
+ informal here, but all details are formalized later in the paper.
246
+ Let us quickly recall how the
247
+ “task” is defined5.
248
+ Following [GKY11, BCV16], a task is defined by an efficiently computable
249
+ utility function u that takes in an input dataset D and a response y such that u(D, y) = 1 if y is
250
+ considered “useful” for D and u(D, y) = 0 otherwise. A mechanism M is said to be α-useful for u
251
+ iff E[u(D, M(D))] ≥ α for all input datasets D. We remark that many well-studied problems—such
252
+ as linear queries with various error metrics—can be written in this form. We will refer to α as the
253
+ usefulness of the mechanism.
254
+ One of our main conceptual contributions is to define a class of tasks that seems to naturally
255
+ circumvent the two earlier-mentioned barriers—tasks where one needs to output a circuit.
256
+ 2.1
257
+ The d-Distance Problem
258
+ Before we detail why tasks that output a circuit might evade the two barriers, let us describe a
259
+ concrete example. We call the following the d-distance problem (where d ∈ N is a parameter):
260
+ ◮ Given: dataset D that consists of n bits
261
+ ◮ Output: circuit C mapping n bits to 1 bit
262
+ ◮ Utility: C is considered useful6 if it outputs
263
+ ⊲ 1 on D, and
264
+ ⊲ 0 on all points at distance greater than d from D.
265
+ Informally, this problem asks to output a circuit that checks if its input is “close” to D. Looking
266
+ ahead, we will ultimately separate CDP from SDP under cryptographic assumptions by considering
267
+ a version of this problem where we only care about datasets in a cryptographically special set.
268
+ We now revisit the two barriers and discuss how the distance problem might circumvent them.
269
+ 1. High-dimensionality: The output of this task is a circuit, which is high-dimensional.
270
+ 2. Exotic cryptography: Because the output of the task is a circuit, it lends itself to a powerful
271
+ class of cryptographic objects: circuit obfuscators [BGI+12]. Roughly speaking, circuit obfusca-
272
+ tors take as input a circuit C and output a scrambled, obfuscated circuit C′ that computes the
273
+ same function as C but which, ideally, has the property that “anything you could do with access
274
+ to the circuit C′, you could do with only black-box access to the function the circuit computes.”
275
+ Importantly, obfuscation is not in the list of primitives ruled out by the barrier in [GKY11].
276
+ 5Please refer to Section 3 for a more formal definition.
277
+ 6One might be concerned about whether this utility function is actually efficiently computable. We will address
278
+ this in Section 2.3 after we describe our final construction.
279
+ 3
280
+
281
+ 2.2
282
+ SDP Lower Bound
283
+ Our starting point for separating CDP from SDP is the d-distance problem described above. Indeed,
284
+ we show that there is no SDP mechanism for this problem for most settings of d.
285
+ Lemma 3. If 0 < d ≤ n.99, then there is no (ε, δ)-SDP mechanism for d-distance that is 0.01-useful
286
+ for ε = O(1) and δ negligible in n.
287
+ In fact, this lower bound is straightforward (Lemma 15) from the well-known blatant non-privacy
288
+ notion (see, e.g., [De12]): no DP algorithm can output a dataset that is (with large probability)
289
+ close to the input dataset. Crucially, our lower bounds are non-constructive, and do not yield an
290
+ efficient adversary (which would imply a similar lower bound against CDP mechanisms). Thus, to
291
+ separate CDP from SDP it suffices to come up with a CDP mechanism M for, say, n.99-distance.
292
+ 2.3
293
+ A CDP Mechanism
294
+ One of our main ideas to help construct a CDP mechanism M is to use obfuscation. In particular,
295
+ we will consider mechanisms where the returned circuit is obfuscated. Recall that in order to prove
296
+ a mechanism M that outputs a circuit C is CDP, one needs to argue that no efficient adversary
297
+ that gets C as input can break the privacy guarantee. By considering mechanisms that return
298
+ obfuscated circuits, we can drastically simplify the type of adversaries we need to prove privacy
299
+ against.
300
+ Instead of proving privacy against adversaries that see the circuit C (i.e., white-box
301
+ setting), sufficiently strong obfuscation means we only need to prove privacy against decision tree
302
+ adversaries that can query the function computed by the circuit (i.e., black-box setting). In other
303
+ words, if we have a mechanism that satisfies DP against black-box adversaries (decision trees) with
304
+ a polynomial number of queries, we can then hope to use sufficiently strong obfuscation to “lift”
305
+ this into a mechanism that is secure against (white-box) computational adversaries with polynomial
306
+ running time.
307
+ Of course, one needs to be careful about whether such “sufficiently strong obfuscation” is even
308
+ possible, but, putting that aside for the moment, the question of whether there is a CDP mechanism
309
+ for n.99-distance (Question 4 below) appears to reduce to whether there is a mechanism for n.99-
310
+ distance that is DP against query (a.k.a. decision-tree) adversaries.
311
+ Question 4. Let ε = O(1) and 0 ≤ d ≤ n.99. Does there exist an ε-CDP mechanism for d-distance
312
+ with constant usefulness?
313
+ While we do not resolve Question 4, we (roughly speaking) show that there is a mechanism that
314
+ is DP against non-adaptive decision tree adversaries, whose queries are fixed a priori. It turns out
315
+ a relatively simple mechanism based on randomized response [War65] works for these less powerful
316
+ adversaries.
317
+ From Non-Adaptive Lower Bound to Computational Lower Bound.
318
+ This switch from
319
+ the usual adaptive query adversaries to non-adaptive query adversaries comes at a price however. It
320
+ is not clear how to use obfuscation to lift a mechanism that is private against non-adaptive queries
321
+ into one that is private against computational adversaries. Indeed, a polynomial-time algorithm
322
+ with even black-box access to a function seems to be an inherently adaptive adversary!
323
+ Surprisingly, we manage to get around this by using another cryptographic object introduced
324
+ by Bitansky, Kalai, and Paneth [BKP18]: collision-resistant keyless hash functions. Informally
325
+ 4
326
+
327
+ speaking, a hash function being collision-resistant and keyless means that “any efficient adversary
328
+ can only generate a number of hash collisions that is at most polynomially larger than the advice
329
+ the adversary gets.”
330
+ We then modify the d-distance problem to only consider datasets that hash to, say, the all
331
+ zeroes string. Formally, zero hash d-distance is the following problem. Let R ⊆ {0, 1}n be the set
332
+ of strings that hash to the all zeroes string.
333
+ ◮ Given: dataset D that consists of n bits
334
+ ◮ Output: circuit C mapping n bits to 1 bit
335
+ ◮ Utility: C is considered useful if D /∈ R or both of the following hold:
336
+ ⊲ it outputs 1 on D
337
+ ⊲ it outputs 0 on all points in R at distance greater than d from D
338
+ In other words, the utility function now completely ignores all points outside of R.
339
+ The high-level intuition behind this change is the following:
340
+ 1. Our CDP mechanism can output a circuit C such that the only inputs where C(x) reveals
341
+ information are those x in the set R (i.e., that hash to zero).
342
+ 2. Any polynomial-time adversary A can only generate fixed polynomial number of elements of
343
+ R by the collision-resistance property of the hash function.
344
+ 3. Combining the above effectively makes the inputs A can query C on “non-adaptive”.
345
+ Finally, in order to “lift” the query separation into the computational realm we use another cryp-
346
+ tographic tool: differing-inputs obfuscation (diO) [BGI+01, BGI+12, ABG+13]. Roughly speaking,
347
+ diO is an obfuscator with the following guarantee: if any efficient adversary can distinguish the
348
+ obfuscation of two circuits C1 and C2, then an efficient adversary can find an input x on which
349
+ C1(x) ̸= C2(x). In particular, the assumption we use is even weaker than public-coin diO [IPS15],
350
+ which is already considered to more plausible than general diO.7
351
+ In summary, diO allows us to reduce computational adversaries to adaptive query adversaries
352
+ and collision-resistant keyless hash functions allows us to reduce adaptive query adversaries to
353
+ non-adaptive query adversaries. Interestingly, to the best of our knowledge, this is the first time
354
+ collision-resistant keyless hash functions are being used together with any obfuscation assumption.
355
+ Making the Utility Function Efficiently Computable.
356
+ Observant readers may have already
357
+ noticed an issue: utility functions that we have considered so far are not necessarily efficiently
358
+ computable. Specifically, a trivial way to implement the utility function would be to enumerate all
359
+ points at distance at least d, feed it into the circuit, and check that the output is as expected; this
360
+ would take 2nΩ(1) time.
361
+ To overcome the above problem, we restrict circuits to only those that are relatively simple, so
362
+ that there is a small “witness” w that certifies that the circuit outputs zero at all points that are
363
+ d-far from D. A naive idea is then to let the CDP mechanism output the circuit C together with
364
+ such a witness w. The utility function can then just efficiently check that w is a valid witness (and
365
+ that C(D) = 0 or x ∈ R). This makes the utility function efficient but unfortunately compromises
366
+ privacy because the witness w itself can leak additional information. To avoid this, we instead use
367
+ non-interactive witness indistinguishable (NIWI) proofs (e.g., [BOV07]). Roughly speaking, this
368
+ allows us to produce a proof π from w (and C and diO), which does not leak any information about
369
+ w (against computationally bounded adversaries), but at the same time still allows us to verify
370
+ 7See Assumption 22 for formal statement of the assumption and Appendix A for comparison with other diO
371
+ assumptions in literature.
372
+ 5
373
+
374
+ that the underlying witness w is valid. The former is sufficient for CDP, while the latter ensures
375
+ that the utility function can be computed efficiently.
376
+ This completes the high-level overview of the constructed task and our CDP mechanism. The
377
+ cryptographic primitives needed for our mechanism are formalized in Assumptions 18, 22 and 26.
378
+ 2.4
379
+ Final Steps
380
+ Finally, we remark that since our problem is now not exactly the original d-distance problem
381
+ anymore, as the utility guarantees are only now meaningful for datasets in R. This means that
382
+ we cannot use the lower bound in Lemma 3 for the d-distance problem directly. Fortunately, we
383
+ can still adapt its proof—a “packing-style” lower bound on each coordinate—to one which applies
384
+ a packing-style argument on each block of coordinates instead. With this, we can prove the lower
385
+ bound for zero hash d-distance as long as the set R has sufficiently large density (≈ 1/n−o(log n)).
386
+ Putting all the ingredients together, we arrive at the following8:
387
+ Theorem 5 (Main Result). Under Assumptions 18, 22 and 26, for any constant εCDP > 0, there
388
+ exists an ensemble u = {un}n∈N of polynomial time computable utility functions such that
389
+ ◮ There is an εCDP-CDP mechanism that is (1 − on(1))-useful for u.
390
+ ◮ For any constant εSDP > 0, there is no εSDP-SDP mechanism that is 0.01-useful for u.
391
+ 2.5
392
+ On the Plausiblility of the Cryptographic Assumptions
393
+ We now discuss the plausiblility of the three cryptographic assumptions we use for our result:
394
+ (i) NIWI: Non-interactive Witness Indistinguishable Proofs (formally, Assumption 26)
395
+ (ii) CRKHF: Collision-Resistant Keyless Hash Functions (formally, Assumption 18)
396
+ (iii) diO-for-pcS: Differing-Inputs Obfuscation for Public-coin Samplers (formally, Assumption 22)
397
+ Regarding (i), NIWI.
398
+ Bitansky and Paneth [BP15a] show that NIWIs exist assuming one-
399
+ way permutations exist and indistinguishability obfuscation (iO) exists. Recently, Jain, Lin, and
400
+ Sahai [JLS21] show that the existence of iO follows from well-founded assumptions; consequently,
401
+ NIWIs exist based on widely-believed assumptions. (We note that other previous works have also
402
+ constructed NIWIs based on other more specific assumptions [BOV07, GOS12].)
403
+ Regarding (ii), CRKHF.
404
+ Bitansky, Kalai, and Paneth [BKP18] defined CRKHFs to model the
405
+ properties of existing hash functions like SHA-2 used in practice. They suggest several candidates
406
+ for CRKHFs, such as hash functions based on AES and Goldreich’s one-way functions. They also
407
+ note that CRKHFs exist in the Random Oracle model, as a random function is a CRKHF. Still,
408
+ it is an open question to base the security of a CRKHF on a standard cryptographic assumption.
409
+ Part of the difficulty of doing this, as [BKP18] describe, is that most cryptographic assumptions
410
+ involve some sort of structure that is useful for constructing cryptographic objects. In contrast,
411
+ the goal of a CRKHF is to have no structure at all. In summary, given the various CRKHF candi-
412
+ dates, the existence in the Random Oracle model, and the fact that CRKHFs exist “in practice,”
413
+ this assumption is quite plausible. For our specific construction, we need a different hash length
414
+ 8We remark that εSDP-SDP mechanism here refers to an ensemble of mechanisms {Mn} which are (εSDP, negl)-SDP.
415
+ (See Definition 7.)
416
+ 6
417
+
418
+ (equivalently, different compression rate) than that used in [BKP18]; please refer to the discussion
419
+ preceding Assumption 18 for the parameters and justification.
420
+ Finally, we remark that, even though the existence of CRKHFs is not known to reduce to any
421
+ “well-founded” assumption, even refuting their existence would answer a longstanding question in
422
+ cryptography: giving non-contrived separations between the Random Oracle model [BR93] and the
423
+ standard model. In the words of Bitansky, Kalai, and Paneth [BKP18]
424
+ “Any attack on the multi-collision resistance of a [keyless] cryptographic hash function
425
+ would constitute a strong and natural separation between the hash and random oracles.
426
+ For several cryptographic hash functions used in practice, the only known separations
427
+ from random oracles are highly contrived [CGH04].”
428
+ Regarding (iii), diO-for-pcS.
429
+ One can think of diO [BGI+01, BGI+12] as an “extractable”
430
+ strengthening of iO. While iO has now become a widely-believed assumption [JLS21], the exis-
431
+ tence of diO is controversial. Several papers (e.g., [BP15b, GGHW17, BSW16]) cast doubt on the
432
+ existence of diO, especially in the case where an arbitrary auxillary input is allowed; we stress that
433
+ all the negative results for diO hold for contrived auxillary inputs and/or distributions. On the
434
+ positive side, [BCP14] show that diO reduces to iO in special cases, such as when the number of
435
+ differing-inputs is bounded by a polynomial. More related to our result, [IPS15] gives a definition
436
+ of public-coin diO that avoids the difficulties presented by earlier negative results regarding auxil-
437
+ iary inputs, although [BP15b] presented some evidence against this definition in special cases. Our
438
+ specific assumption of diO-for-pcS is in fact weaker than the assumption of public-coin diO. In the
439
+ definition of public-coin diO, as in [IPS15], we start with any public-coin sampler (pcS), for which it
440
+ is hard to find an input on which two circuits differ, even given the knowledge of all the randomness
441
+ that underlies the circuits. The security of the obfuscation is required to hold even against adver-
442
+ saries that know all the randomness that underlies the generation of the two circuits. However,
443
+ in our definition, the security of the obfuscation is required to hold only against adversaries that
444
+ observes a single obfuscated circuit, which makes the assumption weaker. See Appendix A for a
445
+ more detailed discussion on comparison of this assumption with other diO assumptions in literature.
446
+ Finally, we only use the existence of diO-for-pcS for a simple circuit family for our result, so even if
447
+ general purpose diO-for-pcS does not exist, we think it is plausible that diO-for-pcS exists for the
448
+ specific family of circuits we need for our result. (See Assumption 22 for the exact pcS family for
449
+ which we require a diO.)
450
+ Final thoughts on our assumptions.
451
+ In conclusion, we view each of our three assumptions
452
+ as plausible. Moreover, each of assumptions has at least some evidence that is hard to refute:
453
+ NIWIs exist based on a widely-believed assumption, refuting CRKHFs would require giving the
454
+ first non-contrived separation between the standard and the Random Oracle model, and despite
455
+ many attempts (e.g., [BP15b, GGHW17, BSW16]) to refute diO, the question is still open, especially
456
+ for the particular diO-for-pcS version.
457
+ 3
458
+ Preliminaries
459
+ A function g : N → R≥0 is said to be negligible if g(n) = n−ω(1). Let PPT be an abbreviation for
460
+ probabilistic polynomial-time Turing machine.
461
+ 7
462
+
463
+ For x ∈ {0, 1}n and r ∈ N, we use Br(x) to denote the (Hamming) ball of radius r around x,
464
+ i.e., {z ∈ {0, 1}n | ∥x − z∥1 ≤ r}. Furthermore, we use diam(S) for a set S ⊆ {0, 1}n to denote the
465
+ (Hamming) diameter of S, i.e., maxx,x′ ∈S ∥x − x′∥1.
466
+ 3.1
467
+ Dataset and Adjacency
468
+ For a domain X, we view a dataset D as a histogram over the domain X, i.e., D ∈ ZX
469
+ ≥0 where Dx
470
+ denotes the number of times x ∈ X appears in the dataset. The size of the dataset is defined as
471
+ ∥D∥1 := �
472
+ x∈X Dx. We write X m as a shorthand for the set of all datasets of size m, and X ∗ for
473
+ the set of all datasets over domain X. Two datasets are adjacent iff ∥D − D′∥1 = 1, i.e., one of the
474
+ datasets is a result of adding or removing a single row from the other dataset.
475
+ 3.2
476
+ Mechanism, Utility Function, and Usefulness
477
+ A mechanism M is a randomized algorithm that takes in a dataset D ∈ X ∗ and outputs an element
478
+ from a set Y. The utility of a mechanism is measured by a utility function u, which is a polynomial-
479
+ time deterministic algorithm that takes in a dataset D ∈ X ∗ together with a response y ∈ Y and
480
+ outputs 0 or 1 (whether the response is good for the dataset). We say that the mechanism M is
481
+ α-useful for utility u iff Pr[u(D, M(D)) = 1] ≥ α.
482
+ Below, we will often discuss an ensemble M = {Mn}n∈N of mechanisms where9 Mn : X ∗
483
+ n → Yn.
484
+ We say that an ensemble of mechanisms is efficient if Mn on input D ∈ X m
485
+ n runs in time poly(n, m).
486
+ For an ensemble u = {un}n∈N of utility functions and α = {αn ∈ [0, 1]}n∈N, we say that M is α-
487
+ useful with respect to u iff Mn is αn-useful with respect to un for all n ∈ N.
488
+ For brevity, we will sometimes refer to “ensemble of mechanisms” simply as “mechanism” and
489
+ “ensemble of utility functions” simply as “utility function” when there is no ambiguity.
490
+ 3.3
491
+ Notions of Differential Privacy
492
+ We now define the notions of DP that will be used throughout the paper.
493
+ (Statistical) Differential Privacy.
494
+ The standard (statistical) notion of DP can be defined in
495
+ terms of the following notion of indistinguishability.
496
+ Definition 6 (Statistical Indistinguishability). Distributions P, Q are said to be (ε, δ)-indistinguishable,
497
+ denoted P ≈ε,δ Q, if for all events (measurable sets) E, it holds that
498
+ Pr
499
+ X∼P[X ∈ E] ≤ eε · Pr
500
+ X∼Q[X ∈ E] + δ,
501
+ and
502
+ Pr
503
+ X∼Q[X ∈ E] ≤ eε · Pr
504
+ X∼P[X ∈ E] + δ.
505
+ For simplicity, we use ≈ε to denote ≈ε,0.
506
+ Definition 7 (Statistical Differential Privacy (SDP) [DMNS06, DKM+06]). For ε, δ > 0, a mecha-
507
+ nism M is said to be (ε, δ)-SDP if and only if for every pair D, D′ of adjacent datasets, we have that
508
+ M(D) ≈ε,δ M(D′). We say that an ensemble M = {Mn}n∈N is ε-SDP for a sequence ε = {εn}n∈N
509
+ if there exists a negligible sequence {δn}n∈N such that Mn is (εn, δn)-SDP for all n ∈ N.
510
+ We note that the above notation, which omits explicit δ for an ensemble of mechanisms, was also
511
+ used by [BCV16].
512
+ 9It is always implicitly assumed that Xn, Yn are of size poly(n).
513
+ 8
514
+
515
+ Computational Differential Privacy.
516
+ The notion of computational DP relaxes the notion
517
+ of indistinguishability to a computational version, where the privacy holds only with respect to
518
+ computationally bounded adversaries.
519
+ Definition 8 (Computational Indistinguishability). Two ensembles of distributions P = {Pn}n∈N
520
+ and Q = {Qn}n∈N, where Pn and Qn are supported over {0, 1}p(n) for some polynomial p(·), are
521
+ said to be ε-computationally-indistinguishable for a sequence ε = {εn}n∈N, denoted P ≈c
522
+ ε Q, if there
523
+ exists a negligible function negl(·) such that for any PPT adversary A, it holds that
524
+ Pr
525
+ X∼Pn[A(X) = 1] ≤ eεn ·
526
+ Pr
527
+ X∼Qn[A(X) = 1] + negl(n), and
528
+ Pr
529
+ X∼Qn[A(X) = 1] ≤ eεn ·
530
+ Pr
531
+ X∼Pn[A(X) = 1] + negl(n).
532
+ In the special case of ε = 0, we suppress the subscript and simply write P ≈c Q.
533
+ Throughout, when we refer to a sequence {(Dn, D′
534
+ n)}n∈N of adjacent datasets, it is always assumed
535
+ that Dn ∈ X mn
536
+ n
537
+ , D′
538
+ n ∈ X m′
539
+ n
540
+ n
541
+ are of sizes mn, m′
542
+ n = poly(n).
543
+ Definition 9 (Computational Differential Privacy (CDP) [MPRV09]). An ensemble M = {Mn}n∈N
544
+ of mechanisms is said to be ε-CDP for a sequence ε = {εn}n∈N, if for any sequence {(Dn, D′
545
+ n)}n∈N
546
+ of adjacent datasets, it holds that {Mn(Dn)}n∈N ≈c
547
+ εn {Mn(D′
548
+ n)}n∈N.
549
+ This definition is often referred to as indistinguishability-based CDP (IND-CDP) in previous
550
+ works [MPRV09, GKY11, BCV16]. Since we only use this notion for our main result, we refer to
551
+ it simply as CDP. The other definition of CDP used in previous works is simulation-based:
552
+ Definition 10 (SIM-CDP [MPRV09]). An ensemble M = (Mn)n∈N of mechanisms is said to be ε-
553
+ SIM-CDP if there exists an (εn, 0)-SDP ensemble {M′
554
+ n}n∈N of mechanisms such that for any sequence
555
+ {Dn ∈ X ∗
556
+ n}n∈N of datasets, with size of Dn being at most poly(n), it holds that Mn(Dn) ≈c M′
557
+ n(Dn).
558
+ It should be noted that SIM-CDP cannot be used for the separation we are looking for. Specif-
559
+ ically, if {Mn}n∈N is ε-SIM-CDP, we may use {M′
560
+ n}n∈N as our ε-SDP mechanism. Since the utility
561
+ function runs in polynomial time, it follows immediately that, if {Mn}n∈N is α-useful, then {M′
562
+ n}n∈N
563
+ is also (α − o(1))-useful. Due to this, we will not consider SIM-CDP again in this paper.
564
+ Calculus of ≈ and ≈c.
565
+ The following properties are well-known.
566
+ Fact 11. The notions of (ε, δ)-indistinguishability and ε-computational-indistinguishability satisfy:
567
+ ◮ Basic Composition: If P0 ≈ε,δ P1 and P1 ≈ε′,δ′ P2, then P0 ≈ε+ε′,δ+δ′ P2.
568
+ Similarly, if
569
+ P0 ≈c
570
+ ε P1 and P1 ≈c
571
+ ε′ P2, then P0 ≈c
572
+ ε+ε′ P2.
573
+ ◮ Post-processing: If P ≈ε,δ Q, then for all (randomized) functions f, it holds that f(P) ≈ε,δ
574
+ f(Q). Similarly, if P ≈c
575
+ ε Q, then for all PPT algorithms A, it holds that A(P) ≈c
576
+ ε A(Q).
577
+ 4
578
+ Low Diameter Set Problem and Nearby Point Problem
579
+ In this section, we introduce the problems that we will use in our separation. Before that, we will
580
+ describe a simplifying assumption that we can make about the inputs.
581
+ 9
582
+
583
+ 4.1
584
+ Simplification of Input Representation
585
+ Recall that so far a dataset may contain multiple copies of an element. Below, however, it will be
586
+ more convenient to only discuss the case where each element appears only once, i.e., D ∈ {0, 1}X .
587
+ This is sufficient since if we have a utility function u : {0, 1}X × Y → {0, 1} defined only on
588
+ D ∈ {0, 1}X , we can easily define the utility function u : NX × Y → {0, 1} by
589
+ u(D, r) =
590
+
591
+ u(D, r)
592
+ if D ∈ {0, 1}X ,
593
+ 1
594
+ otherwise.
595
+ In other words, the utility function considers any response good for datasets with repetition. Clearly,
596
+ if u is efficiently computable, then so is u. Furthermore, suppose that we have an ε-CDP mechanism
597
+ M = {Mn}n∈N for u = {un}n∈N. For every dataset D, let D be defined by Di = min
598
+
599
+ Di, 1
600
+
601
+ .
602
+ Then, we may define M =
603
+
604
+ Mn
605
+
606
+ n∈N by M(D) = M(D). It is simple to check that M remains
607
+ ε-CDP. Furthermore, if M is α-useful for u, then M remains α-useful for u.
608
+ Finally, note that a lower bound for DP algorithms restricted to non-repeated datasets trivially
609
+ implies a lower bound against all datasets.
610
+ Due to this, we will henceforth focus our attention only on the datasets D ∈ {0, 1}X . Further-
611
+ more, throughout the remainder of this paper, we will always pick Xn = [n]. This further simplifies
612
+ the input representation to be just a bit vector x ∈ {0, 1}n. We will define an input of our problem
613
+ in this way. Furthermore, we will henceforth use x instead of D to denote the input dataset.
614
+ 4.2
615
+ Nearby Point Problem
616
+ We will start by defining our first problem, which asks to output a point that is close to the input
617
+ point if the latter belongs to some set R. As we noted in the introduction, when R is the set
618
+ of all points (i.e., Rn = {0, 1}n), this is exactly the same as the problem considered in blatant
619
+ non-privacy [DN03, DMT07]. As we will see later, the presence of the set R is due to our use of
620
+ hashing, which is required in our proof for the CDP mechanism.
621
+ Definition 12 (τ-Nearby R-Point Problem). The nearby point problem parameterized by sequences
622
+ {τn ∈ N}n∈N and {Rn ⊆ {0, 1}n}n∈N is denoted by NBPτ,R.
623
+ For input x ∈ {0, 1}n and output
624
+ y ∈ Yn = {0, 1}n, the utility is defined as:
625
+ uNBP
626
+ τn,Rn(x, y) := 1 {∥x − y∥1 ≤ τn or x /∈ Rn}
627
+ For brevity, we will assume throughout that Rn is efficiently recognizable and henceforth we do
628
+ not state this explicitly. Note that this assumption implies that the utility function defined above
629
+ is efficiently computable. The nearby point problem will be primarily used for proving the lower
630
+ bounds against SDP.
631
+ 4.3
632
+ Verifiable Low Diameter Set Problem
633
+ Next, we define circuit-based tasks for which we will give CDP mechanisms. To do so, we need to
634
+ first define a “τ-diameter verifier”.
635
+ Definition 13 (τ-Diameter Verifier). For a sequence τ = {τn}n∈N of integers, we say that an
636
+ efficiently computable (deterministic) verifier V = {Vn}n∈N is a τ-diameter verifier for circuits of
637
+ size s(n) if it takes as input a circuit C : {0, 1}n → {0, 1} of (polynomial) size s(n) and a proof π
638
+ of size poly(n), and outputs Vn(C, π) = 1 only if diam(C−1(1)) ≤ τn.
639
+ 10
640
+
641
+ We can now define the (verifiable) low diameter set problem as follows:
642
+ Definition 14 (Verifiable τ-Diameter R-Set Problem). The verifiable low diameter set problem
643
+ parameterized by sequences τ = {τn}n∈N, R = {Rn ⊆ {0, 1}n}n∈N, and τ-diameter verifier V =
644
+ {Vn}n∈N is denoted by VLDSτ,R,V . The input, output, and utility are defined as follows:
645
+ ◮ Input: x ∈ {0, 1}n.
646
+ ◮ Output: circuit C and a proof π, both of size poly(n).
647
+ ◮ Utility: uVLDS
648
+ τn,Rn,Vn(x, (C, π)) := 1 {C(x) = 1 or x /∈ Rn} and 1 {Vn(C, π) = 1}.
649
+ For convenience, we also define the following utility function
650
+ ueval
651
+ R (x, C) := 1 {C(x) = 1 or x /∈ R} .
652
+ Note that this does not correspond to a hard task, because a circuit that always outputs one is
653
+ 1-useful. Nonetheless, it will be convenient to state usefulness of some intermediate algorithms via
654
+ this utility function.
655
+ 4.4
656
+ From Low Diameter Set Problem to Nearby Point Problem
657
+ Below we provide a simple observation that reduces the task of proving an SDP lower bound for
658
+ the verifiable low diameter set problem to that of the nearby point problem. (Note here that the
659
+ SDP mechanisms considered below can be computationally inefficient.)
660
+ Lemma 15. If there is an (ε, δ)-SDP α-useful mechanism for the VLDSτ,R,V problem, then there
661
+ is an (ε, δ)-SDP α-useful mechanism for the NBPτ,R problem.
662
+ Proof. Let M be an (ε, δ)-SDP α-useful mechansim for the VLDSτ,R,V problem. We will construct
663
+ an (ε, δ)-SDP α-useful mechanism M′ for the NBPτ,R problem.
664
+ The mechanism M′
665
+ n on input dataset x ∈ {0, 1}n works as follows. First, let (C, π) ← Mn(x).
666
+ If Vn(C, π) = 1, then output the lexicographically first element of C−1(1) (else, output 0n). This
667
+ completes our description of M′.
668
+ Since M is (ε, δ)-SDP, we have that M′ is also (ε, δ)-SDP by post-processing. It remains to
669
+ show that M′ is α-useful. Fix some input x ∈ {0, 1}n. If x /∈ Rn, then any output satisfies utility.
670
+ Thus, it suffices to consider the case where x ∈ Rn. With probability α, we have that Vn(C, π) = 1
671
+ (which implies that C−1(1) has diameter at most τn), and x ∈ C−1(1). Consequently, the distance
672
+ between x and the lexicographically first element of C−1(1) is at most τn. So with probability at
673
+ least α, the output of M′ is useful for x, as desired.
674
+ 5
675
+ CDP Mechanism for Verifiable Low Diameter Set Problem
676
+ In this section we build a CDP mechanism for the verifiable low diameter set problem. We establish
677
+ the following result:
678
+ Theorem 16. Suppose that Assumptions 18, 22 and 26 hold. Then, for all constant εCDP > 0 and
679
+ τ =
680
+
681
+ τn = n0.9�
682
+ n∈N, there exists a τ-diameter verifier V and a sequence R = {Rn}n∈N of sets of
683
+ sizes |Rn| ≥ 2n/no(log n), such that there exists an εCDP-CDP mechanism that is (1 − on(1))-useful
684
+ for uVLDS
685
+ τ,R,V .
686
+ As discussed in the overview, we first build a mechanism that is CDP but without verifiability
687
+ using collision-resistant keyless hash functions and differing-inputs obfuscators (Section 5.1). We
688
+ then turn it into a verifiable one using non-interactive witness indistinguishable proofs (Section 5.2).
689
+ 11
690
+
691
+ 5.1
692
+ CDP Mechanism without Verifiability
693
+ In this section, we construct our first CDP mechanism (Algorithm 3). We depart from the overview
694
+ in Section 2 slightly and do not prove a non-adaptive query lower bound explicitly. Instead, we
695
+ directly show in Section 5.1.2 how to sample the appropriate differing-inputs circuit family. This
696
+ can be then easily turned into our CDP mechanism via diO in Section 5.1.3.
697
+ 5.1.1
698
+ Additional Preliminaries: Cryptographic Primitives
699
+ Throughout this section, we will repeatedly use the so-called randomized response (RR) mecha-
700
+ nism [War65]. Specifically, RRε is an algorithm that takes in x ∈ {0, 1}n and outputs ˜x ∈ {0, 1}n,
701
+ where ˜xi = xi with probability
702
+
703
+ 1+eε independently for each i ∈ [n]. It is well-known (and very
704
+ simple to verify) that RRε is ε-SDP.
705
+ Collision-Resistant Keyless Hash Functions.
706
+ In our construction, we will use the Collision-
707
+ Resistant Keyless Hash Functions (CRKHFs) [BKP18]. The formal definition is as given below.
708
+ Definition 17 (Collision-Resistant Keyless Hash Functions [BKP18]). A sequence of hash func-
709
+ tions
710
+
711
+ Hn : {0, 1}n → {0, 1}γ(n)�
712
+ n∈N is K-collision resistant for advice length ζ for sequences K =
713
+ {Kn}n∈N, ζ = {ζn}n∈N if, for any PPT A and a sequence {zn}n∈N of advices where |zn| = ζn, we
714
+ must have
715
+ Pr
716
+ (Y1,...,YKn)←A(1n;zn) [Y1, . . . , YKn are distinct and Hn(Y1) = · · · = Hn(YKn)] ≤ negl(n).
717
+ We will skip the subscript n whenever it is clear from context.
718
+ In [BKP18], the hash value length γ(n) is assumed to be either linear, i.e., γ(n) = Ω(n), or
719
+ polynomial, i.e., γ(n) = nΘ(1). However, we need a collision-resistant hash function with a much
720
+ smaller γ(n), namely O(log2 n).
721
+ We remark that this is still very much plausible: as long as
722
+ γ(n) is ω(log n), the “guess-and-check” algorithm will only produce a collision with only negligible
723
+ probability. A more precise statement of our assumption is stated below.
724
+ Assumption 18. There is an efficiently computable sequence H = {Hn}n∈N of hash functions with
725
+ hash value length γ(n) = o(log2 n) such that, for any constant c1 > 0, there exists a constant c2 > 0
726
+ such that the hash function sequence is K-collision resistant for advice length ζ where K(n) = nc2
727
+ and ζ(n) = nc1.
728
+ We remark that, for the existence of CDP mechanism (shown in this section), we will only use
729
+ the multi-collision-resistance without relying on the assumption on the value of γ. The latter is
730
+ only used to show that no SDP mechanism exists for the problem (Section 7).
731
+ Differing-Inputs Obfuscators for Public-Coin Samplers.
732
+ For any two circuits C0 and C1,
733
+ a differing-inputs obfuscator diO [BGI+12] guarantees that the non-existence of an efficient ad-
734
+ versary that can find an input on which C0 and C1 differ implies that diO(C0) and diO(C1) are
735
+ computationally indistinguishable. For our application, it even suffices to assume a weaker notion,
736
+ namely that of differing-inputs obfuscator for public-coin samplers, as defined below.
737
+ 12
738
+
739
+ Definition 19 (Public-Coin Differing-Inputs Circuit Sampler). An efficient non-uniform sampling
740
+ algorithm Sampler = {Samplern} is a public-coin differing-inputs sampler for the parameterized
741
+ collection C = {Cn} of circuits if the output of Samplern is distributed over Cn × Cn and for every
742
+ efficient non-uniform algorithm A = {An}, there exists a negligible function negl(·) such that for
743
+ all n ∈ N:
744
+ Pr
745
+ θ [C0(y) ̸= C1(y) : (C0, C1) ← Samplern(θ), y ← An(θ)] ≤ negl(n).
746
+ Here, Samplern is a deterministic algorithm and the only source of randomness is the seed θ.
747
+ Definition 20 (Differing-Inputs Obfuscator for Public-Coin Samplers (cf. [IPS15])). A uniform
748
+ PPT diO is a differing-inputs obfuscator for public-coin samplers for the parameterized circuit
749
+ family C = {Cn} if the following conditions are satisfied:
750
+ ◮ Correctness: For all n ∈ N, for all C ∈ Cn, for all inputs y, we have that
751
+ Pr[C′(y) = C(y) : C′ ← diO(1n, C)] = 1.
752
+ ◮ Polynomial slowdown: There exists a universal polynomial p(·) such that for all C ∈ Cn, it
753
+ holds that
754
+ Pr[|C′| ≤ p(|C|) : C′ ← diO(1n, C)] = 1.
755
+ ◮ Differing-inputs: For every public-coin differing inputs sampler Sampler = {Samplern} for
756
+ C, and every (not necessarily uniform) PPT distinguisher D = {Dn}, there exists a negligible
757
+ function negl such that the following holds for all n ∈ N: For (C0, C1) ← Samplern(θ)
758
+ | Pr
759
+ θ [Dn(diO(1n, C0)) = 1] − Pr
760
+ θ [Dn(diO(1n, C1)) = 1]| ≤ negl(n).
761
+ We note that the notion of diO-for-pcS is in fact weaker than the notion of general public-coin diO
762
+ as given by [IPS15]. We elaborate on this comparison in Appendix A. Whenever n is clear from
763
+ context, we use diO(C) to denote diO(1n, C) for simplicity. When we want to be explicit about the
764
+ randomness ρ (of poly(n) bit length) used by diO we will denote it as diOρ(C).
765
+ We only need the existence of a differing-inputs obfuscator for a specific family of circuits. This cir-
766
+ cuit family will be defined later and therefore we defer formalizing our assumption to Section 5.1.3.
767
+ 5.1.2
768
+ Public-Coin Differing-Inputs Circuits from CRKHFs
769
+ The first step of our proof is to construct a differing-inputs circuit family based on CRKHFs. Our
770
+ sampler is described in Algorithm 1.
771
+ We next prove that the above sampler is a public-coin differing-inputs sampler, which means
772
+ that any efficient adversary, even with the knowledge of ˜x (which is the only source of randomness),
773
+ cannot find an input on which C0 and C1 differ. The proof starts by noticing that any input that
774
+ differentiates C0, C1 must, by definition of the circuits, have hash value υn. Therefore, if there were
775
+ an adversary that can find a differing input, then we could run it multiple times to get Y1, . . . , YK
776
+ that have the same hash value. (See Algorithm 2 below.) However, our proof is not finished yet,
777
+ since it is possible that Y1, . . . , YK are not distinct. Indeed, the crux of the construction is that,
778
+ due to how we select ˜x and define the circuits, a fixed Y will be a differing input with negligible
779
+ probability10. It follows that Y1, . . . , YK must be distinct w.h.p. This is formalized below.
780
+ 10It is also simple to see that this property suffices to prove a non-adaptive query lower bound as discussed in
781
+ Section 2.
782
+ 13
783
+
784
+ Algorithm 1 Differing-Inputs Circuit Family Sampler LDS-Samplern.
785
+ Parameters:
786
+ Adjacent datasets x, x′ ∈ {0, 1}n, hash value υn ∈ {0, 1}γ(n), privacy parameter
787
+ ε > 0, radius r, ˜r > 0.
788
+ Randomness: θ ∼ RRε(0n).
789
+ Output: Circuits C0, C1.
790
+ ˜x ← x ⊕ θ (bit-wise XOR; equivalent to RRε(x))
791
+ C0 ← circuit that takes in z and computes 1
792
+
793
+ z ∈ Br(x) ∩ B˜r(˜x) ∩ H−1
794
+ n (υn)
795
+
796
+ C1 ← circuit that takes in z and computes 1
797
+
798
+ z ∈ Br(x′) ∩ B˜r(˜x) ∩ H−1
799
+ n (υn)
800
+
801
+ return (C0, C1)
802
+ Lemma 21. Let H be as in Assumption 18. For any constant ε > 0, choosing r = 0.5n0.9 and
803
+ ˜r =
804
+ 1
805
+ 1+eε n + n0.6 makes LDS-Samplern (Algorithm 1) a public-coin differing-inputs sampler.
806
+ Proof. Suppose for the sake of contradiction that for some adjacent x, x′ ∈ {0, 1}n, there exists a
807
+ PPT ADI such that
808
+ Pr
809
+ θ [C0(y) ̸= C1(y) : (C0, C1) ← LDS-Samplern(θ), y ← ADI
810
+ n (θ)] ≥ n−c,
811
+ (1)
812
+ for some constant c > 0. Furthermore, let c1 be such that the total size of the descriptions of
813
+ ADI
814
+ n , LDS-Samplern is at most nc1. Finally, let c2 > 0 be as in Assumption 18 and K = nc2.
815
+ Algorithm 2 Collision-Resistant Hash Function Adversary ACRH
816
+ n
817
+ .
818
+ Parameter:
819
+ The target number of collisions K ∈ N, constant c > 0.
820
+ Advice:
821
+ Descriptions of ADI
822
+ n , LDS-Samplern.
823
+ Output: Y1, . . . , YK ∈ {0, 1}n or ⊥.
824
+ i ← 0
825
+ for j = 1, . . . , K · nc+1 do
826
+ θj ← RRε(0n)
827
+ (Cj
828
+ 0, Cj
829
+ 1) ← LDS-Samplern(θj)
830
+ yj ← ADI
831
+ n (θj)
832
+ if Cj
833
+ 0(yj) ̸= Cj
834
+ 1(yj) then
835
+ i ← i + 1
836
+ Yi ← yj
837
+ if i ≥ K then
838
+ break
839
+ if i < K then
840
+ return ⊥
841
+ else
842
+ return Y1, . . . , YK
843
+ Consider the adversary ACRH
844
+ n
845
+ for collision-resistant hash function described in Algorithm 2.
846
+ First, note that by (1) and a standard concentration inequality, the probability that ACRH
847
+ n
848
+ outputs
849
+ ⊥ is on(1). Furthermore, notice that C0, C1 can differ on y only if Hn(y) = υn, meaning that
850
+ Hn(Yi) = υn always. Therefore, it suffices for us to show that the probability that Y1, . . . , YK are
851
+ 14
852
+
853
+ distinct is 1 − on(1). By a union bound, we have that ACRH
854
+ n
855
+ violates the collision-resistance of H
856
+ as desired.
857
+ Thus, we are only left to show that Y1, . . . , YK are not distinct with probability o(1). To see
858
+ that this is the case, notice that
859
+ Pr[Y1, . . . , YK are not distinct] ≤
860
+
861
+ 1≤i1<i2≤K
862
+ Pr[Yi1 = Yi2].
863
+ (2)
864
+ Let us now bound Pr[Yi1 = Yi2] for a fixed pair i1 < i2. Suppose that we fix a value of Yi1 and
865
+ suppose that Yi1 is assigned at step j1 ∈ [1, . . . , K ·nc+1]. Conditioned on these, notice further that
866
+ Pr[Yi2 = Yi1] ≤ Pr[∃j > j1, yj = Yi1]
867
+ ≤ Pr[∃j > j1, Cj
868
+ 0(Yi1) ̸= Cj
869
+ 1(Yi1)]
870
+
871
+
872
+ j>j1
873
+ Pr[Cj
874
+ 0(Yi1) ̸= Cj
875
+ 1(Yi1)].
876
+ (3)
877
+ Now, let us bound the RHS probability for a fixed j > j1. To see this, first observe that Yi1 must
878
+ belong to the symmetric difference Br(x)△Br(x′); otherwise, we must have Cj1
879
+ 0 (Yi1) = Cj1
880
+ 1 (Yi1), a
881
+ contradiction to our definition of Yi1.
882
+ Now, let ˜xj denote the ˜x selected by LDS-Sampler when constructing Cj
883
+ 0, Cj
884
+ 1. We have
885
+ Pr[Cj
886
+ 0(Yi1) ̸= Cj
887
+ 1(Yi1)] ≤ Pr[Yi1 ∈ B˜r(˜xj)].
888
+ (4)
889
+ Let d := ∥Yi1 − x∥1 and ˜d := ∥Yi1 − ˜xj∥1. Since Yi1 ∈ Br(x)△Br(x′), it holds that d ∈ {r, r + 1}.
890
+ Thus, ˜d is distributed as Bin(d,
891
+
892
+ 1+eε ) + Bin(n − d,
893
+ 1
894
+ 1+eε ). We have E˜xj∼RRε(x) ˜d =
895
+ 1
896
+ 1+eε n + eε−1
897
+ eε+1d.
898
+ By Bernstein’s inequality,
899
+ Pr[ ˜d ≤ ˜r] ≤ exp
900
+
901
+
902
+ t2
903
+
904
+ (1+eε)2 n + 2
905
+ 3t
906
+
907
+ ≤ exp(−Ω(n0.8)),
908
+ where t = E˜xj∼RRε(x) ˜d − ˜r ≥ eε−1
909
+ eε+1(0.5n0.9 − 1) − n0.6. Plugging into (4), we have
910
+ Pr[Cj
911
+ 0(Yi1) ̸= Cj
912
+ 1(Yi1)] ≤ exp(−Ω(n0.8)).
913
+ (5)
914
+ Combing (2), (3), (5), we have
915
+ Pr[Y1, . . . , YK are not distinct] ≤ K3nc+1 · exp(−Ω(n0.8)) ≤ exp(−Ω(n0.8)),
916
+ where the last inequality follows from K = nO(1).
917
+ 5.1.3
918
+ From Differing-Inputs Circuits to CDP
919
+ We will next construct CDP mechanism from the previously constructed differing-inputs circuit
920
+ family. First, let us state the assumption we need here:
921
+ Assumption 22. For H as in Assumption 18, any constant ε > 0 and r = 0.5n0.9, ˜r =
922
+ 1
923
+ 1+eε n+n0.6,
924
+ there exists a differing-inputs obfuscator diO for the sampler LDS-Sampler.
925
+ 15
926
+
927
+ Algorithm 3 CDP mechanism MdiO.
928
+ Parameter:
929
+ Differing-inputs obfuscator diO, hash function H, parameters ε, r, ˜r (as in
930
+ Assumption 22), and a hash value υn ∈ {0, 1}γ(n).
931
+ Input: Dataset x ∈ {0, 1}n.
932
+ Output: Circuit : {0, 1}n → {0, 1}.
933
+ ˜x ← RRε(x).
934
+ C ← circuit that takes in z and compute 1
935
+
936
+ z ∈ Br(x) ∩ B˜r(˜x) ∩ H−1
937
+ n (υn)
938
+
939
+ �C ← diOρ(C) for randomness ρ
940
+ return
941
+ �C
942
+ Distribution H0:
943
+ ˜x ← RRε(x)
944
+ C(z) := 1
945
+
946
+ z ∈ Br(x) ∩ B˜r(˜x) ∩ H−1
947
+ n (υn)
948
+
949
+ return diOρ(C)
950
+ Distribution H1:
951
+ ˜x ← RRε(x)
952
+ C(z) := 1
953
+
954
+ z ∈ Br(x′) ∩ B˜r(˜x) ∩ H−1
955
+ n (υn)
956
+
957
+ return diOρ(C)
958
+ Distribution H2:
959
+ ˜x ← RRε(x′)
960
+ C(z) := 1
961
+
962
+ z ∈ Br(x′) ∩ B˜r(˜x) ∩ H−1
963
+ n (υn)
964
+
965
+ return diOρ(C)
966
+ Figure 1: Hybrids in proof of Theorem 23. H0 is precisely MdiO(x) and H2 is precisely MdiO(x′).
967
+ Our mechanism can then be defined by simply applying obfuscator to the circuit generated
968
+ in the same way as C1 in LDS-Samplern. This mechanism MdiO is described more formally in
969
+ Algorithm 3. The CDP property of the mechanism follows rather simply from the definition of diO
970
+ and the fact that RRε is ε-SDP.
971
+ Theorem 23. Under Assumptions 18 and 22, MdiO is ε-CDP.
972
+ Proof. For any adjacent datasets x, x′, we want to show that MdiO(x) ≈c
973
+ ε MdiO(x′). We show this
974
+ using an intermediate hybrid, as shown in Figure 1, where changes from one hybrid to next are
975
+ highlighted in red.
976
+ ◮ Distribution H0 is precisely MdiO(x).
977
+ ◮ Distribution H1 is a variant of H0, where we change x to x′ in the definition of C, but continue
978
+ to sample ˜x ∼ RRε(x).
979
+ ◮ Distribution H2 is a variant of H1, where we sample ˜x ∼ RRε(x′). Note that this is exactly
980
+ MdiO(x′).
981
+ We show that H0 ≈c
982
+ ε H2 by showing that H0 ≈c H1 and H1 ≈ε,0 H2 and using basic
983
+ composition (Fact 11).
984
+ We have from Lemma 21, that under Assumption 18, the joint distri-
985
+ bution of ˜x ∼ RRε(x), and circuits C in H0 and H1 is precisely the output of LDS-Sampler.
986
+ Thus, from Assumption 22, it follows that H0 ≈c H1 by post-processing (Fact 11).
987
+ Next, we
988
+ have that H1 ≈(ε,0) H2, since the only difference between the two is the distribution of ˜x, and
989
+ RRε(x) ≈(ε,0) RRε(x′) (again by post-processing).
990
+ Finally, its utility also follows simply from a standard concentration inequality.
991
+ 16
992
+
993
+ Theorem 24. When choosing ˜r =
994
+ 1
995
+ 1+eε n + n0.6, MdiO is (1 − o(1))-useful for ueval
996
+ H−1
997
+ n (υn).
998
+ Proof. Consider any dataset x. If x /∈ H−1
999
+ n (υn), then, by definition of uLDS
1000
+ H−1
1001
+ n (υn), the utility always
1002
+ evaluates to one. Therefore, we may only consider the case where x ∈ H−1
1003
+ n (υn).
1004
+ In this case, observe that Pr
1005
+
1006
+ ueval
1007
+ H−1
1008
+ n (υn)(x, MdiO(x)) = 1
1009
+
1010
+ = Pr˜x∼RRε(x)[x ∈ B˜r(˜x)]. Notice that
1011
+ ∥x − ˜x∥1 is distributed as Bin(n,
1012
+ 1
1013
+ 1+eε ). Therefore, applying Bernstein’s inequality, we have
1014
+ Pr
1015
+ ˜x∼RRε(x)[x /∈ B˜r(˜x)] ≤ exp
1016
+
1017
+
1018
+ t2
1019
+
1020
+ (1+eε)2 n + 2
1021
+ 3t
1022
+
1023
+ ≤ exp(−Ω(n0.2)),
1024
+ where t = ˜r−
1025
+ n
1026
+ 1+eε = n0.6. This means that Pr
1027
+
1028
+ ueval
1029
+ H−1
1030
+ n (υn)(x, MdiO(x)) = 1
1031
+
1032
+ = 1−o(1) as desired.
1033
+ 5.2
1034
+ CDP Mechanism for VLDS
1035
+ 5.2.1
1036
+ Witness-Indistinguishable Proofs
1037
+ For any NP language L with associated verifier VL, let RL denote the corresponding relation
1038
+ {(x, w) : x ∈ L and VL(x, w) = 1}. Let RL(x) := {w : (x, w) ∈ RL}.
1039
+ Definition 25 (NIWI Proof System). A pair (P, V ) of PPT algorithms is a non-interactive witness
1040
+ indistinguishable (NIWI) proof system for an NP relation RL if it satisfies:
1041
+ ◮ Correctness: for every (x, w) ∈ RL
1042
+ Pr[V (x, π) = 1 : π ← P(x, w)] = 1.
1043
+ ◮ Soundness: there exists a negligible function negl such that for all x /∈ L and π ∈ {0, 1}∗:
1044
+ Pr[V (x, π) = 1] ≤ negl(|x|).
1045
+ ◮ Witness Indistinguishability: There exists a polynomial ζ(·) and a negligible function negl(·),
1046
+ such that for any sequence I = {(x, w0, w1) : w0, w1 ∈ RL(x)} and for all circuits C of size at
1047
+ most ζ(|x|):
1048
+ ����
1049
+ Pr
1050
+ π0←P (x,w0)[C(x, π0) = 1] −
1051
+ Pr
1052
+ π1←P (x,w1)[C(x, π1) = 1]
1053
+ ���� ≤ negl(|x|).
1054
+ Assumption 26 ([BOV07, GOS12, BP15a]). There exists a NIWI proof system for any language
1055
+ in NP.
1056
+ 5.2.2
1057
+ Making Utility Function Efficient Using Witness-Indistinguishable Proofs
1058
+ We consider the NP language �L defined below, and use the corresponding NIWI verifier to define
1059
+ the utility for VLDS.
1060
+ Definition 27. Language �L consists of all circuits �C with a top AND gate, namely of the form
1061
+ �C0 ∧ �C1 such that there exists some x, ˜x and ρ, such that at least one of �C0 or �C1 can be obtained
1062
+ as diOρ(C) where C is a circuit that takes in z and computes 1
1063
+
1064
+ z ∈ Br(x) ∩ B˜r(˜x) ∩ H−1(υ)
1065
+
1066
+ .
1067
+ A “witness” for �C ∈ �L is given by w = (b, x, ˜x, ρ), where b ∈ {0, 1} indicates whether the witness
1068
+ is provided for �C0 or for �C1. Let ( �P, �V ) denote the NIWI proof system for L (guaranteed to exist
1069
+ by Assumption 26).
1070
+ 17
1071
+
1072
+ Algorithm 4 Sub-routine Maux
1073
+ diO.
1074
+ Parameter:
1075
+ Differing-inputs obfuscator diO, hash function H, parameters ε, r, ˜r (as in
1076
+ Assumption 22), and a hash value υ ∈ {0, 1}γ(n).
1077
+ Input: Dataset x ∈ {0, 1}n.
1078
+ Output: Circuit : {0, 1}n → {0, 1}.
1079
+ ˜x ← RRε(x).
1080
+ C ← circuit that takes in z and compute 1
1081
+
1082
+ z ∈ Br(x) ∩ B˜r(˜x) ∩ H−1
1083
+ n (υ)
1084
+
1085
+ �C ← diOρ(C) for randomness ρ
1086
+ return
1087
+ �C, ˜x, ρ
1088
+ Algorithm 5 CDP mechanism Mcdp.
1089
+ Input: Dataset x ∈ {0, 1}n, radius parameters r, ˜r > 0 and privacy parameter ε.
1090
+ Output: Circuit C and a proof string π.
1091
+ �C0, ˜x0, ρ0 ← Maux
1092
+ diO(x)
1093
+ �C1, ˜x1, ρ1 ← Maux
1094
+ diO(x)
1095
+ �C = �C0 ∧ �C1
1096
+ π ← �P( �C, (0, x, ˜x0, ρ0)) (NIWI proof for �C ∈ �L using witness (0, x, ˜x0, ρ0)).
1097
+ return
1098
+ �C, π
1099
+ We consider the verifiable low diameter set problem VLDSτ,H−1(υ),�V . Note that �C ∈ �L auto-
1100
+ matically implies that �C encodes a τ-diameter set (since �C = �C0 ∧ �C1, it suffices to certify that at
1101
+ least one of �C0 or �C1 encodes a τ-diameter set) where τ = 2r = n0.9.
1102
+ Theorem 28. Mcdp (described in Algorithm 5) is 2ε-CDP.
1103
+ Proof. For any adjacent datasets x, x′, we want to show that Mcdp(x) ≈c
1104
+ 2ε Mcdp(x′). We show this
1105
+ through the means of intermediate hybrids, as shown in Figure 2, where changes from one hybrid
1106
+ to next are highlighted in red.
1107
+ ◮ Distribution H0 is precisely Mcdp(x).
1108
+ ◮ Distribution H1 is a variant of H0, where �C1 is generated through x′ instead of x.
1109
+ ◮ Distribution H2 is a variant of H1, where we switch π from corresponding to witness (0, x, ˜x0, ρ0)
1110
+ to the witness (1, x′, ˜x1, ρ1).
1111
+ ◮ Distribution H3 is a variant of H2, where �C0 is also generated through x′ instead of x.
1112
+ ◮ Distribution H4 is a variant of H3, where we switch π from corresponding to witness (1, x′, ˜x1, ρ1)
1113
+ to the witness (0, x′, ˜x0, ρ0). Note that this is exactly Mcdp(x′).
1114
+ From Assumption 26 and post-processing (Fact 11), we have that H1 ≈c H2, and similarly H3 ≈c
1115
+ H4.
1116
+ Next, we show that H0 ≈c
1117
+ ε H1. Note that the output of H0 and H1 do not depend on ˜x1
1118
+ and ρ1. Thus the only material change between H0 and H1 is that �C1 ∼ MdiO(x) in H0 versus
1119
+ �C1 ∼ MdiO(x′) in H1. From Theorem 23, we have that MdiO(x) ≈c
1120
+ ε MdiO(x′). Thus, it follows
1121
+ that H0 ≈c
1122
+ ε H1 by post-processing (Fact 11). Similarly, it follows that H2 ≈c
1123
+ ε H3 (here we use that
1124
+ ˜x0 and ρ0 are immaterial to the final output of H2 and H3).
1125
+ Combining these using basic composition (Fact 11), we get that H0 ≈c
1126
+ 2ε H4, thus implying that
1127
+ Mcdp is 2ε-CDP.
1128
+ 18
1129
+
1130
+ Distribution H0:
1131
+ �C0, ˜x0, ρ0 ← Maux
1132
+ diO(x)
1133
+ �C1, ˜x1, ρ1 ← Maux
1134
+ diO(x)
1135
+ �C = �C0 ∧ �C1
1136
+ π ← �P( �C, (0, x, ˜x0, ρ0))
1137
+ return
1138
+ �C, π
1139
+ Distribution H1:
1140
+ �C0, ˜x0, ρ0 ← Maux
1141
+ diO(x)
1142
+ �C1, ˜x1, ρ1 ← Maux
1143
+ diO(x′)
1144
+ �C = �C0 ∧ �C1
1145
+ π ← �P( �C, (0, x, ˜x0, ρ0))
1146
+ return
1147
+ �C, π
1148
+ Distribution H2:
1149
+ �C0, ˜x0, ρ0 ← Maux
1150
+ diO(x)
1151
+ �C1, ˜x1, ρ1 ← Maux
1152
+ diO(x′)
1153
+ �C = �C0 ∧ �C1
1154
+ π ← �P( �C, (1, x′, ˜x1, ρ1)).
1155
+ return
1156
+ �C, π
1157
+ Distribution H3:
1158
+ �C0, ˜x0, ρ0 ← Maux
1159
+ diO(x′)
1160
+ �C1, ˜x1, ρ1 ← Maux
1161
+ diO(x′)
1162
+ �C = �C0 ∧ �C1
1163
+ π ← �P( �C, (1, x′, ˜x1, ρ1)).
1164
+ return
1165
+ �C, π
1166
+ Distribution H4:
1167
+ �C0, ˜x0, ρ0 ← Maux
1168
+ diO(x′)
1169
+ �C1, ˜x1, ρ1 ← Maux
1170
+ diO(x′)
1171
+ �C = �C0 ∧ �C1
1172
+ π ← �P( �C, (0, x′, ˜x0, ρ0))
1173
+ return
1174
+ �C, π
1175
+ Figure 2: Hybrids in proof of Theorem 28. H0 is precisely Mcdp(x) and H4 is precisely Mcdp(x′).
1176
+ Corollary 29. Mcdp is (1 − o(1))-useful for uVLDS
1177
+ τ,H−1(υ),�V .
1178
+ Proof. The utility for x /∈ H−1(υ) is trivially 1. Consider x ∈ H−1(υ). Suppose the mechanism
1179
+ MdiO is (1 − η)-useful for ueval
1180
+ H−1(υ). Since we sample �C0 and �C1 from MdiO independently we have
1181
+ that �C(x) = 1 with probability at least 1 − 2η. Finally, note that the proof π in the output of
1182
+ Mcdp is always accepted by �V . From Theorem 24, we have that η = o(1), and hence Mcdp is
1183
+ 1 − 2η = 1 − o(1) useful for uVLDS
1184
+ τ,H−1(υ),�V .
1185
+ We end this section by proving Theorem 16. The proof is essentially a straightforward combina-
1186
+ tion of the previous two results. The only choice left to make is to select the hash value υ; we select
1187
+ it so that the size of the preimage H−1(υ) is maximized. This ensures that the set R = H−1(υ)
1188
+ has enough density as required in Theorem 16. (Note: the density requirement in Theorem 16 is
1189
+ not important for showing the existence of a CDP mechanism, but instead is later used to show the
1190
+ non-existence of SDP mechanisms.)
1191
+ Proof of Theorem 16. Let H, τ, �V be as defined above. Furthermore, let υ be such that H−1(υ) is
1192
+ maximized and ε = εCDP/2. The fact that there exists an εCDP-CDP mechanism that is (1 − o(1))-
1193
+ useful for uVLDS
1194
+ τ,R,�V follows immediately from Theorem 28 and Corollary 29.
1195
+ Furthermore, by our
1196
+ choice of υ, notice that |R| = |H−1(υ)| ≤ 2n/2γ(n) ≥ 2n/no(log n), where the latter comes from our
1197
+ assumption on γ in Assumption 18.
1198
+ 6
1199
+ SDP Lower Bounds for the Nearby Point Problem
1200
+ In this section, we will show that there is no O(1)-SDP algorithm for the nearby point problem
1201
+ with target threshold n0.99 as long as the set Rn is fairly dense, as formalized below.
1202
+ Theorem 30. Let τ = {τn}n∈N and R = {Rn ⊆ {0, 1}n}n∈N be such that τn ≤ n0.99 and |Rn| ≥
1203
+ 2n/no(log n). Then, for any constant ε > 0 and any negligible function negl, there exists a sufficiently
1204
+ large n ∈ N such that there is no (ε, negl(n))-SDP algorithm that is 0.01-useful for unear
1205
+ τ,R .
1206
+ 19
1207
+
1208
+ To prove Theorem 30, let us first recall the standard “blatant non-privacy implies non-DP”
1209
+ proof11, which corresponds to the case Rn = {0, 1}n. At a high-level, these proofs proceed by
1210
+ showing that the error in each coordinate is large by “matching” each x ∈ {0, 1}n with another
1211
+ point x′ which is the same as x except with the i-th bit flipped; a basic calculation then shows that
1212
+ (on average) the i-th bit is predicted incorrectly with large probability. Summing this up over all
1213
+ the coordinates yield the desired bound.
1214
+ As we are in the case where Rn ̸= {0, 1}n, we cannot use the proof above directly. Nonetheless,
1215
+ we can still adapt the above proof. More specifically, instead of looking at each coordinate at a
1216
+ time, we look at a block of coordinates. For each block, we try to find a matching in the same spirit
1217
+ as above, but we now allow the x, x′ to have a larger distance; simple calculations give us a lower
1218
+ bound on being incorrect in this block (Section 6.2). We then “sum up” across all blocks to get
1219
+ a large distance (Section 6.3). Even though we get a large distance τ via this approach, the error
1220
+ probability (i.e. one minus usefulness) is small (i.e. o(1)). Fortunately, we can overcome this using
1221
+ the so-called DP hyperparameter tuning algorithm [LT19, PS21] (Section 6.4). This concludes our
1222
+ proof overview.
1223
+ 6.1
1224
+ Additional Preliminaries: Tools from Differential Privacy
1225
+ We will require several additional tools from DP literature, which we list below for completeness.
1226
+ Laplace Mechanism.
1227
+ The Laplace distribution with scale parameter b > 0, denoted by Lap(b),
1228
+ is the probability distribution over R with probability mass function z �→ 1
1229
+ 2b exp(−|z|/b).
1230
+ Given a function f : X ∗ → R, its sensitivity is defined as ∆(f) := maxD,D′ |f(D) − f(D′)|,
1231
+ where the maximum is over all pair D, D′ of adjacent datasets.
1232
+ The Laplace mechanism [DMNS06] is an ε-SDP mechanism that simply outputs f(X)+Lap(∆(f)/ǫ).
1233
+ Basic Composition.
1234
+ We will also use the so-called basic composition theorem: an algorithm that
1235
+ just runs an (ε1, δ1)-SDP and an (ε2, δ2)-SDP algorithms as subroutines, is (ε1 + ε2, δ1 + δ2)-SDP.
1236
+ Group Privacy.
1237
+ The following fact is well-known and is often referred to as group privacy.
1238
+ Fact 31 (Group Privacy (e.g., [SU16])). Let M : X ∗ → Y be an (ε, δ)-SDP mechanism and let
1239
+ D, D′ ∈ X ∗ be such that ∥D−D′∥ ≤ t, then, for all S ⊆ Y we have Pr[M(D) ∈ S] ≤ eε′·Pr[M(D′) ∈
1240
+ S] + δ′, where ε′ = tε and δ′ = etε−1
1241
+ eε−1 · δ.
1242
+ DP Hyperparameter Tuning.
1243
+ We will also use the following result of Liu and Talwar [LT19] on
1244
+ DP hyperparameter tuning. We remark that some improvements in the constants has been made in
1245
+ [PS21], by using a different distribution of the number of repetitions. Nonetheless, since we are only
1246
+ interested in an asymptotic bound, we choose to work with the slightly simpler hyperparameter
1247
+ tuning algorithm from [LT19].
1248
+ The hyperparameter tuning algorithm from [LT19] allows us to take any DP “base” mechanism
1249
+ Mbase, which outputs a candidate y and a score q ∈ R, run it multiple times and output a candidate
1250
+ with score that is above a certain threshold. The precise description is in Algorithm 6.
1251
+ We will use the following DP guarantee of Mtuning, which was shown in [LT19]12.
1252
+ 11Here we follow the proofs in [Sur19, Man22].
1253
+ 12Note that this is a simplified version of [LT19, Theorem 3.1] where we simply set ε0 = 1.
1254
+ 20
1255
+
1256
+ Algorithm 6 DP Hyperparameter Tuning Mtuning.
1257
+ Parameters: Mechanism Mbase, Threshold s, Number of Steps T, Stopping Probability γ.
1258
+ Input: Dataset D
1259
+ for j = 1, . . . , T do
1260
+ Let (y, q) ← Mbase(D).
1261
+ if q ≥ s then
1262
+ return x (and halt)
1263
+ With probability γ:
1264
+ return ⊥ (and halt)
1265
+ Theorem 32 (DP Hyperparameter Tuning [LT19]). Let ε > 0 and δ, γ ∈ [0, 1]. Suppose that Mbase
1266
+ is (ε, δ)-SDP and T ≥ 2/γ. Then, the DP Hyperparameter Tuning mechanism Mtuning defined in
1267
+ Algorithm 6 is (2ε + 1, 10e2ε · δ/γ).
1268
+ 6.2
1269
+ Weak Hardness
1270
+ We start with a relatively weak hardness for the case of τ = 0, i.e., the answer is considered correct
1271
+ iff it is the same as the input. To prove this, we recall a couple of facts.
1272
+ The first is a simple relation between independent set and maximum matching. Let ind(G)
1273
+ denote the size of the maximum independent set of G.
1274
+ Fact 33. For any graph G = (V, E), there exists matching of size at least (|V | − ind(G))/2.
1275
+ Let Hd denote the distance-d graph on the hypercube, i.e., Hd = ({0, 1}n, E) where (x, x′) ∈ E
1276
+ iff ∥x − x′∥1 ≤ d. Let
1277
+ � n
1278
+ ≤d
1279
+
1280
+ = �d
1281
+ i=0
1282
+ �n
1283
+ i
1284
+
1285
+ . The following is the “packing” lower bound.
1286
+ Fact 34. For any d ∈ N, ind(H2d+1) ≤ 2n/
1287
+ � n
1288
+ ≤d
1289
+
1290
+ .
1291
+ We are now ready to prove a lower bound for the nearby problem.
1292
+ Theorem 35. For any R ⊆ {0, 1}n, d, ε, δ, let ε′ = (2d + 1)ε and δ′ = eε′−1
1293
+ eε−1 δ. Then, for any
1294
+ (ε, δ)-SDP algorithm M, we have
1295
+
1296
+ x∈R
1297
+ Pr[M(x) ̸= x] ≥ 0.5e−ε′(1 − δ′)
1298
+
1299
+ |R| − 2n
1300
+ � n
1301
+ ≤d
1302
+
1303
+
1304
+ .
1305
+ Proof. Let H2d+1[R] denote the subgraph of H2d+1 induced on R. Notice that ind(H2d+1[R]) ≤
1306
+ ind(H2d+1). Therefore, by Fact 33 and Fact 34, we can conclude H2d+1[R] contains a matching of
1307
+ size at least m ≥
1308
+
1309
+ |R| − 2n/
1310
+ � n
1311
+ ≤d
1312
+ ��
1313
+ /2. Let the matching be (x1, ˜x1), . . . , (xm, ˜xm).
1314
+ For each i ∈ [m], we have
1315
+ Pr[M(xi) ̸= xi] + Pr[M(˜xi) ̸= ˜xi]
1316
+ ≥ Pr[M(xi) = ˜xi] + Pr[M(˜xi) ̸= ˜xi]
1317
+ (Group privacy, Fact 31)
1318
+ ≥ e−ε′(Pr[M(˜xi) = ˜xi] − δ′) + Pr[M(˜xi) ̸= ˜xi]
1319
+ ≥ e−ε′(Pr[M(˜xi) = ˜xi] + Pr[M(˜xi) ̸= ˜xi] − δ′)
1320
+ = e−ε′(1 − δ′).
1321
+ Adding this over all i ∈ [m] yields the claimed bound.
1322
+ 21
1323
+
1324
+ 6.3
1325
+ Boosting the Distance
1326
+ We can now prove a hardness for larger τ by dividing the coordinates into groups and applying the
1327
+ previously derived weak hardness result on each group. We note that the “non-usefulness” we get
1328
+ on the right hand side is still insufficient for Theorem 30; this will be dealt with in Section 6.4.
1329
+ Theorem 36. Let n = n′ · b′ for some n′, b′ ∈ N. For any R ⊆ {0, 1}n, d, ε, δ, ζ, let ε′ = (2d + 1)ε
1330
+ and δ′ = eε′−1
1331
+ eε−1 δ. Then, for any (ε, δ)-SDP algorithm M, there exists x ∈ R such that
1332
+ Pr[unear
1333
+ ζ·b′,R(M(x), x) = 0] ≥
1334
+
1335
+ 0.5e−ε′(1 − δ′)
1336
+
1337
+ 1 −
1338
+ 2n
1339
+ |R| ·
1340
+ � n′
1341
+ ≤d
1342
+
1343
+ ��
1344
+ − ζ.
1345
+ Proof. Let Bi := {(i − 1)n′ + 1, . . . , in′} for all i ∈ [b′]. Furthermore, let R(Bi,z−Bi) denote the set
1346
+ the set of all x ∈ R such that x−Bi = z−Bi.
1347
+ First, notice that
1348
+
1349
+ x∈R
1350
+ Pr[unear
1351
+ ζ·b′,R(M(x), x) = 0] =
1352
+
1353
+ x∈R
1354
+ Ey←M(x) 1
1355
+ �|{i ∈ [n] | yi ̸= xi}|
1356
+ b′
1357
+ > ζ
1358
+
1359
+
1360
+
1361
+ x∈R
1362
+ Ey←M(x) 1
1363
+ �|{i ∈ [b′] | yBi ̸= xBi}|
1364
+ b′
1365
+ > ζ
1366
+
1367
+
1368
+
1369
+ x∈R
1370
+ Ey←M(x)
1371
+
1372
+ Pr
1373
+ i∈[b′][yBi ̸= xBi] − ζ
1374
+
1375
+ =
1376
+
1377
+  1
1378
+ b′
1379
+
1380
+ i∈[b′]
1381
+
1382
+ x∈R
1383
+ Pr[M(x)Bi ̸= xBi]
1384
+
1385
+  − ζ|R|
1386
+
1387
+
1388
+
1389
+  1
1390
+ b′
1391
+
1392
+ i∈[b′]
1393
+
1394
+ z−Bi∈{0,1}[n]\Bi
1395
+
1396
+ x∈R(Bi,z−Bi )
1397
+ Pr[M(x)Bi ̸= xBi]
1398
+
1399
+
1400
+  − ζ|R|.
1401
+ For each fixed z−Bi ∈ {0, 1}[n]\Bi, consider the mechanism M′ : {0, 1}Bi → {0, 1}Bi defined by
1402
+ M′(xBi) := Mi(xBi ◦ z−Bi)|Bi.
1403
+ It is clear that M′ is (ε, δ)-SDP.
1404
+ Furthermore, observe that
1405
+ Pr[M(x)Bi ̸= xBi] = Pr[M′(x) ̸= xBi] for all x ∈ R(Bi,z−Bi). Therefore, by applying Theorem 35
1406
+ and plugging it back into the above, we get
1407
+
1408
+ x∈R
1409
+ Pr[unear
1410
+ ζ·b′,R(M(x), x) = 0]
1411
+
1412
+
1413
+
1414
+  1
1415
+ b′
1416
+
1417
+ i∈[b′]
1418
+
1419
+ z−Bi∈{0,1}[n]\Bi
1420
+ 0.5e−ε′(1 − δ′)
1421
+
1422
+ |R(Bi,z−Bi)| − 2n′
1423
+ � n′
1424
+ ≤d
1425
+
1426
+ �
1427
+
1428
+  − ζ|R|
1429
+ =
1430
+
1431
+  1
1432
+ b′
1433
+
1434
+ i∈[b′]
1435
+ 0.5e−ε′(1 − δ′)
1436
+
1437
+ |R| − 2n−n′ · 2n′
1438
+ � n′
1439
+ ≤d
1440
+
1441
+ �
1442
+  − ζ|R|
1443
+ =
1444
+
1445
+ 0.5e−ε′(1 − δ′)
1446
+
1447
+ |R| − 2n
1448
+ � n′
1449
+ ≤d
1450
+
1451
+ ��
1452
+ − ζ|R|.
1453
+ Dividing by |R| then gives us the claimed bound.
1454
+ 22
1455
+
1456
+ 6.4
1457
+ Boosting the Failure Probability
1458
+ We will now prove the last part of the lower bound, which is to show that the existence of even
1459
+ slightly useful mechanism also leads to an existence of a highly useful mechanism, albeit at a slight
1460
+ increanse in the distance threshold. The formal statement and its proof are given below; the proof
1461
+ uses the DP hyperparameter tuning algorithm (Theorem 32).
1462
+ Theorem 37. Suppose that there exists an (ε, δ)-SDP mechanism M : {0, 1}n → {0, 1}n that is
1463
+ α-useful for unear
1464
+ τ,R . Then, there exists an (ε′, δ′)-SDP mechanism M′ : {0, 1}n → {0, 1}n that is
1465
+ (1 − 1/n1000)-useful for unear
1466
+ τ ′,R where ε′ = 4ε + 1, δ′ = O
1467
+
1468
+ n11e2ε
1469
+ α
1470
+ · δ
1471
+
1472
+ and τ ′ = τ + O
1473
+ � ln n
1474
+ α
1475
+
1476
+ .
1477
+ Proof. First, let us construct the mechanism Mbase : {0, 1}n → {0, 1}n × R as follows:
1478
+ ◮ On input x ∈ {0, 1}n, first let y ← M(x).
1479
+ ◮ Then, let q = ∥x − y∥1 + z where z ∼ Lap(1/ε).
1480
+ ◮ Output (x, q).
1481
+ Since M is (ε, δ)-SDP and the Laplace mechanism is ε-SDP, the basic composition theorem implies
1482
+ that the entire Mbase mechanism is (2ε, δ)-SDP.
1483
+ Let �T = ln(5n1000)/α. Let τ ′ = τ − log(10n1000 �T)/ε. We now apply Algorithm 6 with γ =
1484
+ 0.5/(n1000 �T), T = 2/γ and threshold s = τ ′ − log(10n1000 �T)/ε.
1485
+ Theorem 32 ensures that the
1486
+ resulting algorithm Mtuning is (4ε + 1, 10e2εδ/γ)-SDP. Our final mechanism �
1487
+ M is the mechanism
1488
+ that runs Mtuning.
1489
+ If the output is not ⊥, �
1490
+ M returns that output.
1491
+ Otherwise, �
1492
+ M returns an
1493
+ arbritrary element of {0, 1}n. Since �
1494
+ M is simple a post-processing of Mtuning, we have �
1495
+ M is also
1496
+ (4ε + 1, 10e2εδ/γ)-SDP.
1497
+ We will next show that Mtuning is (1 − 1/n1000)-useful for unear
1498
+ τ ′,R. By definition of the utility
1499
+ function, this immediately holds for any x /∈ R. Therefore, we may only consider any x ∈ R.
1500
+ Consider Mtuning on such an x. Let yi, zi, qi denote the corresponding values of y, z, q in the ith
1501
+ run of Mbase.
1502
+ We will consider the following three events:
1503
+ ◮ Let E1 denote the event that |∥xi − yi∥1 − qi| > log(10n1000 �T)/ε for some i ∈ [ �T].
1504
+ ◮ Let E2 denote the event that uτ,R(yi) = 0 for all i ∈ [ �T].
1505
+ ◮ Let E3 denote the event that Mtuning halts in the first �T steps.
1506
+ Before we bound the probability of each events, notice that, if none of E1, E2, E3 occurs, we must
1507
+ have unear
1508
+ τ ′,R(y) = 1 (where y denote the output of �
1509
+ M), since s − τ, τ ′ − s ≥ log(10n1000 �T)/ε. That is,
1510
+ Pr
1511
+ y←�
1512
+ M(x)
1513
+ [unear
1514
+ τ ′,R(y) = 0] ≤ Pr[E1 ∨ E2 ∨ E3] ≤ Pr[E1] + Pr[E2] + Pr[E3].
1515
+ We will now bound the probability for each event. For E1, it immediately follows from the
1516
+ Laplace tail bound together with a union bound that
1517
+ Pr[E1] ≤ �T · 2/(10n1000 �T) = 0.2/n1000.
1518
+ For E2, the α-usefulness of M implies that
1519
+ Pr[E2] ≤ (1 − α)
1520
+ �T ≤ 0.2/n1000.
1521
+ Finally, for E3, we may simply use a union bound, which gives
1522
+ Pr[E3] ≤ γ · �T ≤ 0.5/n1000.
1523
+ 23
1524
+
1525
+ By combining the four inequalities above, we have
1526
+ Pr
1527
+ y←�
1528
+ M(x)
1529
+ [unear
1530
+ τ ′,R(y) = 0] < 1/n1000,
1531
+ as desired.
1532
+ 6.5
1533
+ Putting Things Together: Proof of Theorem 30
1534
+ Proof of Theorem 30. Suppose for the sake of contradiction that, for some constant ε > 0 and
1535
+ negligible function negl, there exists an (ε, negl(n))-SDP mechanism Mn that is 0.01-useful for
1536
+ unear
1537
+ τn,Rn for every n ∈ N.
1538
+ By Theorem 37, there is a (4ε + 1, δ′(n)) mechanism M′
1539
+ n that is (1 − 1/n1000)-useful for unear
1540
+ τ ′n,Rn
1541
+ where δ′(n) is a negligible function and τ ′
1542
+ n = τn + O(log n) = O(n0.99).
1543
+ Plugging this into
1544
+ Theorem 36 with R = Rn, n′ = n0.005, b′ = n0.995, ζ = τ ′
1545
+ n/b′ ≤ O(n−0.005), ε = 4ε + 1, δ = δ′(n), d =
1546
+ (log n0.004)/3ε (which gives ε′ ≤ log(2n0.004) and δ′ = 1/nω(1) in Theorem 36), we have
1547
+ 1
1548
+ n1000 ≥
1549
+
1550
+ 0.5 · e− log(2n0.004)(1 − n−ω(1)) (1 − o(1))
1551
+
1552
+ − O(n−0.005)
1553
+ = O(n−0.004) · (1 − o(1)) − O(n−0.005),
1554
+ which is a contradiction for any sufficiently large n.
1555
+ 7
1556
+ Putting Things Together: Proof of Theorem 5
1557
+ Our main theorem follows from trivially combining the main results from the previous two sections.
1558
+ Proof of Theorem 5. Let u = uVLDS
1559
+ τ,R,V be as given in Theorem 16, which immediately yields the
1560
+ existence of an εCDP-CDP mechanism that is (1 − o(1))-useful. Furthermore, by |R| ≥ 2n/no(n),
1561
+ Theorem 30 implies that there is no εSDP-SDP mechanism that is 0.01-useful for {unear
1562
+ τ,R }. Finally,
1563
+ applying Lemma 15, we can conclude that there is no εSDP-SDP mechanism that is 0.01-useful for
1564
+ {uVLDS
1565
+ τ,R,V }. This concludes our proof.
1566
+ 8
1567
+ Conclusion and Discussion
1568
+ In this work, we give a first task that, under certain assumptions, admits an efficient CDP algorithm
1569
+ but does not admit an (even inefficient) SDP algorithm. As mentioned in Section 1, perhaps the
1570
+ most intriguing next direction would be to see if there are more “natural” tasks for which CDP
1571
+ algorithms can go beyond known SDP lower bounds.
1572
+ On the technical front, there are also a few interesting directions. For example, it would be
1573
+ interesting to see if the three assumptions in our paper can be removed, relaxed, or replaced (by
1574
+ perhaps more widely believed assumptions). Alternatively, we can ask the opposite question: what
1575
+ are the (cryptographic) assumptions necessary for separating CDP and SDP?
1576
+ Such a question
1577
+ has been extensively studied in the multiparty model [HMST22, GMPS13, GKM+16, HMSS19,
1578
+ HNO+18]; for example, it is known that key-agreement is necessary and sufficient to get better-
1579
+ than-local-DP protocol for inner product in the two-party setting [HMST22]. Achieving such a
1580
+ 24
1581
+
1582
+ result in our setting would significantly deepen our understanding of the CDP-vs-SDP question in
1583
+ the central model.
1584
+ Another possible improvement is to strengthen the hardness of the adversary. In this paper,
1585
+ we only consider polynomial-time adversaries. Indeed, our CDP mechanism does not remain CDP
1586
+ against quasi-polynomial adversary. The reason is that we choose the hash value length to be only
1587
+ o(log2 λ) in Assumption 18, so a trivial “guess-and-check” algorithm can break this assumption in
1588
+ time λO(log λ). However, as far as we are aware, there is no inherent barrier in proving a separation
1589
+ with CDP that holds even against, e.g., sub-exponential time adversaries. Achieving such a result
1590
+ (potentially under stronger or different assumptions) would definitely be interesting.
1591
+ Furthermore, our task (or more precisely the utility function) is non-uniform (through the choice
1592
+ of υn). It would also be interesting to have a uniform task.
1593
+ Acknowledgments
1594
+ We thank Prabhanjan Ananth for helpful discussions about differing-inputs obfuscation, and anony-
1595
+ mous reviewers for helpful comments.
1596
+ 25
1597
+
1598
+ References
1599
+ [ABG+13]
1600
+ Prabhanjan Ananth, Dan Boneh, Sanjam Garg, Amit Sahai, and Mark Zhandry.
1601
+ Differing-inputs obfuscation and applications. IACR Cryptol. ePrint Arch., page 689,
1602
+ 2013.
1603
+ [Abo18]
1604
+ John M Abowd. The US Census Bureau adopts differential privacy. In KDD, pages
1605
+ 2867–2867, 2018.
1606
+ [App17]
1607
+ Apple Differential Privacy Team.
1608
+ Learning with privacy at scale.
1609
+ Apple Machine
1610
+ Learning Journal, 2017.
1611
+ [BCP14]
1612
+ Elette Boyle, Kai-Min Chung, and Rafael Pass. On extractability obfuscation. In TCC,
1613
+ pages 52–73, 2014.
1614
+ [BCV16]
1615
+ Mark Bun, Yi-Hsiu Chen, and Salil P. Vadhan. Separating computational and statis-
1616
+ tical differential privacy in the client-server model. In TCC, pages 607–634, 2016.
1617
+ [BGI+01]
1618
+ Boaz Barak, Oded Goldreich, Russell Impagliazzo, Steven Rudich, Amit Sahai, Salil P.
1619
+ Vadhan, and Ke Yang. On the (im)possibility of obfuscating programs. In CRYPTO,
1620
+ pages 1–18, 2001.
1621
+ [BGI+12]
1622
+ Boaz Barak, Oded Goldreich, Russell Impagliazzo, Steven Rudich, Amit Sahai, Salil P.
1623
+ Vadhan, and Ke Yang.
1624
+ On the (im)possibility of obfuscating programs.
1625
+ J. ACM,
1626
+ 59(2):6:1–6:48, 2012.
1627
+ [BKP18]
1628
+ Nir Bitansky, Yael Tauman Kalai, and Omer Paneth.
1629
+ Multi-collision resistance: a
1630
+ paradigm for keyless hash functions. In STOC, pages 671–684, 2018.
1631
+ [BNO08]
1632
+ Amos Beimel, Kobbi Nissim, and Eran Omri. Distributed private data analysis: Si-
1633
+ multaneously solving how and what. In CRYPTO, pages 451–468, 2008.
1634
+ [BOV07]
1635
+ Boaz Barak, Shien Jin Ong, and Salil P. Vadhan. Derandomization in cryptography.
1636
+ SIAM J. Comput., 37(2):380–400, 2007.
1637
+ [BP15a]
1638
+ Nir Bitansky and Omer Paneth. Zaps and non-interactive witness indistinguishability
1639
+ from indistinguishability obfuscation. In TCC, pages 401–427, 2015.
1640
+ [BP15b]
1641
+ Elette Boyle and Rafael Pass. Limits of extractability assumptions with distributional
1642
+ auxiliary input. In ASIACRYPT, pages 236–261, 2015.
1643
+ [BR93]
1644
+ Mihir Bellare and Phillip Rogaway. Ccs. pages 62–73, 1993.
1645
+ [BSW16]
1646
+ Mihir Bellare, Igors Stepanovs, and Brent Waters. New negative results on differing-
1647
+ inputs obfuscation. In EUROCRYPT, pages 792–821, 2016.
1648
+ [CGH04]
1649
+ Ran Canetti, Oded Goldreich, and Shai Halevi.
1650
+ The random oracle methodology,
1651
+ revisited. J. ACM, 51(4):557–594, 2004.
1652
+ [De12]
1653
+ Anindya De. Lower bounds in differential privacy. In TCC, pages 321–338, 2012.
1654
+ 26
1655
+
1656
+ [DKM+06] Cynthia Dwork, Krishnaram Kenthapadi, Frank McSherry, Ilya Mironov, and Moni
1657
+ Naor. Our data, ourselves: Privacy via distributed noise generation. In EUROCRYPT,
1658
+ pages 486–503, 2006.
1659
+ [DKY17]
1660
+ Bolin Ding, Janardhan Kulkarni, and Sergey Yekhanin.
1661
+ Collecting telemetry data
1662
+ privately. In NeurIPS, pages 3571–3580, 2017.
1663
+ [DMNS06]
1664
+ Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam D. Smith. Calibrating
1665
+ noise to sensitivity in private data analysis. In TCC, pages 265–284, 2006.
1666
+ [DMT07]
1667
+ Cynthia Dwork, Frank McSherry, and Kunal Talwar. The price of privacy and the
1668
+ limits of LP decoding. In STOC, pages 85–94, 2007.
1669
+ [DN03]
1670
+ Irit Dinur and Kobbi Nissim. Revealing information while preserving privacy. In PODS,
1671
+ pages 202–210, 2003.
1672
+ [EPK14]
1673
+ ´Ulfar Erlingsson, Vasyl Pihur, and Aleksandra Korolova. RAPPOR: Randomized ag-
1674
+ gregatable privacy-preserving ordinal response. In CCS, pages 1054–1067, 2014.
1675
+ [GGHW17] Sanjam Garg, Craig Gentry, Shai Halevi, and Daniel Wichs. On the implausibility of
1676
+ differing-inputs obfuscation and extractable witness encryption with auxiliary input.
1677
+ Algorithmica, 79(4):1353–1373, 2017.
1678
+ [GKM+16] Vipul Goyal, Dakshita Khurana, Ilya Mironov, Omkant Pandey, and Amit Sahai. Do
1679
+ distributed differentially-private protocols require oblivious transfer? In ICALP, pages
1680
+ 29:1–29:15, 2016.
1681
+ [GKY11]
1682
+ Adam Groce, Jonathan Katz, and Arkady Yerukhimovich. Limits of computational
1683
+ differential privacy in the client/server setting. In TCC, pages 417–431, 2011.
1684
+ [GMPS13]
1685
+ Vipul Goyal, Ilya Mironov, Omkant Pandey, and Amit Sahai. Accuracy-privacy trade-
1686
+ offs for two-party differentially private protocols. In CRYPTO, pages 298–315, 2013.
1687
+ [GOS12]
1688
+ Jens Groth, Rafail Ostrovsky, and Amit Sahai.
1689
+ New techniques for noninteractive
1690
+ zero-knowledge. J. ACM, 59(3):11:1–11:35, 2012.
1691
+ [Gre16]
1692
+ Andy Greenberg. Apple’s “differential privacy” is about collecting your data – but not
1693
+ your data. Wired, June, 13, 2016.
1694
+ [GT08]
1695
+ Ben Green and Terence Tao. The primes contain arbitrarily long arithmetic progres-
1696
+ sions. Annals of Mathematics, 167(2):481–547, 2008.
1697
+ [HMSS19]
1698
+ Iftach Haitner, Noam Mazor, Ronen Shaltiel, and Jad Silbak. Channels of small log-
1699
+ ratio leakage and characterization of two-party differentially private computation. In
1700
+ TCC, pages 531–560, 2019.
1701
+ [HMST22]
1702
+ Iftach Haitner, Noam Mazor, Jad Silbak, and Eliad Tsfadia. On the complexity of
1703
+ two-party differential privacy. In STOC, pages 1392–1405, 2022.
1704
+ [HNO+18]
1705
+ Iftach Haitner, Kobbi Nissim, Eran Omri, Ronen Shaltiel, and Jad Silbak. Compu-
1706
+ tational two-party correlation: A dichotomy for key-agreement protocols. In FOCS,
1707
+ pages 136–147, 2018.
1708
+ 27
1709
+
1710
+ [HT10]
1711
+ Moritz Hardt and Kunal Talwar. On the geometry of differential privacy. In STOC,
1712
+ pages 705–714, 2010.
1713
+ [IPS15]
1714
+ Yuval Ishai, Omkant Pandey, and Amit Sahai. Public-coin differing-inputs obfuscation
1715
+ and its applications. In TCC, pages 668–697, 2015.
1716
+ [JLS21]
1717
+ Aayush Jain, Huijia Lin, and Amit Sahai. Indistinguishability obfuscation from well-
1718
+ founded assumptions. In STOC, pages 60–73, 2021.
1719
+ [KT18]
1720
+ Krishnaram Kenthapadi and Thanh T. L. Tran. PriPeARL: A framework for privacy-
1721
+ preserving analytics and reporting at LinkedIn. In CIKM, pages 2183–2191, 2018.
1722
+ [LT19]
1723
+ Jingcheng Liu and Kunal Talwar. Private selection from private candidates. In STOC,
1724
+ pages 298–309, 2019.
1725
+ [Man22]
1726
+ Pasin Manurangsi. Tight bounds for differentially private anonymized histograms. In
1727
+ SOSA, pages 203–213, 2022.
1728
+ [MMP+10] Andrew McGregor, Ilya Mironov, Toniann Pitassi, Omer Reingold, Kunal Talwar, and
1729
+ Salil P. Vadhan. The limits of two-party differential privacy. In FOCS, pages 81–90,
1730
+ 2010.
1731
+ [MPRV09]
1732
+ Ilya Mironov, Omkant Pandey, Omer Reingold, and Salil P. Vadhan. Computational
1733
+ differential privacy. In CRYPTO, pages 126–142, 2009.
1734
+ [PS21]
1735
+ Nicolas Papernot and Thomas Steinke. Hyperparameter tuning with renyi differential
1736
+ privacy. CoRR, abs/2110.03620, 2021.
1737
+ [RSP+21]
1738
+ Ryan Rogers, Subbu Subramaniam, Sean Peng, David Durfee, Seunghyun Lee, San-
1739
+ tosh Kumar Kancha, Shraddha Sahay, and Parvez Ahammad. LinkedIn’s audience
1740
+ engagements API: A privacy preserving data analytics system at scale. J. Priv. Con-
1741
+ fiden., 11(3), 2021.
1742
+ [RTTV08]
1743
+ Omer Reingold, Luca Trevisan, Madhur Tulsiani, and Salil P. Vadhan. Dense subsets
1744
+ of pseudorandom sets. In FOCS, pages 76–85, 2008.
1745
+ [Sha14]
1746
+ Stephen Shankland. How Google tricks itself to protect Chrome user privacy. CNET,
1747
+ October, 2014.
1748
+ [SU16]
1749
+ Thomas Steinke and Jonathan R. Ullman. Between pure and approximate differential
1750
+ privacy. J. Priv. Confidentiality, 7(2), 2016.
1751
+ [Sur19]
1752
+ Ananda Theertha Suresh. Differentially private anonymized histograms. In NeurIPS,
1753
+ pages 7969–7979, 2019.
1754
+ [TZ08]
1755
+ Terence Tao and Tamar Ziegler. The primes contain arbitrarily long polynomial pro-
1756
+ gressions. Acta Mathematica, 201(2):213 – 305, 2008.
1757
+ [Vad17]
1758
+ Salil P. Vadhan. The complexity of differential privacy. In Tutorials on the Foundations
1759
+ of Cryptography, pages 347–450. Springer International Publishing, 2017.
1760
+ [War65]
1761
+ Stanley L Warner. Randomized response: A survey technique for eliminating evasive
1762
+ answer bias. JASA, 60(309):63–69, 1965.
1763
+ 28
1764
+
1765
+ A
1766
+ Comparison of various diO assumptions
1767
+ We review and compare the various notions of differing inputs obfuscation, showing that the notion
1768
+ of diO-for-pcS (Definition 20) is in fact weaker (or at least, no stronger) than all notions of differing
1769
+ inputs obfuscation studied in literature.
1770
+ The definition of diO as given by [BGI+12] did not include the notion of a sampler. Instead for
1771
+ any circuits C0 and C1, if an adversary A can distinguish diO(C0) and diO(C1) with non-negligible
1772
+ advantage, then there exists an adversary A′ that, given any circuits C′
1773
+ 0 and C′
1774
+ 1, where C′
1775
+ b is
1776
+ functionally equivalent to Cb for b ∈ {0, 1}, A′(C′
1777
+ 0, C′
1778
+ 1) can return x such that C0(x) ̸= C1(x).
1779
+ This notion is stronger than the corresponding notion involving samplers. Since most applica-
1780
+ tions of differing-inputs obfuscation in literature are stated using differing-inputs samplers, we will
1781
+ only refer to diO notions that involve these.
1782
+ Definition 38 (Differing-Inputs Circuit Sampler [ABG+13]). An efficient non-uniform sampling
1783
+ algorithm Sampler = {Samplern} is a differing-inputs sampler for the parameterized collection
1784
+ C = {Cn} of circuits if the output of Samplern is distributed over Cn × Cn × {0, 1}∗ and for every
1785
+ efficient non-uniform algorithm A = {An}, there exists a negligible function negl(·) such that for
1786
+ all n ∈ N:
1787
+ Pr
1788
+ θ [C0(y) ̸= C1(y) : (C0, C1, aux) ← Samplern(θ), y ← An(C0, C1, aux)] ≤ negl(n).
1789
+ Plain Sampler. We call a differing-inputs sampler as a Plain Sampler if aux is always ⊥.
1790
+ Public-Coin Sampler. We call a differing-inputs sampler as Public-Coin Sampler if aux is equal
1791
+ to θ (precisely Definition 19).
1792
+ General Sampler. We call a differing-inputs sampler as a General Sampler whenever we want to
1793
+ emphasize that aux is allowed to be any function of θ. In particular, plain and public-coin
1794
+ samplers are special cases of general samplers.
1795
+ Note that, the more information that aux is allowed to contain, the more restricted the distribution
1796
+ over circuit pairs (C0, C1) gets. In particular, any public-coin Sampler remains a differing-inputs
1797
+ Sampler if we set aux to be some function of θ (instead of being all of θ), and similarly, any general
1798
+ differing-inputs Sampler can be converted to a plain-Sampler by simply setting aux = ⊥.
1799
+ We can consider two notions of security of differing inputs obfuscators, depending on whether or
1800
+ not the distinguisher has access to aux. Recall that the “differing-inputs” condition in Definition 20
1801
+ was
1802
+ | Pr
1803
+ θ [Dn(diO(1n, C0)) = 1] − Pr
1804
+ θ [Dn(diO(1n, C1)) = 1]| ≤ negl(n).
1805
+ (6)
1806
+ On the other hand, we could consider a different notion where for any general sampler Sampler, for
1807
+ (C0, C1, aux) ← Samplern(θ), we replace the “differing-inputs” condition with
1808
+ | Pr
1809
+ θ [Dn(diO(1n, C0), aux) = 1] − Pr
1810
+ θ [Dn(diO(1n, C1), aux) = 1]| ≤ negl(n).
1811
+ (7)
1812
+ Depending on the type of sampler (plain or public-coin or general) and the notion of security
1813
+ for differing inputs obfuscators ((6) or (7)), we get various kinds of diO assumptions, which we list
1814
+ below.
1815
+ 29
1816
+
1817
+ plain-diO
1818
+ pc-diO
1819
+ gen-diO
1820
+ diO-for-genS
1821
+ diO-for-pcS
1822
+ Figure 3: Comparisons between different diO assumptions, where A → B denotes that existence of
1823
+ A implies existence of B, or in other words, existence of A is a stronger assumption than existence of
1824
+ B. Existence of diO-for-pcS (assumption used in this paper) is the weakest among all the notions.
1825
+ Plain diO. We refer to plain-diO as the notion of diO that holds only against plain samplers. Note,
1826
+ there is no difference here between the security notions of (6) and (7), since aux = ⊥ anyway.
1827
+ Public-Coin diO. We refer to pc-diO, as the notion of public-coin diO defined by [IPS15], cor-
1828
+ responding to the notion of diO that holds only against public-coin samplers, where the
1829
+ distinguisher also has access to aux = θ, as in (7).
1830
+ General diO. We refer to gen-diO, as the notion of general diO defined by [ABG+13], correspond-
1831
+ ing to the notion of diO that holds for general samplers, and where the distinguisher also has
1832
+ access to aux, as in (7).
1833
+ diO for General Samplers. We define diO-for-genS as the notion of diO that holds only against
1834
+ general samplers, but where the distinguisher does not have access to aux = θ, as in (6).
1835
+ diO for Public-Coin Samplers. This is precisely Definition 20, where the security of diO holds
1836
+ only for public-coin samplers, where the distinguisher does not have access to aux, as in (6).
1837
+ Comparison between different diO assumptions.
1838
+ The comparison between the assumptions
1839
+ asserting existence of each type of diO is illustrated in Figure 3, with justification for each arrow
1840
+ given as follows:
1841
+ ◮ Existence of gen-diO implies existence of plain-diO and pc-diO, since both are special cases
1842
+ corresponding to plain samplers and public-coin samplers respectively.
1843
+ ◮ To the best of knowledge, it is unknown whether the assumptions of existence of plain-diO
1844
+ and the existence of pc-diO are comparable or not.
1845
+ ◮ Existence of plain-diO implies existence of diO-for-genS since any general sampler can be
1846
+ converted to a plain sampler by simply setting aux = ⊥; note that the distinguisher (in the
1847
+ definition of diO) does not have access to aux in either case.
1848
+ 30
1849
+
1850
+ ◮ Existence of diO-for-genS implies existence of plain-diO and diO-for-pcS since both are special
1851
+ cases corresponding to plain samplers and public-coin samplers respectively.
1852
+ ◮ Existence of pc-diO implies existence of diO-for-pcS, since the distinguisher in the definition
1853
+ of diO-for-pcS does not have access to θ and hence is less powerful.
1854
+ Finally, one may wonder, what was special about the application of diO in this paper that only
1855
+ required diO-for-pcS and not gen-diO or pc-diO as in prior work in cryptography. The main reason is
1856
+ that, in cryptographic applications, an aux is provided to adversaries to enable certain cryptographic
1857
+ functionality (such as by revealing some public key parameters), and thus, it is required that the diO
1858
+ is secure even given knowledge of this aux information. In applications of pc-diO, the distinguisher
1859
+ typically does not have access to all of θ (such as some secret key parameters may be hidden), but
1860
+ security given knowledge of entire θ implies security given partial knowledge of θ. In the setting of
1861
+ this paper, there wasn’t any particular functionality that needed to be enabled, other than basic
1862
+ circuit evaluation, and the particular circuit samplers of interest were public-coin differing inputs
1863
+ samplers, which is why it suffices to only assume diO-for-pcS.
1864
+ 31
1865
+
I9AyT4oBgHgl3EQfTfcL/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
KdE4T4oBgHgl3EQfiA17/content/tmp_files/2301.05130v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
KdE4T4oBgHgl3EQfiA17/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
LtAyT4oBgHgl3EQfgPig/content/tmp_files/2301.00356v1.pdf.txt ADDED
@@ -0,0 +1,1175 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Multiorbital effects in high-order harmonic emission from CO2
2
+ Andres Mora, Lauren Bauerle, Yuqing Xia, Agnieszka Jaron1
3
+ 1JILA and Department of Physics, University of Colorado, Boulder, CO 80309-0440, USA
4
+ We study the ellipticity of high-order harmonics emitted from CO2 molecule driven by linearly po-
5
+ larized laser fields using numerical simulations within the time-dependent density functional theory.
6
+ We find that the overall ellipticity of the harmonics is small, which is in agreement with experimen-
7
+ tal data. On the other hand, our analysis of the numerical results indicates that several valence
8
+ orbitals contribute significantly to the harmonic emission and some of these contributions show a
9
+ strong ellipticity of the harmonics. The small ellipticity in the total harmonics signal arises from a
10
+ combination of factors, namely, the fact that harmonic emission from different orbitals is strongest
11
+ at different alignment angles of the molecular axis with respect to the laser polarization direction,
12
+ as well as interference effects and a strong laser coupling between several inner valence orbitals.
13
+ PACS numbers: 32.80.Fb,32.80.Wr
14
+ I.
15
+ INTRODUCTION
16
+ High-order harmonic generation (HHG) is one of the
17
+ highly nonlinear, nonperturbative processes that occur
18
+ when an atom or molecule is irradiated by an intense
19
+ laser field [1, 2]. It results from the distortion of the elec-
20
+ tron density in the presence of the strong electromagnetic
21
+ field of the laser and the power spectrum of the emitted
22
+ harmonic radiation corresponds to the Fourier transform
23
+ of the electron dipole acceleration. The intensity spectra
24
+ of the emitted high harmonics shows some general char-
25
+ acteristic features, such as a fast decrease of the signal
26
+ over the first few harmonics followed by a region with
27
+ fairly constant plateau harmonic intensities ending by a
28
+ sharp cutoff, beyond which the harmonic intensity drops
29
+ quickly.
30
+ Over the past few decades HHG has been an active area
31
+ of research since it provides a source for coherent short-
32
+ wavelength light, extending into the soft-X-ray regime
33
+ [3], and for ultrashort laser pulses and waveforms in the
34
+ attosecond [4, 5]. Furthermore, it has been shown that
35
+ HHG spectra contain information about the atomic and
36
+ electronic structure of the target (e.g., [6–11]), ultrafast
37
+ molecular and intra-molecular electron dynamics (e.g.,
38
+ [12–15]) as well as time resolution of chemical processes
39
+ (e.g., [16]).
40
+ Basic intuitive picture of HHG is provided by the semi-
41
+ classical three-step model [17–19], according to which an
42
+ electron tunnels through the barrier created by the laser
43
+ field and the Coulomb field into the continuum, is accel-
44
+ erated by the electric field of the laser first away from
45
+ and then back to the parent ion.
46
+ Upon return it re-
47
+ combines, emitting excess energy in the form of high-
48
+ order harmonic radiation. Since the process occurs every
49
+ half cycle of the driving laser field, an attosecond pulse
50
+ train is produced.
51
+ In recent years, it has been shown
52
+ that, in particular for high-order harmonic generation
53
+ from molecules, the generated harmonic spectra incorpo-
54
+ rate more features than predicted by the basic three-step
55
+ single-active-electron model. One example are polarime-
56
+ try measurements of high-order harmonic emission from
57
+ aligned diatomic and linear triatomic molecules driven by
58
+ linearly polarized laser fields. Surprisingly, strong ellipti-
59
+ cally polarized harmonics were observed for N2 [20, 21],
60
+ while in contrast CO2 exhibited a much lower ellipticity
61
+ in the harmonic emission [20]. Structural effects [22–26],
62
+ such as the symmetry of the Highest Occupied Molecular
63
+ Orbital (HOMO) as well as interference effects, and ultra-
64
+ fast multielectron dynamics involving lower-lying orbitals
65
+ in the molecule [27] or in the molecular ion [21] have been
66
+ put forward as potential origins for the observed elliptic-
67
+ ity.
68
+ In this article we focus on the role of multielectron
69
+ and multiorbital effects in the neutral CO2 molecule on
70
+ the polarization state of high-order harmonics. We have
71
+ shown previously [27], that results based on the time-
72
+ dependent density functional theory (TDDFT) are in
73
+ excellent agreement with the experimental data for N2
74
+ [20, 21], if contributions from at least three Kohn-Sham
75
+ orbitals are taken into account. Similar strong influence
76
+ of inner shell contributions has been observed and pre-
77
+ dicted for other strong-field processes as well [28–37].
78
+ Our results of numerical TDDFT simulations show
79
+ that indeed the contributions from several valence or-
80
+ bitals contribute to the higher-order harmonic emission
81
+ from CO2. Moreover, we find that the emission from each
82
+ of the orbitals is elliptically polarized. However, our re-
83
+ sults for the total high-order harmonic spectrum, which
84
+ includes the contributions of up to six orbitals, surpris-
85
+ ingly shows, in agreement with the experimental data
86
+ [20], almost no ellipticity.
87
+ Thus, despite the fact that
88
+ high-order harmonic generation from CO2 appears to be
89
+ a multielectron process with several orbitals actively in-
90
+ volved, signatures in the ellipticity of the harmonic emis-
91
+ sion from the different orbitals fade away in the total
92
+ signal.
93
+ The article is organized as follows: In the next sec-
94
+ tion we briefly outline the basics of the time-dependent
95
+ density functional approach used for our numerical sim-
96
+ ulations. We then discuss the application to calculations
97
+ of the ellipticity of high-order harmonic generation of
98
+ molecules, including the proper account of the distribu-
99
+ tion of alignment in the molecular ensemble. Next, we
100
+ arXiv:2301.00356v1 [physics.atm-clus] 1 Jan 2023
101
+
102
+ 2
103
+ compare the results of our calculations with the experi-
104
+ mental data and analyze the contributions from the dif-
105
+ ferent valence orbitals to the total harmonic spectra. We
106
+ end with a brief summary of our results.
107
+ II.
108
+ THEORY
109
+ In the nonperturbative intensity regime the theoret-
110
+ ical study of the interaction of multielectron targets,
111
+ e.g. molecules, with ultrashort laser pulses is challenging.
112
+ An approximative approach to analyze multielectron and
113
+ multiorbital effects in strong-field processes utilizes the
114
+ framework of the time-dependent density functional the-
115
+ ory (TDDFT). In this section we outline the application
116
+ of TDDFT to the calculation of high-harmonic genera-
117
+ tion in molecules, focusing in particular on the evaluation
118
+ of the ellipticity of the radiation in an ensemble of aligned
119
+ molecules.
120
+ A.
121
+ TDDFT for strong-field induced molecular
122
+ processes
123
+ The TDDFT approach is based on the one-to-one cor-
124
+ respondence between the time-dependent electron den-
125
+ sity ρ(r, t) and the time-dependent potential in multi-
126
+ electron Schr¨odinger equation [38]. The density is calcu-
127
+ lated from the time-dependent multielectron Schr¨odinger
128
+ equation expressed as system of auxiliary time-dependent
129
+ noninteracting single-electron Kohn-Sham equations:
130
+ i ∂
131
+ ∂tφk(r, t) =
132
+
133
+ −∇2
134
+ 2 + VKS(r, t)
135
+
136
+ φk(r, t)
137
+ (1)
138
+ with
139
+ ρ(r, t) =
140
+ n
141
+
142
+ k=1
143
+ fk|φk(r, t)|2
144
+ (2)
145
+ where r is the electronic coordinate, fk is the electron
146
+ population in the k-th Kohn-Sham orbital φk(r, t) and
147
+ n is the number of orbitals. For a molecule interacting
148
+ with a time-dependent intense laser field the Kohn-Sham
149
+ potential
150
+ VKS(r, t) = Vext(r, t) +
151
+
152
+ ρ(r′, t)
153
+ |r − r′|dr′ + Vxc(r)
154
+ (3)
155
+ includes the external potential due to the interaction of
156
+ the electron with the N nuclei in the molecule and with
157
+ the time-dependent electric field:
158
+ Vext(r, t) =
159
+ N
160
+
161
+ i=1
162
+
163
+ Zi
164
+ |Ri − r| + E0(t) sin(ωt)
165
+ n
166
+
167
+ k=1
168
+ rk · ˆϵ (4)
169
+ where Zi is the charge of the ith nucleus, ˆϵ is the polar-
170
+ ization direction, ω and E0(t) are the angular frequency
171
+ and the time-dependent amplitude of the laser field. In
172
+ the present calculations we considered a sin2-shaped en-
173
+ velope.
174
+ The exact form of the exchange-correlation potential
175
+ Vxc, which takes account of the multielectron effects, is
176
+ unknown. To use TDDFT for practical calculations, dif-
177
+ ferent approaches have been proposed to design density
178
+ functionals for the exchange-correlation energy (for an
179
+ overview, see e.g., [39]). For the present calculations, we
180
+ have performed systematic studies with various function-
181
+ als and found that functionals based on the local density
182
+ approximation (LDA),
183
+ ELDA
184
+ xc
185
+ [ρ] =
186
+
187
+ ρ(r)Vxc(r)dr ,
188
+ (5)
189
+ provide, in general, good results.
190
+ An improvement
191
+ is to take into account the correct asymptotic behav-
192
+ ior (1/r), which can be done, for example, via the
193
+ exchange-correlation potential proposed by van Leeuwen
194
+ and Baerends [40],
195
+ V LB
196
+ xc (α, β; r) = αV LDA
197
+ x
198
+ (r) + βV LDA
199
+ c
200
+ (r)
201
+ (6)
202
+
203
+ βx2(r)ρ1/3(r)
204
+ 1 + 3βx(r) ln[x2(r) + (x2(r) + 1)1/2],
205
+ where V LDA
206
+ x
207
+ and V LDA
208
+ c
209
+ are the LDA exchange and cor-
210
+ relation potentials and x(r) = |∇ρ(r)|/[ρ(r)]4/3. α and
211
+ β are parameters obtained by fit to the exact exchange-
212
+ correlation function of a certain atomic or molecular sys-
213
+ tem. A similar TDDFT approach for the interaction of
214
+ molecules with strong fields has been used recently by
215
+ Chu and co-workers [41, 42].
216
+ In order to solve the Kohn-Sham equations, Eq. (1),
217
+ we have discretized the wavefunction in space and time
218
+ with uniform step ∆x = 0.03 a.u. and ∆t = 0.03 a.u.,
219
+ which converts the ansatz into a matrix equation using
220
+ the Octopus code [43, 44]. The initial wavefunctions for
221
+ the molecules considered in our study have been obtained
222
+ by solving the eigenvalue problem self-consistently using
223
+ an initial guess and geometry optimized using Octopus
224
+ code as well (this ensures consistency and minimizes risk
225
+ for errors). The wavefunction for each orbital is prop-
226
+ agated forward in time using the enforced time-reversal
227
+ symmetry method. We used grids that extend over 120
228
+ a.u. in polarization direction and 36 a.u. in the trans-
229
+ verse directions. To suppress reflection of the wavefunc-
230
+ tions at the boundary of the grid an imaginary absorbing
231
+ potential has been applied.
232
+ B.
233
+ High-order harmonic generation from an
234
+ ensemble of aligned molecules
235
+ High-order harmonic generation is determined through
236
+ the Fourier transform of the laser induced dipole moment
237
+ in the target. Within the TDDFT formalism, the laser
238
+
239
+ 3
240
+ FIG. 1: Configuration of pump (aligning) pulse in the y − z
241
+ plane, probe (driver) pulse along the ˆz-direction and molecu-
242
+ lar axis.
243
+ induced dipole moment is given by:
244
+ dtot =
245
+ n
246
+
247
+ k=1
248
+ dk,
249
+ (7)
250
+ where dk is the contribution to the dipole moment from
251
+ the kth Kohn-Sham orbital,
252
+ dk = ⟨φk(r, t)|r|φk(r, t)⟩ .
253
+ (8)
254
+ The HHG spectrum is then found using the Fourier trans-
255
+ form of the dipole moment, d(ω):
256
+ P(ω) =
257
+ ω4
258
+ 12πϵ0c3 d(ω) · d∗(ω) .
259
+ (9)
260
+ For the molecules studied below, the laser induced dipole
261
+ moment has two components, parallel (d||) and perpen-
262
+ dicular (d⊥) with respect to the direction of the electric
263
+ field of the driving laser. The ellipticity of a given har-
264
+ monic is then determined by:
265
+ ϵ =
266
+
267
+ 1 + r2 −
268
+
269
+ 1 + 2r2 cos(2δ) + r4
270
+ 1 + r2 +
271
+
272
+ 1 + 2r2 cos(2δ) + r4
273
+ (10)
274
+ where
275
+ r = |d⊥(ω)|
276
+ |d||(ω)|
277
+ (11)
278
+ is the amplitude ratio and
279
+ δ = arg[d⊥(ω)] − arg[d||(ω)]
280
+ (12)
281
+ is the relative phase between the two components. Maxi-
282
+ mum ellipticity, i.e. circular polarization, occurs for r = 1
283
+ and δ = π.
284
+ In the experimental observations of the ellipticity in
285
+ high-order harmonic generation of linear molecules, the
286
+ molecules are often aligned by a pump laser pulse. The
287
+ distribution of the alignment, achieved in the experi-
288
+ ments, is typically measured via ⟨cos2(θ)⟩, where θ is
289
+ the angle between the polarization direction of the pump
290
+ laser and the internuclear axis of the molecule (see Fig.
291
+ 1).
292
+ In our simulations we have accounted for the ex-
293
+ perimental alignment of molecular ensemble by solving
294
+ the Kohn-Sham equations for different alignment angles.
295
+ For each angle, we obtained the parallel and perpendicu-
296
+ lar components of the dipole moment and then averaged
297
+ them using the reported alignment distributions.
298
+ III.
299
+ RESULTS
300
+ In this section we present our results for the po-
301
+ larization and ellipticity of high-order harmonics from
302
+ molecules H+
303
+ 2 , H2, and CO2.
304
+ The data for the differ-
305
+ ent molecules provide us with the opportunity to com-
306
+ pare our results with those from other theoretical anal-
307
+ ysis (for the one-electron system H+
308
+ 2 ) and demonstrate
309
+ how multielectron effects and inner valence shell contri-
310
+ butions influence the harmonics’ ellipticity for the larger
311
+ molecules.
312
+ A.
313
+ Harmonic generation from H+
314
+ 2 and H2
315
+ In order to test our numerical calculations, we first
316
+ present results for the one-electron system H+
317
+ 2 . In Fig.
318
+ 2 we show results for the amplitudes (upper panel) and
319
+ the phase difference (lower panel) for the 57th harmonics
320
+ emitted from H+
321
+ 2 as a function of the alignment angle be-
322
+ tween the molecular axis and the polarization direction of
323
+ a driving laser pulse at 800 nm and 3×1014 W/cm2 with
324
+ a pulse duration of 30 fs. The laser parameters are cho-
325
+ sen to be the same as in a recent work by Son et al. [26],
326
+ who studied the ellipticity of high-order harmonic gen-
327
+ eration from H+
328
+ 2 using the time-dependent generalized
329
+ pseudospectral method. Our results are in good agree-
330
+ ment with those previously reported for the overall shape
331
+ of the components with a minimum at about 50o for the
332
+ parallel component and a phase jump at the same align-
333
+ ment angle. It has been shown before [24–26], that these
334
+ characteristic features are related to the two-center in-
335
+ terference effect occurring in the parallel component.
336
+ In order to get an impression of the influence of multi-
337
+ electron effects on the ellipticity of high-order harmonics,
338
+ we compare results for H+
339
+ 2 (Fig. 3, (a-c)) and H2 (Fig. 3,
340
+ (d-f)) obtained at the same set of laser parameters (800
341
+ nm, 3×1014 W/cm2). In each case we present theoretical
342
+ predictions for four consecutive odd harmonics. For the
343
+ single-electron molecule we observe, in agreement with
344
+ our results in Fig. 2, a maximum close to 1 in the ratio
345
+ of the amplitude in parallel and perpendicular direction
346
+ (a), a rapid change in the phase difference (b) and corre-
347
+
348
+ Probe
349
+ Pump
350
+ aser
351
+ a
352
+ molecule
353
+ 1
354
+ m
355
+ x4
356
+ FIG. 2: Amplitudes of parallel and perpendicular components
357
+ (a) and phase difference (b) of 57th harmonic order of H+
358
+ 2 as
359
+ a function of the alignment angle. Laser parameters: 800 nm,
360
+ 3 × 1014 W/cm2 and 30 fs.
361
+ spondingly a maximum in the ellipticity (c) around the
362
+ alignment angle, at which the interference minimum in
363
+ the specific harmonic occurs. For H2, one would expect
364
+ a similar pattern for the amplitude and the phase dif-
365
+ ference, since both electrons are in the same molecular
366
+ orbital as in the case of H+
367
+ 2 . Indeed, some features in
368
+ the overall trend of the results in Fig. 3 are similar, in
369
+ particular we still note a maximum amplitude ratio (d)
370
+ and a quick phase change (e) at about the same angles
371
+ as for H+
372
+ 2 .
373
+ However, for the ratio we observe a much
374
+ narrower structure and for the lowest harmonic a second
375
+ maximum. On the other hand, the data for the phase dif-
376
+ ference are not as smooth as those for the single-electron
377
+ molecule.
378
+ As a result, we observe a much more com-
379
+ plex pattern for the ellipticity of the harmonics generated
380
+ from H2 (f), although some maxima in the structures still
381
+ occur near the alignment angle for the interference min-
382
+ imum. Thus, the comparison for the simplest molecules
383
+ indicates that the ellipticity of high-order harmonics can
384
+ be strongly influenced by multielectron effects. For larger
385
+ molecules we may therefore expect even more complex
386
+ features in the overall ellipticity patterns, since interfer-
387
+ ences from orbitals with different symmetry as well as
388
+ coupling between different orbitals [27, 36] may play ad-
389
+ ditional role.
390
+ FIG. 3: Comparison of amplitude ratio r (a, d), phase differ-
391
+ ence δ (b, e), and ellipticity (c, f) of high order harmonics from
392
+ H+
393
+ 2 (a-c) and H2 (d-f) as a function of the alignment angle:
394
+ 27th (solid lines), 29th (dashed lines), 31st (dashed-dotted
395
+ lines), and 33th harmonic (dotted lines). Laser parameters as
396
+ in Fig. 2.
397
+ FIG. 4:
398
+ TDDFT results for the intensity ratio of perpendic-
399
+ ular to parallel component of four consecutive harmonics in
400
+ CO2 as a function of the angle between the pump and the 30
401
+ fs probe laser pulse at 800 nm and 1.5 × 1014 W/cm2: 17th
402
+ (red line), 19th (blue line), 21st (green line) and 23rd har-
403
+ monic (black line). For each angle, the experimental reported
404
+ alignment distribution [20] was considered in the calculations.
405
+ B.
406
+ Harmonic generation from CO2
407
+ Next, we analyze the results of our calculations for the
408
+ ellipticity in the harmonic generation from the more com-
409
+ plex but linear triatomic molecule CO2, which has been
410
+ also studied experimentally [20].
411
+ In order to compare
412
+ with the experimental data, we have obtained the inten-
413
+
414
+ 0.008
415
+ Amplitude [arb. units]
416
+ 0.007
417
+ (a)
418
+ 0.006
419
+ 0.005
420
+ 0.004
421
+ 0.003
422
+ 0.002
423
+ 0.001
424
+ 0.000
425
+ 10
426
+ 30
427
+ 50
428
+ 70
429
+ 90
430
+ Alignment angle [degree]
431
+ 1.0
432
+ (b)
433
+ Phase difference [π]
434
+ 0.5
435
+ 0.0
436
+ 0.5
437
+ -1.0
438
+ 10
439
+ 30
440
+ 50
441
+ 70
442
+ 90
443
+ Alignment angle [degree]2.5
444
+ 16
445
+ (a)
446
+ (d)
447
+ amplitude ratio
448
+ 2.0
449
+ amplitude ratio
450
+ 12
451
+ 1.5
452
+ 8
453
+ 1.0
454
+ 4
455
+ 0.5
456
+ 0.0
457
+ 0
458
+ 10
459
+ 30
460
+ 50
461
+ 70
462
+ 90
463
+ 10
464
+ 30
465
+ 50
466
+ 70
467
+ 90
468
+ 1.0
469
+ 1.0
470
+ (b)
471
+ (e)
472
+ 0.5
473
+ 0.5
474
+ 0.0
475
+ 0.0
476
+ -0.5
477
+ -0.5
478
+ -1.0
479
+ -1.0
480
+ 10
481
+ 30
482
+ 50
483
+ 70
484
+ 90
485
+ 10
486
+ 30
487
+ 50
488
+ 70
489
+ 90
490
+ 1.0
491
+ 1.0
492
+ (c)
493
+ (f)
494
+ 0.8
495
+ 0.8
496
+ ellipticity
497
+ ellipticity
498
+ 0.6
499
+ 0.6
500
+ 0.4
501
+ 0.4
502
+ 0.2
503
+ 0.2
504
+ 0.0
505
+ 0.0
506
+ 10
507
+ 30
508
+ 50
509
+ 70
510
+ 90
511
+ 10
512
+ 30
513
+ 50
514
+ 70
515
+ 90
516
+ alignment angle (degree)
517
+ alignment angle (degree)0.075
518
+ H17
519
+ -H19
520
+ H21
521
+ Intensity Ratio
522
+ 0.05
523
+ 一H23
524
+ 0.025
525
+ -100
526
+ -50
527
+ 0
528
+ 50
529
+ 100
530
+ Pump-probe
531
+ angle (degree)5
532
+ FIG. 5:
533
+ Comparison of results for the ellipticity of high-order
534
+ harmonics as a function of alignment angle for CO2: without
535
+ (a) and with averaging (b). Laser parameters as in Fig. 4.
536
+ sity ratio of the perpendicular to parallel component of
537
+ the harmonic emission as a function of the angle between
538
+ the pump and probe laser pulse.
539
+ For each orientation
540
+ angle considered, we have taken into account the exper-
541
+ imentally reported alignment distribution by performing
542
+ an average over the simulation results for the respective
543
+ alignment angles in the distribution. Our results in Fig. 4
544
+ show a rather small intensity ratio and, hence, relatively
545
+ small ellipticity with a maximum at about a relative an-
546
+ gle of about 60o between polarization direction of pump
547
+ and probe pulse for each of the harmonics studied exper-
548
+ imentally. The absolute values as well as the position of
549
+ the maxima are in very good agreement with the obser-
550
+ vations by Zhao et al. [20]. The observed and calculated
551
+ rather weak perpendicular component of the harmonics
552
+ in CO2 is in contrast to results for N2, for which both
553
+ experiment [20, 21] and TDDFT [27] as well as other
554
+ calculations [21, 24, 26] show a strong ellipticity for the
555
+ emitted harmonics at certain alignment angles.
556
+ Part of the explanation for the weak ellipticity is due
557
+ to the ensemble angle average effect, which reduces the
558
+ overall ellipticity, as observed before in N2 [27]. The ef-
559
+ fect can be seen from the comparison of the harmonics
560
+ ellipticity as a function of the alignment angle without (a)
561
+ and with (b) average in Fig. 5. It is clearly seen that,
562
+ in particular for the lower-order harmonics (below 15th
563
+ harmonics), without averaging there is a strong elliptic-
564
+ ity for certain alignment angles which disappears after
565
+ alignment average is taken into account. In contrast, for
566
+ the experimentally reported data in the range of 17th to
567
+ 23rd harmonics the averaging process does have a smaller
568
+ effect only.
569
+ In this latter range of harmonics from CO2 the main
570
+ origin for the weak ellipticity is actually the role of mul-
571
+ tielectron effects involving contributions from several or-
572
+ bitals. In order to analyze these contributions, we com-
573
+ pare in Fig. ?? the ellipticity of the harmonic response
574
+ from the HOMO only (a) with those when adding subse-
575
+ quently the contributions from the inner valence orbitals
576
+ up to HOMO-5 (f). The comparison shows that the ellip-
577
+ ticity of high-order harmonics from CO2 is influenced by
578
+ the six valence orbitals considered. While the ellipticity
579
+ of 17th to 23rd harmonics generated from the HOMO is
580
+ rather large for certain alignments angles, the ellipticity
581
+ FIG. 6:
582
+ Ellipticity of high-order harmonics as a func-
583
+ tion
584
+ of
585
+ alignment
586
+ angle
587
+ for
588
+ CO2.
589
+ Starting
590
+ with
591
+ the
592
+ results
593
+ from
594
+ HOMO
595
+ only
596
+ (a):
597
+ (1πg)4,
598
+ contribu-
599
+ tions
600
+ from
601
+ inner
602
+ valence
603
+ orbitals
604
+ are
605
+ added
606
+ subse-
607
+ quently
608
+ in
609
+ the
610
+ other
611
+ panels:
612
+ (b)
613
+ (3σu)2(1πg)4,
614
+ (c)
615
+ (1πu)4(3σu)2(1πg)4,
616
+ (d)
617
+ (4σg)2(1πu)4(3σu)2(1πg)4,
618
+ (e)
619
+ (2σu)2(4σg)2(1πu)4(3σu)2(1πg)4,
620
+ (f)
621
+ (3σg)2(2σu)2(4σg)2
622
+ (1πu)4(3σu)2(1πg)4. Laser parameters as in Fig. 4.
623
+ gradually gets weaker as more contributions are added.
624
+ In contrast, for harmonics around the cutoff there re-
625
+ mains a strong ellipticity at some alignment angles.
626
+ The ellipticity of higher-order harmonics at certain
627
+ alignment angles from the HOMO (3πg) can be under-
628
+ stood based on the two-center interference effect, similar
629
+ as in the case of H+
630
+ 2 and H2 above. The importance of
631
+ such orbital structure effect for the harmonic generation
632
+ from the HOMO of CO2 has been pointed out before
633
+ [24]. The strong contributions from the inner valence or-
634
+ bitals originate on a variety of effects. Both, HOMO-1
635
+ (2σu) and HOMO-2 (1πu) have a different orbital symme-
636
+ try than the HOMO of CO2. Therefore, ionization and,
637
+ hence, harmonic generation, from HOMO is suppressed
638
+ due to destructive interference at alignment angles of 0◦
639
+ and 90◦ while it is at maximum around 45◦ [45]. In con-
640
+ trast, the ionization rate is largest at 0◦ for HOMO-1 and
641
+ 90◦ for HOMO-2.
642
+ Consequently, high-order harmonic
643
+ generation from these two orbitals contributes strongly
644
+ close to alignment angles at which the signal from the
645
+ HOMO is weakest, despite the fact that the ionization
646
+
647
+ 30
648
+ 0.9
649
+ 25
650
+ 0.8
651
+ 0.7
652
+ Harmonic order
653
+ 20
654
+ 0.6
655
+ 0.5
656
+ 15
657
+ 0.4
658
+ 10
659
+ 0.3
660
+ 0.2
661
+ 5
662
+ 0.1
663
+ 0
664
+ 20
665
+ 40
666
+ 60
667
+ 80
668
+ Angle(Degrees25
669
+ 0.9
670
+ 20
671
+ 0.8
672
+ 0.7
673
+ Harmonic order
674
+ 15
675
+ 0.6
676
+ 0.5
677
+ 0.4
678
+ 10
679
+ 0.3
680
+ 0.2
681
+ 5
682
+ 0.1
683
+ 0
684
+ 20
685
+ 40
686
+ 60
687
+ 80
688
+ Angle (Degrees)25
689
+ 25
690
+ (a)
691
+ (d)
692
+ Harmonic order
693
+ 20
694
+ 0.8
695
+ Harmonicorder
696
+ 20
697
+ 0.8
698
+ 15
699
+ 0.6
700
+ 15
701
+ 0.6
702
+ 10
703
+ 0.4
704
+ 10
705
+ 0.4
706
+ 0.2
707
+ 0.2
708
+ 0
709
+ 20
710
+ 40
711
+ 60
712
+ 80
713
+ 0
714
+ 20
715
+ 40
716
+ 60
717
+ 80
718
+ Angle[degrees]
719
+ Angle[degrees]
720
+ 25
721
+ 25
722
+ (e)
723
+ (b)
724
+ 0.8
725
+ 20
726
+ 0.8
727
+ Harmonic
728
+ 15
729
+ 0.6
730
+ 15
731
+ 0.6
732
+ 10
733
+ 0.4
734
+ 10
735
+ 0.4
736
+ 0.2
737
+ 0.2
738
+ 0
739
+ 0
740
+ 20
741
+ 40
742
+ 60
743
+ 80
744
+ 0
745
+ 20
746
+ 40
747
+ 60
748
+ 80
749
+ Angledegrees
750
+ Angle [degrees]
751
+ 25
752
+ 25
753
+ (c)
754
+ (f)
755
+ Harmonic order
756
+ 20
757
+ 0.8
758
+ order
759
+ 20
760
+ 0.8
761
+ 15
762
+ 0.6
763
+ Harmonic
764
+ 15
765
+ 0.6
766
+ 10
767
+ 0.4
768
+ 10
769
+ 0.4
770
+ 0.2
771
+ 0.2
772
+ 0
773
+ 20
774
+ 40
775
+ 60
776
+ 80
777
+ 0
778
+ 20
779
+ 40
780
+ 60
781
+ 80
782
+ Angle[degrees
783
+ Angle[degrees6
784
+ FIG. 7: Rotational averaging assuming distribution 1.The rest
785
+ of the notation and parameters as in fig.6.
786
+ potential for the inner valence orbitals is smaller than
787
+ that of the HOMO.
788
+ As for the other inner valence orbitals, that have an
789
+ even higher ionization potential, we have found that these
790
+ are either strongly coupled to one of the higher lying
791
+ states or among each other by the driving field. In the
792
+ case of the HOMO-3 state (2σg), the projection onto the
793
+ HOMO-1 state is shown in Fig. 9(a). We observe a strong
794
+ coupling driven by the field although the frequency is
795
+ non-resonant.
796
+ This explains the significant change in
797
+ the ellipticity pattern upon inclusion of the HOMO-3
798
+ state (Fig. ??(d)). Finally, HOMO-4 and HOMO-5 states
799
+ slightly contribute to the 17th to 23rd harmonic genera-
800
+ tion at the given parameters and, hence, to the ellipticity
801
+ pattern, since these two orbitals are coupled with each
802
+ other, leading to a population transfer of about 40% (see
803
+ Fig. 9(b)).
804
+ To summarize, our results obtained within the time-
805
+ dependent density functional theory indicate that high-
806
+ order harmonic generation from CO2 is influenced by
807
+ multielectron effects with contributions from a significant
808
+ number of inner-valence orbitals, besides the contribution
809
+ from the HOMO. The harmonic emission from these or-
810
+ bitals is strongest at different alignment angles due to
811
+ interference effects arising from the specific orbital struc-
812
+ tures and there is a strong laser driven coupling between
813
+ certain orbitals. As a result, the overall ellipticity of the
814
+ FIG. 8: Rotational averaging assuming disributions 2. The
815
+ rest of the notation and parameters as in fig.6.
816
+ higher-order harmonics is rather small, except for the
817
+ cutoff harmonics. The partial alignment and the related
818
+ averaging of the results for different orientation angles
819
+ further diminishes the ellipticity.
820
+ Acknowledgments
821
+ This work was supported by the U.S. National Science
822
+ Foundation (Grants Nos. Grant No. PHY-1734006 and
823
+ Grant No. PHY-2110628). This work utilized the Sum-
824
+ mit supercomputer, which was supported by the U.S. Na-
825
+ tional Science Foundation and the University of Colorado
826
+ Boulder.
827
+ [1] A. McPherson, G. Gibson, H. Jara, T.S. Luk, I.A. McIn-
828
+ tyre, K. Boyer and C.K. Rhodes, J. Opt. Soc. Am. B 4,
829
+ 595 (1987).
830
+ [2] M. Ferray, A. L’Huillier, X.F. Li, L.A. Lompre, G. Main-
831
+
832
+ 25
833
+ 25
834
+ a
835
+ (d)
836
+ Harmonic order
837
+ 20
838
+ 0.8
839
+ Harmonicorder
840
+ 20
841
+ 0.8
842
+ 5
843
+ 0.6
844
+ 15
845
+ 0.6
846
+ 10
847
+ 0.4
848
+ 10
849
+ 0.4
850
+ 0.2
851
+ 5
852
+ 0.2
853
+ 0
854
+ 0
855
+ 20
856
+ 40
857
+ 60
858
+ 80
859
+ 0
860
+ 20
861
+ 40
862
+ 60
863
+ 80
864
+ Angle[degrees]
865
+ Angle[degrees
866
+ 25
867
+ 25
868
+ (b)
869
+ e
870
+ 0.8
871
+ Harmonic order
872
+ 20
873
+ 0.8
874
+ 15
875
+ 0.6
876
+ 15
877
+ 0.6
878
+ 10
879
+ 0.4
880
+ 10
881
+ 0.4
882
+ 0.2
883
+ 0.2
884
+ 0
885
+ 20
886
+ 40
887
+ 60
888
+ 80
889
+ 0
890
+ 20
891
+ 40
892
+ 60
893
+ 80
894
+ Angle [degrees]
895
+ Angle[degrees
896
+ 25
897
+ 25
898
+ (c)
899
+ (f)
900
+ 0.8
901
+ Harmonicorder
902
+ 20
903
+ 0.8
904
+ Harmonic
905
+ 15
906
+ 0.6
907
+ 15
908
+ 0.6
909
+ 10
910
+ 0.4
911
+ 10
912
+ 0.4
913
+ 5
914
+ 0.2
915
+ 5
916
+ 0.2
917
+ 0
918
+ 20
919
+ 40
920
+ 60
921
+ 80
922
+ 0
923
+ 20
924
+ 40
925
+ 60
926
+ 80
927
+ Angle[degrees
928
+ Angle[degrees25
929
+ 25
930
+ a
931
+ (d)
932
+ Harmonic order
933
+ 20
934
+ 0.8
935
+ 20
936
+ 0.8
937
+ Harmonic orde
938
+ 15
939
+ 0.6
940
+ 15
941
+ 0.6
942
+ 10
943
+ 0.4
944
+ 10
945
+ 0.4
946
+ 0.2
947
+ 5
948
+ 0.2
949
+ 0
950
+ 0
951
+ 20
952
+ 40
953
+ 60
954
+ 80
955
+ 0
956
+ 20
957
+ 40
958
+ 60
959
+ 80
960
+ Angle[degrees]
961
+ Angle[degrees]
962
+ 25
963
+ 25
964
+ (b)
965
+ le
966
+ 0.8
967
+ 20
968
+ 0.8
969
+ 15
970
+ 0.6
971
+ 15
972
+ 0.6
973
+ 10
974
+ 0.4
975
+ 10
976
+ 0.4
977
+ 0.2
978
+ 0.2
979
+ 0
980
+ 0
981
+ 20
982
+ 40
983
+ 60
984
+ 80
985
+ 0
986
+ 20
987
+ 40
988
+ 60
989
+ 80
990
+ Angle [degrees]
991
+ Angle[degrees
992
+ 25
993
+ 25
994
+ (c)
995
+ (f)
996
+ 0.8
997
+ e 20
998
+ 0.8
999
+ 0.6
1000
+ Harmonic
1001
+ 15
1002
+ 0.6
1003
+ 10
1004
+ 0.4
1005
+ 10
1006
+ 0.4
1007
+ 5
1008
+ 0.2
1009
+ 0.2
1010
+ 0
1011
+ 20
1012
+ 40
1013
+ 60
1014
+ 80
1015
+ 0
1016
+ 20
1017
+ 40
1018
+ 60
1019
+ 80
1020
+ Angle[degrees
1021
+ Angle[degrees]7
1022
+ FIG. 9:
1023
+ Projection of coupled inner valence orbitals (a)
1024
+ HOMO-3 (4σg) to HOMO-1 (3πu) (a) and (b) HOMO-5 (3σg)
1025
+ to HOMO-4 (2πu) (b) for an alignment angle of 20o. Laser
1026
+ parameters as in Fig. 4.
1027
+ fray and C. Manus, J. Phys. B 21, L31 (1988).
1028
+ [3] T. Popmintchev, M.-C. Chen, D. Popmintchev, P. Arpin,
1029
+ S. Brown, S. Alisauskas, G. Andriukaitis, T. Balciu-
1030
+ nas, O.D. M¨ucke, A. Pugzlys, A. Baltuska, B. Shim,
1031
+ S.E. Schrauth, A. Gaeta, C. Hernandez-Garcia, L. Plaja,
1032
+ A. Becker, A. Jaron-Becker, M.M. Murnane and H.C.
1033
+ Kapteyn H C, Science 91, 1287 (2012).
1034
+ [4] M. Hentschel, R. Kienberger, C. Spielmann, G.A. Reider,
1035
+ N. Milosevic, T. Brabec, P. Corkum, U. Heinzmann, M.
1036
+ Drescher and F. Krausz F, Nature 414, 509 (2001).
1037
+ [5] P.M. Paul, E.S. Toma, P. Breger, G. Mullot, F. Auge,
1038
+ Ph. Balcou, H.G. Muller and P. Agostini, Science 292,
1039
+ 1689 (2001).
1040
+ [6] M. Lein, N. Hay, R. Velotta, J.P. Marangos and P.L.
1041
+ Knight, Phys. Rev. Lett. 88, 183903 (2002).
1042
+ [7] J. Itatani, J. Levesque, D. Zeidler, H. Niikura, H. Pepin,
1043
+ J.C. Kieffer, P.B. Corkum and D.M. Villeneuve, Nature
1044
+ 432, 867 (2004).
1045
+ [8] H.J. W¨orner, H. Niikura, J.B. Bertrand, P.B. Corkum
1046
+ and D.M. Villeneuve, Phys. Rev. Lett. 102, 103901
1047
+ (2009).
1048
+ [9] S. Haessler, J. Caillat, W. Boutu, C. Giovanetti-Teixeira,
1049
+ T. Ruchon, T. Auguste, Z. Diveki, P. Breger, A. Maquet,
1050
+ B. Carre, R. Taieb and P. Salieres, Nat. Phys. 6, 200
1051
+ (2010).
1052
+ [10] C. Vozzi, M. Negro, F. Calegari, G. Sansone, M. Nisoli,
1053
+ S. De Silvestri and S. Stagira, Nat. Phys. 7, 822 (2011).
1054
+ [11] S.B. Schoun, R. Chirla, J. Wheeler, C. Roedig, P. Agos-
1055
+ tini, L.F. DiMauro, K.J. Schafer and M.B. Gaarde, Phys.
1056
+ Rev. Lett. 112, 153001 (2014).
1057
+ [12] S. Baker, J.S. Robinson, C.A. Haworth, H. Teng, R.A.
1058
+ Smith, C.C. Chirila, M. Lein, J.W.G. Tisch and J.P.
1059
+ Marangos, Science 312, 424 (2006).
1060
+ [13] W. Li, X. Zhou, R. Lock, S. Patchkovskii, A. Stolow, H.C.
1061
+ Kapteyn and M.M. Murnane, Science 322, 1207 (2008).
1062
+ [14] C.I. Blaga, J. Xu, A.D. DiChiara, E. Sistrunk, K. Zhang,
1063
+ P. Agostini, T.A. Miller, L.F. DiMauro and C.D. Lin,
1064
+ Nature 483, 194 (2012).
1065
+ [15] M. Miller, A. Jaron-Becker and A. Becker, Phys Rev A
1066
+ 93 013406 (2016).
1067
+ [16] H.J. W¨orner,
1068
+ J.B. Bertrand,
1069
+ D.V. Kartashov,
1070
+ P.B.
1071
+ Corkum and D.M. Villeneuve, Nature 466, 604 (2010).
1072
+ [17] K.J. Schafer, B. Yang, L.F. DiMauro and K.C. Kulander,
1073
+ Phys. Rev. Lett. 70, 1599 (1993).
1074
+ [18] P.B. Corkum, Phys. Rev. Lett. 71, 1994 (1993).
1075
+ [19] M. Lewenstein, Ph. Balcou, M.Yu. Ivanov, A. L’Huillier
1076
+ and P.B. Corkum, Phys. Rev. A 49, 2117 (1994).
1077
+ [20] X. Zhao, R. Lock, N. Wagner, W. Li, H.C. Kapteyn and
1078
+ M.M. Murnane, Phys. Rev. Lett. 102, 073902 (2009).
1079
+ [21] Y. Mairesse, J. Higuet, N. Dudovich, D. Shafir, B. Fabre,
1080
+ E. Mevel, E. Constant, S. Patchkovskii, Z. Walters,
1081
+ M.Yu. Ivanov and O. Smirnova, Phys. Rev. Lett. 104,
1082
+ 213601 (2010).
1083
+ [22] A. Etches, C.B. Madsen and L.B. Madsen, Phys. Rev. A
1084
+ 81, 013409 (2010).
1085
+ [23] S. Ramakrishna, P.A.J. Sherratt, A.D. Dutoi and T. Sei-
1086
+ deman, Phys. Rev. A 81, 021802(R) (2010).
1087
+ [24] A.-T. Le, R.R. Lucchese and C.D. Lin, Phys. Rev. A 82,
1088
+ 023814 (2010).
1089
+ [25] E.V. van der Zwan and M. Lein Phys. Rev. A 82, 033405
1090
+ (2010).
1091
+ [26] S.-K. Son, D.A. Telnov and S.I. Chu, Phys. Rev. A 82,
1092
+ 043829 (2010).
1093
+ [27] Y. Xia and A. Jaron-Becker, Opt. Lett. 39, 1461 (2014).
1094
+ [28] G.N. Gibson, R.R. Freeman and T.J. McIllrath, Phys.
1095
+ Rev. Lett. 67, 1230 (1991).
1096
+ [29] A. Talebpour, S. Petit and S.L. Chin, Opt. Commun.
1097
+ ,171 285 (1999).
1098
+ [30] A. Becker, A.D. Bandrauk and S.L. Chin, Chem. Phys.
1099
+ 343, 345 (2001).
1100
+ [31] A. Jaron-Becker, A. Becker and F.H.M. Faisal, Phys.
1101
+ Rev. A 69, 023410 (2004).
1102
+ [32] B. McFarland, J.P. Farrell, P.H. Bucksbaum and M.
1103
+ G¨uhr, Science 322, 1232 (2008).
1104
+ [33] A.T. Le, R.R. Lucchese and C.D. Lin, J. Phys. B: At.
1105
+ Mol. Opt. Phys. 42, 211001 (2009).
1106
+ [34] O. Smirnova, Y. Mairesse, S. Patchkovskii, N. Dudovich,
1107
+ D. Villeneuve, P.B. Corkum and M.Yu. Ivanov, Nature
1108
+ 460, 972 (2009).
1109
+ [35] Z. Diveki, A. Camper, S. Haessler, T. Auguste, T. Ru-
1110
+ chon, B. Carre, P. Salieres, R. Guichard, J. Caillat, A.
1111
+ Maquet and R. Taieb, New J. Phys. 14, 023062 (2012).
1112
+ [36] Y. Xia and A. Jaron-Becker, Opt. Express 24, 4689
1113
+ (2016).
1114
+ [37] D.L.
1115
+ Smith
1116
+ and
1117
+ G.N.
1118
+ Gibson,
1119
+ Phys.
1120
+ Rev.
1121
+ A
1122
+ 97,
1123
+ 021401(R) (2018).
1124
+ [38] E. Runge and E.K.U. Gross, Phys. Rev. Lett. 52, 997
1125
+ (1984).
1126
+ [39] E.S. Gustavo and N.S. Viktor, in Theory and Appli-
1127
+ cations of Computational Chemistry (Elsevier, 2005) p
1128
+ 669ff.
1129
+ [40] R. van Leeuwen and E.J. Baerends, Phys. Rev. A 49,
1130
+ 2421 (1994).
1131
+ [41] X. Chu and S.I. Chu, Phys. Rev. A 63, 023411 (2001).
1132
+ [42] X. Chu and S.I. Chu, Phys. Rev. A 64, 063404 (2001).
1133
+ [43] M.A.L. Marques, A. Castro, G. F. Bertsch, and A. Rubio,
1134
+
1135
+ 1.0
1136
+ (a)
1137
+ 0.8
1138
+ Projection
1139
+ 0.6
1140
+ 0.4
1141
+ 0.2
1142
+ 0.0
1143
+ 0
1144
+ 5
1145
+ 10
1146
+ 15
1147
+ 20
1148
+ 25
1149
+ 30
1150
+ 35
1151
+ Time [fs]
1152
+ 1.0
1153
+ (b)
1154
+ 0.8
1155
+ Projection
1156
+ 0.6
1157
+ 0.4
1158
+ 0.2
1159
+ 0.0
1160
+ 0
1161
+ 5
1162
+ 10
1163
+ 15
1164
+ 20
1165
+ 25
1166
+ 30
1167
+ 35
1168
+ Time [fs]8
1169
+ Comput. Phys. Commun. 151, 60 (2003).
1170
+ [44] A. Castro, H. Appel, M. Oliveira, C.A. Rozzi, X. An-
1171
+ drade, F. Lorenzen, M.A.L. Marques, E.K.U. Gross, and
1172
+ A. Rubio, Phys. Stat. Sol. B 243, 2465 (2006).
1173
+ [45] A.-T. Le, R.R. Lucchese, M.T. Lee and C.D. Lin, Phys.
1174
+ Rev. Lett. 102, 203001 (2009).
1175
+
LtAyT4oBgHgl3EQfgPig/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
QtE3T4oBgHgl3EQfDAk3/content/2301.04281v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec0d46acbb62950030c1782c3c9961aed5775a8411032bd4c974ee682aef601d
3
+ size 2121406
TdE5T4oBgHgl3EQfAQ7r/content/tmp_files/2301.05378v1.pdf.txt ADDED
@@ -0,0 +1,2484 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Dynamical Signatures of Liouvillian Flat Band
2
+ Yu-Guo Liu1 and Shu Chen1, 2, 3, ∗
3
+ 1Beijing National Laboratory for Condensed Matter Physics,
4
+ Institute of Physics, Chinese Academy of Sciences, Beijing 100190, China
5
+ 2School of Physical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China
6
+ 3Yangtze River Delta Physics Research Center, Liyang, Jiangsu 213300, China
7
+ (Dated: January 16, 2023)
8
+ Although flat-band structures have attracted intensive studies in condensed matter and optical physics due to
9
+ their eigenstates exhibiting huge degeneracy and allowing for the localization of wave packet, it is not clear how
10
+ the flat band of Liouvillian influences the relaxation dynamics of open quantum systems. To this end, we study
11
+ the dynamical signatures of Liouvillian flat band in the scheme of Lindblad master equation. Considering a chain
12
+ model with gain and loss, we demonstrate three kinds of band dispersion of Liouvillian: flat bland, dispersionless
13
+ only in the real part and imaginary part, and capture their dynamical signatures: when the rapidity spectrum of
14
+ Liouvillian is flat, the particle numbers in different sites relax to its steady state value with the same decay rate;
15
+ when the real or imaginary part of rapidity spectrum is dispersionless, the relaxation behaviors have oscillating
16
+ or forked characteristics. We also unveil that the Liouvillian flat band can lead to dynamical localization, which
17
+ is characterized by the halt of propagation of a local perturbation on the steady state.
18
+ Introduction.— Band structure of a Hamiltonian plays an
19
+ important role in understanding the motion of particles in pe-
20
+ riodic crystals. Usually, special band structures may give rise
21
+ to exotic quantum phenomena, for example, low-energy ex-
22
+ citations of electrons on a linear dispersive band in graphene
23
+ behave like massless Dirac fermions [1, 2]. Another instance
24
+ is the flat band (FB) in which all electrons carry the same en-
25
+ ergy regardless of their momentum. Due to the dispersionless
26
+ band structure, particles in FB have arbitrarily large effective
27
+ mass, so they will be localized in real space. Especially, in
28
+ strongly correlated systems, heavy degeneracy and zero ki-
29
+ netic energy in FB can increase density of electronic states
30
+ and highlight Coulomb interaction, leading to rich many-body
31
+ phenomena [3–5].
32
+ In open quantum systems, dynamics of density matrix ρ is
33
+ described by Lindblad master equation (LME) under Born-
34
+ Markov approximation [6–8]:
35
+
36
+ dt = L(ρ) := −i[H, ρ] +
37
+
38
+ µ
39
+
40
+ LµρL†
41
+ µ − 1
42
+ 2{L†
43
+ µLµ, ρ}
44
+
45
+ ,
46
+ (1)
47
+ where L is called the Liouvillian superoperator, H is the
48
+ Hamiltonian of system, and Lµ are Lindblad operators which
49
+ reflect the coupling between system and environment. The
50
+ Planck constant ℏ is set to unity throughout this Letter. There
51
+ have been several methods developed to obtain the spectrum
52
+ of L, especially for quadratic systems [9–15]. In Ref.[15],
53
+ a route for realizing dispersionless bands is proposed based
54
+ on the underlying mechanism with the emergence of a dis-
55
+ sipationless dark space. Generally speaking, the short-time
56
+ dynamics is related to the Liouvillian eigenvalues with large
57
+ modulus of the real part, whereas the long-time relaxation to
58
+ the smallest modulus beyond zero (the so called Liouvillian
59
+ gap) [16–22]. However, how the structure of Liouvillian, es-
60
+ pecially the Liouvillian flat band (LFB), influences dynamics
61
+ is still a subtle and unexplored question.
62
+ In this Letter, we focus on the dynamics of open quan-
63
+ tum systems with LFB. In comparison with the real spectrum
64
+ of Hamiltonian system, the Liouvillian spectrum is complex,
65
+ and thus the corresponding rapidity spectrum can exhibit more
66
+ rich structures with dispersionless band in both imaginary and
67
+ real part or either of them. To make our study concrete, we
68
+ shall first apply a geometrically intuitive method to construct
69
+ lattice with correlated gain and loss, which supports LFB, and
70
+ explore the generality of dynamical signatures associated with
71
+ the structure of Liouvillian spectrum. We show that the ra-
72
+ pidity spectra from Liouvillian and damping-matrix spectra
73
+ of correlation functions have the same dispersion characteris-
74
+ tics, which lead to different signatures of damping dynamics
75
+ of local particle number distribution: oscillating, forked, syn-
76
+ chronous damping are related to the band dispersionless only
77
+ in imaginary part, real part and in both parts, respectively. Fur-
78
+ thermore, we exactly solve the model and show that the LFB
79
+ can induce dynamical localization, which is characterized by
80
+ the halt of the propagation of a local perturbation on the non-
81
+ equilibrium steady state (NESS).
82
+ Formalism.— The density matrix ρ and Liouvillian super-
83
+ operator L in Eq. (1) can be formally expressed as
84
+ ρ =
85
+
86
+ IJ
87
+ ρIJ|I⟩a⟨J|a, L(ρ) =
88
+
89
+ i j
90
+ Fi(a, a†) ρ Fj(a, a†), (2)
91
+ where a is the set of fermionic annihilation operators i.e. a =
92
+ (a1, a2, · · · ), Fi(a, a†) is a function with variables among a
93
+ and a†, I = (I1, I2, · · · ), J = (J1, J2, · · · ) and
94
+ |I⟩a⟨J|a = (a†
95
+ 1)I1(a†
96
+ 2)I2 · · · (a†
97
+ L)IL|0⟩a⟨0|a(aL)JL · · · (a1)J1, (3)
98
+ where |0⟩a is the vacuum state for all a−fermions. For the
99
+ convenience of analysis and calculation, we map fermionic
100
+ LME into a new representation referred to as C by following
101
+ the method in Ref. [10]:
102
+ ρ → | ρ⟩C =
103
+
104
+ IJ
105
+ ρIJ(a†
106
+ 1)I1 · · · (a†
107
+ L)IL(c†
108
+ 1 ˆP)J1 · · · (c†
109
+ L ˆP)JL |0⟩,
110
+ (4a)
111
+ L → ˆLC =
112
+
113
+ i j
114
+ Fi(a, a†) F T
115
+ j ( ˆPc, c† ˆP),
116
+ (4b)
117
+ arXiv:2301.05378v1 [cond-mat.other] 13 Jan 2023
118
+
119
+ 2
120
+ where c = (c1, c2, · · · ) is the set of annihilation operators
121
+ of c−fermions, which is a one-to-one mapping from a, T
122
+ means matrix transpose, and |0⟩ is the vacuum state of both
123
+ a− and c−fermions.
124
+ ˆP is the parity operator defined by
125
+ ˆP = exp
126
+
127
+ iπ �
128
+ j(a†
129
+ jaj + c†
130
+ jcj)
131
+
132
+ , which is introduced to ensure
133
+ fermionic anticommutation relations between a−fermions and
134
+ c−fermions. Full mapping process is shown in the Supple-
135
+ mental Material [25].
136
+ Model.— We consider a Liouvillian in a periodic chain:
137
+ L(ρ) = −i[H, ρ] + (1 − w)DL(ρ) + (1 + w)DR(ρ),
138
+ (5)
139
+ where H = �
140
+ l J(a†
141
+ l+1al + h.c.), w ∈ [−1, 1], and
142
+ DL(ρ) =
143
+
144
+ l
145
+
146
+ 2AlρA†
147
+ l − A†
148
+ l Alρ − ρA†
149
+ l Al
150
+
151
+ ,
152
+ DR(ρ) =
153
+
154
+ l
155
+
156
+ 2A†
157
+ l ρAl − AlA†
158
+ l ρ − ρAlA†
159
+ l
160
+
161
+ ,
162
+ (6)
163
+ where Al = √γ1a†
164
+ l + √γ2al+1. The operators Al and A†
165
+ l tie the
166
+ gain and loss of neighboring sites together, which could be re-
167
+ alized by optical superlattice with Bose-Einstein condensate
168
+ reservoir [23]. The role of w ∈ [−1, 1] is analogous to the sta-
169
+ tistical distribution from temperature [24]. Mapping Eq. (5)
170
+ into the representation C, we get a ladder model consisting of
171
+ a−fermion chain and c−fermion chain (see the Supplemental
172
+ Material [25]). The L is mapped to ˆL = ˆH + (1 − w) ˆDL + (1 +
173
+ w) ˆDR, where ˆH = �
174
+ l
175
+
176
+ −iJ(a†
177
+ l+1al + h.c.) + iJ(c†
178
+ l+1cl + h.c.)
179
+
180
+ .
181
+ ˆDL and ˆDR are illustrated in Fig. 1 (a) and (b), which have
182
+ leftward and rightward hoppings, respectively, along two di-
183
+ agonals of every plaquette in the ladder. The cross-stitch-type
184
+ hopping is crucial for generating FB because it can form a
185
+ destructive-interference structure, which consists with our ex-
186
+ perience in the FB ladder models [34–37].
187
+ In momentum space,
188
+ ˆL can be expressed in BdG
189
+ form
190
+ as
191
+ ˆL
192
+ =
193
+ 0.5 ˆLk=0 + �π−
194
+ k=0+ ˆLk,
195
+ where
196
+ ˆLk
197
+ =
198
+ (a†
199
+ k c†
200
+ k a−k c−k) Lk (ak ck a†
201
+ −k c†
202
+ −k)T − 4γ and γ = γ1 + γ2.
203
+ Due to parity conservation in ˆL, the operator ˆP can be substi-
204
+ tuted by a constant P which equals 1(−1) when ˆL acts on the
205
+ state with even (odd) fermions. Then we have
206
+ Lk = −i2J cos kσz ⊗ σz − 4 √γ1γ2 cos kPσz ⊗ σx −
207
+ 2γPσy ⊗ σy + 2w
208
+
209
+ (γ2 − γ1)σz ⊗ I + 2 √γ1γ2 sin kσy ⊗ σz
210
+ +i(γ2 − γ1)Pσx ⊗ σy + i2 √γ1γ2 sin kPI ⊗ σx
211
+
212
+ ,
213
+ (7)
214
+ where I and σi are identity and Pauli matrices.
215
+ ˆLk can
216
+ be diagonalized as ˆLk
217
+ = λ−(k)
218
+
219
+ ζ
220
+
221
+ 1(k)ζ1(k) + ζ
222
+
223
+ 4(k)ζ4(k)
224
+
225
+ +
226
+ λ+(k)
227
+
228
+ ζ
229
+
230
+ 2(k)ζ2(k) + ζ
231
+
232
+ 3(k)ζ3(k)
233
+
234
+ , where ζ
235
+
236
+ i(k) and ζj(k
237
+ ′) ful-
238
+ fill anticommutation relations: {ζ
239
+
240
+ i(k), ζj(k
241
+ ′)} = δi jδkk′ and
242
+
243
+
244
+ i(k), ζ
245
+
246
+ j(k
247
+ ′)} = {ζi(k), ζj(k
248
+ ′)} = 0 [9]. The λ±(k) is called
249
+ rapidity spectrum given by λ±(k) = −2γ ± 2mk for both odd
250
+ and even parity [38], where
251
+ mk =
252
+ �����
253
+
254
+ (4γ1γ2 − J2) cos2 k,
255
+ J2 ≤ 4γ1γ2,
256
+ i
257
+
258
+ (J2 − 4γ1γ2) cos2 k,
259
+ J2 > 4γ1γ2.
260
+ (8)
261
+ FIG. 1. ˆDL, ˆDR and ˆL = ˆH + ˆDL + ˆDR are sketched by (a), (b) and
262
+ (c), where the color ovals, straight lines (with or without arrow) and
263
+ wavy lines represent onsite loss, particle hopping and pair production
264
+ and annihilation. The bule, red and orange ovals are corresponding
265
+ to terms (γ1 − γ2)ˆna/c, l − γ1, (γ2 − γ1)ˆna/c, l − γ2 and constant loss
266
+ −γ.
267
+ ˆna/c, l is the particle number operator of a− or c−fermion on
268
+ the site l. Horizontal black wavy lines represent ± √γ1γ2 ˆP(alal+1 +
269
+ h.c.) or ± √γ1γ2 ˆP(clcl+1 +h.c.). The black arrows indicate directional
270
+ hoppings with strength −2 √γ1γ2 ˆP. The bule, red and orange vertical
271
+ wavy lines are corresponding to 2γ1 ˆPa†
272
+ l c†
273
+ l + 2γ2 ˆPclal, 2γ2 ˆPa†
274
+ l c†
275
+ l +
276
+ 2γ1 ˆPclal and 2γ ˆP(a†
277
+ l c†
278
+ l + clal). (d) shows the (c) in even parity and
279
+ under flat band condition, where J = 2 √γ1γ2 = 1. The dashed wavy
280
+ lines indicate the pairing terms have no effect on single particle- or
281
+ hole- excitation on its steady state.
282
+ The λ±(k) is independent with w and we show it in Fig. 2.
283
+ When J2 = 4 √γ1γ2, λ is a FB of k. When J2 < 4 √γ1γ2
284
+ (J2 > 4 √γ1γ2), λ is dispersionless in its imaginary (real) part.
285
+ Especially, in Fig. 2 (c) and (f) the spectrum is pure real, which
286
+ indicates Lk possessing a pseudo-Hermiticity [39–41], while
287
+ in Fig. 2 (a) and (d) the complex spectrum shows the break-
288
+ ing of pseudo-Hermiticity. Since the Liouvillian spectrum is
289
+ obtained by sum of different number of λ±(k), it inherits the
290
+ characteristics of rapidity spectrum, as shown in Fig. 2 (g)∼(i).
291
+ When J2 = 4 √γ1γ2, Liouvillian spectrum consists of some
292
+ highly degenerate discrete points (Fig. 2 (h)), corresponding
293
+ to different occupations of the FB of rapidity spectrum, so we
294
+ call this kind of Liouvillian spectrum as the LFB.
295
+ Two-operator correlation functions.— By making Fourier
296
+ transform, Eq. (5) becomes
297
+ L(ρ) =
298
+ π
299
+
300
+ k=−π
301
+
302
+ −i2J cos k[ˆnk, ρ]+(1−w)DL
303
+ k(ρ)+(1+w)DR
304
+ k (ρ)
305
+
306
+ ,
307
+ (9)
308
+ where DL
309
+ k(ρ) = 2BkρB†
310
+ k − {B†
311
+ kBk, ρ}, DR
312
+ k (ρ) = 2B†
313
+ kρBk −
314
+ {BkB†
315
+ k, ρ} and Bk = √γ1eika†
316
+ k + √γ2a−k.
317
+ We define two-
318
+ operator correlation functions: Gk1, k2 = Tr(a†
319
+ k1ak2ρ), Dk1, k2 =
320
+ Tr(ak1ak2ρ), and D∗
321
+ k1, k2 = Tr(a†
322
+ k2a†
323
+ k1ρ). In terms of the correla-
324
+ tion function vector Ψk1k2 = (Gk1,k2,G−k2,−k1, Dk2,−k1, D∗
325
+ k1,−k2)T,
326
+ the dynamical evolution is governed by the following closed
327
+
328
+ (a)
329
+ (b)
330
+ Y102.
331
+ Y12P
332
+ a
333
+ a
334
+ a
335
+ C
336
+ C
337
+ C
338
+ h121
339
+ (c)
340
+ (d)
341
+ iJ
342
+ 2J
343
+ 2
344
+ 2
345
+ a
346
+ a
347
+ a
348
+ a
349
+ a
350
+ a
351
+ c
352
+ c
353
+ C
354
+ c
355
+ C
356
+ iJ
357
+ iJ
358
+ 2
359
+ 23
360
+ (b)
361
+ (a)
362
+ (d)
363
+ (e)
364
+ (c)
365
+ (f)
366
+ (g)
367
+ (h)
368
+ (i)
369
+ FIG. 2. (a)∼(c) the real part of rapidity spectra λ±(k). (d)∼(f) the
370
+ imaginary part of λ±(k). (g)∼(i) the Liouvillian spectra obtained by
371
+ exactly diagonalizing 6-site lattice with w = 0, J = 1 and γ1 = 0.25
372
+ for all subfigures. γ2 = 0.5 in (a), (d) and (g). γ2 = 1 in (b), (e) and
373
+ (h). γ2 = 1.5 in (c), (f) and (i).
374
+ equation:
375
+ d
376
+ dtΨk1k2 = Xk1k2Ψk1k2 + Vk1k2,
377
+ (10)
378
+ where
379
+ Xk1k2 = −4γI ⊗ I + i2J cos k1σz ⊗ σz − i2J cos k2I ⊗ σz
380
+ + 4 √γ1γ2 cos k1σx ⊗ σz − 4 √γ1γ2 cos k2σy ⊗ σy (11)
381
+ and Vk1k2
382
+ =
383
+ δk1,k2
384
+
385
+ 2γ + 2w(γ2 − γ1), 2γ + 2w(γ2 −
386
+ γ1), i4w √γ1γ2 sin k1, −i4w √γ1γ2 sin k1
387
+ �T.
388
+ The
389
+ damping
390
+ matrix
391
+ Xk1k2
392
+ has
393
+ four
394
+ eigenstates
395
+ which
396
+ fulfill
397
+ the
398
+ equation
399
+ Xk1k2|Γ±±
400
+ k1k2⟩
401
+ =
402
+ Γ±±
403
+ k1k2|Γ±±
404
+ k1k2⟩
405
+ with
406
+ the
407
+ eigenvalues
408
+ given
409
+ by
410
+ Γ±±
411
+ k1k2
412
+ =
413
+ −4γ ±
414
+ 2
415
+
416
+ 4γ1γ2 − J2 �
417
+ (| cos k1| ± | cos k2|sgn(4γ1γ2 − J2))2,
418
+ where
419
+ sgn(x) is a sign function. Γ also has a transition from the
420
+ complex to the real by decreasing J due to the PT −symmetry
421
+ of Xk1k2.
422
+ In the Supplemental Material [25] we show that
423
+ Xk1k2 has higher symmetry than ˆLk, which makes Xk1k2 have a
424
+ similar band structure as ˆLk. In Fig. 3, we see that Γ±±
425
+ k1k2 fully
426
+ inherits the dispersion characteristics of real and imaginary
427
+ part from the rapidity spectra in Fig. 2.
428
+ (b)
429
+ (a)
430
+ (d)
431
+ (e)
432
+ (c)
433
+ (f)
434
+ FIG. 3. (a)∼(c) the real part of Γ±±
435
+ k1k2. (d)∼(f) the imaginary part of
436
+ Γ±±
437
+ k1k2. J = 1 and γ1 = 0.25 are for all subfigures. γ2 = 0.5 in (a) and
438
+ (d). γ2 = 1 in (b) and (e). γ2 = 1.5 in (c) and (f).
439
+ Flat-band damping dynamics.— Damping dynamics dis-
440
+ plays the converging processes from initial state to NESS [45].
441
+ Here, we show that the “flat band” in real or imaginary
442
+ or both parts will effectively influence the damping behav-
443
+ iors in real space.
444
+ We concentrate on the vector Ψl1l2 =
445
+ (Gl1,l2,Gl2,l1, Dl2,l1, D∗
446
+ l1,l2)T consisting of real-space correlation
447
+ functions:
448
+ Gl1, l2 = Tr(a†
449
+ l1al2ρ), Dl1, l2 = Tr(al1al2ρ), D∗
450
+ l1, l2 = Tr(a†
451
+ l2a†
452
+ l1ρ).
453
+ Introduce the deviating expectation of operator ˆO as �
454
+ O(t) =
455
+ ⟨ ˆO⟩(t) − ⟨ ˆO⟩S to describe the deviation from steady state ex-
456
+ pectation value ⟨ ˆO⟩S
457
+ = ⟨ ˆO⟩(∞).
458
+ From Eq. (10), we get
459
+ d
460
+ dt �Ψk1k2 = Xk1k2�Ψk1k2.
461
+ Making Fourier transformation, we
462
+ have �Ψl1l2(t) = �
463
+ k1k2 ei(−k1l1+k2l2)�Ψk1k2(t).
464
+ Decomposing ar-
465
+ bitrary initial state �Ψk1k2(0) by the eigenstates of Xk1k2 i.e.
466
+ �Ψk1k2(0) = �
467
+ αβ Cαβ
468
+ k1k2|Γαβ
469
+ k1k2⟩, where α and β take ±, then we
470
+ have
471
+ �Ψl1l2(t) =
472
+
473
+ k,µ
474
+ eik·˜rCµ
475
+ ket Γµ
476
+ k |Γµ
477
+ k⟩,
478
+ (12)
479
+ where k = (k1, k2), ˜r = (−l1, l2) and µ = (α, β). For non-zero
480
+ Liouvillian gap, the system exponentially decays to NESS
481
+ with time, so we can define instantaneous decay rate K(t) of
482
+ the j component of �Ψl1l2(t) as
483
+ K j
484
+ l1l2 = d
485
+ dt log
486
+
487
+ |�Ψj
488
+ l1l2(t)|
489
+
490
+ .
491
+ (13)
492
+ Below we unveil how K(t) is affected by the dispersion of
493
+ Γµ
494
+ k through Fig. 4, in which the damping behaviors of local
495
+ deviating particle number �nl = �
496
+ Gll from the initial state with a
497
+ single excitation on site 1 are shown:
498
+ (i) When FB appears, Γµ
499
+ k becomes a constant, denoted
500
+ by Γ0. Then we have �Ψl1l2(t) = eΓ0t �
501
+ k,µ eik·˜rCµ
502
+ k|Γµ
503
+ k⟩ and
504
+ K j
505
+ l1l2(t) = Re(Γ0), which means for arbitrary initial state dif-
506
+ ferent two-operator correlation functions will synchronously
507
+ relax to their steady state expectation values with the same
508
+ decay rate, as demonstrated in Fig. 4 (b) and (e), where dif-
509
+ ferent curves of log(˜nl) as a function with γt have the same
510
+ constant slope, i.e. K1
511
+ ll = 4γ for all l.
512
+ (ii) When Γµ
513
+ k is only dispersionless in its real part, we set
514
+ Γµ
515
+ k = −x0 ��� iyµ(k), where x0 and yµ(k) are real. Then we
516
+ have �Ψl1l2(t) = e−x0t �
517
+ k,µ Cµ
518
+ keik·˜re−iyµ(k)t|Γµ
519
+ k⟩ and
520
+ K j
521
+ l1l2 = −x0 + d
522
+ dt log
523
+ ��������
524
+ ���
525
+
526
+ k,µ
527
+
528
+ k|Γµ
529
+ k⟩ jei�
530
+ k·˜r−yµ(k)t����
531
+ �������� .
532
+ (14)
533
+ The right side of Eq. (14) contains sum of a series of plane
534
+ waves, which leads to K j
535
+ l1l2(t) oscillating around x0, as shown
536
+ in Fig. 4 (d). The oscillating slopes lead to continuously inter-
537
+ secting curves in Fig. 4 (a).
538
+ (iii) When Γµ
539
+ k is only dispersionless in its imaginary part,
540
+ we set Γµ
541
+ k = −(xc + δxµ(k)) − iy0, where xc and δxµ(k)
542
+ are the central value and the offset function of Re(Γµ
543
+ k),
544
+
545
+ -1.5
546
+ -2-2
547
+ -2.5
548
+ -3k(元)k(元)J2
549
+ >
550
+ 412J2
551
+ = 412J2
552
+ 412Im(入±ReReReIm2
553
+ -4Re(X±5
554
+ 0
555
+ 5
556
+ -20
557
+ -10
558
+ 05
559
+ 5
560
+ -30
561
+ -15
562
+ 05
563
+ 5
564
+ -40
565
+ -20
566
+ 02
567
+ 0
568
+ 2
569
+ 0
570
+ 0.5
571
+ 10.5
572
+ 0
573
+ -0.5
574
+ 0
575
+ 0.5
576
+ 10.5
577
+ 0
578
+ -0.5
579
+ 0
580
+ 0.5k(元)2
581
+ 0
582
+ -2
583
+ 1
584
+ 0
585
+ 0
586
+ 1
587
+ -11
588
+ 0
589
+ 1
590
+ 1
591
+ 0
592
+ 0
593
+ -1 -1J2
594
+ 412k1(元)k2(元)k1(元)k1(元)k1(元)k2(元)k1(元)k1(元)k2(元)1
595
+ 0
596
+ 1
597
+ 1
598
+ 0
599
+ 0k2(元)k2(元)k2(元)-2
600
+ -3
601
+ -4
602
+ 1
603
+ 1
604
+ 0
605
+ 0
606
+ .1.4
607
+ -5
608
+ 6
609
+ 1
610
+ 1
611
+ 0
612
+ 0
613
+ 1-5
614
+ .10
615
+ 1
616
+ 1
617
+ 0
618
+ 0
619
+ -1
620
+ -1HH
621
+ Re(T)
622
+ k1k2HH
623
+ k1k2J2
624
+ >
625
+ 412J2
626
+ = 4124
627
+ and y0 is the imaginary part.
628
+ Then we have �Ψl1l2(t) =
629
+ e−(xc+iy0)t �
630
+ k,µ Cµ
631
+ keik·˜re−δxµ(k)t|Γµ
632
+ k⟩ and
633
+ K j
634
+ l1l2 = −xc + d
635
+ dt log
636
+ ��������
637
+ ���
638
+
639
+ k,µ
640
+
641
+ k|Γµ
642
+ k⟩ jeik·˜re−δxµ(k)t���
643
+ �������� .
644
+ (15)
645
+ Since δxµ(k) is real, the relaxation process does not display
646
+ oscillating decay rates (see Fig. 4 (f)). This induces the forked
647
+ damping curves typically as shown in Fig. 4 (c).
648
+ (b)
649
+ (a)
650
+ (c)
651
+ (d)
652
+ (e)
653
+ (f)
654
+ FIG. 4. The damping of particle number at different sites. The lattice
655
+ has 15 sites under the periodic boundary condition. Initial state is a
656
+ single excitation on the first site from vacuum. The time evolutions of
657
+ log(|˜nl|) are shown in (a), (b), (c), and their derivatives K1
658
+ ll are shown
659
+ in (d), (e) and (f). The blue, red and orange lines are corresponding
660
+ to l = 1, l = 2 and l = 3, respectively. In (a) and (d), γ2 is set as 0.5.
661
+ In (b) and (e), γ2 = 1. In (c) and (f), γ2 = 1.5. Others parameters are
662
+ the same in all subfigures with J = 1, γ1 = 0.25 and w = 0.25. The
663
+ black dashed line represents a constant decay rate as ˜nl ∝ e−4γt.
664
+ The above damping dynamics is directly related to disper-
665
+ sion of damping-matrix spectra. The damping-matrix spec-
666
+ tra reflect the decay of correlation functions, however, the Li-
667
+ ovillian spectra reflect the decay of the whole system. We
668
+ prove that the damping-matrix spectra are included in Liouvil-
669
+ lian spectra in the Supplemental Material [25]. Therefore, for
670
+ more general models with closed evolution equations of two-
671
+ operator correlation functions, the dispersionless Liouvillian
672
+ bands will lead to dispersionless damping-matrix spectra, and
673
+ then give rise to the same dynamical signatures as shown in
674
+ our model.
675
+ Localized
676
+ normal
677
+ master
678
+ modes
679
+ and
680
+ dynamic
681
+ localization.— In isolated system, FBs lead to localized
682
+ eigenstates by destructive interference.
683
+ Now, we exactly
684
+ solve our model (see the Supplemental Material [25]) to show
685
+ that the LFB can induce dynamic localization by localized
686
+ normal master modes (LNMMs), which suppress propagation
687
+ of local perturbation on NESS.
688
+ Usually, the odd parity part of ˆL has no effect on the ex-
689
+ pectation value of observation in pure fermionic system [25].
690
+ Therefore, we focus on the balanced model (w = 0) with even
691
+ parity (P = 1), whose Liouvillian is illustrated in Fig. 1 (c).
692
+ By solving the equation ζi(k)|Ω⟩ = 0 for i = 1 ∼ 4, we get the
693
+ steady state |Ω⟩ as
694
+ |Ω⟩ = 1
695
+ N
696
+ π
697
+
698
+ k=−π
699
+ (1 + a†
700
+ kc†
701
+ −k)|0⟩ = 1
702
+ N
703
+ L
704
+
705
+ l=1
706
+ (1 + a†
707
+ l c†
708
+ l )|0⟩,
709
+ (16)
710
+ where N = 2L and this state is independent with γ1 and γ2.
711
+ At the FB point with J = 2 √γ1γ2, the exceptional degeneracy
712
+ occurs in the non-Hermitian matrix Lk of Eq. (7) with four
713
+ eigenstates coalescing into two. Then ˆLk is reduced to ˆLk =
714
+ −2γ
715
+
716
+ ζ
717
+
718
+ A(k)ζA(k) + ζ
719
+
720
+ B(k)ζB(k)
721
+
722
+ , where
723
+ ζ
724
+
725
+ A(k) = −a†
726
+ k + c−k,
727
+ ζA(k) = 1
728
+ 2(−ak + ick + ia†
729
+ −k + c†
730
+ −k),
731
+ ζ
732
+
733
+ B(k) = ak + c†
734
+ −k,
735
+ ζB(k) = 1
736
+ 2(a†
737
+ k − ic†
738
+ k + ia−k + c−k).
739
+ (17)
740
+ Making Fourier transformation, we get
741
+ ζ
742
+
743
+ A(l) =
744
+
745
+ k
746
+ e−iklζ
747
+
748
+ A(k) = −a†
749
+ l +cl, ζ
750
+
751
+ B(l) =
752
+
753
+ k
754
+ eiklζ
755
+
756
+ B(k) = al+c†
757
+ l ,
758
+ (18)
759
+ which create local eigenstate ζ
760
+
761
+ A,B(l)|Ω⟩ of ˆL with eigenvalue
762
+ −2γ and are coined as LNMMs.
763
+ We can also understand LNMMs intuitively from the per-
764
+ spective of destructive interference. Writing the real-space Li-
765
+ ouvillian with w = 0, J = 2 √γ1γ2 = 1 as ˆL = �
766
+ l(ˆhl + ˆfl −2γ),
767
+ where the hopping term hl is defined as ˆhl = −i(a†
768
+ l+1al +
769
+ h.c.) + i(c†
770
+ l+1cl + h.c.) − (a†
771
+ l+1cl + c†
772
+ l+1al + h.c) and the pair-
773
+ ing term ˆfl is defined as fl = 2γ(a†
774
+ l c†
775
+ l + clal), we can check
776
+ that ˆflal|Ω⟩ =
777
+ ˆflcl|Ω⟩ =
778
+ ˆfla†
779
+ l |Ω⟩ =
780
+ ˆflc†
781
+ l |Ω⟩ = 0. This im-
782
+ plies that the pairing terms do not affect a single particle or
783
+ hole excited on the NESS. Therefore, for these states only
784
+ hopping terms make sense. We schematically plot this re-
785
+ duced ladder in Fig. 1 (d). It is easy to find another LNMM
786
+ as ζ
787
+
788
+ C(l) = a†
789
+ l − ic†
790
+ l from the view of destructive interference,
791
+ which forbids the state ζ
792
+
793
+ C(l)|Ω transferring to other sites. We
794
+ can also check that ˆL ζ
795
+
796
+ C(l)|Ω⟩ = −2γ ζ
797
+
798
+ C(l)|Ω⟩.
799
+ The LNMMs contain decay information of quantum
800
+ jumps. To see it clearly, we map the C−representation state
801
+ ζ
802
+
803
+ A(l)ζ
804
+
805
+ B(l)|Ω⟩, for example, back to density-matrix representa-
806
+ tion:
807
+ ζ
808
+
809
+ A(l)ζ
810
+
811
+ B(l)|Ω⟩ → −a†
812
+ l alρs + ρsala†
813
+ l + alρsa†
814
+ l − a†
815
+ l ρsal,
816
+ (19)
817
+ where ρs is the density matrix of NESS. The terms alρsa†
818
+ l
819
+ and a†
820
+ l ρsal are exactly corresponding to local quantum jumps
821
+ on NESS. Eq. (19) implies that the local perturbation on
822
+ NESS from quantum jumps will relax to NESS without ex-
823
+ panding its territory. To see it clearly, we simulate the evo-
824
+ lution from an initial state described by the density matrix
825
+ ρ0 = a†
826
+ 1ρsa1/Tr(a†
827
+ 1ρsa1), which is created by a quantum jump
828
+ on the first site of NESS. In Fig 5, we demonstrate the time
829
+ evolution of particle numbers of the first three sites in a lattice
830
+ with 15 sites. In the initial time, a jump occurs on the first
831
+ site of NESS, increasing only the particle number on the first
832
+ site n1 to 1 with others sites keeping their steady state value
833
+
834
+ 2
835
+ -6
836
+ 0
837
+ 2-2
838
+ -4
839
+ -6
840
+ 0
841
+ 1
842
+ 2K=
843
+ -4ttK=
844
+ -4K=
845
+ -4tttK
846
+ 1
847
+ 1log
848
+ (1i
849
+ nt
850
+ 1)-2
851
+ -6
852
+ 0
853
+ 1
854
+ 20
855
+ n1
856
+ n2
857
+ n3
858
+ -5
859
+ -10
860
+ 0
861
+ 1
862
+ 20
863
+ n1
864
+ n2
865
+ n3
866
+ -5
867
+ -10
868
+ 0
869
+ 1
870
+ 20
871
+ n1
872
+ n2
873
+ n3
874
+ ~
875
+ -5
876
+ -10
877
+ 0
878
+ 1
879
+ 2J2
880
+ >
881
+ 412J2
882
+ = 412J2
883
+ 412t5
884
+ (a)
885
+ (b)
886
+ (c)
887
+ FIG. 5. The time evolution of particle number on the first site n1
888
+ shown in (a), second site n2 in (b) and third site n3 in (c). Initial
889
+ state is a†
890
+ 1ρsa1/Tr(a†
891
+ 1ρsa1) corresponding to a quantum jump on the
892
+ first site of steady state. The periodic lattice has 15 sites with w = 0,
893
+ J = 1, γ1 = 0.25 in all subfigures. The black dotted, red solid, and
894
+ blue dashed lines are corresponding to γ2 = 0.5, γ2 = 1 and γ2 = 1.5,
895
+ respectively.
896
+ 0.5. The red solid line, black dotted line and blue dashed line
897
+ are corresponding to the situation with J2 = 4γ1γ2 (LFB),
898
+ J2 > 4γ1γ2, and J2 < 4γ1γ2, respectively. We can see that
899
+ when J2 � 4γ1γ2, the perturbation can spread from n1 to n3.
900
+ However, for the case with LFB, the perturbation excitation
901
+ decays locally without going through to n2 and n3, indicating
902
+ the occurrence of dynamical localization.
903
+ Final remarks.— (i) We use a geometrically intuitive
904
+ method to construct flat band models in open system and
905
+ demonstrate that the dispersion of Liouvillian band can ef-
906
+ fectively affect the damping dynamics of local particle num-
907
+ ber, intermediated by damping matrix of correlation function
908
+ vector. When the Liouvillian flat band appears, the particle
909
+ number in different sites will relax to their stable values syn-
910
+ chronously. When only the real or imaginary part of rapidity
911
+ spectrum is dispersionless, the damping behaviors show the
912
+ oscillating or forked characteristic.
913
+ (ii) We show flat-band Liouvillian can induce dynamical
914
+ localization on NESS by the localized normal master modes,
915
+ which halt the propagation of perturbation from other sites to
916
+ the target sites.
917
+ (iii) Our model does not exhibit non-Hermitian skin ef-
918
+ fect [42, 43], which was uncovered to cause many abnormal
919
+ phenomena such as boundary sensitivity [44], chiral and he-
920
+ lical damping [45, 46] and slowing down of relaxation pro-
921
+ cesses [20]. The interplay between Liouvillian flat band and
922
+ non-Hermitian skin effect is an interesting topic for future
923
+ studies.
924
+ Acknowledgments.— We thank X. L. Wang, Z. Y. Zheng
925
+ and C. X. Guo for helpful discussions.The work is supported
926
+ by National Key National Key Research and Development
927
+ Program of China (Grant No.2021YFA1402104), the NSFC
928
+ under Grants No.
929
+ 12174436 and No.
930
+ T2121001, and the
931
+ Strategic Priority Research Program of Chinese Academy of
932
+ Sciences under Grant No.XDB33000000.
933
934
+ [1] S.-Q. Shen, Topological insulators, Vol. 174 (Springer,2012).
935
+ [2] A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov,
936
+ and A. K. Geim, The electronic properties of graphene, Rev.
937
+ Mod. Phys. 81, 109 (2009).
938
+ [3] Z. Liu, F. Liu, and Y.-S. Wu, Exotic electronic states in the world
939
+ of flat bands: From theory to material, Chinese Physics B 23,
940
+ 077308 (2014).
941
+ [4] Congjun Wu, Doron Bergman, Leon Balents, and S. Das Sarma,
942
+ Flat bands and wigner crystallization in the honeycomb optical
943
+ lattice, Phys. Rev. Lett. 99, 070401 (2007).
944
+ [5] N. Regnault and B. Andrei Bernevig, Fractional Chern Insulator,
945
+ Phys. Rev. X 1, 021014 (2011).
946
+ [6] Lindblad, G. On the generators of quantum dynamical semi-
947
+ groups, Commun.Math. Phys. 48, 119-130 (1976).
948
+ [7] C. W. Gardiner and M. J. Collett, Input and output in damped
949
+ quantum systems: Quantum stochastic differential equations and
950
+ the master equation, Phys. Rev. A 31, 3761 (1985).
951
+ [8] D. Manzano, A short introduction to the Lindblad master equa-
952
+ tion, AIP Advances 10, 025106 (2020).
953
+ [9] T. Prosen, Third quantization: a general method to solve mas-
954
+ ter equations for quadratic open Fermi systems, New Journal of
955
+ Physics 10, 043026 (2008).
956
+ [10] C. Guo and D. Poletti, Solutions for bosonic and fermionic
957
+ dissipative quadratic open systems, Phys. Rev. A 95, 052107
958
+ (2017).
959
+ [11] N. Shibata and H. Katsura, Dissipative spin chain as a non-
960
+ Hermitian Kitaev ladder, Phys. Rev. B 99, 174303 (2019).
961
+ [12] B. Horstmann, J. I. Cirac, and G. Giedke, Noise-driven dynam-
962
+ ics and phase transitions in fermionic systems, Phys. Rev. A 87,
963
+ 012108 (2013).
964
+ [13] Y. Zhang and T. Barthel, Criticality and Phase Classification for
965
+ Quadratic Open Quantum Many-Body Systems, Phys. Rev. Lett.
966
+ 129, 120401 (2022).
967
+ [14] T. Barthel and Y. Zhang, Solving quasi-free and quadratic Lind-
968
+ blad master equations for open fermionic and bosonic systems,
969
+ arXiv:2112.08344 [quant-ph] (2022).
970
+ [15] S. Talkington and M. Claassen, Dissipation Induced Flat Bands,
971
+ Phys. Rev. B 106, L161109 (2022).
972
+ [16] Z. Cai and T. Barthel, Algebraic versus Exponential Decoher-
973
+ ence in Dissipative Many-Particle Systems, Phys. Rev. Lett. 111,
974
+ 150403 (2013).
975
+ [17] M. ˇZnidariˇc, Relaxation times of dissipative many-body quan-
976
+ tum systems, Phys. Rev. E 92, 042143 (2015).
977
+ [18] F. Minganti, A. Biella, N. Bartolo, and C. Ciuti, Spectral theory
978
+ of Liouvillians for dissipative phase transitions, Phys. Rev. A 98,
979
+ 042118 (2018).
980
+ [19] T. Mori and T. Shirai, Resolving a Discrepancy between Liou-
981
+ villian Gap and Relaxation Time in Boundary-Dissipated Quan-
982
+ tum Many-Body Systems, Phys. Rev. Lett. 125, 230604 (2020).
983
+ [20] T. Haga, M. Nakagawa, R. Hamazaki, and M. Ueda, Liouvillian
984
+
985
+ ---J2 > 412
986
+ J2 = 412
987
+ J2 < 412
988
+ 0.50.51
989
+ 0.50.501
990
+ 0.5
991
+ 0
992
+ 0.5
993
+ 1.5
994
+ 2n
995
+ 1n
996
+ 2n
997
+ 3t6
998
+ Skin Effect: Slowing Down of Relaxation Processes without Gap
999
+ Closing, Phys. Rev. Lett. 127, 070402 (2021).
1000
+ [21] Y. Nakanishi and T. Sasamoto, PT phase transition in open
1001
+ quantum systems with Lindblad dynamics, Phys. Rev. A 105,
1002
+ 022219 (2022).
1003
+ [22] Y.-N. Zhou, L. Mao, and H. Zhai, R´enyi entropy dynamics and
1004
+ Lindblad spectrum for open quantum systems, Phys. Rev. Re-
1005
+ search 3, 043060 (2021).
1006
+ [23] S. Diehl, E. Rico, M. Baranov, and P. Zoller, Topology by dissi-
1007
+ pation in atomic quantum wires, Nature Phys 7, 971-977 (2011).
1008
+ [24] T. L. Landi, D. Poletti, and G. Schaller, Non-equilibrium
1009
+ boundary driven quantum systems: models, methods and prop-
1010
+ erties, arXiv:2104.14350 [quant-ph] (2022).
1011
+ [25] See Supplemental Material for (i) Mapping of Lindblad master
1012
+ equation, (ii) Diagonalization, exceptional point and symmetry
1013
+ of the Liouvillian, (iii) Exactly solution and discussion on parity,
1014
+ (iv) Evolution equations of correlation functions and the symme-
1015
+ try of damping matrix, (v) Particle number distribution of steady
1016
+ state, (vi) The relationship between the damping-matrix spectra
1017
+ and the Liouvillian spectra. The supplemental materias include
1018
+ also references [26–33].
1019
+ [26] M.-D. Choi, Completely positive linear maps on complex ma-
1020
+ trices, Linear Algebra Applications 10, 285 (1975).
1021
+ [27] A. Jamiołkowski, Linear transformations which preserve trace
1022
+ and positive semidefiniteness of operators, Rep. Math. Phys. 3,
1023
+ 275 (1972).
1024
+ [28] J. E. Tyson, Operator-Schmidt decompositions and the Fourier
1025
+ transform, with applications to the operator-Schmidt numbers of
1026
+ unitaries, J. Phys. A: Math. Gen. 36, 10101 (2003).
1027
+ [29] M. Zwolak and G. Vidal, Mixed-State Dynamics in One-
1028
+ Dimensional Quantum Lattice Systems: A Time-Dependent Su-
1029
+ peroperator Renormalization Algorithm, Phys. Rev. Lett. 93,
1030
+ 207205 (2004).
1031
+ [30] K. Kawabata, K. Shiozaki, M. Ueda, and M. Sato, Symmetry
1032
+ and Topology in Non-Hermitian Physics, Phys. Rev. X 9, 041015
1033
+ (2019).
1034
+ [31] A. W. W. Ludwig, Topological phases:
1035
+ classification of
1036
+ topological insulators and superconductors of noninteracting
1037
+ fermions, and beyond, Physica Scripta T168, 014001 (2015).
1038
+ [32] C.-H. Liu, H. Jiang, and S. Chen, Topological classification of
1039
+ non-Hermitian systems with reflection symmetry, Phys. Rev. B
1040
+ 99, 125103 (2019).
1041
+ [33] C.-H. Liu and S. Chen, Topological classification of defects in
1042
+ non-Hermitian systems, Phys. Rev. B 100, 144106 (2019).
1043
+ [34] M. Creutz, End States, Ladder Compounds, and Domain-Wall
1044
+ Fermions, Phys. Rev. Lett. 83, 2636 (1999).
1045
+ [35] W. Maimaiti, A. Andreanov, H. C. Park, O. Gendelman, and S.
1046
+ Flach, Compact localized states and flat-band generators in one
1047
+ dimension, Phys. Rev. B 95, 115135 (2017).
1048
+ [36] W. Maimaiti and A. Andreanov, Non-Hermitian flatband gener-
1049
+ ator in one dimension, Phys. Rev. B 104, 035115 (2021).
1050
+ [37] Y. Kuno, T. Orito, and I. Ichinose, Flat-band many-body local-
1051
+ ization and ergodicity breaking in the Creutz ladder, New Journal
1052
+ of Physics 22, 013032 (2020).
1053
+ [38] Although the coefficient of ˆLk=0 is 0.5 in the equation ˆL =
1054
+ 0.5 ˆLk=0 + �π−
1055
+ k=0+ ˆLk, the rapidity for k = 0 can be still described
1056
+ by λ±(0) due to the degeneracy as ζ
1057
+
1058
+ i (k = 0) = ζ
1059
+
1060
+ 5−i(k = 0) and
1061
+ ζi(k = 0) = ζ5−i(k = 0).
1062
+ [39] A. Mostafazadeh, Pseudo-Hermiticity versus PT-symmetry. II.
1063
+ A complete characterization of non-Hermitian Hamiltonians
1064
+ with a real spectrum, Journal of Mathematical Physics 43, 2814
1065
+ (2002).
1066
+ [40] A. Mostafazadeh, Pseudo-Hermiticity for a class of nondiag-
1067
+ onalizable Hamiltonians, Journal of Mathematical Physics 43,
1068
+ 6343 (2002).
1069
+ [41] Y. Ashida, Z. Gong, and M. Ueda, Non-Hermitian physics, Ad-
1070
+ vances in Physics 69, 249 (2020).
1071
+ [42] S. Yao and Z. Wang, Edge States and Topological Invariants of
1072
+ Non-Hermitian Systems, Phys. Rev. Lett. 121, 086803 (2018).
1073
+ [43] F. K. Kunst, E. Edvardsson, J. C. Budich, and E. J. Bergholtz,
1074
+ Biorthogonal Bulk-Boundary Correspondence in Non-Hermitian
1075
+ Systems, Phys. Rev. Lett. 121, 026808 (2018).
1076
+ [44] C.-X. Guo, C.-H. Liu, X.-M. Zhao, Y. Liu, and S. Chen, Exact
1077
+ Solution of Non-Hermitian Systems with Generalized Bound-
1078
+ ary Conditions: Size-Dependent Boundary Effect and Fragility
1079
+ of the Skin Effect, Phys. Rev. Lett. 127, 116801 (2021).
1080
+ [45] F. Song, S. Yao, and Z. Wang, Non-Hermitian Skin Effect and
1081
+ Chiral Damping in Open Quantum Systems, Phys. Rev. Lett.
1082
+ 123, 170401 (2019).
1083
+ [46] C.-H. Liu, K. Zhang, Z. Yang, and S. Chen, Helical damp-
1084
+ ing and dynamical critical skin effect in open quantum systems,
1085
+ Physical Review Research 2, 043167 (2020).
1086
+
1087
+ 7
1088
+ SUPPLEMENTAL MATERIAL: Dynamics Signatures of Liouvillian Flat Band
1089
+ S1. Mapping of Lindblad master equation
1090
+ FIG. S1. Mapping of Lindblad master equation.
1091
+ The Lindblad master equation, formalized density matrix ρ and Liouvillian superoperator L is shown in Eq. (1) and Eq. (2) in
1092
+ the main text. First we carry out the Choi-Jamiolkwski isomorphism [1–4] to map the fermionic LME into representation B as
1093
+ d
1094
+ dt| ρ⟩B = ˆLB| ρ⟩B,
1095
+ (S1)
1096
+ where | ρ⟩B is vectorized from ρ and ˆLB is mapped from L. Specifically, the mapping is
1097
+ ρ → | ρ⟩B =
1098
+
1099
+ IJ
1100
+ ρIJ|I⟩a ⊗ |J⟩b,
1101
+ (S2a)
1102
+ L → ˆLB =
1103
+
1104
+ i j
1105
+ Fi(a, a†) ⊗ F T
1106
+ j (b, b†),
1107
+ (S2b)
1108
+ where b = (b1, b2, · · · ) is the set of annihilation operators of b−fermions, which is one-to-one mapping from a, and T means
1109
+ matrix transpose. |I⟩a and |J⟩b are defined as
1110
+ |I⟩a = (a†
1111
+ 1)I1(a†
1112
+ 2)I2 · · · (a†
1113
+ L)IL|0⟩a,
1114
+ (S3a)
1115
+ |J⟩b = (b†
1116
+ 1)J1(b†
1117
+ 2)J2 · · · (b†
1118
+ L)JL|0⟩b,
1119
+ (S3b)
1120
+ where |0⟩a and |0⟩b are vacuum state of all a−fermions and b−fermions, respectively. In this representation, the expectation
1121
+ value of observable becomes
1122
+ ⟨ ˆOa⟩ =B ⟨S0| ˆOa ⊗ Ib| ρ⟩B,
1123
+ (S4)
1124
+ where B⟨S0| is a special state defined as:
1125
+ B⟨S0| =
1126
+
1127
+ S
1128
+ ⟨S|a ⊗ ⟨S|b =
1129
+
1130
+ S
1131
+
1132
+ ⟨0|a(aL)S L · ·(a1)S 1 ⊗ ⟨0|b(bL)S L · ·(b1)S 1�
1133
+ ,
1134
+ (S5)
1135
+ and Ib is a unit operator of all b−fermions. The element S i of S = (S 1, S 2, · · · ) can take 0 or 1, and �
1136
+ S requires a sum over all
1137
+ possible configurations of S. Let us prove Eq. (S4):
1138
+ ⟨ ˆOa⟩ =
1139
+
1140
+ IJS
1141
+ ρIJ ⟨S|a ˆOa|I⟩a⟨S|bIb|J⟩b
1142
+ =
1143
+
1144
+ IJS
1145
+ ρIJ a⟨0|aS L
1146
+ L · · · aS 1
1147
+ 1 ˆOa(a†
1148
+ 1)I1 · · · (a†
1149
+ L)IL|0⟩a δSJ
1150
+ =
1151
+
1152
+ IJ
1153
+ ρIJ a⟨0|aJL
1154
+ L · · · aJ1
1155
+ 1 ˆOa(a†
1156
+ 1)I1 · · · (a†
1157
+ L)IL|0⟩a
1158
+ =
1159
+
1160
+ IJ
1161
+ ρIJ a⟨0| ˆOa|0⟩a = Tr( ˆOaρ).
1162
+ (S6)
1163
+
1164
+ p=p [I)a<Jla
1165
+ d
1166
+ 0= L(p)= F;(a,at)pF,(a,at)
1167
+ Tr(Oap
1168
+ IJ
1169
+ ii
1170
+ IJ
1171
+ [p>c
1172
+ 278
1173
+ In representation B, operators satisfy the following relations:
1174
+ {ai, a†
1175
+ j} = {bi, b†
1176
+ j} = δi j,
1177
+ {a†
1178
+ i , a†
1179
+ j} = {ai, a j} = {b†
1180
+ i , b†
1181
+ j} = {bi, bj} = 0,
1182
+ (S7a)
1183
+ [a†
1184
+ i , b j] = [a†
1185
+ i , b†
1186
+ j] = [ai, bj] = [ai, b†
1187
+ j] = 0.
1188
+ (S7b)
1189
+ The commutation relations in Eq. (S7b) are from the direct product between a−fermions and b−fermions, which are unfavorable
1190
+ for further analysis. To enforce fermionic anticommutation relations over all operators, we define operators of c−fermions as
1191
+ c† = b† ˆP and c = ˆPb, where ˆP is a parity operator defined as
1192
+ ˆP := exp
1193
+
1194
+
1195
+
1196
+ l
1197
+ (a†
1198
+ l al + b†
1199
+ l bl)
1200
+
1201
+ = exp
1202
+
1203
+
1204
+
1205
+ l
1206
+ (a†
1207
+ l al + c†
1208
+ l cl)
1209
+
1210
+ .
1211
+ (S8)
1212
+ It is easy to check the fermionic anticommutation relations in a−fermions and c−fermions:
1213
+ {ci, c†
1214
+ j} = δi j,
1215
+ {c†
1216
+ i , c†
1217
+ j} = {ci, c j} = 0
1218
+ (S9a)
1219
+ {a†
1220
+ i , cj} = {a†
1221
+ i , c†
1222
+ j} = {ai, cj} = {ai, c†
1223
+ j} = 0
1224
+ (S9b)
1225
+ By c we can fully fermionize system from representation B to representation C. The mapping is
1226
+ | ρ⟩B → | ρ⟩C =
1227
+
1228
+ IJ
1229
+ ρIJ(a†
1230
+ 1)I1 · · · (a†
1231
+ L)IL(c†
1232
+ 1 ˆP)J1 · · · (c†
1233
+ L ˆP)JL |0⟩,
1234
+ (S10a)
1235
+ ˆLB → ˆLC =
1236
+
1237
+ ij
1238
+ Fi(a, a†) F T
1239
+ j ( ˆPc, c† ˆP).
1240
+ (S10b)
1241
+ The LME and the expectation value of observable in representation C are
1242
+ d
1243
+ dt| ρ⟩C = ˆLC| ρ⟩C,
1244
+ (S11a)
1245
+ ⟨ ˆOa⟩ =C ⟨S0| ˆOa| ρ⟩C,
1246
+ (S11b)
1247
+ where C⟨S0| is defined as:
1248
+ C⟨S0| =
1249
+
1250
+ S
1251
+ ⟨0| ( ˆPcL)S L · · · ( ˆPc1)S 1aS L
1252
+ L · · · aS 1
1253
+ 1 .
1254
+ (S12)
1255
+ Combining the mappings in Eq. (S2) and Eq. (S10), we get the final mapping, i.e., Eq. (4) in the main text. The mapping
1256
+ process is schematically shown in Fig. S1.
1257
+ S2. Model in representation C: diagonalization, exceptional point and symmetry
1258
+ In this section we map our Liouvillian in Eq. (5) in the main text into representation C and get its BdG form in momentum
1259
+ space. Based on the BdG form, we show the exceptional point and symmetry of our Liouvillian.
1260
+ A. The derivation of ˆL
1261
+ Our Liouvillian L in Eq. (5) is mapped into ˆL by the mapping (4b) in the main text:
1262
+ L(·) = −i[H, ·] + (1 − w)DL(·) + (1 + w)DR(·) → ˆL = ˆH + (1 − w) ˆDL + (1 + w) ˆDR.
1263
+ (S13)
1264
+ Note that our matrix representation of creation and annihilation operator is real, thus we have aT = a†, cT = c†, ˆPT = ˆP. Then
1265
+ we get
1266
+ − i[H, ·] → ˆH = −iH(a, a†) + iHT( ˆPc, c† ˆP) = −iJ
1267
+
1268
+ l
1269
+ (a†
1270
+ l+1al + a†
1271
+ l al+1) + iJ
1272
+
1273
+ l
1274
+ (c†
1275
+ l cl+1 + c†
1276
+ l+1cl).
1277
+ (S14)
1278
+
1279
+ 9
1280
+ Note that our matrix representation of Al = √γ1a†
1281
+ l + √γ2al+1 is real, thus we have AT
1282
+ l = A†
1283
+ l . Then we get
1284
+ DL(·) → ˆDL =
1285
+
1286
+ l
1287
+
1288
+ 2Al(a, a†)Al( ˆPc, c† ˆP) − A†
1289
+ l (a, a†)Al(a, a†) − A†
1290
+ l ( ˆPc, c† ˆP)Al( ˆPc, c† ˆP)
1291
+
1292
+ =
1293
+
1294
+ l
1295
+
1296
+ 2( √γ1a†
1297
+ l + √γ2al+1)( √γ1c†
1298
+ l ˆP + √γ2 ˆPcl+1) − ( √γ1al + √γ2a†
1299
+ l+1)( √γ1a†
1300
+ l + √γ2al+1)
1301
+ − ( √γ1 ˆPcl + √γ2c†
1302
+ l+1 ˆP)( √γ1c†
1303
+ l ˆP + √γ2 ˆPcl+1)
1304
+
1305
+ =
1306
+
1307
+ l
1308
+
1309
+ − 2 √γ1γ2 ˆP(a†
1310
+ l cl+1 + c†
1311
+ l al+1) + 2γ1 ˆPa†
1312
+ l c†
1313
+ l + 2γ2 ˆPcl+1al+1 − √γ1γ2(alal+1 + a†
1314
+ l+1a†
1315
+ l )
1316
+ + √γ1γ2(clcl+1 + c†
1317
+ l+1c†
1318
+ l ) − γ2(a†
1319
+ l+1al+1 + c†
1320
+ l+1cl+1) − γ1(ala†
1321
+ l + clc†
1322
+ l )
1323
+
1324
+ ,
1325
+ (S15)
1326
+ DR(·) → ˆDR =
1327
+
1328
+ l
1329
+
1330
+ 2A†
1331
+ l (a, a†)A†
1332
+ l ( ˆPc, c† ˆP) − Al(a, a†)A†
1333
+ l (a, a†) − Al( ˆPc, c† ˆP)A†
1334
+ l ( ˆPc, c† ˆP)
1335
+
1336
+ =
1337
+
1338
+ l
1339
+
1340
+ 2( √γ1al + √γ2a†
1341
+ l+1)( √γ1 ˆPcl + √γ2c†
1342
+ l+1 ˆP) − ( √γ1a†
1343
+ l + √γ2al+1)( √γ1al + √γ2a†
1344
+ l+1)
1345
+ − ( √γ1c†
1346
+ l ˆP + √γ2 ˆPcl+1)( √γ1 ˆPcl + √γ2c†
1347
+ l+1 ˆP)
1348
+
1349
+ =
1350
+
1351
+ l
1352
+
1353
+ − 2 √γ1γ2 ˆP(a†
1354
+ l+1cl + c†
1355
+ l+1al) + 2γ1 ˆPclal + 2γ2 ˆPa†
1356
+ l+1c†
1357
+ l+1 + √γ1γ2(alal+1 + a†
1358
+ l+1a†
1359
+ l )
1360
+ − √γ1γ2(clcl+1 + c†
1361
+ l+1c†
1362
+ l ) − γ2(al+1a†
1363
+ l+1 + cl+1c†
1364
+ l+1) − γ1(a†
1365
+ l al + c†
1366
+ l cl)
1367
+
1368
+ .
1369
+ (S16)
1370
+ Due to [ ˆP, ˆL] = 0, the state will keep its parity in the evolution governed by the Lindblad master equation. Therefore, ˆP can
1371
+ reduce to a constant P, which equals 1 in even parity channel and −1 in odd parity channel.
1372
+ By Fourier transformation
1373
+ a†
1374
+ l =
1375
+ π
1376
+
1377
+ k=−π
1378
+ e−ikla†
1379
+ k,
1380
+ al =
1381
+ π
1382
+
1383
+ k=−π
1384
+ eiklak,
1385
+ c†
1386
+ l =
1387
+ π
1388
+
1389
+ k=−π
1390
+ e−iklc†
1391
+ k,
1392
+ cl =
1393
+ π
1394
+
1395
+ k=−π
1396
+ eiklck,
1397
+ (S17)
1398
+ we get ˆL in BdG form as
1399
+ ˆL = 1
1400
+ 2
1401
+ ˆLk=0 +
1402
+ π−
1403
+
1404
+ k=0+
1405
+ ˆLk,
1406
+ (S18)
1407
+ where
1408
+ ˆLk = (a†
1409
+ k c†
1410
+ k a−k c−k) Lk (ak ck a†
1411
+ −k c†
1412
+ −k)T − 4γ,
1413
+ (S19)
1414
+ and
1415
+ Lk = −i2J cos kσz ⊗ σz − 4 √γ1γ2 cos kPσz ⊗ σx − 2γPσy ⊗ σy + 2w
1416
+
1417
+ +(γ2 − γ1)σz ⊗ I + 2 √γ1γ2 sin kσy ⊗ σz
1418
+ +i(γ2 − γ1)Pσx ⊗ σy + i2 √γ1γ2 sin kPI ⊗ σx
1419
+
1420
+ .
1421
+ (S20)
1422
+ B. Diagonalization of ˆLk
1423
+ We make a similarity transformation for ˆLk by matrix W:
1424
+ ˆLk = (a†
1425
+ k c†
1426
+ k a−k c−k) W W−1 Lk W W−1 (ak ck a†
1427
+ −k c†
1428
+ −k)T − 4γ
1429
+ = (ζ
1430
+
1431
+ 1(k) ζ
1432
+
1433
+ 2(k) ζ3(k) ζ4(k)) Λ (ζ1(k) ζ2(k) ζ
1434
+
1435
+ 3(k) ζ
1436
+
1437
+ 4(k))T − 4γ
1438
+ = λ1(k)ζ
1439
+
1440
+ 1(k)ζ1(k) + λ2(k)ζ
1441
+
1442
+ 2(k)ζ2(k) + λ3(k)ζ3(k)ζ
1443
+
1444
+ 3(k) + λ4(k)ζ4(k)ζ
1445
+
1446
+ 4(k) − 4γ,
1447
+ (S21)
1448
+ where
1449
+ (a†
1450
+ k c†
1451
+ k a−k c−k) W = (ζ
1452
+
1453
+ 1(k) ζ
1454
+
1455
+ 2(k) ζ3(k) ζ4(k)),
1456
+ W−1 (ak ck a†
1457
+ −k c†
1458
+ −k)T = (ζ1(k) ζ2(k) ζ
1459
+
1460
+ 3(k) ζ
1461
+
1462
+ 4(k))T
1463
+ (S22)
1464
+
1465
+ 10
1466
+ and Λ is a diagonal matrix given by
1467
+ Λ = W−1 Lk W = diag(λ1(k), λ2(k), λ3(k), λ4(k)).
1468
+ (S23)
1469
+ We write W and W−1 as
1470
+ W = (⃗v1 ⃗v2 ⃗v3 ⃗v4),
1471
+ W−1 =
1472
+ ��������������
1473
+ ⃗u t
1474
+ 1
1475
+ ⃗u t
1476
+ 2
1477
+ ⃗u t
1478
+ 3
1479
+ ⃗u t
1480
+ 4
1481
+ ��������������
1482
+ ,
1483
+ (S24)
1484
+ where the column vector ⃗vi and row vector ⃗u t
1485
+ j satisfy ⃗u t
1486
+ j · ⃗vi = δi j. Then we have
1487
+ ζ
1488
+
1489
+ 1(k) = (a†
1490
+ k c†
1491
+ k a−k c−k) · ⃗v1,
1492
+ ζ
1493
+
1494
+ 2(k) = (a†
1495
+ k c†
1496
+ k a−k c−k) · ⃗v2,
1497
+ ζ
1498
+
1499
+ 3(k) = (ak ck a†
1500
+ −k c†
1501
+ −k) · ⃗u3,
1502
+ ζ
1503
+
1504
+ 4(k) = (ak ck a†
1505
+ −k c†
1506
+ −k) · ⃗u4,
1507
+ ζ1(k) = (ak ck a†
1508
+ −k c†
1509
+ −k) · ⃗u1,
1510
+ ζ2(k) = (ak ck a†
1511
+ −k c†
1512
+ −k) · ⃗u2,
1513
+ ζ3(k) = (a†
1514
+ k c†
1515
+ k a−k c−k) · ⃗v3,
1516
+ ζ4(k) = (a†
1517
+ k c†
1518
+ k a−k c−k) · ⃗v4.
1519
+ (S25)
1520
+ ζ
1521
+
1522
+ i(k) and ζj(k) hold anticommutation relations:
1523
+
1524
+
1525
+ i(k), ζj(k)} = δi j,
1526
+
1527
+
1528
+ i(k), ζ
1529
+
1530
+ j(k)} = {ζi(k), ζj(k)} = 0
1531
+ (S26)
1532
+ Calculating the eigenvalues of Eq. (S20), we get the same values for both even and odd parity: λ1(k) = −2γ − 2mk, λ2(k) =
1533
+ −2γ + 2mk, λ3(k) = 2γ − 2mk and λ4(k) = 2γ + 2mk , where
1534
+ mk =
1535
+ �����
1536
+
1537
+ (4γ1γ2 − J2) cos2 k,
1538
+ 4γ1γ2 ≥ J2
1539
+ i
1540
+
1541
+ (J2 − 4γ1γ2) cos2 k,
1542
+ 4γ1γ2 < J2
1543
+ (S27)
1544
+ Then Lk can be diagonalized as
1545
+ ˆLk = λ−(k)
1546
+
1547
+ ζ
1548
+
1549
+ 1(k)ζl(k) + ζ
1550
+
1551
+ 4(k)ζ4(k)
1552
+
1553
+ + λ+(k)
1554
+
1555
+ ζ
1556
+
1557
+ 2(k)ζ2(k) + ζ
1558
+
1559
+ 3(k)ζ3(k)
1560
+
1561
+ ,
1562
+ (S28)
1563
+ where
1564
+ λ±(k) = −2γ ± mk.
1565
+ (S29)
1566
+ C. Exceptional point
1567
+ When J2 = 4γ1γ2, the exceptional point of Lk emerges. To see it clearly, we show real and imaginary part of the rapidity
1568
+ λ±(k) in Fig. S2. When the flat band condition is satisfied (γ2 = 1), it occurs exceptional degeneracy between λ+ and λ−.
1569
+ (a)
1570
+ (b)
1571
+ FIG. S2. The real (a) and imaginary (b) part of λ±(k) as a function with k and γ2. Other parameters are taken as J = 1 and γ1 = 0.25
1572
+
1573
+ 2
1574
+ 0
1575
+ -2
1576
+ 0
1577
+ 0
1578
+ 0.5
1579
+ 1
1580
+ 2
1581
+ 10
1582
+ -5
1583
+ 0
1584
+ 0
1585
+ 0.5
1586
+ 1
1587
+ 2
1588
+ 1Im(入±Re(入±)k(元)k(元)11
1589
+ D. The symmetry of Liouvillian
1590
+ Due to ˆLk = ˆL−k, we can write ˆL in Eq. (S18) as ˆL = 1
1591
+ 2
1592
+ �π
1593
+ k=−π ˆLk. Therefore, we can study the symmetry of the Liouvillian
1594
+ from Lk with k ∈ (−π, π). It is easy to check that Lk in Eq. (S20) has time-reversal symmetry (TRS), particle-hole symmetry
1595
+ (PHS) and chiral symmetry (CS)[5–8]:
1596
+ TRS : T+ L∗
1597
+ k T −1
1598
+ +
1599
+ = L−k
1600
+ =⇒ T+ = σz ⊗ σx; T+T ∗
1601
+ + = 1
1602
+ PHS : C− LT
1603
+ k C−1
1604
+ − = −L−k
1605
+ =⇒ C− = σx ⊗ I; C−C∗
1606
+ − = 1
1607
+ CS : Γ L†
1608
+ k Γ−1 = −Lk
1609
+ =⇒ Γ = σy ⊗ σx; Γ2 = 1.
1610
+ (S30)
1611
+ Due to that Lk has full real spectrum in the region J2 < 4γ1γ2, the mathematical theorem ensures the Liouvillian having pseudo-
1612
+ Hermiticity i.e. there exists a Hermitian matrix η that η L†
1613
+ kη−1 = Lk. Especially, when w = 0, the system will additionally have
1614
+ inversion symmetry (IS) and the pseudo-Hermiticity will be enhanced to the parity-time symmetry (PTS):
1615
+ IS : P Lk P−1 = L−k
1616
+ =⇒ P = σy ⊗ σy
1617
+ PTS : PT L∗
1618
+ k PT −1 = Lk
1619
+ =⇒ PT = σx ⊗ σz.
1620
+ (S31)
1621
+ S3. Exactly solution of the model when w = 0
1622
+ In this section, we exactly solve our model both in even and odd channels. We show steady state and all the excited states
1623
+ of the open system. In addition, we prove that the odd parity states have no contribution on observations with even fermionic
1624
+ operators. Last, we calculate the correlation functions of steady state and local-quantum-jump states beyond the steady state.
1625
+ A. All the eigenstates of ˆL
1626
+ First, we diagonalize ˆLk in even channel (P = 1). Then normal master modes are show in Eq. (S25). The vectors ⃗v and ⃗u can
1627
+ be solved as
1628
+ ⃗v1 = 1
1629
+ 2
1630
+
1631
+ − 1 − iJ cos k/mk, −2 √γ1γ2 cos k/mk, −2 √γ1γ2 cos k/mk, 1 + iJ cos k/mk
1632
+ �T
1633
+ ⃗v2 = 1
1634
+ 2
1635
+
1636
+ − 1 + iJ cos k/mk, 2 √γ1γ2 cos k/mk, 2 √γ1γ2 cos k/mk, 1 − iJ cos k/mk
1637
+ �T
1638
+ ⃗v3 = 1
1639
+ 2
1640
+
1641
+ 1, −iJ + mk/ cos k
1642
+ 2 √γ1γ2
1643
+ , iJ − mk/ cos k
1644
+ 2 √γ1γ2
1645
+ , 1
1646
+ �T
1647
+ ⃗v4 = 1
1648
+ 2
1649
+
1650
+ 1, −iJ − mk/ cos k
1651
+ 2 √γ1γ2
1652
+ , iJ + mk/ cos k
1653
+ 2 √γ1γ2
1654
+ , 1
1655
+ �T
1656
+ ⃗u1 = 1
1657
+ 2
1658
+
1659
+ − 1, iJ − mk/ cos k
1660
+ 2 √γ1γ2
1661
+ , iJ − mk/ cos k
1662
+ 2 √γ1γ2
1663
+ , 1
1664
+ �T
1665
+ ⃗u2 = 1
1666
+ 2
1667
+
1668
+ − 1, iJ + mk/ cos k
1669
+ 2 √γ1γ2
1670
+ , iJ + mk/ cos k
1671
+ 2 √γ1γ2
1672
+ , 1
1673
+ �T
1674
+ ⃗u3 = 1
1675
+ 2
1676
+
1677
+ 1 + iJ cos k/mk, 2 √γ1γ2 cos k/mk, −2 √γ1γ2 cos k/mk, 1 + iJ cos k/mk
1678
+ �T
1679
+ ⃗u4 = 1
1680
+ 2
1681
+
1682
+ 1 − iJ cos k/mk, −2 √γ1γ2 cos k/mk, 2 √γ1γ2 cos k/mk, 1 − iJ cos k/mk
1683
+ �T.
1684
+ (S32)
1685
+ We make an ansatz for steady state |Ω⟩ as
1686
+ |Ω⟩ =
1687
+ π
1688
+
1689
+ k=0
1690
+ (z1 + z2a†
1691
+ kc†
1692
+ −k)(z3 + z4a†
1693
+ −kc†
1694
+ k)|0⟩.
1695
+ (S33)
1696
+ Solving the steady state equations: ζi|Ω⟩ = 0 for i = 1 ∼ 4, we get z1 = z2 and z3 = z4. Therefore, the solution of steady state
1697
+ (the Eq. (16) in the main text) is given by
1698
+ |Ω⟩ = 1
1699
+ N
1700
+ π
1701
+
1702
+ k=−π
1703
+ (1 + a†
1704
+ kc†
1705
+ −k)|0⟩.
1706
+ (S34)
1707
+
1708
+ 12
1709
+ By using Tr(ρs) = 1, we get the normalization factor N as
1710
+ N =C ⟨S0|
1711
+ π
1712
+
1713
+ k=−π
1714
+ (1 + a†
1715
+ kc†
1716
+ −k)|0⟩ = 2L,
1717
+ (S35)
1718
+ where L is the length of the chain. The details of N = 2L is given in subsection D. In addition, we get steady state in real space
1719
+ given by
1720
+ |Ω⟩ = 1
1721
+ N exp
1722
+
1723
+ π
1724
+
1725
+ k=−π
1726
+ a†
1727
+ kc†
1728
+ −k
1729
+
1730
+ |0⟩ = 1
1731
+ N exp
1732
+
1733
+ L
1734
+
1735
+ l=1
1736
+ a†
1737
+ l c†
1738
+ l
1739
+
1740
+ |0⟩ = 1
1741
+ N
1742
+ L
1743
+
1744
+ l=1
1745
+ (1 + a†
1746
+ l c†
1747
+ l )|0⟩.
1748
+ (S36)
1749
+ Under the parity constraint, valid eigenstates in even parity channel are |Ω⟩, ζ
1750
+
1751
+ α1(ki)ζ
1752
+
1753
+ α2(kj)|Ω⟩, ζ
1754
+
1755
+ α1(ki)ζ
1756
+
1757
+ α2(kj)ζ
1758
+
1759
+ α3(km)ζ
1760
+
1761
+ α4(kn)|Ω⟩, · · ·
1762
+ Secondly, we diagonalize ˆLk in the odd channel (P = −1). The process of diagonalization is the same as it in the even channel,
1763
+ however, the eigenvectors ⃗v and ⃗u of odd channel are different from them in even channel. We mark the eigenvectors and normal
1764
+ master modes of the odd channel with ’∗’:
1765
+ ζ
1766
+
1767
+ 1∗(k) = (a†
1768
+ k c†
1769
+ k a−k c−k) · ⃗v1∗,
1770
+ ζ
1771
+
1772
+ 2∗(k) = (a†
1773
+ k c†
1774
+ k a−k c−k) · ⃗v2∗,
1775
+ ζ
1776
+
1777
+ 3∗(k) = (ak ck a†
1778
+ −k c†
1779
+ −k) · ⃗u3∗,
1780
+ ζ
1781
+
1782
+ 4∗(k) = (ak ck a†
1783
+ −k c†
1784
+ −k) · ⃗u4∗,
1785
+ ζ1∗(k) = (ak ck a†
1786
+ −k c†
1787
+ −k) · ⃗u1∗,
1788
+ ζ2∗(k) = (ak ck a†
1789
+ −k c†
1790
+ −k) · ⃗u2∗,
1791
+ ζ3∗(k) = (a†
1792
+ k c†
1793
+ k a−k c−k) · ⃗v3∗,
1794
+ ζ4∗(k) = (a†
1795
+ k c†
1796
+ k a−k c−k) · ⃗v4∗,
1797
+ (S37)
1798
+ where
1799
+ ⃗v1∗ = 1
1800
+ 2
1801
+
1802
+ 1 + iJ cos k/mk, −2 √γ1γ2 cos k/mk, 2 √γ1γ2 cos k/mk, 1 + iJ cos k/mk
1803
+ �T
1804
+ ⃗v2∗ = 1
1805
+ 2
1806
+
1807
+ 1 − iJ cos k/mk, 2 √γ1γ2 cos k/mk, −2 √γ1γ2 cos k/mk, 1 − iJ cos k/mk
1808
+ �T
1809
+ ⃗v3∗ = 1
1810
+ 2
1811
+
1812
+ − 1, −iJ + mk/ cos k
1813
+ 2 √γ1γ2
1814
+ , −iJ + mk/ cos k
1815
+ 2 √γ1γ2
1816
+ , 1
1817
+ �T
1818
+ ⃗v4∗ = 1
1819
+ 2
1820
+
1821
+ − 1, −iJ − mk/ cos k
1822
+ 2 √γ1γ2
1823
+ , −iJ − mk/ cos k
1824
+ 2 √γ1γ2
1825
+ , 1
1826
+ �T
1827
+ ⃗u1∗ = 1
1828
+ 2
1829
+
1830
+ 1, iJ − mk/ cos k
1831
+ 2 √γ1γ2
1832
+ , −iJ + mk/ cos k
1833
+ 2 √γ1γ2
1834
+ , 1
1835
+ �T
1836
+ ⃗u2∗ = 1
1837
+ 2
1838
+
1839
+ 1, iJ + mk/ cos k
1840
+ 2 √γ1γ2
1841
+ , −iJ − mk/ cos k
1842
+ 2 √γ1γ2
1843
+ , 1
1844
+ �T
1845
+ ⃗u3∗ = 1
1846
+ 2
1847
+
1848
+ − 1 − iJ cos k/mk, 2 √γ1γ2 cos k/mk, 2 √γ1γ2 cos k/mk, 1 + iJ cos k/mk
1849
+ �T
1850
+ ⃗u4∗ = 1
1851
+ 2
1852
+
1853
+ − 1 + iJ cos k/mk, −2 √γ1γ2 cos k/mk, −2 √γ1γ2 cos k/mk, 1 − iJ cos k/mk
1854
+ �T.
1855
+ (S38)
1856
+ Solving the equation, ζi∗|Ω∗⟩ = 0 for i = 1 ∼ 4, we get
1857
+ |Ω∗⟩ = 1
1858
+ N
1859
+ π
1860
+
1861
+ k=−π
1862
+ (1 − a†
1863
+ kc†
1864
+ −k)|0⟩ = 1
1865
+ N
1866
+ L
1867
+
1868
+ l=1
1869
+ (1 − a†
1870
+ l c†
1871
+ l )|0⟩.
1872
+ (S39)
1873
+ Note that |Ω∗⟩ is even parity ( ˆP|Ω∗⟩ = +1|Ω∗⟩). Therefore, the valid eigenstates in odd parity channel are the states with odd
1874
+ numbers of excitations on the |Ω∗⟩, i.e. ζ
1875
+
1876
+ α1∗(ki)|Ω∗⟩, ζ
1877
+
1878
+ α1∗(ki)ζ
1879
+
1880
+ α2∗(k j)ζ
1881
+
1882
+ α3∗(km)|Ω∗⟩, · · ·
1883
+ In summary, the full eigenstates of ˆL are
1884
+ Steady state:
1885
+ |Ω⟩
1886
+ Single excitation:
1887
+ ζ
1888
+
1889
+ α1∗(ki) |Ω∗⟩
1890
+ Double excitation:
1891
+ ζ
1892
+
1893
+ α1(ki)ζ
1894
+
1895
+ α2(k j) |Ω⟩
1896
+ Triple excitation:
1897
+ ζ
1898
+
1899
+ α1∗(ki)ζ
1900
+
1901
+ α2∗(kj)ζ
1902
+
1903
+ α3∗(km) |Ω∗⟩
1904
+ Quadruple excitation:
1905
+ ζ
1906
+
1907
+ α1(ki)ζ
1908
+
1909
+ α2(kj)ζ
1910
+
1911
+ α3(km)ζ
1912
+
1913
+ α4(kn) |Ω⟩
1914
+ · · ·
1915
+ (S40)
1916
+
1917
+ 13
1918
+ B. Flat band condition
1919
+ When the condition J2 = 4γ1γ2 is satisfied, Liouvillian flat band occurs. We have λ1 = λ2 = −2γ, λ3 = λ4 = 2γ and mk = 0,
1920
+ which leads to divergence of eigenvectors ⃗v1, ⃗v2, ⃗u3, ⃗u4, ⃗v1∗, ⃗v2∗, ⃗u3∗ and ⃗u4∗. This indicates the exceptional point of ˆL. However,
1921
+ we can eliminate divergence by summing of these eigenvectors. Setting J = 2 √γ1γ2, we can get the normal master modes in
1922
+ even parity
1923
+ ζ
1924
+
1925
+ A(k) = (a†
1926
+ k c†
1927
+ k a−k c−k) · (⃗v1 + ⃗v2) = −a†
1928
+ k + c−k
1929
+ ζA(k) = (ak ck a†
1930
+ −k c†
1931
+ −k) · (⃗u1 + ⃗u2)/2 = 1
1932
+ 2(−ak + ick + ia†
1933
+ −k + c†
1934
+ −k)
1935
+ ζ
1936
+
1937
+ B(k) = ak ck a†
1938
+ −k c†
1939
+ −k) · (⃗u3 + ⃗u4) = ak + c†
1940
+ −k
1941
+ ζB(k) = (a†
1942
+ k c†
1943
+ k a−k c−k) · (⃗v3 + ⃗v4)/2 = 1
1944
+ 2(a†
1945
+ k − ic†
1946
+ k + ia−k + c−k),
1947
+ (S41)
1948
+ and in odd parity
1949
+ ζ
1950
+
1951
+ A∗(k) = (a†
1952
+ k c†
1953
+ k a−k c−k) · (⃗v1∗ + ⃗v2∗) = a†
1954
+ k + c−k
1955
+ ζA∗(k) = (ak ck a†
1956
+ −k c†
1957
+ −k) · (⃗u1∗ + ⃗u2∗)/2 = 1
1958
+ 2(ak + ick − ia†
1959
+ −k + c†
1960
+ −k)
1961
+ ζ
1962
+
1963
+ B∗(k) = ak ck a†
1964
+ −k c†
1965
+ −k) · (⃗u3∗ + ⃗u4∗) = −ak + c†
1966
+ −k
1967
+ ζB∗(k) = (a†
1968
+ k c†
1969
+ k a−k c−k) · (⃗v3∗ + ⃗v4∗)/2 = 1
1970
+ 2(−a†
1971
+ k − ic†
1972
+ k − ia−k + c−k).
1973
+ (S42)
1974
+ C. Ineffectiveness of odd parity
1975
+ Given an arbitrary state | ρ⟩, it can be decomposed into even and odd eigenstate of ˆL:
1976
+ | ρ⟩ =
1977
+ � �
1978
+ i
1979
+ Ce
1980
+ i |i⟩e
1981
+
1982
+ +
1983
+ � �
1984
+ j
1985
+ Co
1986
+ j | j⟩o
1987
+
1988
+ ,
1989
+ (S43)
1990
+ where |i⟩e and | j⟩o represents even and odd parity state in Eq. (S40). The expectation value of observation ˆO is
1991
+ C⟨S0| ˆO| ρ⟩ =
1992
+ � �
1993
+ i
1994
+ Ce
1995
+ i C⟨S0| ˆO|i⟩e
1996
+
1997
+ +
1998
+ � �
1999
+ j
2000
+ Co
2001
+ j C⟨S0| ˆO| j⟩o
2002
+
2003
+ .
2004
+ (S44)
2005
+ When ˆO has even fermionic operators, we have C⟨S0| ˆO| j⟩o = 0. When ˆO has odd fermionic operators, we have C⟨S0| ˆO|i⟩e = 0.
2006
+ Usually, in pure fermionic system, fermionic operators appear in pairs, so the odd parity part of ˆL does not influence the
2007
+ expectation value of observation.
2008
+ D. Correlation functions of steady state and quantum jump states
2009
+ Firstly, we show the details for the calculation of normalization factor N:
2010
+ N =C ⟨S0|
2011
+ L
2012
+
2013
+ l=1
2014
+ (1 + a†
2015
+ l c†
2016
+ l )|0⟩
2017
+ =
2018
+
2019
+ S
2020
+ ⟨0|( ˆPcL)S L · · · ( ˆPc1)S 1aS L
2021
+ L · · · aS 1
2022
+ 1 (1 + a†
2023
+ 1c†
2024
+ 1) · · · (1 + a†
2025
+ Lc†
2026
+ L)|0⟩
2027
+ =
2028
+
2029
+ S
2030
+ ⟨0|( ˆPcLaL)S L · · · ( ˆPc1a1)S 1 (1 + a†
2031
+ 1c†
2032
+ 1) · · · (1 + a†
2033
+ Lc†
2034
+ L)|0⟩
2035
+ = ⟨0|(1 + ˆPcLaL) · · · (1 + ˆPc1a1) (1 + a†
2036
+ 1c†
2037
+ 1) · · · (1 + a†
2038
+ Lc†
2039
+ L)|0⟩
2040
+ = ⟨0|
2041
+ L
2042
+
2043
+ l=1
2044
+
2045
+ (1 + ˆPclal)(1 + a†
2046
+ l c†
2047
+ l )
2048
+
2049
+ |0⟩
2050
+ = 2L.
2051
+ (S45)
2052
+
2053
+ 14
2054
+ Secondly, we show the particle number distribution of the steady state ns
2055
+ j
2056
+ ns
2057
+ j =C ⟨S0|a†
2058
+ jaj|Ω⟩
2059
+ = 1
2060
+ N
2061
+
2062
+ S
2063
+ ⟨0|( ˆPcL)S L · · · ( ˆPc1)S 1aS L
2064
+ L · · · aS 1
2065
+ 1 a†
2066
+ ja j (1 + a†
2067
+ 1c†
2068
+ 1) · · · (1 + a†
2069
+ Lc†
2070
+ L)|0⟩
2071
+ = 2L−1
2072
+ N ⟨0|(1 + ˆPc jaj)a†
2073
+ jaj(1 + a†
2074
+ jc†
2075
+ j)|0⟩
2076
+ = 1
2077
+ 2
2078
+ (S46)
2079
+ The other correlation functions of steady state can be calculated by the same method. The results are
2080
+ Gs
2081
+ j1,j2 = 0 (j1 � j2), Ds
2082
+ j1,j2 = 0, Ds∗
2083
+ j1, j2 = 0.
2084
+ (S47)
2085
+ Thirdly, we focus on a state from a quantum jump on the site l of the steady state. We denote this state as |φl⟩:
2086
+ |φl⟩ :=
2087
+ a†
2088
+ l ρsal
2089
+ Tr(a†
2090
+ l ρsal)
2091
+ =
2092
+ a†
2093
+ l c†
2094
+ l |Ω⟩
2095
+ C⟨S0|a†
2096
+ l c†
2097
+ l |Ω⟩
2098
+ .
2099
+ (S48)
2100
+ The particle number on site j of |φl⟩, denoted as nl
2101
+ j:
2102
+ nl
2103
+ j=l =C ⟨S0|a†
2104
+ l al|φl⟩
2105
+ =
2106
+ ⟨0|(1 + ˆPclal) a†
2107
+ l al a†
2108
+ l c†
2109
+ l (1 + a†
2110
+ l c†
2111
+ l )|0⟩
2112
+ ⟨0|(1 + ˆPclal)a†
2113
+ l c†
2114
+ l (1 + a†
2115
+ l c†
2116
+ l )|0⟩
2117
+ = 1.
2118
+ (S49)
2119
+ nl
2120
+ j�l =C ⟨S0|a†
2121
+ jaj|φl⟩
2122
+ =
2123
+ ⟨0|(1 + ˆPclal)a†
2124
+ l c†
2125
+ l (1 + a†
2126
+ l c†
2127
+ l ) (1 + ˆPcjaj)a†
2128
+ jc†
2129
+ j(1 + a†
2130
+ jc†
2131
+ j)|0⟩
2132
+ ⟨0|(1 + ˆPclal)a†
2133
+ l c†
2134
+ l (1 + a†
2135
+ l c†
2136
+ l ) (1 + ˆPcja j)(1 + a†
2137
+ jc†
2138
+ j)|0⟩
2139
+ = 1
2140
+ 2.
2141
+ (S50)
2142
+ By the same way, we get the other correlation functions of |φl⟩. The results are
2143
+ Gl
2144
+ j1,j2 = 0 (j1 � j2), Dl
2145
+ j1,j2 = 0, Dl∗
2146
+ j1, j2 = 0.
2147
+ (S51)
2148
+ S4. Evolution equations of correlation functions
2149
+ In this section, we derive the evolution equations of two-operator correlation functions both in real space and momentum
2150
+ space and show the symmetry of damping matrix in momentum space.
2151
+ A. Evolution equations of correlation functions in real space
2152
+ The evolution equation of the expectation value of operator ˆO in the open system is
2153
+ d
2154
+ dtTr( ˆOρ(t)) = Tr( ˆO d
2155
+ dtρ) = Tr( ˆOL(ρ)).
2156
+ (S52)
2157
+ By considering the Liouvillian L in Eq. (5) in the main text, the equation becomes
2158
+ d
2159
+ dtTr( ˆOρ(t)) = −iTr( ˆO[H, ρ]) + (1 − w)Tr( ˆODL(ρ)) + (1 + w)Tr( ˆODR(ρ)).
2160
+ (S53)
2161
+
2162
+ 15
2163
+ Using the relation Tr(ABC) = Tr(CAB), we have
2164
+ Tr( ˆO[H, ρ]) = Tr([ ˆO, H] ρ) = J
2165
+
2166
+ l
2167
+ Tr([ ˆO, a†
2168
+ l+1al + a†
2169
+ l al+1] ρ),
2170
+ (S54)
2171
+ Tr( ˆODL(ρ)) =
2172
+
2173
+ l
2174
+
2175
+ Tr( ˆO2Al ρ A†
2176
+ l ) − Tr( ˆOA†
2177
+ l Al ρ) − Tr( ˆO ρ A†
2178
+ l Al)
2179
+
2180
+ =
2181
+
2182
+ l
2183
+
2184
+ Tr([A†
2185
+ l , ˆO]Al ρ) + Tr(A†
2186
+ l [ ˆO, Al] ρ)
2187
+
2188
+ ,
2189
+ (S55)
2190
+ Tr( ˆODR(ρ)) =
2191
+
2192
+ l
2193
+
2194
+ Tr( ˆO2A†
2195
+ l ρ Al) − Tr( ˆOAlA†
2196
+ l ρ) − Tr( ˆO ρ AlA†
2197
+ l )
2198
+
2199
+ =
2200
+
2201
+ l
2202
+
2203
+ Tr([Al, ˆO]A†
2204
+ l ρ) + Tr(Al[ ˆO, A†
2205
+ l ] ρ)
2206
+
2207
+ .
2208
+ (S56)
2209
+ Substituting ˆO = a†
2210
+ l1al2, ˆO = al1al2 and ˆO = a†
2211
+ l2a†
2212
+ l1 into Eq.(S52) ∼ Eq.(S55), we get the evolution equations of Gl1,l2, Dl1,l2 and
2213
+ D∗
2214
+ l1,l2, respectively. Namely, the evolution equations of correlation functions in real space are
2215
+ d
2216
+ dtGl1,l2 = − 4γGl1,l2 + iJ(Gl1−1,l2 + Gl1+1,l2 − Gl1,l2−1 − Gl1,l2+1) + 2�γ + w(γ2 − γ1)� δl1,l2
2217
+ + √γ1γ2 (−Dl1−1,l2 − Dl1+1,l2 + Dl2,l1−1 + Dl2,l1+1) + √γ1γ2 (D∗
2218
+ l1,l2−1 + D∗
2219
+ l1,l2+1 − D∗
2220
+ l2−1,l1 − D∗
2221
+ l2+1,l1),
2222
+ (S57)
2223
+ d
2224
+ dt Dl1,l2 = + 2 √γ1γ2(−Gl1−1,l2 − Gl1+1,l2 + Gl2−1,l1 + Gl2+1,l1) + 2w √γ1γ2
2225
+ �δl1,l2−1 − δl2,l1−1
2226
+
2227
+ − 4γDl1,l2 − iJ/2 (Dl1−1,l2 + Dl1+1,l2 + Dl1,l2−1 + Dl1,l2+1) + iJ/2 (Dl2,l1−1 + Dl2,l1+1 + Dl2−1,l1 + Dl2+1,l1),
2228
+ (S58)
2229
+ d
2230
+ dt D∗
2231
+ l1,l2 = + 2 √γ1γ2(−Gl2,l1−1 − Gl2,l1+1 + Gl1,l2−1 + Gl1,l2+1) + 2w √γ1γ2
2232
+ �δl1,l2−1 − δl2,l1−1
2233
+
2234
+ − 4γD∗
2235
+ l1,l2 + iJ/2 (D∗
2236
+ l1−1,l2 + D∗
2237
+ l1+1,l2 + D∗
2238
+ l1,l2−1 + D∗
2239
+ l1,l2+1) − iJ/2 (D∗
2240
+ l2,l1−1 + D∗
2241
+ l2,l1+1 + D∗
2242
+ l2−1,l1 + D∗
2243
+ l2+1,l1).
2244
+ (S59)
2245
+ B. Evolution equations of correlation functions in momentum space
2246
+ The Liouvillian of our model in momentum space is shown in Eq. (9) in the main text. Substituting this equation into Eq. (S52),
2247
+ we have
2248
+ d
2249
+ dtTr( ˆOρ(t)) =
2250
+ π
2251
+
2252
+ k=−π
2253
+
2254
+ − i2J cos kTr( ˆO[ˆnk, ρ]) + (1 − w)Tr( ˆODL
2255
+ k(ρ)) + (1 + w)DR
2256
+ k (ρ)
2257
+
2258
+ =
2259
+ π
2260
+
2261
+ k=−π
2262
+
2263
+ − i2J cos kTr([ ˆO, ˆnk]ρ) + (1 − w)� Tr([B†
2264
+ k, ˆO]Bkρ) + Tr(B†
2265
+ k[ ˆO, Bk]ρ) �
2266
+ + (1 + w)� Tr([Bk, ˆO]B†
2267
+ kρ) + Tr(Bk[ ˆO, B†
2268
+ k]ρ) ��
2269
+ .
2270
+ (S60)
2271
+ Substituting ˆO = a†
2272
+ k1ak2, ˆO = a†
2273
+ −k2a−k1, ˆO = ak2a−k1 and ˆO = a†
2274
+ −k2a†
2275
+ k1 into Eq.(S60), we get the evolution equations of correlation
2276
+ functions Gk1,k2, G−k2,−k1, Dk2,−k1 and D∗
2277
+ k1,−k2:
2278
+ d
2279
+ dt
2280
+ ���������������
2281
+ Gk1,k2
2282
+ G−k2,−k1
2283
+ Dk2,−k1
2284
+ D∗
2285
+ k1,−k2
2286
+ ���������������
2287
+ = Xk1k2
2288
+ ���������������
2289
+ Gk1,k2
2290
+ G−k2,−k1
2291
+ Dk2,−k1
2292
+ D∗
2293
+ k1,−k2
2294
+ ���������������
2295
+ + Vk1k2,
2296
+ (S61)
2297
+ where
2298
+ Xk1k2 =
2299
+ ��������������
2300
+ −4γ + i2J(cos k1 − cos k2)
2301
+ 0
2302
+ 4 √γ1γ2 cos k1
2303
+ 4 √γ1γ2 cos k2
2304
+ 0
2305
+ −4γ + i2J(cos k2 − cos k1)
2306
+ −4 √γ1γ2 cos k2
2307
+ −4 √γ1γ2 cos k1
2308
+ 4 √γ1γ2 cos k1
2309
+ −4 √γ1γ2 cos k2
2310
+ −4γ − i2J(cos k1 + cos k2)
2311
+ 0
2312
+ 4 √γ1γ2 cos k2
2313
+ −4 √γ1γ2 cos k1
2314
+ 0
2315
+ −4γ + i2J(cos k1 + cos k2)
2316
+ ��������������
2317
+ (S62)
2318
+
2319
+ 16
2320
+ and
2321
+ Vk1k2 = δk1,k2
2322
+ ��������������
2323
+ 2γ + 2w(γ2 − γ1)
2324
+ 2γ + 2w(γ2 − γ1)
2325
+ i4w √γ1γ2 sin k1
2326
+ −i4w √γ1γ2 sin k1
2327
+ ��������������
2328
+ .
2329
+ (S63)
2330
+ Eq. (S61) ∼ Eq. (S63) are just the same equations as Eq. (10) and Eq. (11) in the main text.
2331
+ C. Symmetry of damping matrix
2332
+ Denoting k = (k1, k2), then we can check the damping matrix Xk has TRS, PHS, CS, IS and PTS:
2333
+ TRS : UT X∗
2334
+ k U−1
2335
+ T = X−k
2336
+ =⇒
2337
+ UT = σz ⊗ σx; UT U∗
2338
+ T = 1
2339
+ PHS : UC XT
2340
+ k U−1
2341
+ C = −X−k
2342
+ =⇒ UC = I ⊗ σx; UCU∗
2343
+ C = 1
2344
+ CS : UΓ X†
2345
+ k U−1
2346
+ Γ = −Xk
2347
+ =⇒
2348
+ UΓ = σz ⊗ I; U2
2349
+ Γ = 1
2350
+ IS : UP Xk U−1
2351
+ P = X−k
2352
+ =⇒
2353
+ UP = I ⊗ I
2354
+ PTS : UPT X∗
2355
+ k U−1
2356
+ PT = Xk
2357
+ =⇒ UPT = σz ⊗ σx.
2358
+ (S64)
2359
+ Compared with the symmetry of the Liouvillian in Eq.(S30), Xk has higher symmetry.
2360
+ S5. Particle number distribution of steady state
2361
+ As for steady state, the Eq. (S61) equals to 0, so we can get the correlation functions of steady state by
2362
+
2363
+ Gs
2364
+ k1,k2 Gs
2365
+ −k2,−k1 Ds
2366
+ k2,−k1 Ds∗
2367
+ k1,−k2
2368
+ �T = −X−1
2369
+ k1k2Vk1k2
2370
+ (S65)
2371
+ When k1 = k2 = k, we get particle number distribution of steady state in momentum space ns
2372
+ k:
2373
+ ns
2374
+ k = Gs
2375
+ kk = (1 − w)γ1 + (1 + w)γ2
2376
+
2377
+
2378
+ 2Jwγ1γ2 cos2 k sin k
2379
+ γ3 + γ(J2 − 4γ1γ2) cos2 k.
2380
+ (S66)
2381
+ Due to the translation invariance of our system, the particle number distributes uniformly on each site. Therefore, particle
2382
+ number on site l in the thermodynamic limit can be calculated by
2383
+ ns
2384
+ l = 1
2385
+ L
2386
+ L
2387
+
2388
+ j=1
2389
+ ns
2390
+ j = 1
2391
+ L
2392
+
2393
+ k
2394
+ ns
2395
+ k = 1
2396
+
2397
+ � π
2398
+ k=−π
2399
+ dk ns
2400
+ k = 1
2401
+ 2 + w(γ2 − γ1)
2402
+
2403
+ .
2404
+ (S67)
2405
+ S6. The relationship between the damping-matrix spectra and the Liouvillian spectra
2406
+ In this section, we demonstrate that for a real physical process with closed evolution equations of correlation functions the
2407
+ damping-matrix spectra are the subset of the Liouvillian spectra.
2408
+ The general form of closed evolution equations of correlation functions is
2409
+ d
2410
+ dtΨ = X Ψ + V,
2411
+ (S68)
2412
+ where
2413
+ X
2414
+ is
2415
+ the
2416
+ damping
2417
+ matrix,
2418
+ Ψ
2419
+ is
2420
+ the
2421
+ vector
2422
+ of
2423
+ correlation
2424
+ functions,
2425
+ for
2426
+ example,
2427
+ Ψ
2428
+ is
2429
+ taken
2430
+ as
2431
+ (Gk1,k2,G−k2,−k1, Dk2,−k1, D∗
2432
+ k1,−k2)T in our model. The vector V induces the correlation function vector of steady state ΨS as
2433
+ ΨS = −X−1V. By deducting ΨS , we have
2434
+ d
2435
+ dt(Ψ(t) − ΨS ) = X (Ψ(t) − ΨS ).
2436
+ (S69)
2437
+
2438
+ 17
2439
+ If the correlation function vector ΨΓ is governed by the eigen equation of damping matrix, we have
2440
+ X (ΨΓ(t) − ΨS ) = Γ (ΨΓ(t) − ΨS ),
2441
+ (S70)
2442
+ where Γ is the eigenvalue of X. The equation in the initial time is
2443
+ X (ΨΓ(0) − ΨS ) = Γ (ΨΓ(0) − ΨS ),
2444
+ (S71)
2445
+ Then from Eq. (S69), we obtain
2446
+ ΨΓ(t) − ΨS = eΓt (ΨΓ(0) − ΨS ).
2447
+ (S72)
2448
+ If ΨΓ is in a real physical process, we will have
2449
+ ΨΓ(t) = C⟨S0| ˆΨ e ˆLC t| ρ(0)⟩C,
2450
+ (S73a)
2451
+ ΨS = C⟨S0| ˆΨ e ˆLC t|Ω⟩C = C⟨S0| ˆΨ |Ω⟩C,
2452
+ (S73b)
2453
+ where ˆLC is the Liouvillian of system in representation C, | ρ(0)⟩C is the initial state of system and |Ω⟩C is the steady state
2454
+ of system.
2455
+ ˆΨ is the vector of operators in terms of correlation function vector Ψ, for example, in our model ˆΨ equals to
2456
+ (a†
2457
+ k1ak2, a†
2458
+ −k2a−k1, a−k1ak2, a†
2459
+ −k2a†
2460
+ k1)T. Substituting Eq.(S73) into Eq.(S72), we obtain
2461
+ C⟨S0| ˆΨ e ˆLC t� | ρ(0)⟩C − |Ω⟩C
2462
+ � = C⟨S0| ˆΨ eΓ t� | ρ(0)⟩C − |Ω⟩C
2463
+ �.
2464
+ (S74)
2465
+ Comparing the two sides of the above equation, we have
2466
+ e ˆLC t� | ρ(0)⟩C − |Ω⟩C
2467
+ � = eΓ t� | ρ(0)⟩C − |Ω⟩C
2468
+ �,
2469
+ (S75)
2470
+ and thus the eigenvalue Γ of damping matrix X is also the eigenvalue of Liouvillian ˆLC.
2471
2472
+ [1] M.-D. Choi, Completely positive linear maps on complex matrices, Linear Algebra Applications 10, 285 (1975).
2473
+ [2] A. Jamiołkowski, Linear transformations which preserve trace and positive semidefiniteness of operators, Rep. Math. Phys. 3, 275 (1972).
2474
+ [3] J. E. Tyson, Operator-Schmidt decompositions and the Fourier transform, with applications to the operator-Schmidt numbers of unitaries,
2475
+ J. Phys. A: Math. Gen. 36, 10101 (2003).
2476
+ [4] M. Zwolak and G. Vidal, Mixed-State Dynamics in One-Dimensional Quantum Lattice Systems: A Time-Dependent Superoperator Renor-
2477
+ malization Algorithm, Phys. Rev. Lett. 93, 207205 (2004).
2478
+ [5] K. Kawabata, K. Shiozaki, M. Ueda, and M. Sato, Symmetry and Topology in Non-Hermitian Physics, Phys. Rev. X 9, 041015 (2019).
2479
+ [6] A. W. W. Ludwig, Topological phases: classification of topological insulators and superconductors of noninteracting fermions, and beyond,
2480
+ Physica Scripta T168, 014001 (2015).
2481
+ [7] C.-H. Liu, H. Jiang, and S. Chen, Topological classification of non-Hermitian systems with reflection symmetry, Phys. Rev. B 99, 125103
2482
+ (2019).
2483
+ [8] C.-H. Liu and S. Chen, Topological classification of defects in non-Hermitian systems, Phys. Rev. B 100, 144106 (2019).
2484
+
TdE5T4oBgHgl3EQfAQ7r/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
VdAyT4oBgHgl3EQfhfjt/content/tmp_files/2301.00380v1.pdf.txt ADDED
@@ -0,0 +1,535 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Ultrafast switching of persistent electron and hole currents in ring molecules
2
+ Tennesse Joyce and Agnieszka Jaron
3
+ JILA and Department of Physics, University of Colorado, Boulder, CO-80309, USA
4
+ (Dated: January 3, 2023)
5
+ A circularly polarized laser pulse can induce persistent intra-molecular currents by either exciting
6
+ or ionizing molecules. These two cases are identified as electron currents and hole currents, respec-
7
+ tively, and up to now they have been studied only separately. We report ab initio time-dependent
8
+ density-functional theory (TDDFT) simulations of currents during resonance-enhanced two-photon
9
+ ionization of benzene, which reveal for the first time that both electron and hole currents can be
10
+ present simultaneously. By adjusting the intensity of the laser pulse, the balance between the two
11
+ types of current can be controlled, and the overall sign of the current can be switched. We provide
12
+ a physical explanation for the effect in terms of complex molecular orbitals which is consistent with
13
+ the TDDFT simulations.
14
+ It has long been understood that, in response to an ap-
15
+ plied magnetic field, the delocalized electrons of an aro-
16
+ matic molecule circulate in so-called aromatic ring cur-
17
+ rent [1, 2]. This effect is important in nuclear magnetic
18
+ resonance spectroscopy, where the internal magnetic field
19
+ generated by the ring current is responsible for diamag-
20
+ netic shielding [3]. In 2006, it was proposed that ring
21
+ currents in molecules could also be induced by ultra-
22
+ short laser pulses with circular or elliptical polarization
23
+ [4, 5]. The basic mechanism is that angular momentum
24
+ carried by light is transfered to electrons in a molecule.
25
+ Due to conservation of angular momentum, the current
26
+ persists after the pulse has ended—even without an ex-
27
+ ternal magnetic field.
28
+ Various experiments on atomic
29
+ targets have confirmed the existence of the effect [6, 7],
30
+ although no direct observational data is available in the
31
+ case of molecules. Recent interest in photoinduced ring
32
+ currents is motivated by the rapid technological advances
33
+ in polarization control of high-harmonic radiation made
34
+ in the last few years [8–10], which may enable experimen-
35
+ tal study of these phenomena in the near future [11].
36
+ There are several major advantages of photoinduced
37
+ ring currents compared to those induced by static mag-
38
+ netic fields. First, the current is expected to be orders of
39
+ magnitude stronger, and so is the induced magnetic field
40
+ [12]. Second, they enable femtosecond (or even attosec-
41
+ ond) time-resolved studies of aromaticity and magnetism
42
+ [13, 14]. Lastly, they establish the possibility for coherent
43
+ control of ring currents [15], which may have applications
44
+ for controlling chemical reactions or the operation of ad-
45
+ vanced opto-electronic devices.
46
+ In this Letter we predict a novel effect which causes
47
+ the dominant charge carrier of the ring current to transi-
48
+ tion from electrons to holes as the peak laser intensity in-
49
+ creases past around 1012 W/cm2. We illustrate the effect
50
+ with a series of ab initio time-dependent density func-
51
+ tional theory (TDDFT) simulations of benzene (C6H6),
52
+ which is the prototypical aromatic molecule. Lastly, we
53
+ demonstrate that the effect is not accounted for in the
54
+ commonly used few level model of ring currents, due to
55
+ the fact that it neglects ionization. This calls into ques-
56
+ tion the results of several previous studies (e.g. [4, 5, 15])
57
+ where it was assumed that the few level model is accurate
58
+ for laser intensities on the order of 1012 W/cm2.
59
+ We begin by introducing the distinction between elec-
60
+ tron and hole current: when an electron is promoted to
61
+ an orbital with nonzero angular momentum, this creates
62
+ an electron current; when an electron is removed (e.g.,
63
+ ionized) from an orbital with nonzero angular momen-
64
+ tum, this creates a hole current. So far, hole currents
65
+ have mostly been studied in the context of strong field
66
+ ionization of atoms by circularly polarized laser pulses,
67
+ and it was recently confirmed experimentally that a hole
68
+ can be created with a specific angular momentum relative
69
+ to the laser polarization [16–19]. Electron currents on the
70
+ other hand do not involve ionization, only excitation.
71
+ However, in the interaction of atoms and molecules
72
+ with strong laser fields, excitation and ionization are of-
73
+ ten closely related and occur together. A typical example
74
+ is resonance-enhanced multiphoton ionization (REMPI)
75
+ [20, 21], a two-step ionization process wherein an atom
76
+ or molecule is first excited to an intermediate state (that
77
+ must be resonant with some multiple of the laser fre-
78
+ quency) and then subsequently ionized.
79
+ Now consider
80
+ REMPI in a system where the intermediate excited state
81
+ corresponds to an electron current, and the final ionized
82
+ state corresponds to a hole current (we will show that
83
+ benzene is such a system). The balance between excita-
84
+ tion and ionization (and therefore electron and hole cur-
85
+ rent) will depend on the laser intensity because the pro-
86
+ cesses involve different numbers of photons (and therefore
87
+ scale with different powers of intensity). In particular at
88
+ low intensities we expect electron current to dominate
89
+ (excitation), and at high intensities we expect hole cur-
90
+ rent to dominate (ionization).
91
+ Our main theoretical method is TDDFT, as imple-
92
+ mented by Octopus [22–24], which provides a fully non-
93
+ perturbative description of the light-matter interaction.
94
+ As a reference point to compare against the full TDDFT
95
+ simulations, we also consider the few level model of ring
96
+ currents (e.g. [5]). We discuss the implementations of
97
+ both models in [25]. Because the few level model does
98
+ not include ionization, we expect the two models to di-
99
+ verge at high enough laser intensities.
100
+ The laser pulse in our simulations is described in the
101
+ arXiv:2301.00380v1 [physics.chem-ph] 1 Jan 2023
102
+
103
+ 2
104
+ FIG. 1.
105
+ (a) Visualization of the current density based on the component passing through a plane bisecting the molecule as
106
+ shown (averagea over all possible orientations of that plane [see Eq. (2)]) (b) Cross sections of the current density taken at
107
+ the end of the laser pulse (t = 200 a.u.) for several different simulations. At low laser intensity the co-rotating current (red)
108
+ dominates, while at high intensity the counter-rotating current (blue) dominates. Note: Each plot is scaled individually relative
109
+ to the maximum absolute value within that plot. The nuclei lie in the plane z = 0 with the carbon ring at x = ±2.63 a.u. and
110
+ the hydrogen ring at x = ±4.69 a.u..
111
+ dipole approximation by the following electric field,
112
+ E(t) =
113
+
114
+ E sin2 (πt/T) Re
115
+
116
+ ˆϵeiω(t−T/2)�
117
+ ,
118
+ 0 < t < T,
119
+ 0,
120
+ otherwise,
121
+ (1)
122
+ with central frequency ω = 6.76 eV (183 nm), dura-
123
+ tion T = 16π/ω = 202 a.u. = 4.9 fs, circular polar-
124
+ ization ˆϵ = (ˆx + iˆy)/
125
+
126
+ 2 (with the molecule in the xy-
127
+ plane), and a variable peak amplitude E.
128
+ The central
129
+ frequency was chosen to be resonant with the doubly
130
+ degenerate E1u state (as computed with linear response
131
+ TDDFT [25]), which is predominantly associated with
132
+ the HOMO-LUMO transition (HOMO = Highest Occu-
133
+ pied Molecular Orbital; LUMO = Lowest Unoccupied
134
+ Molecular Orbital).
135
+ Because the computed ionization
136
+ threshold is 9.0 eV < 2ω, this laser pulse is designed
137
+ to drive 1+1 REMPI where one photon is enough to
138
+ promote electron to the excited state and one additional
139
+ photon to ionize.
140
+ After interacting with the laser pulse (t > T), the ben-
141
+ zene molecule is in a superposition of the A1g ground
142
+ state and the E1u excited state and also, to an extent,
143
+ ionized. This causes oscillations in the charge and current
144
+ densities ρ(r, t) and J(r, t), respectively, with period 612
145
+ as (corresponding to the energy difference between the
146
+ ground state and excited states), which are an example
147
+ of attosecond charge migration [26].
148
+ In order to visualize the current we isolate the sta-
149
+ tionary component of the current density, by computing
150
+ an angle averaged cross section defined by the following
151
+ integral (in cylindrical coordinates ρ, z, φ),
152
+ J(x, z, t) = 1
153
+
154
+ � 2π
155
+ 0
156
+ ˆφ · J(|x|, z, φ)dφ.
157
+ (2)
158
+ The geometric interpretation of this integral is given in
159
+ Fig.
160
+ 1.
161
+ The angle averaging procedure for the few
162
+ level model causes that the fast-oscillating component
163
+ effectively vanishes. Within the few-level model, the fast-
164
+ oscillating component of the current density is zeroed out
165
+ by this averaging procedure because of its parity.
166
+ It has
167
+ similar effect on TDDFT results, and therefore J(x, z, y)
168
+ has only a very gradual time dependence for t > T. The
169
+ same is true for TDDFT. These integrated current densi-
170
+ ties are plotted in Fig. 1b. At low intensities the current
171
+ is a combination of a strong co-rotating current (red) and
172
+ a weak counter-rotating current (blue), while at high in-
173
+ tensities the counter-rotating current dominates. As we
174
+ explain below (see Fig. 4), the reversal is a signature of
175
+ the transition from electron to hole current regime.
176
+ The oscillatory component of the charge motion is best
177
+ visualized by plotting the charge displacement,
178
+ ∆ρ(r, t) = ρ(r, t) − ρ(r, 0),
179
+ (3)
180
+ shown in Fig. 2. The cloud of displaced charge circulates
181
+ around the molecule with the expected period of 612 as,
182
+ and this continues even after the pulse ends.
183
+ Overall,
184
+ both the magnitude and shape of the charge displace-
185
+ ment are remarkably similar between the two models,
186
+ however there are some subtle differences.
187
+ First, long
188
+ after the laser pulse the two models gradually become
189
+ desynchronized. Second, in TDDFT there appears to be
190
+ a rearrangement of charge in the plane of the molecule,
191
+ whereas the few level model only predicts the dynamics
192
+ above and below the plane.
193
+
194
+ (a)
195
+ (b)
196
+ 3.8 × 1011 W/cm2
197
+ Few-level Model
198
+ 4.0
199
+ 2.0
200
+ ('n
201
+ -0.0
202
+ N
203
+ -2.0
204
+ -4.0
205
+ 5 × 1012 W/cm²
206
+ 1013 W/cm²
207
+ 4.0
208
+ 2.0
209
+ -0.0
210
+ -2.0
211
+ -4.0
212
+ -6.0
213
+ -3.0
214
+ 0.0
215
+ 3.0
216
+ -6.0
217
+ -3.0
218
+ 0.0
219
+ 3.0
220
+ 6.0
221
+ x (a.u.)
222
+ x (a.u.)3
223
+ FIG. 2.
224
+ Snapshots of the charge displacement induced by a circularly-polarized laser pulse with peak intensity 5×1012 W/cm2
225
+ taken around the peak of the laser pulse (first three columns t ≈ 100 a.u.) and after the laser pulse (last three columns t ≈ 400
226
+ a.u.). Light areas indicate excess electrons while dark areas indicate fewer electrons, as compared to the ground state charge
227
+ density before the laser pulse. We compare the results between the two theoretical models, TDDFT (top row) and the few
228
+ level model (bottom row).
229
+ FIG. 3.
230
+ Comparison of full TDDFT simulations (solid blue
231
+ line) to the few level model (orange dashed line). For peak
232
+ intensities, when ionization (dotted green line) becomes non-
233
+ negligible, the two models begin to disagree.
234
+ The smooth
235
+ lines have been interpolated between the calculated intensities
236
+ using the method described in [25].
237
+ Another important observation about the density dif-
238
+ ference is that the dark areas are generally larger than
239
+ the light areas. In the TDDFT results one reason for this
240
+ is ionization, with the ionization probability given by
241
+ P ionize = −
242
+
243
+ ∆ρ(r, 2T)d3r,
244
+ (4)
245
+ where the integral ranges over the simulation box. Unex-
246
+ pectedly, the few level model also appears to have dark
247
+ areas larger than light areas even though it does not
248
+ include ionization, and in fact the charge displacement
249
+ must integrate to zero in that model. The reason for this
250
+ is that the E1u is of mixed character, part of which in-
251
+ volves excitation to LUMO + 3 [25].
252
+ Note: The excess
253
+ of darker areas in the TDDFT model is a combination of
254
+ both ionization and excitation to LUMO +3 orbital.
255
+ The intensity dependence of the dynamics is illustrated
256
+ in Fig. 3. using the current. Note: this current is directly
257
+ proportional to z-component of the magnetic moment as
258
+ well as z-component of electronic angular momentum),
259
+ Since the domain of integration is the simulation box,
260
+ ionized electrons are not included.
261
+ For this reason we
262
+ plot Lz(2T) so that the ionizing wavepacket has enough
263
+ time to leave the box. Whereas in the few level model
264
+ the magnetic moment increases monotonically with the
265
+ laser intensity (up to about 1013 W/cm2, after which
266
+ the system Rabi oscillates back to the ground state), in
267
+ TDDFT the current starts to decrease already around
268
+ 1012 W/cm2, and reverses sign for even higher intensities.
269
+ We also plot the ionization probability (defined in Eq. 4),
270
+ and conclude that the reversal occurs precisely when the
271
+ ionization probability becomes non-negligible.
272
+ The implications of the transition from electron to hole
273
+ current on the charge dynamics, and the underlying phys-
274
+ ical mechanism responsible for that transition, can be
275
+ understood in more detail describe in more detail using
276
+ complex molecular orbitals, as illustrated schematically
277
+ in Fig. 4. These orbitals represent a change of basis from
278
+ the usual real-valued Kohn-Sham orbitals ψn(r) (defined
279
+ in [25]),
280
+ ψHOMO
281
+ ±
282
+ (r) = [ψ14(r) ± iψ15(r)] /
283
+
284
+ 2,
285
+ (5)
286
+ ψLUMO
287
+ ±
288
+ (r) = [ψ16(r) ± iψ17(r)] /
289
+
290
+ 2.
291
+ (6)
292
+ The advantage of using complex orbitals is that they are
293
+ eigenfunctions of the 6-fold symmetry operator (rotation
294
+
295
+ t = 105 a.u.
296
+ t = 110 a.u.
297
+ t = 390 a.u.
298
+ t = 395 a.u.
299
+ t = 100 a.u.
300
+ t = 400 a.u.
301
+ TDDFT
302
+ Few
303
+ Level
304
+ Model0.8
305
+ Lz (TDDFT)
306
+ Lz (Few level)
307
+ 0.6
308
+ lonizationprobability(TDDFT)
309
+ or probability
310
+ 0.4
311
+ 0.2
312
+ (a.u.)
313
+ 0.0
314
+ 0.2
315
+ 1011
316
+ 1012
317
+ 1013
318
+ Peak intensity (W/cm2)4
319
+ FIG. 4.
320
+ Schematic illustrating the complex molecular or-
321
+ bitals and the physical mechanism for the transition from
322
+ electron to hole current. Color indicates the complex phase.
323
+ about the molecular axis by 60◦),
324
+ exp
325
+
326
+ − iπ
327
+ 3¯h
328
+ ˆLz
329
+ ¯h
330
+
331
+ ψHOMO
332
+ ±
333
+ (r) = exp
334
+
335
+ ∓iπ
336
+ 3
337
+
338
+ ψHOMO
339
+ ±
340
+ (r),(7)
341
+ exp
342
+
343
+ −iπ
344
+ 3
345
+ ˆLz
346
+ ¯h
347
+
348
+ ψLUMO
349
+ ±
350
+ (r) = exp
351
+
352
+ ∓2iπ
353
+ 3
354
+
355
+ ψLUMO
356
+ ±
357
+ (r).(8)
358
+ The complex orbitals have magnetic quantum numbers
359
+ m defined modulo 6: ψHOMO
360
+ ±
361
+ have m = ±1 and ψLUMO
362
+ ±
363
+ have m = ±2. Just as for atomic orbitals, the sign of
364
+ m indicates the direction the electron circulates around
365
+ the molecule, and the magnitude indicates more-or-less
366
+ the angular speed. We have chosen our conventions such
367
+ that m > 0 electrons are co-rotating with the laser field,
368
+ and m < 0 electrons are counter-rotating.
369
+ Using the notation of complex orbitals, Fig. 4 illus-
370
+ trates how in the ground state, both ψHOMO
371
+ ±
372
+ are dou-
373
+ bly occupied, and consequently there is zero net cur-
374
+ rent. When the benzene molecule is exposed to a cir-
375
+ cularly polarized laser pulse, the usual selection rule
376
+ ∆m = 1 applies (here we assume the laser is polar-
377
+ ized in the molecular plane, see [25] for the more gen-
378
+ eral case), so that the only dipole-allowed transition is
379
+ ψHOMO
380
+ +
381
+ to ψLUMO
382
+ +
383
+ , which is the dominant component of
384
+ the E1u excited state. The electron excited to LUMO
385
+ contributes a strong co-rotating current (m = +2), but
386
+ the imbalance of electrons in the HOMO contributes a
387
+ weaker counter-rotating current (m = −1). This can al-
388
+ ternatively be interpreted as a positively charged hole
389
+ occupying ψHOMO
390
+ +
391
+ producing a co-rotating hole current
392
+ (rather than a counter-rotating electron current). This
393
+ is precisely what we see in the top row of Fig. 1b, two
394
+ components to the current with opposite sign (red and
395
+ blue).
396
+ In order to explain the reversal of the current at higher
397
+ intensity (bottom row of Fig. 1b), we simply recognize
398
+ that the electron previously excited to ψLUMO
399
+ +
400
+ can ab-
401
+ sorb a second photon from the same laser pulse, ionizing,
402
+ and leaving behind only the hole current. The balance
403
+ between the one-photon excitation and the two-photon
404
+ ionization processes can be controlled by varying the laser
405
+ intensity, because the first process scales with I while the
406
+ second process scales with I2 (with I ∝ E2 the laser in-
407
+ tensity). Furthermore, it is now apparent that the sign
408
+ reversal can be interpreted as a change in the dominant
409
+ charge carrier from electrons to holes.
410
+ In conclusion, we have shown that both electron and
411
+ hole currents are present during resonance-enhanced two-
412
+ photon ionization of benzene, and the balance between
413
+ the two current regimes can be controlled by varying the
414
+ peak laser intensity. We have proposed a simple expla-
415
+ nation for the effect in terms of molecular orbitals, which
416
+ is consistent with the results of full TDDFT simulations.
417
+ Variants of complex orbital model should apply to a wide
418
+ variety of molecules other than benzene, meaning that
419
+ the structure of the complex molecular orbitals can be
420
+ used to predict the interplay between electron and hole
421
+ currents during REMPI. In order to measure this effect
422
+ in experiment, several pump-probe schemes have been
423
+ proposed that are sensitive to the magnitude and direc-
424
+ tion of the ring current [7, 11]. In [25], we demonstrate
425
+ that the reversal is independent of the orientation of the
426
+ molecule, which greatly simplifies any potential exper-
427
+ iment.
428
+ Finally, our results suggest that the few level
429
+ model typically used to study photoinduced ring currents
430
+ may be insufficient even for moderate laser intensities
431
+ around 1012 W/cm2. A more ab initio nonperturbative
432
+ theory such as TDDFT, as used in present paper, is more
433
+ appropriate for this regime.
434
+ This work was supported by the NSF Grant No. PHY-
435
+ 1734006 and Grant No. PHY-2110628. This work uti-
436
+ lized resources from the University of Colorado Boulder
437
+ Research Computing Group, which is supported by the
438
+ National Science Foundation.
439
+ [1] J. A. N. F. Gomes and R. B. Mallion, Chemical Reviews
440
+ 101, 1349 (2001).
441
+ [2] T. M. Krygowski, H. Szatylowicz, O. A. Stasyuk, J. Do-
442
+ minikowska,
443
+ and M. Palusiak, Chemical Reviews 114,
444
+ 6383 (2014).
445
+ [3] T. Heine, C. Corminboeuf, and G. Seifert, Chemical Re-
446
+ views 105, 3889 (2005).
447
+ [4] I. Barth and J. Manz, Angewandte Chemie International
448
+
449
+ 2元
450
+ m=-2
451
+ m = +2
452
+ LUMO
453
+ Phase
454
+ 7元
455
+ 0
456
+ m=-1
457
+ m=+1
458
+ HOMO5
459
+ Edition 45, 2962 (2006).
460
+ [5] I. Barth, J. Manz, Y. Shigeta, and K. Yagi, Journal of
461
+ the American Chemical Society 128, 7043 (2006).
462
+ [6] M. Wollenhaupt, M. Krug, J. K¨ohler, T. Bayer, C. Sarpe-
463
+ Tudoran,
464
+ and T. Baumert, Applied Physics B 95, 245
465
+ (2009).
466
+ [7] S. Eckart, M. Kunitski, M. Richter, A. Hartung, J. Rist,
467
+ F. Trinter, K. Fehre, N. Schlott, K. Henrichs, L. P. H.
468
+ Schmidt, T. Jahnke, M. Sch¨offler, K. Liu, I. Barth,
469
+ J. Kaushal, F. Morales, M. Ivanov, O. Smirnova,
470
+ and
471
+ R. D¨orner, Nature Physics 14, 701 (2018).
472
+ [8] A. Fleischer, O. Kfir, T. Diskin, P. Sidorenko, and O. Co-
473
+ hen, Nature Photonics 8, 543 (2014).
474
+ [9] D. D. Hickstein, F. J. Dollar, P. Grychtol, J. L. Ellis,
475
+ R. Knut, C. Hern´andez-Garc´ıa, D. Zusin, C. Gentry,
476
+ J. M. Shaw, T. Fan, K. M. Dorney, A. Becker, A. Jaro´n-
477
+ Becker, H. C. Kapteyn, M. M. Murnane,
478
+ and C. G.
479
+ Durfee, Nature Photonics 9, 743 (2015).
480
+ [10] P.-C. Huang, C. Hern´andez-Garc´ıa, J.-T. Huang, P.-Y.
481
+ Huang, C.-H. Lu, L. Rego, D. D. Hickstein, J. L. Ellis,
482
+ A. Jaron-Becker, A. Becker, S.-D. Yang, C. G. Durfee,
483
+ L. Plaja, H. C. Kapteyn, M. M. Murnane, A. H. Kung,
484
+ and M.-C. Chen, Nature Photonics 12, 349 (2018).
485
+ [11] O. Neufeld and O. Cohen, Physical Review Letters 123,
486
+ 103202 (2019).
487
+ [12] K.-J. Yuan and A. D. Bandrauk, Physical Review A 88,
488
+ 013417 (2013).
489
+ [13] I. S. Ulusoy and M. Nest, Journal of the American Chem-
490
+ ical Society 133, 20230 (2011).
491
+ [14] K.-J. Yuan, C.-C. Shu, D. Dong,
492
+ and A. D. Ban-
493
+ drauk, The Journal of Physical Chemistry Letters 8, 2229
494
+ (2017).
495
+ [15] H. Mineo and Y. Fujimura, The Journal of Chemical
496
+ Physics 147, 224301 (2017).
497
+ [16] I. Barth and O. Smirnova, Physical Review A 84, 063415
498
+ (2011).
499
+ [17] T. Herath, L. Yan, S. K. Lee, and W. Li, Physical Review
500
+ Letters 109, 043004 (2012).
501
+ [18] X. Zhu, P. Lan, K. Liu, Y. Li, X. Liu, Q. Zhang, I. Barth,
502
+ and P. Lu, Optics Express 24, 4196 (2016).
503
+ [19] K. Liu and I. Barth, Physical Review A 94, 043402
504
+ (2016).
505
+ [20] V. S. Antonov, I. N. Knyazev, V. S. Letokhov, V. M.
506
+ Matiuk, V. G. Movshev,
507
+ and V. K. Potapov, Optics
508
+ Letters 3, 37 (1978).
509
+ [21] W. Dietz, H. J. Neusser, U. Boesl, E. W. Schlag,
510
+ and
511
+ S. H. Lin, Chemical Physics 66, 105 (1982).
512
+ [22] M. A. L. Marques, A. Castro, G. F. Bertsch, and A. Ru-
513
+ bio, Computer Physics Communications 151, 60 (2003).
514
+ [23] A. Castro, H. Appel, M. Oliveira, C. A. Rozzi, X. An-
515
+ drade, F. Lorenzen, M. a. L. Marques, E. K. U. Gross,
516
+ and A. Rubio, physica status solidi (b) 243, 2465 (2006).
517
+ [24] X. Andrade, D. Strubbe, U. D. Giovannini, A. H. Larsen,
518
+ M. J. T. Oliveira,
519
+ J. Alberdi-Rodriguez,
520
+ A. Varas,
521
+ I. Theophilou, N. Helbig, M. J. Verstraete, L. Stella,
522
+ F. Nogueira, A. Aspuru-Guzik, A. Castro, M. A. L.
523
+ Marques,
524
+ and A. Rubio, Physical Chemistry Chemical
525
+ Physics 17, 31371 (2015).
526
+ [25] See Supplemental Material for details on the TDDFT
527
+ and few-level models, orientation dependence, and the
528
+ method for interpolating over intensity.
529
+ [26] H. J. W¨orner, C. A. Arrell, N. Banerji, A. Cannizzo,
530
+ M. Chergui, A. K. Das, P. Hamm, U. Keller, P. M. Kraus,
531
+ E. Liberatore, P. Lopez-Tarifa, M. Lucchini, M. Meuwly,
532
+ C. Milne, J.-E. Moser, U. Rothlisberger, G. Smolentsev,
533
+ J. Teuscher, J. A. van Bokhoven, and O. Wenger, Struc-
534
+ tural Dynamics 4, 061508 (2017).
535
+
VdAyT4oBgHgl3EQfhfjt/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,447 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf,len=446
2
+ page_content='Ultrafast switching of persistent electron and hole currents in ring molecules Tennesse Joyce and Agnieszka Jaron JILA and Department of Physics, University of Colorado, Boulder, CO-80309, USA (Dated: January 3, 2023) A circularly polarized laser pulse can induce persistent intra-molecular currents by either exciting or ionizing molecules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
3
+ page_content=' These two cases are identified as electron currents and hole currents, respec- tively, and up to now they have been studied only separately.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
4
+ page_content=' We report ab initio time-dependent density-functional theory (TDDFT) simulations of currents during resonance-enhanced two-photon ionization of benzene, which reveal for the first time that both electron and hole currents can be present simultaneously.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
5
+ page_content=' By adjusting the intensity of the laser pulse, the balance between the two types of current can be controlled, and the overall sign of the current can be switched.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
6
+ page_content=' We provide a physical explanation for the effect in terms of complex molecular orbitals which is consistent with the TDDFT simulations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
7
+ page_content=' It has long been understood that, in response to an ap- plied magnetic field, the delocalized electrons of an aro- matic molecule circulate in so-called aromatic ring cur- rent [1, 2].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
8
+ page_content=' This effect is important in nuclear magnetic resonance spectroscopy, where the internal magnetic field generated by the ring current is responsible for diamag- netic shielding [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
9
+ page_content=' In 2006, it was proposed that ring currents in molecules could also be induced by ultra- short laser pulses with circular or elliptical polarization [4, 5].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
10
+ page_content=' The basic mechanism is that angular momentum carried by light is transfered to electrons in a molecule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
11
+ page_content=' Due to conservation of angular momentum, the current persists after the pulse has ended—even without an ex- ternal magnetic field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
12
+ page_content=' Various experiments on atomic targets have confirmed the existence of the effect [6, 7], although no direct observational data is available in the case of molecules.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
13
+ page_content=' Recent interest in photoinduced ring currents is motivated by the rapid technological advances in polarization control of high-harmonic radiation made in the last few years [8–10], which may enable experimen- tal study of these phenomena in the near future [11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
14
+ page_content=' There are several major advantages of photoinduced ring currents compared to those induced by static mag- netic fields.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
15
+ page_content=' First, the current is expected to be orders of magnitude stronger, and so is the induced magnetic field [12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
16
+ page_content=' Second, they enable femtosecond (or even attosec- ond) time-resolved studies of aromaticity and magnetism [13, 14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
17
+ page_content=' Lastly, they establish the possibility for coherent control of ring currents [15], which may have applications for controlling chemical reactions or the operation of ad- vanced opto-electronic devices.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
18
+ page_content=' In this Letter we predict a novel effect which causes the dominant charge carrier of the ring current to transi- tion from electrons to holes as the peak laser intensity in- creases past around 1012 W/cm2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
19
+ page_content=' We illustrate the effect with a series of ab initio time-dependent density func- tional theory (TDDFT) simulations of benzene (C6H6), which is the prototypical aromatic molecule.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
20
+ page_content=' Lastly, we demonstrate that the effect is not accounted for in the commonly used few level model of ring currents, due to the fact that it neglects ionization.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
21
+ page_content=' This calls into ques- tion the results of several previous studies (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
22
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
23
+ page_content=' [4, 5, 15]) where it was assumed that the few level model is accurate for laser intensities on the order of 1012 W/cm2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
24
+ page_content=' We begin by introducing the distinction between elec- tron and hole current: when an electron is promoted to an orbital with nonzero angular momentum, this creates an electron current;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
25
+ page_content=' when an electron is removed (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
26
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
27
+ page_content=', ionized) from an orbital with nonzero angular momen- tum, this creates a hole current.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
28
+ page_content=' So far, hole currents have mostly been studied in the context of strong field ionization of atoms by circularly polarized laser pulses, and it was recently confirmed experimentally that a hole can be created with a specific angular momentum relative to the laser polarization [16–19].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
29
+ page_content=' Electron currents on the other hand do not involve ionization, only excitation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
30
+ page_content=' However, in the interaction of atoms and molecules with strong laser fields, excitation and ionization are of- ten closely related and occur together.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
31
+ page_content=' A typical example is resonance-enhanced multiphoton ionization (REMPI) [20, 21], a two-step ionization process wherein an atom or molecule is first excited to an intermediate state (that must be resonant with some multiple of the laser fre- quency) and then subsequently ionized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
32
+ page_content=' Now consider REMPI in a system where the intermediate excited state corresponds to an electron current, and the final ionized state corresponds to a hole current (we will show that benzene is such a system).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
33
+ page_content=' The balance between excita- tion and ionization (and therefore electron and hole cur- rent) will depend on the laser intensity because the pro- cesses involve different numbers of photons (and therefore scale with different powers of intensity).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
34
+ page_content=' In particular at low intensities we expect electron current to dominate (excitation), and at high intensities we expect hole cur- rent to dominate (ionization).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
35
+ page_content=' Our main theoretical method is TDDFT, as imple- mented by Octopus [22–24], which provides a fully non- perturbative description of the light-matter interaction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
36
+ page_content=' As a reference point to compare against the full TDDFT simulations, we also consider the few level model of ring currents (e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
37
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
38
+ page_content=' [5]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
39
+ page_content=' We discuss the implementations of both models in [25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
40
+ page_content=' Because the few level model does not include ionization, we expect the two models to di- verge at high enough laser intensities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
41
+ page_content=' The laser pulse in our simulations is described in the arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
42
+ page_content='00380v1 [physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
43
+ page_content='chem-ph] 1 Jan 2023 2 FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
44
+ page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
45
+ page_content=' (a) Visualization of the current density based on the component passing through a plane bisecting the molecule as shown (averagea over all possible orientations of that plane [see Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
46
+ page_content=' (2)]) (b) Cross sections of the current density taken at the end of the laser pulse (t = 200 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
47
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
48
+ page_content=') for several different simulations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
49
+ page_content=' At low laser intensity the co-rotating current (red) dominates, while at high intensity the counter-rotating current (blue) dominates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
50
+ page_content=' Note: Each plot is scaled individually relative to the maximum absolute value within that plot.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
51
+ page_content=' The nuclei lie in the plane z = 0 with the carbon ring at x = ±2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
52
+ page_content='63 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
53
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
54
+ page_content=' and the hydrogen ring at x = ±4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
55
+ page_content='69 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
56
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
57
+ page_content='. dipole approximation by the following electric field, E(t) = � E sin2 (πt/T) Re � ˆϵeiω(t−T/2)� , 0 < t < T, 0, otherwise, (1) with central frequency ω = 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
58
+ page_content='76 eV (183 nm), dura- tion T = 16π/ω = 202 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
59
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
60
+ page_content=' = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
61
+ page_content='9 fs, circular polar- ization ˆϵ = (ˆx + iˆy)/ √ 2 (with the molecule in the xy- plane), and a variable peak amplitude E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
62
+ page_content=' The central frequency was chosen to be resonant with the doubly degenerate E1u state (as computed with linear response TDDFT [25]), which is predominantly associated with the HOMO-LUMO transition (HOMO = Highest Occu- pied Molecular Orbital;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
63
+ page_content=' LUMO = Lowest Unoccupied Molecular Orbital).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
64
+ page_content=' Because the computed ionization threshold is 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
65
+ page_content='0 eV < 2ω, this laser pulse is designed to drive 1+1 REMPI where one photon is enough to promote electron to the excited state and one additional photon to ionize.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
66
+ page_content=' After interacting with the laser pulse (t > T), the ben- zene molecule is in a superposition of the A1g ground state and the E1u excited state and also, to an extent, ionized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
67
+ page_content=' This causes oscillations in the charge and current densities ρ(r, t) and J(r, t), respectively, with period 612 as (corresponding to the energy difference between the ground state and excited states), which are an example of attosecond charge migration [26].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
68
+ page_content=' In order to visualize the current we isolate the sta- tionary component of the current density, by computing an angle averaged cross section defined by the following integral (in cylindrical coordinates ρ, z, φ), J(x, z, t) = 1 2π � 2π 0 ˆφ · J(|x|, z, φ)dφ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
69
+ page_content=' (2) The geometric interpretation of this integral is given in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
70
+ page_content=' 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
71
+ page_content=' The angle averaging procedure for the few level model causes that the fast-oscillating component effectively vanishes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
72
+ page_content=' Within the few-level model, the fast- oscillating component of the current density is zeroed out by this averaging procedure because of its parity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
73
+ page_content=' It has similar effect on TDDFT results, and therefore J(x, z, y) has only a very gradual time dependence for t > T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
74
+ page_content=' The same is true for TDDFT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
75
+ page_content=' These integrated current densi- ties are plotted in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
76
+ page_content=' 1b.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
77
+ page_content=' At low intensities the current is a combination of a strong co-rotating current (red) and a weak counter-rotating current (blue), while at high in- tensities the counter-rotating current dominates.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
78
+ page_content=' As we explain below (see Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
79
+ page_content=' 4), the reversal is a signature of the transition from electron to hole current regime.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
80
+ page_content=' The oscillatory component of the charge motion is best visualized by plotting the charge displacement, ∆ρ(r, t) = ρ(r, t) − ρ(r, 0), (3) shown in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
81
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
82
+ page_content=' The cloud of displaced charge circulates around the molecule with the expected period of 612 as, and this continues even after the pulse ends.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
83
+ page_content=' Overall, both the magnitude and shape of the charge displace- ment are remarkably similar between the two models, however there are some subtle differences.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
84
+ page_content=' First, long after the laser pulse the two models gradually become desynchronized.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
85
+ page_content=' Second, in TDDFT there appears to be a rearrangement of charge in the plane of the molecule, whereas the few level model only predicts the dynamics above and below the plane.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
86
+ page_content=' (a) (b) 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
87
+ page_content='8 × 1011 W/cm2 Few-level Model 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
88
+ page_content='0 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
89
+ page_content="0 ('n 0." metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
90
+ page_content='0 N 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
91
+ page_content='0 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
92
+ page_content='0 5 × 1012 W/cm² 1013 W/cm² 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
93
+ page_content='0 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
94
+ page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
95
+ page_content='0 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
96
+ page_content='0 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
97
+ page_content='0 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
98
+ page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
99
+ page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
100
+ page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
101
+ page_content='0 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
102
+ page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
103
+ page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
104
+ page_content='0 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
105
+ page_content='0 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
106
+ page_content='0 x (a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
107
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
108
+ page_content=') x (a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
109
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
110
+ page_content=' )3 FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
111
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
112
+ page_content=' Snapshots of the charge displacement induced by a circularly-polarized laser pulse with peak intensity 5×1012 W/cm2 taken around the peak of the laser pulse (first three columns t ≈ 100 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
113
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
114
+ page_content=') and after the laser pulse (last three columns t ≈ 400 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
115
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
116
+ page_content=').' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
117
+ page_content=' Light areas indicate excess electrons while dark areas indicate fewer electrons, as compared to the ground state charge density before the laser pulse.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
118
+ page_content=' We compare the results between the two theoretical models, TDDFT (top row) and the few level model (bottom row).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
119
+ page_content=' FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
120
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
121
+ page_content=' Comparison of full TDDFT simulations (solid blue line) to the few level model (orange dashed line).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
122
+ page_content=' For peak intensities, when ionization (dotted green line) becomes non- negligible, the two models begin to disagree.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
123
+ page_content=' The smooth lines have been interpolated between the calculated intensities using the method described in [25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
124
+ page_content=' Another important observation about the density dif- ference is that the dark areas are generally larger than the light areas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
125
+ page_content=' In the TDDFT results one reason for this is ionization, with the ionization probability given by P ionize = − � ∆ρ(r, 2T)d3r, (4) where the integral ranges over the simulation box.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
126
+ page_content=' Unex- pectedly, the few level model also appears to have dark areas larger than light areas even though it does not include ionization, and in fact the charge displacement must integrate to zero in that model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
127
+ page_content=' The reason for this is that the E1u is of mixed character, part of which in- volves excitation to LUMO + 3 [25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
128
+ page_content=' Note: The excess of darker areas in the TDDFT model is a combination of both ionization and excitation to LUMO +3 orbital.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
129
+ page_content=' The intensity dependence of the dynamics is illustrated in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
130
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
131
+ page_content=' using the current.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
132
+ page_content=' Note: this current is directly proportional to z-component of the magnetic moment as well as z-component of electronic angular momentum), Since the domain of integration is the simulation box, ionized electrons are not included.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
133
+ page_content=' For this reason we plot Lz(2T) so that the ionizing wavepacket has enough time to leave the box.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
134
+ page_content=' Whereas in the few level model the magnetic moment increases monotonically with the laser intensity (up to about 1013 W/cm2, after which the system Rabi oscillates back to the ground state), in TDDFT the current starts to decrease already around 1012 W/cm2, and reverses sign for even higher intensities.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
135
+ page_content=' We also plot the ionization probability (defined in Eq.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
136
+ page_content=' 4), and conclude that the reversal occurs precisely when the ionization probability becomes non-negligible.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
137
+ page_content=' The implications of the transition from electron to hole current on the charge dynamics, and the underlying phys- ical mechanism responsible for that transition, can be understood in more detail describe in more detail using complex molecular orbitals, as illustrated schematically in Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
138
+ page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
139
+ page_content=' These orbitals represent a change of basis from the usual real-valued Kohn-Sham orbitals ψn(r) (defined in [25]), ψHOMO ± (r) = [ψ14(r) ± iψ15(r)] / √ 2, (5) ψLUMO ± (r) = [ψ16(r) ± iψ17(r)] / √ 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
140
+ page_content=' (6) The advantage of using complex orbitals is that they are eigenfunctions of the 6-fold symmetry operator (rotation t = 105 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
141
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
142
+ page_content=' t = 110 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
143
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
144
+ page_content=' t = 390 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
145
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
146
+ page_content=' t = 395 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
147
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
148
+ page_content=' t = 100 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
149
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
150
+ page_content=' t = 400 a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
151
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
152
+ page_content=' TDDFT Few Level Model0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
153
+ page_content='8 Lz (TDDFT) Lz (Few level) 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
154
+ page_content='6 lonizationprobability(TDDFT) or probability 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
155
+ page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
156
+ page_content='2 (a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
157
+ page_content='u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
158
+ page_content=') 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
159
+ page_content='0 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
160
+ page_content='2 1011 1012 1013 Peak intensity (W/cm2)4 FIG.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
161
+ page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
162
+ page_content=' Schematic illustrating the complex molecular or- bitals and the physical mechanism for the transition from electron to hole current.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
163
+ page_content=' Color indicates the complex phase.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
164
+ page_content=' about the molecular axis by 60◦), exp � − iπ 3¯h ˆLz ¯h � ψHOMO ± (r) = exp � ∓iπ 3 � ψHOMO ± (r),(7) exp � −iπ 3 ˆLz ¯h � ψLUMO ± (r) = exp � ∓2iπ 3 � ψLUMO ± (r).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
165
+ page_content=' (8) The complex orbitals have magnetic quantum numbers m defined modulo 6: ψHOMO ± have m = ±1 and ψLUMO ± have m = ±2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
166
+ page_content=' Just as for atomic orbitals, the sign of m indicates the direction the electron circulates around the molecule, and the magnitude indicates more-or-less the angular speed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
167
+ page_content=' We have chosen our conventions such that m > 0 electrons are co-rotating with the laser field, and m < 0 electrons are counter-rotating.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
168
+ page_content=' Using the notation of complex orbitals, Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
169
+ page_content=' 4 illus- trates how in the ground state, both ψHOMO ± are dou- bly occupied, and consequently there is zero net cur- rent.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
170
+ page_content=' When the benzene molecule is exposed to a cir- cularly polarized laser pulse, the usual selection rule ∆m = 1 applies (here we assume the laser is polar- ized in the molecular plane, see [25] for the more gen- eral case), so that the only dipole-allowed transition is ψHOMO + to ψLUMO + , which is the dominant component of the E1u excited state.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
171
+ page_content=' The electron excited to LUMO contributes a strong co-rotating current (m = +2), but the imbalance of electrons in the HOMO contributes a weaker counter-rotating current (m = −1).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
172
+ page_content=' This can al- ternatively be interpreted as a positively charged hole occupying ψHOMO + producing a co-rotating hole current (rather than a counter-rotating electron current).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
173
+ page_content=' This is precisely what we see in the top row of Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
174
+ page_content=' 1b, two components to the current with opposite sign (red and blue).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
175
+ page_content=' In order to explain the reversal of the current at higher intensity (bottom row of Fig.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
176
+ page_content=' 1b), we simply recognize that the electron previously excited to ψLUMO + can ab- sorb a second photon from the same laser pulse, ionizing, and leaving behind only the hole current.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
177
+ page_content=' The balance between the one-photon excitation and the two-photon ionization processes can be controlled by varying the laser intensity, because the first process scales with I while the second process scales with I2 (with I ∝ E2 the laser in- tensity).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
178
+ page_content=' Furthermore, it is now apparent that the sign reversal can be interpreted as a change in the dominant charge carrier from electrons to holes.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
179
+ page_content=' In conclusion, we have shown that both electron and hole currents are present during resonance-enhanced two- photon ionization of benzene, and the balance between the two current regimes can be controlled by varying the peak laser intensity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
180
+ page_content=' We have proposed a simple expla- nation for the effect in terms of molecular orbitals, which is consistent with the results of full TDDFT simulations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
181
+ page_content=' Variants of complex orbital model should apply to a wide variety of molecules other than benzene, meaning that the structure of the complex molecular orbitals can be used to predict the interplay between electron and hole currents during REMPI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
182
+ page_content=' In order to measure this effect in experiment, several pump-probe schemes have been proposed that are sensitive to the magnitude and direc- tion of the ring current [7, 11].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
183
+ page_content=' In [25], we demonstrate that the reversal is independent of the orientation of the molecule, which greatly simplifies any potential exper- iment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
184
+ page_content=' Finally, our results suggest that the few level model typically used to study photoinduced ring currents may be insufficient even for moderate laser intensities around 1012 W/cm2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
185
+ page_content=' A more ab initio nonperturbative theory such as TDDFT, as used in present paper, is more appropriate for this regime.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
186
+ page_content=' This work was supported by the NSF Grant No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
187
+ page_content=' PHY- 1734006 and Grant No.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
188
+ page_content=' PHY-2110628.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
189
+ page_content=' This work uti- lized resources from the University of Colorado Boulder Research Computing Group, which is supported by the National Science Foundation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
190
+ page_content=' [1] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
191
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
192
+ page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
193
+ page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
194
+ page_content=' Gomes and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
195
+ page_content=' B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
196
+ page_content=' Mallion, Chemical Reviews 101, 1349 (2001).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
197
+ page_content=' [2] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
198
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
199
+ page_content=' Krygowski, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
200
+ page_content=' Szatylowicz, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
201
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
202
+ page_content=' Stasyuk, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
203
+ page_content=' Do- minikowska, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
204
+ page_content=' Palusiak, Chemical Reviews 114, 6383 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
205
+ page_content=' [3] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
206
+ page_content=' Heine, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
207
+ page_content=' Corminboeuf, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
208
+ page_content=' Seifert, Chemical Re- views 105, 3889 (2005).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
209
+ page_content=' [4] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
210
+ page_content=' Barth and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
211
+ page_content=' Manz, Angewandte Chemie International 2元 m=-2 m = +2 LUMO Phase 7元 0 m=-1 m=+1 HOMO5 Edition 45, 2962 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
212
+ page_content=' [5] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
213
+ page_content=' Barth, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
214
+ page_content=' Manz, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
215
+ page_content=' Shigeta, and K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
216
+ page_content=' Yagi, Journal of the American Chemical Society 128, 7043 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
217
+ page_content=' [6] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
218
+ page_content=' Wollenhaupt, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
219
+ page_content=' Krug, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
220
+ page_content=' K¨ohler, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
221
+ page_content=' Bayer, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
222
+ page_content=' Sarpe- Tudoran, and T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
223
+ page_content=' Baumert, Applied Physics B 95, 245 (2009).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
224
+ page_content=' [7] S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
225
+ page_content=' Eckart, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
226
+ page_content=' Kunitski, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
227
+ page_content=' Richter, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
228
+ page_content=' Hartung, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
229
+ page_content=' Rist, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
230
+ page_content=' Trinter, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
231
+ page_content=' Fehre, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
232
+ page_content=' Schlott, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
233
+ page_content=' Henrichs, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
234
+ page_content=' P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
235
+ page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
236
+ page_content=' Schmidt, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
237
+ page_content=' Jahnke, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
238
+ page_content=' Sch¨offler, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
239
+ page_content=' Liu, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
240
+ page_content=' Barth, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
241
+ page_content=' Kaushal, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
242
+ page_content=' Morales, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
243
+ page_content=' Ivanov, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
244
+ page_content=' Smirnova, and R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
245
+ page_content=' D¨orner, Nature Physics 14, 701 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
246
+ page_content=' [8] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
247
+ page_content=' Fleischer, O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
248
+ page_content=' Kfir, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
249
+ page_content=' Diskin, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
250
+ page_content=' Sidorenko, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
251
+ page_content=' Co- hen, Nature Photonics 8, 543 (2014).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
252
+ page_content=' [9] D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
253
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
254
+ page_content=' Hickstein, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
255
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
256
+ page_content=' Dollar, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
257
+ page_content=' Grychtol, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
258
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
259
+ page_content=' Ellis, R.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
260
+ page_content=' Knut, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
261
+ page_content=' Hern´andez-Garc´ıa, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
262
+ page_content=' Zusin, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
263
+ page_content=' Gentry, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
264
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
265
+ page_content=' Shaw, T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
266
+ page_content=' Fan, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
267
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
268
+ page_content=' Dorney, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
269
+ page_content=' Becker, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
270
+ page_content=' Jaro´n- Becker, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
271
+ page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
272
+ page_content=' Kapteyn, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
273
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
274
+ page_content=' Murnane, and C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
275
+ page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
276
+ page_content=' Durfee, Nature Photonics 9, 743 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
277
+ page_content=' [10] P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
278
+ page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
279
+ page_content=' Huang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
280
+ page_content=' Hern´andez-Garc´ıa, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
281
+ page_content='-T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
282
+ page_content=' Huang, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
283
+ page_content='-Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
284
+ page_content=' Huang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
285
+ page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
286
+ page_content=' Lu, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
287
+ page_content=' Rego, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
288
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
289
+ page_content=' Hickstein, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
290
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
291
+ page_content=' Ellis, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
292
+ page_content=' Jaron-Becker, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
293
+ page_content=' Becker, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
294
+ page_content='-D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
295
+ page_content=' Yang, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
296
+ page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
297
+ page_content=' Durfee, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
298
+ page_content=' Plaja, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
299
+ page_content=' C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
300
+ page_content=' Kapteyn, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
301
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
302
+ page_content=' Murnane, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
303
+ page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
304
+ page_content=' Kung, and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
305
+ page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
306
+ page_content=' Chen, Nature Photonics 12, 349 (2018).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
307
+ page_content=' [11] O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
308
+ page_content=' Neufeld and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
309
+ page_content=' Cohen, Physical Review Letters 123, 103202 (2019).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
310
+ page_content=' [12] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
311
+ page_content='-J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
312
+ page_content=' Yuan and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
313
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
314
+ page_content=' Bandrauk, Physical Review A 88, 013417 (2013).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
315
+ page_content=' [13] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
316
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
317
+ page_content=' Ulusoy and M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
318
+ page_content=' Nest, Journal of the American Chem- ical Society 133, 20230 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
319
+ page_content=' [14] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
320
+ page_content='-J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
321
+ page_content=' Yuan, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
322
+ page_content='-C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
323
+ page_content=' Shu, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
324
+ page_content=' Dong, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
325
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
326
+ page_content=' Ban- drauk, The Journal of Physical Chemistry Letters 8, 2229 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
327
+ page_content=' [15] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
328
+ page_content=' Mineo and Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
329
+ page_content=' Fujimura, The Journal of Chemical Physics 147, 224301 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
330
+ page_content=' [16] I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
331
+ page_content=' Barth and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
332
+ page_content=' Smirnova, Physical Review A 84, 063415 (2011).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
333
+ page_content=' [17] T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
334
+ page_content=' Herath, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
335
+ page_content=' Yan, S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
336
+ page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
337
+ page_content=' Lee, and W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
338
+ page_content=' Li, Physical Review Letters 109, 043004 (2012).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
339
+ page_content=' [18] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
340
+ page_content=' Zhu, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
341
+ page_content=' Lan, K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
342
+ page_content=' Liu, Y.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
343
+ page_content=' Li, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
344
+ page_content=' Liu, Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
345
+ page_content=' Zhang, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
346
+ page_content=' Barth, and P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
347
+ page_content=' Lu, Optics Express 24, 4196 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
348
+ page_content=' [19] K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
349
+ page_content=' Liu and I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
350
+ page_content=' Barth, Physical Review A 94, 043402 (2016).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
351
+ page_content=' [20] V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
352
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
353
+ page_content=' Antonov, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
354
+ page_content=' N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
355
+ page_content=' Knyazev, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
356
+ page_content=' S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
357
+ page_content=' Letokhov, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
358
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
359
+ page_content=' Matiuk, V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
360
+ page_content=' G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
361
+ page_content=' Movshev, and V.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
362
+ page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
363
+ page_content=' Potapov, Optics Letters 3, 37 (1978).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
364
+ page_content=' [21] W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
365
+ page_content=' Dietz, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
366
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
367
+ page_content=' Neusser, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
368
+ page_content=' Boesl, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
369
+ page_content=' W.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
370
+ page_content=' Schlag, and S.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
371
+ page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
372
+ page_content=' Lin, Chemical Physics 66, 105 (1982).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
373
+ page_content=' [22] M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
374
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
375
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
376
+ page_content=' Marques, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
377
+ page_content=' Castro, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
378
+ page_content=' F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
379
+ page_content=' Bertsch, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
380
+ page_content=' Ru- bio, Computer Physics Communications 151, 60 (2003).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
381
+ page_content=' [23] A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
382
+ page_content=' Castro, H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
383
+ page_content=' Appel, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
384
+ page_content=' Oliveira, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
385
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
386
+ page_content=' Rozzi, X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
387
+ page_content=' An- drade, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
388
+ page_content=' Lorenzen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
389
+ page_content=' a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
390
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
391
+ page_content=' Marques, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
392
+ page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
393
+ page_content=' U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
394
+ page_content=' Gross, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
395
+ page_content=' Rubio, physica status solidi (b) 243, 2465 (2006).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
396
+ page_content=' [24] X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
397
+ page_content=' Andrade, D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
398
+ page_content=' Strubbe, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
399
+ page_content=' D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
400
+ page_content=' Giovannini, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
401
+ page_content=' H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
402
+ page_content=' Larsen, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
403
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
404
+ page_content=' T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
405
+ page_content=' Oliveira, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
406
+ page_content=' Alberdi-Rodriguez, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
407
+ page_content=' Varas, I.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
408
+ page_content=' Theophilou, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
409
+ page_content=' Helbig, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
410
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
411
+ page_content=' Verstraete, L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
412
+ page_content=' Stella, F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
413
+ page_content=' Nogueira, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
414
+ page_content=' Aspuru-Guzik, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
415
+ page_content=' Castro, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
416
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
417
+ page_content=' L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
418
+ page_content=' Marques, and A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
419
+ page_content=' Rubio, Physical Chemistry Chemical Physics 17, 31371 (2015).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
420
+ page_content=' [25] See Supplemental Material for details on the TDDFT and few-level models, orientation dependence, and the method for interpolating over intensity.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
421
+ page_content=' [26] H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
422
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
423
+ page_content=' W¨orner, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
424
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
425
+ page_content=' Arrell, N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
426
+ page_content=' Banerji, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
427
+ page_content=' Cannizzo, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
428
+ page_content=' Chergui, A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
429
+ page_content=' K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
430
+ page_content=' Das, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
431
+ page_content=' Hamm, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
432
+ page_content=' Keller, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
433
+ page_content=' M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
434
+ page_content=' Kraus, E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
435
+ page_content=' Liberatore, P.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
436
+ page_content=' Lopez-Tarifa, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
437
+ page_content=' Lucchini, M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
438
+ page_content=' Meuwly, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
439
+ page_content=' Milne, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
440
+ page_content='-E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
441
+ page_content=' Moser, U.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
442
+ page_content=' Rothlisberger, G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
443
+ page_content=' Smolentsev, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
444
+ page_content=' Teuscher, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
445
+ page_content=' A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
446
+ page_content=' van Bokhoven, and O.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
447
+ page_content=' Wenger, Struc- tural Dynamics 4, 061508 (2017).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/VdAyT4oBgHgl3EQfhfjt/content/2301.00380v1.pdf'}
WNAyT4oBgHgl3EQfu_m7/vector_store/index.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1bb1ce92669537ffd3c2fb588842ec0edd29960fa52396311baf26e1727f63b4
3
+ size 299618
WtAyT4oBgHgl3EQfWPdu/content/tmp_files/2301.00159v1.pdf.txt ADDED
@@ -0,0 +1,2049 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT
2
+ EPIDEMIC MODELS
3
+ I: THE REPLACEMENT NUMBER DYNAMICS
4
+ FLORIAN NILL
5
+ 31-DEC-2022
6
+ Abstract. As shown recently by the author, constant population SI(R)S models map to
7
+ Hethcote’s classic endemic model originally proposed in 1973. This unifies a whole class
8
+ of models with up to 10 parameters being all isomorphic to a simple 2-parameter master
9
+ model for endemic bifurcation. In this work this procedure is extended to a 14-parameter
10
+ SSISS Model, including social behavior parameters, a (diminished) susceptibility of the
11
+ R-compartment and unbalanced constant per capita birth and death rates, thus covering
12
+ many prominent models in the literature. Under mild conditions, in the dynamics for
13
+ fractional variables in this model all vital parameters become redundant at the cost of
14
+ possibly negative incidence rates. There is a symmetry group GS acting on parameter
15
+ space A, such that systems with GS-equivalent parameters are isomorphic and map to the
16
+ same normalized system. Using (Xrep, I) as canonical coordinates, Xrep the replacement
17
+ number, normalization reduces to parameter space A/GS with 5 parameters only. This
18
+ approach reveals unexpected relations between various models in the literature. Part two
19
+ of this work will analyze equilibria, stability and backward bifurcation and part three
20
+ will further reduce the number of essential parameters from 5 to 3.
21
+ Contents
22
+ 1.
23
+ Introduction
24
+ 2
25
+ 2.
26
+ The SSISS model
27
+ 6
28
+ 2.1.
29
+ Constant population
30
+ 8
31
+ 2.2.
32
+ Time varying population
33
+ 9
34
+ 2.3.
35
+ Classifying parameter space
36
+ 10
37
+ 2.4.
38
+ Examples from the literature
39
+ 12
40
+ 2.5.
41
+ Absence of periodic solutions
42
+ 14
43
+ 3.
44
+ Normalization
45
+ 15
46
+ 3.1.
47
+ Phase space
48
+ 15
49
+ 3.2.
50
+ Canonical coordinates
51
+ 16
52
+ 3.3.
53
+ Main results
54
+ 17
55
+ 3.4.
56
+ Examples revisited
57
+ 22
58
+ 4.
59
+ Summary and outlook
60
+ 23
61
+ Appendix A.
62
+ Normalizing linear vital dynamics
63
+ 24
64
+ Appendix B.
65
+ Scaling the SI(R)S model
66
+ 24
67
+ Appendix C.
68
+ The case α1 = α2 = 0
69
+ 27
70
+ References
71
+ 27
72
+ E-mail address: [email protected].
73
+ 2020 Mathematics Subject Classification. 34C23, 34C26, 37C25, 92D30.
74
+ Key words and phrases. SIRS model, SSISS model, normalization, symmetry, stability, endemic bifur-
75
+ cation, backward bifurcation.
76
+ The author is retired physicist, Dr.rer.nat.habil., formerly senior research fellow at Inst. theor. Physik,
77
+ Freie Universität Berlin.
78
+ 1
79
+ arXiv:2301.00159v1 [q-bio.PE] 31 Dec 2022
80
+
81
+ 2
82
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
83
+ 1. Introduction
84
+ Building mathematical models to describe phenomena in natural sciences one typically
85
+ encounters dynamical variables and external parameters. Within the model values for
86
+ external parameters are considered to be given from outside, like fundamental natural
87
+ constants (speed of light c, Planck’s constant ℏ), parameters describing material or bi-
88
+ ological properties (spring constant κ, birth rate δ, recovery rate γ) or social behavior
89
+ (contact rate β). Naturally, reducing the number of essential parameters is always a goal
90
+ to detect redundancies within parameter space and to simplify computations by unload-
91
+ ing formulas. In the simplest case a pure dimensional scale parameter may without loss
92
+ be put equal to one by choosing dimensional units appropriately. For example, putting
93
+ c = 1 amounts to measuring spatial distances by light running times and masses in units
94
+ of energies, putting ℏ = 1 amounts to measuring energies by angular frequencies and
95
+ putting γ = 1 amounts to measuring time in units of the recovery time in an epidemic
96
+ model.
97
+ More generally a normalization program consists of finding appropriate coordinate
98
+ transformations in variable+parameter space such that the transformed system only de-
99
+ pends on a maximally reduced subset of transformed parameters. Examples are1
100
+ Harmonic oscillator
101
+ Predator-prey model
102
+ ˙u
103
+ =
104
+ v
105
+ ˙u
106
+ =
107
+ −uv + c1u
108
+ ˙v
109
+ =
110
+ −u
111
+ ˙v
112
+ =
113
+ uv − v
114
+ Classic SIR model
115
+ Classic endemic model
116
+ ˙u
117
+ =
118
+ −uv
119
+ ˙u
120
+ =
121
+ −uv − c1u + c2
122
+ ˙v
123
+ =
124
+ uv − v
125
+ ˙v
126
+ =
127
+ uv − v
128
+ (1.1)
129
+ Following this strategy the 6-parameter SI(R)S model (≡ combined SIRS/SIS model)
130
+ with standard incidence, constant vaccination and immunity waning rates and a balanced
131
+ birth and death rate has recently been shown by the author (Nill 2022) to admit a nor-
132
+ malized version looking like the classic endemic model above2.
133
+ In this work (including two follow ups to be denoted as parts II and III (Nill n.d.[b],[c]))
134
+ this method is extended to the case where immunity after recovery (or vaccination) is
135
+ incomplete right from the onset and where also compartment dependent constant per
136
+ capita birth and death rates lead to a time varying population size N.
137
+ In this way
138
+ one is naturally lead to replacing the SI(R)S model by a SSISS model, where in place
139
+ of the usual S, I and R compartments we have two susceptible compartments S1 and
140
+ S2 and one infectious compartment I. Infection transmission from I to S2 is diminished
141
+ as compared to transmission to S1. There is a vaccination flow from S1 to S2 and an
142
+ immunity waning flow from S2 to S1. The model could also be interpreted by considering
143
+ 1The variables in these examples are:
144
+ - Harmonic oscillator: u = q, v = p/
145
+
146
+ mk, where q, p, κ, m are coordinate, momentum, spring constant
147
+ and particle mass and where the oscillation period is normalized to T = 2π by putting m/k = 1.
148
+ - Predator-prey model: (u, v) denote appropriately rescaled prey and predator populations, respectively,
149
+ and the predator mortality rate is normalized to one.
150
+ - SIR model: u = r0S, v = r0I, where r0 is the basic reproduction number, (S, I) are susceptible and
151
+ infectious fractions of the population and where the recovery rate is normalized to γ = 1.
152
+ - Endemic model:
153
+ (u, v, r0, γ) as above, c1 = δ/(γ + δ) and c2 = r0c1, where δ is the balanced
154
+ birth/mortality rate and where now time scale is normalized to γ + δ = 1.
155
+ 2Aapart from allowing also values u ∈ R and an enlarged parameter range (c1, c2) ∈ R+ × R ∪ {0, 0}.
156
+
157
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
158
+ 3
159
+ S2 as the “lock-down” fraction and S1 as the “freedom fraction”. In this picture flows from
160
+ S1 to S2 and vice-versa are described by an I-linear (respectively (N −I)-linear) flow with
161
+ rate parameters θi, i = 1, 2, modeling social behavior in reaction to published prevalence
162
+ data.
163
+ Combining both interpretations it turns out to be convenient to start with an
164
+ abstract version of a SSISS model staying completely symmetric under interchanging S1
165
+ and S2, see Fig. 1.
166
+ The present part I provides a normalization prescription reducing the number of inde-
167
+ pendent parameters in this model from initially fourteen to essentially five (four in the
168
+ SI(R)S model sub-case). Based on this approach, part II will give a complete review on
169
+ equilibria and stability in the master SSISS model, thereby also recovering an exceptional
170
+ scenario which had been overlooked in the literature so far. In part III the scaling sym-
171
+ metry for SI(R)S models mentioned above will be generalized to the full SSISS model,
172
+ thereby reducing the number of parameters again by two. So, the total reduction from
173
+ fourteen to three reveals a great hidden redundancy in parameter space. It also provides
174
+ a unifying view on results in the literature concerning equilibrium states, endemic bifur-
175
+ cation and stability properties for all kinds of sub-classes of this model. Put differently,
176
+ in the presence of a common normalized version presenting basically repeated arguments
177
+ for various subsets of non-vanishing parameters becomes obsolete.
178
+ Relating this work to the literature, let me focus on deterministic SIR-type 3-compartment
179
+ dynamical systems, which conveniently may be classified according to
180
+ A) constant vs. time-varying total population size N,
181
+ B) infection transmission only from I to S vs. also from I to R (in which case it makes
182
+ sense to rename S ≡ S1 and R ≡ S2).
183
+ Also, I will restrict this survey to models with standard bi-linear incidence flows βiSiI/N,
184
+ such that the vector field ˙Y = V(Y), Y = (S1, S2, I), is homogeneous of first order. This
185
+ applies to diseases where the number of effective contacts per capita is independent of N.
186
+ ad A) Endemic models with constant population have first been constructed by adding
187
+ a non-zero balanced birth and death rate to the classic SIR model of (Kermack and
188
+ McKendrick 1927). As shown by (Hethcote 1974) (see also (Hethcote 1976, 1989)), in
189
+ this way already the simplest model without vaccination and loss of immunity shows
190
+ a bifurcation from a stable disease-free equilibrium point (DFE) to a stable endemic
191
+ scenario when raising the basic reproduction number R0 above one. Nowadays this is
192
+ considered as Hethcote’s classic endemic model. Including linear vaccination and/or loss
193
+ of immunity terms and optionally also considering recovery without immunity one ends up
194
+ with various types of constant population SI(R)S models without changing this picture,
195
+ see for example (Batistela et al. 2021; Chauhan, Misra, and Dhar 2014; Korobeinikov and
196
+ Wake 2002; O’Regan et al. 2010). As remarked above (and reviewed in more detail in
197
+ Appendix B), the true reason lies in the fact that constant population SI(R)S models with
198
+ up to 10 parameters all map to the same normalized 2-parameter version of the classic
199
+ endemic model as given in Eq. (1.1).
200
+ Models with variable population are mostly studied under the assumption of a constant
201
+ (i.e. N-independent) birth flow. Heuristically this may be justified by assuming that
202
+ N varies slowly on characteristic epidemic time scales. But truly speaking, as already
203
+ pointed out by (Mena-Lorca and Hethcote 1992), this Ansatz rather models a constant
204
+ immigration scenario. So in this work I will follow the more natural proposal of modeling
205
+ vital dynamics by possibly department dependent constant per capita birth and death
206
+
207
+ 4
208
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
209
+ rates. Note that, unless fine tuning parameters, this implies that either N(t) → ∞ or
210
+ N(t) → 0 as t → ∞. So in this type of models one always analyzes the dynamics of
211
+ fractional variables Si := Si/N, I := I/N, which is well known to be independent of N(t).
212
+ Apparently, this stream of models has been initiated by (Busenberg and Driessche 1990,
213
+ 1991; Derrick and Driessche 1993). (Razvan 2001) has studied a SIRS model in this sense
214
+ with infection transmission also from outside and a SIS-version with varying population
215
+ size has been analyzed by (J. Li and Ma 2002). For generalizations to SEIR models see
216
+ e.g. (Greenhalgh 1997; M. Y. Li et al. 1999; G. Lu and Z. Lu 2018; Sun and Hsieh 2010).
217
+ ad B) A different approach to modeling partial and/or waning immunity consists of
218
+ introducing a diminished incidence flow with rate βR ≡ β2 > 0 directly from R ≡ S2
219
+ to I. This has presumably first been proposed in the so-called SIRI model of (Derrick
220
+ and Driessche 1993), see above. In addition, the authors also introduced a time varying
221
+ population size N(t) and an excess mortality ∆µI in compartment I to this model. In turn,
222
+ they didn’t use linear vaccination nor immunity waning terms. In this way they identified
223
+ a range of parameters in the domain R0 < 1, for which besides the locally asymptotically
224
+ stable disease free equilibrium there also coexist two endemic equilibria, one being a
225
+ saddle and the other one also being locally asymptotically stable. Later (Hadeler and
226
+ Castillo-Chavez 1995) found the same phenomenon in their combined SIS/SIRS core group
227
+ model with linear vaccination, constant population and also two incidence rates βi for
228
+ S → I and R → I. Meanwhile it is well known that models with infection incidents
229
+ from several compartments may show a so-called backward bifurcation from the disease-
230
+ free to an endemic scenario (Hadeler and Driessche 1997). This means that two locally
231
+ asymptotically stable equilibrium states may coexist for some range below threshold,
232
+ causing also hysteresis effects upon varying parameters. Apparently, a varying population
233
+ size is not needed for this. In (Kribs-Zaleta and Velasco-Hernandez 2000) the authors have
234
+ improved and extended these results by adding also a linear immunity waning rate to the
235
+ model of (Hadeler and Driessche 1997).
236
+ One may also distinguish vaccinated and recovered people into separate compartments.
237
+ This leads to 4-compartment models, where similar results have been obtained by, e.g.
238
+ (J. Arino, Mccluskey, and Driessche 2003; Yang, Sun, and Julien Arino 2010).
239
+ Backward bifurcation has lately also been observed in SEIRS-type models for Covid-
240
+ 19 by considering two distinguished susceptible compartments.
241
+ In (Nadim and Chat-
242
+ topadhyay 2020) the less susceptible compartment had been interpreted as an incomplete
243
+ lockdown and in (Diagne et al. 2021) as an incomplete vaccination efficacy.
244
+ More recently, in (Avram, Adenane, Basnarkov, et al. 2021; Avram, Adenane, Bianchin,
245
+ et al. 2022) the authors have given a thorough stability analysis of an eight parameter
246
+ SIRS-type model by adding a varying population size to the model of (Kribs-Zaleta and
247
+ Velasco-Hernandez 2000) (apparently without being aware of that paper).
248
+ Closing this overview I should also remark that backward bifurcation is also observed
249
+ when considering I-dependent contact or recovery rates to model reactive behavior or
250
+ infection treatment. However the list of papers on this topic over the last 20 years becomes
251
+ too huge to be quoted at this place.
252
+ This paper extends the normalization algorithm for constant population SI(R)S models to
253
+ models as above, i.e. with time varying population size and/or a non-zero incidence rate
254
+ βR ≡ β2 from R ≡ S2 to I. As a starting observation, there is an ambiguity in deriving the
255
+ dynamics ˙y = F(y) for fractional variables y = (S1, S2, I), see Appendix A. This allows
256
+ choosing the vector field F such that all vital dynamics parameters become redundant,
257
+
258
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
259
+ 5
260
+ provided the birth-minus-death rates νi = δi −µi in S1 and S2 coincide, ν1 = ν2 = ν. This
261
+ redundancy already reduces the number of parameters in the master SSISS model from
262
+ fourteen to eight. More than that, F depends on the incidence rates βi only as a function
263
+ of ˜βi = βi + νI − ν, where νI = δI − µI is the birth-minus-death rate in I. Assuming for
264
+ simplicity compartment independent birth rates gives ˜βi = βi − ∆µI, where ∆µI denotes
265
+ the excess mortality in I. In this way models with variable population, ∆µI > 0, and
266
+ absence of a incidence rate from R, β2 = 0, look like models with constant population,
267
+ ∆µI = 0, and a negative incidence rate β2 = ˜β2 < 0. Conversely, models with positive
268
+ incidence rates βi > 0 and excess mortality ∆µI < min{β1, β2} behave like models with
269
+ constant population size and incidence rates βi = ˜βi > 0. So, the above classification
270
+ schemes A) and B) become blurred and, instead, it is more expedient to view all models
271
+ as if they had constant population size and two distinguished and possibly also negative
272
+ incidence rates ˜βi ∈ R.
273
+ In this way most of the above bench marking 3-compartment models (if necessary after
274
+ imposing the constraint ν1 = ν2) become comparable as sub-cases of the master SISS
275
+ model, with tilde parameters swallowing all birth and death rates and possibly with
276
+ negative incidence rates ˜βi ∈ R. As an example, the models of (Hadeler and Castillo-
277
+ Chavez 1995) and (Kribs-Zaleta and Velasco-Hernandez 2000) become isomorphic and
278
+ they completely cover the sub-case µ1 = µ2 and 0 < min{˜β1, ˜β2} in (Avram, Adenane,
279
+ Bianchin, et al. 2022). Also, apart from an irrelevant boundary case, the complementary
280
+ sub-case µ1 = µ2 and 0 > min{˜β1, ˜β2} in (Avram, Adenane, Bianchin, et al. 2022) is
281
+ covered by the model of (J. Li and Ma 2002). So, applying the normalization procedure
282
+ of this paper, all results in Section 5 and 6 of (Avram, Adenane, Bianchin, et al. 2022)
283
+ already follow from the previous literature. A more detailed list of unexpected relations
284
+ between the above models is given in Section 2.4.
285
+ The plan of this paper is as follows. In Sections 2.1 and 2.2 we pass to fractional com-
286
+ partment variables, Si = Si/N and I = I/N, and prove redundancy of all vital dynamics
287
+ parameters at the cost of possibly negative incidence rates ˜βi. For convenience, time scale
288
+ is also normalized by putting the total expected waiting time in compartment I equal to
289
+ one. In this way the number of essential parameters is already reduced from fourteen to
290
+ seven. Thus, denoting A the space of essential parameters, we have dim A = 7.
291
+ Section 2.3 classifies various useful subsets in parameter space like Aphys ⊂ A, guaran-
292
+ teeing forward invariance of the physical triangle
293
+ Tphys := {(S1, S2, I) ∈ R3
294
+ ≥0 | S1 + S2 + I = 1},
295
+ and Abio ⊂ Aphys, guaranteeing an epidemiological interpretation of parameters by re-
296
+ quiring in particular θ1 ≥ 0 ≥ θ2.
297
+ Section 2.4 identifies eight examples from the above list of models as sub-cases of the
298
+ master SSISS model. In this way we obtain various relations between these models as
299
+ indicated above, which apparently have not been recognized before.
300
+ In Section 2.5 we adapt methods from (Busenberg and Driessche 1990) to prove ab-
301
+ sence of periodic solutions for all parameters non-negative, except βi. The extension to
302
+ parameters a ∈ Abio (requiring θ2 ≤ 0) heavily relies on the symmetry results in Section
303
+ 3 and will be proven in Section 3.3.
304
+ Section 3 starts from the observation, that the time-normalized equation of motion for
305
+ I takes the generic form ˙I = (Xrep − 1)I, where Xrep = β1S1 + β2S2 is the replacement
306
+
307
+ 6
308
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
309
+ number (Hethcote 2000), i.e.
310
+ the expected number of secondary cases produced by a
311
+ typical infectious individual during its time of infectiousness (nowadays mostly called
312
+ effective reproduction number).
313
+ A coordinate free formulation of the model naturally
314
+ leads to taking (Xrep, I) as independent canonical coordinates3 in the physical triangle
315
+ Tphys. In this way, we arrive at formulating the SSISS model as a dynamical system in
316
+ (Xrep, I)-space, called the replacement number (RN) dynamics (Section 3.2).
317
+ ˙Xrep = f(Xrep, I),
318
+ ˙I = (Xrep − 1)I.
319
+ (1.2)
320
+ Since f(Xrep, I) turns out to be a 5-parameter quadratic polynomial with no term ∼ X2
321
+ rep,
322
+ the number of free parameters is now reduced from seven to five.
323
+ The main results of this paper are derived in Section 3.3. Denoting D the new parameter
324
+ set, dim D = 5, the above approach yields a surjective submersion A ∋ a �→ x(a) ∈ D.
325
+ Moreover, A becomes a principal fibre bundle with respect to a group right action ◁ :
326
+ A × GS → A such that x(a ◁ g) = x(a) and D ∼= A/GS. Here GS ⊂ GL+(R2) is the
327
+ group acting on (S1, S2) ∈ R2 and leaving S1 + S2 invariant. Eq. (1.2) implies that SSISS
328
+ dynamical systems at parameter values a, a′ ∈ A are isomorphic whenever a and a′ are
329
+ GS-equivalent, i.e. x(a) = x(a′) or equivalently a′ = a ◁ g for some g ∈ GS. In this way
330
+ we also get
331
+ -
332
+ Absence of periodic solutions also for parameters a ∈ Abio,
333
+ -
334
+ Conditions under which the social behavior parameters θi can be “gauged to zero”, i.e.
335
+ there exists g ∈ GS such that a ◁ g ∈ Aθ=0.
336
+ Section 3.4 revisits the examples from the literature within the new formalism and Sec-
337
+ tion 4 gives a summary and outlook to parts II and III of this work. Finally, Appen-
338
+ dix A provides a normalization prescription for the dynamics of fractional variables in
339
+ n-compartment models with linear (i.e. constant per capita) birth and death rates, Ap-
340
+ pendix B reviews the scaling symmetry in SI(R)S models introduced in (Nill 2022) and
341
+ Appendix C discusses a boundary case in parameter space.
342
+ Acknowledgement I would like to thank Florin Avram for encouraging interest and
343
+ useful discussions.
344
+ 2. The SSISS model
345
+ This Section starts with proposing an abstract completely symmetrized SSISS model
346
+ consisting of three compartments, S1, S2 and I, with total population N = S1 + S2 + I.
347
+ Members of I are infectious, members of S1 are highly susceptible (socially active or not
348
+ immune) and members of S2 are less susceptible (partly immune or reducing contacts).
349
+ The flow diagram between compartments is depicted in Fig. 1.
350
+ The parameters in this model may be given the following interpretations
351
+ 3Here “canonical” is not meant in the sense of Hamiltonian systems.
352
+
353
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
354
+ 7
355
+ Figure 1. Completely symmetric flow diagram of the SSISS model.
356
+ All pa-
357
+ rameters are nonnegative except θ2 ∈ [−α2, 0]. Also q1 + q2 = 1, γ1 + γ2 > 0
358
+ and β1 > β2. Generalizing to compartment dependent birth rates amounts to
359
+ replacing δN by δ1S1 + δ2S2 + δII.
360
+ α1
361
+ :
362
+ Vaccination rate of susceptibles moving from S1 → S2 (assuming
363
+ θ1 = θ2 = 0, see below).
364
+ α2
365
+ :
366
+ Immunity waning rate inducing a flow from S2 → S1 (assuming
367
+ θ2 = 0, see below).
368
+ βi
369
+ :
370
+ Number of effective contacts per unit time of a susceptible from Si.
371
+ γi
372
+ :
373
+ Recovery rate from I → Si.
374
+ θ1
375
+ :
376
+ Willingness to get vaccinated (alternatively to reduce contacts)
377
+ given the actual prevalence I/N.
378
+ In reality only one of the two
379
+ parameters α1 and θ1 should be chosen non-zero.
380
+ θ2
381
+ :
382
+ Epidemiologically one should restrict to θ2 = 0 or (θ2 = −α2 < 0
383
+ and α1 = 0). In this latter case the meaning of the S2-compartment
384
+ is “contact reducing” and α2 = −θ2 parametrizes the readiness to
385
+ increase contacts proportional to 1 − I/N.
386
+ µi
387
+ :
388
+ Mortality rate in Si.
389
+ µI
390
+ :
391
+ Mortality rate in I. One could also consider vertical transmission,
392
+ in which case µI would be the mortality rate diminished by the rate
393
+ of infected newborns.
394
+ ∆µI
395
+ :
396
+ Mortality excess ∆µI = µI − µ in case µ1 = µ2 = µ, which will be
397
+ assumed most of the time.
398
+ δ
399
+ :
400
+ Rate of not infected newborns. Generalizing to compartment de-
401
+ pendent birth rates amounts to replacing δN = δ1S1 + δ2S2 + δII.
402
+ qi
403
+ :
404
+ Split ratio of newborns between S1 and S2, q1 + q2 = 1. In the
405
+ reduced-immunity interpretation q2 would be the portion of vacci-
406
+ nated newborns.
407
+
408
+ 8
409
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
410
+ So in total this model counts 15 independent parameters (12 if we require constant total
411
+ population, δi = µi, δI = µI). Epidemiologically all parameters except 0 ≥ θ2 ≥ −α2
412
+ are assumed non-negative and also β2 < β1. A more technical classification of admissible
413
+ parameter ranges will be given below. Here is a list of prominent examples in the literature
414
+ -
415
+ Hethcotes classic 3-parameter endemic model (Hethcote 1974, 1976, 1989) by putting
416
+ δ = µi = µI > 0, q1 = 1, β1 > 0, γ2 > 0 and all other parameters vanishing.
417
+ -
418
+ The 7-parameter SIRS model with time varying population size in (Busenberg and
419
+ Driessche 1990), adding to Hethcote’s model an immunity waning rate α2 and allowing
420
+ different (constant per capita) mortality and birth rates.
421
+ -
422
+ The 6-parameter SIRI model of (Derrick and Driessche 1993), replacing the immunity
423
+ waning rate α2 in (Busenberg and Driessche 1990) by the incidence rate β2 > 0 and
424
+ also requiring µ1 = µ2.
425
+ -
426
+ An extended 10-parameter constant population SI(R)S (i.e. mixed SIRS/SIS) model
427
+ with constant and I-linear vaccination rates α1, θ1, an immunity waning rate α2 and
428
+ two recovery flows I ← Si. Hence δi = µi, δI = µI and θ2 = β2 = 04.
429
+ -
430
+ The 6-parameter isolated core system in (Hadeler and Castillo-Chavez 1995), with
431
+ two incidence and recovery rates, βi, γi > 0, a vaccination term α1 > 0 and a constant
432
+ population with balanced birth and death rates, δ = µi = µI > 0 and q1 = 1.
433
+ -
434
+ The 7-parameter vaccination models of (Kribs-Zaleta and Velasco-Hernandez 2000)
435
+ adding an immunity waning rate α2 > 0 to the model of (Hadeler and Castillo-Chavez
436
+ 1995). As we will see in Eq. (2.24) below, due to a redundancy of parameters the two
437
+ models actually stay isomorphic.
438
+ -
439
+ The 8-parameter SIS-model with vaccination and varying population size of (J. Li and
440
+ Ma 2002) keeping only θi = γ2 = β2 = 0 and assuming µ1 = µ2 = µ.5 As we will see
441
+ in (2.25), after a parameter transformation this model becomes isomorphic to the case
442
+ where only θi = 0 and β2 ≤ 0.
443
+ -
444
+ The 8-parameter SIRS-type model analyzed recently by (Avram, Adenane, Bianchin,
445
+ et al. 2022), keeping only γ1 = θ1 = θ2 = q2 = 0 and all other parameters positive.
446
+ The authors allow a varying population size by first discussing the general case of all
447
+ mortality rates being different and then concentrate on µ1 = µ2 ̸= δ and ∆µI > 0.
448
+ Their paper is closest to the present work and in fact initiated it.
449
+ In a “zeroth normalization” step I will now show that passing to fractional variables and
450
+ requiring δ1 − µ1 = δ2 − µ2 all vital dynamic parameters in the SSISS model become
451
+ redundant6. In this way the number of essential parameters reduces from 14 to 8. The
452
+ price to pay in the non-constant population case is possibly getting negative incidence
453
+ rates βi.
454
+ 2.1. Constant population. To get a constant population N the birth rates have to obey
455
+ δi = µi and δI = µI, or more generally
456
+ δ = (µ1S1 + µ2S2 + µII)/N .
457
+ (2.1)
458
+ In case µ1 = µ2 = µ this would read δ = µ+I∆µI. Heuristically this should be understood
459
+ as an approximation for ∆µI/µ ≪ 1. Under this assumption, denoting fractions of the
460
+ 4Here I have chosen enlarge the conventional setting for SI(R)S models by also allowing θ1 > 0.
461
+ 5Actually the authors let µ be a function of N, which however disappears when passing to fractional
462
+ variables.
463
+ 6Redundancy of constant per capita birth and death rates may in fact be shown under quite general
464
+ assumptions in n-compartment models, see Appendix A.
465
+
466
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
467
+ 9
468
+ total population by Si = Si/N and I = I/N and introducing the notations
469
+ ˜α1
470
+ :=
471
+ α1 + q2µ1 ,
472
+ ˜γ1
473
+ :=
474
+ γ1 + q1µI ,
475
+ ˜α2
476
+ :=
477
+ α2 + q1µ2 ,
478
+ ˜γ2
479
+ :=
480
+ γ2 + q2µI ,
481
+ (2.2)
482
+ S =
483
+
484
+ S1
485
+ S2
486
+
487
+ ,
488
+ D(β) =
489
+
490
+ β1
491
+ 0
492
+ 0
493
+ β2
494
+
495
+ ,
496
+ E(α) =
497
+
498
+ α1
499
+ −α2
500
+ −α1
501
+ α2
502
+
503
+ ,
504
+ ˜γ =
505
+
506
+ ˜γ1
507
+ ˜γ2
508
+
509
+ (2.3)
510
+ the dynamical system described by the flow diagram Fig. 1 becomes
511
+ ˙S
512
+ =
513
+ − [E( ˜α) + IE(θ) + ID(β)] S + I ˜γ ,
514
+ (2.4)
515
+ ˙I
516
+ =
517
+ ˜γ(Xrep − 1)I ,
518
+ ˜γ = ˜γ1 + ˜γ2
519
+ (2.5)
520
+ Xrep
521
+ :=
522
+ (β1S1 + β2S2)/˜γ .
523
+ (2.6)
524
+ Note that ˜γ−1 ≡ (γ1 + γ2 + µI)−1 is the expected waiting time in I and hence Xrep is the
525
+ replacement number (Hethcote 2000), i.e. the expected number of secondary cases pro-
526
+ duced by a typical infectious individual during its time of infectiousness. In conventional
527
+ SI(R)S models, i.e. for β2 = θ2 = 0, the replacement number in the limit S1 = 1 would
528
+ become the basic reproduction number r0 = β1/γ. This is why nowadays the replacement
529
+ number is mostly called effective reproduction number. Later we will also have the notion
530
+ of a reduced reproduction number R0 as the value of Xrep at the disease-free equilibrium.
531
+ To avoid misunderstandings, I prefer to keep the various notions of “reproduction num-
532
+ bers” for parameters, whereas the replacement number Xrep is considered as a dynamical
533
+ variable.
534
+ Now obviously, by (2.2), all vital dynamics parameters become redundant and may be
535
+ absorbed by redefining αi and γi. Note that this observation is independent of the choice
536
+ of βi and θi, i.e. it already holds in a combined SI(R)S model.
537
+ 2.2. Time varying population. To derive the equations of motion in case of a time vary-
538
+ ing population keep compartment dependent per capita birth and death rates δi, δI, µi, µI
539
+ constant and put Y = (S1, S2, I), y = N−1Y and
540
+ ν ≡ (ν1, ν2, νI) := (δ1 − µ1, δ2 − µ2, δI − µI).
541
+ Then ˙y = ˙Y/N − y ˙N/N and ˙N/N = ⟨ν | y⟩. Using S1 + S2 + I = 1 we may rewrite
542
+ S1 ˙N/N = S1[ν1 + (ν2 − ν1)S2 + (νI − ν1)I]
543
+ S2 ˙N/N = S2[ν2 + (ν1 − ν2)S1 + (νI − ν2)I]
544
+ I ˙N/N = I[νI + (ν1 − νI)S1 + (ν2 − νI)S2].
545
+ So now introduce
546
+ ˜α1
547
+ :=
548
+ α1 + q2δ1 ,
549
+ ˜α2
550
+ :=
551
+ α2 + q1δ2 ,
552
+ ˜γ1
553
+ :=
554
+ γ1 + q1δI ,
555
+ ˜γ2
556
+ :=
557
+ γ2 + q2δI ,
558
+ ˜β1
559
+ :=
560
+ β1 + νI − ν1 ,
561
+ ˜β2
562
+ :=
563
+ β2 + νI − ν2 .
564
+ (2.7)
565
+ With the same notation as in Eq. (2.3) and e(ν) :=
566
+
567
+ ν1 − ν2
568
+ ν2 − ν1
569
+
570
+ we then get
571
+ ˙S
572
+ =
573
+
574
+
575
+ E( ˜α) + IE(θ) + ID( ˜β)
576
+
577
+ S + I ˜γ + S1S2e(ν) ,
578
+ (2.8)
579
+ ˙I
580
+ =
581
+ ˜γ(Xrep − 1)I ,
582
+ (2.9)
583
+ Xrep
584
+ :=
585
+ (˜β1S1 + ˜β2S2)/˜γ ,
586
+ ˜γ := ˜γ1 + ˜γ2.
587
+ (2.10)
588
+
589
+ 10
590
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
591
+ So, imposing the condition ν1 = ν2 =: ν and putting ∆νI := ν−νI we get e(ν) = 0 and the
592
+ equations of motion look exactly as in the case of constant population (2.4)-(2.6). Again
593
+ all vital dynamics parameters become redundant and may be absorbed by redefining βi,
594
+ αi and γi. The difference this time is that ˜βi = βi − ∆νI may become negative! Thus we
595
+ arrive at
596
+ Proposition 2.1. Assume ν1 = ν2.
597
+ i)
598
+ If ∆νI ≤ min{β1, β2} the SSISS model with variable population maps to the model with
599
+ constant population.
600
+ ii) If ∆νI > min{β1, β2} it maps to the model with min{β1, β2} = 0 and variable popula-
601
+ tion with �
602
+ ∆νI = ∆νI − min{β1, β2}.
603
+ iii) If ∆νI = β2 < β1 and θ2 = 0 it becomes the extended SI(R)S model with θ1 ≥ 0 and
604
+ two recovery flows I → S1 and I → S2.
605
+ Remark 2.2. Note that under the usual assumptions δi = δI = δ and µ1 = µ2 = µ, ∆νI
606
+ coincides with the excess mortality in the infectious compartment, ∆νI = µI −µ = ∆µI.
607
+ Remark 2.3. The observation that on the level of fractional variables in both scenarios
608
+ (constant vs. variable population, the latter provided ν1 = ν2) all vital dynamics param-
609
+ eters are redundant seems to be new7. Essential for this is allowing all four parameters
610
+ (αi, γi) being positive and βi possibly being negative. The introduction of parameters θi
611
+ is not needed to assure this. Redundancy of constant per capita birth and death rates
612
+ may in fact be shown under quite general assumptions in n-compartment models, see
613
+ Appendix A.
614
+ 2.3. Classifying parameter space. In this subsection assume ν1 = ν2. Then the re-
615
+ formulation in terms of possibly negative incidence rates ˜βi leads to a new classification
616
+ scheme identifying seven sectors in this model. For θi = 0 these are labeled by the signa-
617
+ tures of ˜β1 + ˜β2 and ˜β1 ˜β2 (in case of a compartment independent birth rate δ equivalently
618
+ by the size of the excess mortality ∆µI), see Table 1. For θi ̸= 0 this classification will be
619
+ refined in Section 3, Table 3.
620
+ To simplify notation, in what follows let me drop the tilde above parameters. The case
621
+ β1 = β2 will be ignored, since in this case putting S = S1 + S2 one easily checks that
622
+ (S, I) obeys the dynamics of a SIS model, which can immediately be solved by separation
623
+ of variables. Also, due to the permutation symmetry 1 ↔ 2, there is no loss assuming
624
+ β1 > β2. Next, choosing time scale to be measured in units of γ−1, we may without
625
+ loss also put γ = 1. Thus, assume γi ∈ [0, 1] and γ1 + γ2 = 1. So, having started from
626
+ fourteen, essentially we are now left with seven free parameters (think of all greek symbols
627
+ of dimension [time]−1 being divided by γ).
628
+ To further classify the space of admissible parameters some formalism will be needed. Put
629
+ C := {(αi, γi, θi) ∈ R6 | α1 + α2 > 0 ∧ γ1 + γ2 = 1}
630
+ (2.11)
631
+ C+ := C ∩ {(αi, γi) ∈ R4
632
+ ≥0}
633
+ (2.12)
634
+ Csplit := C ∩ {θ1 ≥ 0 ≥ θ2}
635
+ (2.13)
636
+ Cphys := C+ ∩ {θi + αi ≥ 0 , i = 1, 2}
637
+ (2.14)
638
+ Cbio := Csplit ∩ Cphys
639
+ (2.15)
640
+ 7As communicated privately this had also been realized recently in a talk by Florin Avram.
641
+
642
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
643
+ 11
644
+ Table 1. Seven sectors in the SSISS-model at θi = 0 and for compartment
645
+ independent birth rate δ. By Corollary 2.9 Sector I is isomorphic to the models
646
+ of (Hadeler and Castillo-Chavez 1995; Kribs-Zaleta and Velasco-Hernandez 2000)
647
+ and Sectors III-VII are largely covered by (J. Li and Ma 2002). Sector II is a
648
+ mixed SI(R)S model with two recovery flows I → R and I → S.
649
+ Sector
650
+ sign(˜β1 + ˜β2)
651
+ sign(˜β1 ˜β2)
652
+ Interval [˜β1, ˜β2]
653
+ Excess mortality ∆µI
654
+ I
655
+ +
656
+ +
657
+ 0 < ˜β2 < ˜β1
658
+ ∆µI < β2
659
+ II (SIRS)
660
+ +
661
+ 0
662
+ 0 = ˜β2 < ˜β1
663
+ ∆µI = β2
664
+ III
665
+ +
666
+
667
+ 0 < −˜β2 < ˜β1
668
+ β2 < ∆µI < (β1 + β2)/2
669
+ IV
670
+ 0
671
+
672
+ 0 < −˜β2 = ˜β1
673
+ ∆µI = (β1 + β2)/2
674
+ V
675
+
676
+
677
+ ˜β2 < −˜β1 < 0
678
+ (β1 + β2)/2 < ∆µI < β1
679
+ VI
680
+
681
+ 0
682
+ ˜β2 < ˜β1 = 0
683
+ β1 = ∆µI
684
+ VII
685
+
686
+ +
687
+ ˜β2 < ˜β1 < 0
688
+ β1 < ∆µI
689
+ Note that for θi = 0 we have C+ = Cphys = Cbio. Denoting
690
+ B := {β = (β1, β2) ∈ R2 | β2 < β1}.
691
+ (2.16)
692
+ the full parameter sets are then given by A := C × B or Ax := Cx × B, respectively. I will
693
+ also use obvious notations like Aθ=0 := A ∩ {θi = 0} and Aα≥0 := A ∩ {αi ≥ 0}.
694
+ Remark 2.4. In the definition of C in (2.11) the border case α1 = α2 = 0 (i.e. absence of
695
+ constant vaccination and waning immunity rates) has been excluded, see Appendix C for
696
+ a short discussion. For the body of this paper I will stick with the assumption α1+α2 > 0.
697
+ Next, it is easy to check, that for a ∈ Aphys the physical triangle
698
+ Tphys := {(S1, S2, I) ∈ R3
699
+ ≥0 | S1 + S2 + I = 1}
700
+ (2.17)
701
+ stays forward invariant under the dynamics (2.8)-(2.9), i.e. on Tphys we have I = 0 ⇒ ˙I =
702
+ 0 and Si = 0 ⇒ ˙Si ≥ 0. Note that θi + αi ≥ 0 in (2.14) is sufficient but not necessary to
703
+ assure this.
704
+ Lemma 2.5. In the SSISS model (2.8)-(2.9) the physical triangle stays forward invariant
705
+ for all parameters (αi, βi, γi, θi) ∈ Aphys, also including the border case α1 = α2 = 0.
706
+
707
+ We are now ready to state a main result of this paper. Assuming ν1 = ν2 the normaliza-
708
+ tion procedure to be introduced in Section 3 will further reduce the number of essential
709
+ parameters from seven to five. This means, SSISS models fall into isomorphy classes map-
710
+ ping to the same normalized system. It turns out, that these isomorphy classes coincide
711
+ with orbits under a parameter symmetry group GS acting simultaneously on phase P
712
+ and parameter space A, such that parameters for the normalized system are naturally
713
+ identified as elements of A/GS.
714
+ Theorem 2.6. For y = (S1, S2, I)T ∈ R3 and parameter values a = (α, β, γ, θ) ∈ A
715
+ denote ˙y = Fa(y) the dynamical system (2.8)-(2.9) with vector field Fa : R3 → R3. Let
716
+ GS ⊂ GL+(R2) be the subgroup acting on S ∈ R2 from the left and leaving S1 + S2
717
+ invariant.
718
+
719
+ 12
720
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
721
+ i)
722
+ Then there exists a free right action ◁ : A × GS → A such that A becomes a principal
723
+ GS-bundle and
724
+ Fa ◦ Tg = Tg ◦ Fa◁ g ,
725
+ Tg :=
726
+
727
+ � g
728
+ 0
729
+ 0
730
+ 0
731
+ 0
732
+ 1
733
+
734
+ � ,
735
+ ∀(a, g) ∈ A × GS.
736
+ (2.18)
737
+ ii) Put j := ( 0 1
738
+ 1 0 ) and for g ∈ GS denote ¯g := jgj ∈ GS. Viewing α, γ, θ ∈ R2 as column
739
+ vectors and β ∈ B as a row vector and writing a ◁ g = a′ = (α′, β′, γ′, θ′) we have
740
+ α′ = ¯g−1α,
741
+ θ′ = ¯g−1θ + ϑ
742
+ γ′ = g−1γ,
743
+ ϑ =
744
+ 1
745
+ β′
746
+ 1 − β′
747
+ 2
748
+
749
+ −(β1 − β′
750
+ 1)(β2 − β′
751
+ 1)
752
+ (β1 − β′
753
+ 2)(β2 − β′
754
+ 2)
755
+
756
+ β′ = βg
757
+ iii) The GS-right action B × GS ∋ (β, g) �→ βg ∈ B is free and transitive and A ∼=
758
+ A/GS × B as trivial principal fiber bundles.
759
+ iv) Put S′ = g−1S. Then ⟨β|S⟩ = ⟨β′|S′⟩ ≡ Xrep and therefore ˙Xrep = fa(Xrep, I) where
760
+ fa = fa◁ g is GS-invariant, i.e. it only depends on A/GS.
761
+ v) If θ1 ≥ θ2 or θ1θ2 > 08, then there exists g ∈ GS such that a′ := a ◁ g ∈ Aθ=0, i.e. the
762
+ parameters θi may be “gauged to zero”. If in this case a ∈ Abio then also a′ ∈ Abio.
763
+ Remark 2.7. As we will see, although the linear transformation Tg preserves the condition
764
+ S1 + S2 + I = 1, it does not necessarily leave R3
765
+ ≥0 (and hence Tphys) invariant.
766
+ Remark 2.8. Since dim GS = 2 we have dim A/GS = dim A − 2.
767
+ So, using (Xrep, I)
768
+ as independent coordinates in Tphys, the number of essential parameters of the SSISS
769
+ dynamical system reduces from seven to five.
770
+ Parts i)-iv) of Theorem 2.6 will be proven in Corollary 3.7 and Lemma 3.8 and part v) in
771
+ Lemma 3.18. Before coming to this let me close this Section
772
+ -
773
+ in Subsection 2.4 with shortly revisiting some bench-marking models in the literature
774
+ within the present framework,
775
+ -
776
+ in Subsection 2.5 with proving absence of periodic solutions by optimizing the methods
777
+ of (Busenberg and Driessche 1990).
778
+ 2.4. Examples from the literature. For simplicity, in this subsection let me assume a
779
+ compartment independent birth rate δ. Formulating the dynamics for fractional variables
780
+ y = (S1, S2, I) there always remains an ambiguity by adding a vectorfield vanishing on
781
+ Tphys. In Eqs. (2.8)-(2.9) the vector field F ≡ Fa has the special form
782
+ F(y) = My + Γ(y ⊗ y),
783
+ ⟨1|M = ⟨1|Γ = 0,
784
+ (2.19)
785
+ where M ∈ R3×3, 1 = (1, 1, 1) and Γ ∈ Hom (R3 ⊗ R3, R3). As is shown in Appendix A,
786
+ n-compartment models with at most quadratic terms and population size varying only
787
+ due to constant per capita birth and death rates may always be normalized in this way.
788
+ Using different conventions bears the risk of overlooking redundancies in parameter space.
789
+ Moreover, it also makes it tedious to pin down the differences between (or equivalence of)
790
+ various models in the literature. Table 2 shows how the examples quoted at the beginning
791
+ of this Section9 compare with each other when mapped to the present set of parameters.
792
+ 8Actually these conditions are sufficient but not necessary. For a weaker condition see Section 3.3.
793
+ 9Heth = (Hethcote 1974, 1976, 1989); SIRI = (Derrick and Driessche 1993); BuDr = (Busenberg
794
+ and Driessche 1990); SI(R)S = 10-parameter mixed SIRS/SIS model with constant population size and
795
+
796
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
797
+ 13
798
+ Table 2. Mapping models in the literature9 expressed in non-normalized vari-
799
+ ables (S1, S2, I) to the present choice of parameters. The column # counts the
800
+ number of free parameters in the original models.
801
+ After passing to fractional
802
+ variables (S1, S2, I) and tilde parameters, Eq. (2.2) or Eq. (2.7), and resetting
803
+ time scale to ˜γ = 1, the column #eff counts the number of effectively independent
804
+ parameters as determined in Eqs. (2.20)-(2.26).
805
+ α1
806
+ α2
807
+ β1
808
+ β2
809
+ γ1
810
+ γ2
811
+ δ
812
+ µ1
813
+ µ2
814
+ µI
815
+ q1
816
+ q2
817
+ #
818
+ #eff
819
+ Heth
820
+ 0
821
+ 0
822
+
823
+ 0
824
+ 0
825
+
826
+ δ = µ1 = µ2 = µI
827
+ 1
828
+ 0
829
+ 3
830
+ 2
831
+ SIRI1
832
+ 0
833
+ 0
834
+
835
+
836
+ 0
837
+
838
+
839
+ µ1 = µ2
840
+
841
+ 1
842
+ 0
843
+ 6
844
+ 3
845
+ SIRI2
846
+ 0
847
+ 0
848
+
849
+
850
+
851
+ 0
852
+
853
+ µ1 = µ2
854
+
855
+ 0
856
+ 1
857
+ 6
858
+ 3
859
+ BuDr
860
+ 0
861
+
862
+
863
+ 0
864
+ 0
865
+
866
+
867
+
868
+
869
+
870
+ 1
871
+ 0
872
+ 7
873
+ 5
874
+ SI(R)S
875
+
876
+
877
+
878
+ 0
879
+
880
+
881
+ δ = µ1 = µ2 = µI
882
+
883
+
884
+ 7
885
+ 4
886
+ HaCa
887
+
888
+ 0
889
+
890
+
891
+
892
+
893
+ δ = µ1 = µ2 = µI
894
+ 1
895
+ 0
896
+ 6
897
+ 5
898
+ KZVH
899
+
900
+
901
+
902
+
903
+
904
+
905
+ δ = µ1 = µ2 = µI
906
+ 1
907
+ 0
908
+ 7
909
+ 5
910
+ LM
911
+
912
+
913
+
914
+ 0
915
+
916
+ 0
917
+
918
+ µi = f(N)
919
+
920
+
921
+
922
+ 8
923
+ 5
924
+ AABH1
925
+
926
+
927
+
928
+
929
+ 0
930
+
931
+
932
+ µ1 = µ2
933
+ 10
934
+
935
+ 1
936
+ 0
937
+ 8
938
+ 5
939
+ AABH2
940
+
941
+
942
+
943
+
944
+
945
+ 0
946
+
947
+ µ1 = µ2
948
+ 10
949
+
950
+ 0
951
+ 1
952
+ 8
953
+ 5
954
+ Applying the transformations (2.2) or (2.7), respectively, maps the above 11-parameter
955
+ set to the redundancy-free 6-parameter set (˜αi, ˜βi, ˜γi). After resetting time scale to ˜γ ≡
956
+ ˜γ1 + ˜γ2 = 1 the classification of the above models looks as follows:
957
+ AHeth = Abio ∩ Aθ=0 ∩ {˜α1 = 0 ∧ ˜γ2 > 0 ∧ ˜γ1 = ˜α2 ∧ ˜β2 = 0}
958
+ (2.20)
959
+ ASIRIi = Abio ∩ Aθ=0 ∩ {˜αi = 0 ∧ ˜γj > 0 ∧ ˜γi = ˜αj, j ̸= i}
960
+ (2.21)
961
+ ABuDr = Abio ∩ Aθ=0 ∩ {˜α1 = 0 ∧ ˜γ2 > 0 ∧ ˜β2 < 0}11
962
+ (2.22)
963
+ ASIRS = Abio ∩ Aθ2=0 ∩ {˜β2 = 0}
964
+ (2.23)
965
+ AKZVH = Abio ∩ Aθ=0 ∩ {˜β2 > 0} = AHaCa
966
+ (2.24)
967
+ ALM = Abio ∩ Aθ=0 ∩ {˜β2 < 0 ∧ ˜γ1 > 0}
968
+ (2.25)
969
+ AAABHi = Abio ∩ Aθ=0 ∩ {˜γj > 0, j ̸= i}
970
+ (2.26)
971
+ The dimensions of these parameter spaces are displayed in the last column of Table 211.
972
+ To verify Eqs. (2.20)-(2.26) the following explanations should suffice.
973
+
974
+ The SIRI model of (Derrick and Driessche 1993) with varying population requires
975
+ αi = γ1 = 0. Since for βR > βS the mapping to the SISS model permutes 1 ↔ 2 (i.e.
976
+ maps R → S1 and S → S2), if βR < βS we get ˜α1 = 0, ˜α2 = ˜γ1 = δ and ˜γ2 = γ2 > 0,
977
+ and if βR > βS we get ˜α2 = 0, ˜α1 = ˜γ2 = δ and ˜γ1 = γ1 > 0.
978
+
979
+ The SIRS model of (Busenberg and Driessche 1990) differs from SIRI by allowing
980
+ α2 > 0 and µ1 < µ2, but in turn it requires βS > βR = 0. Thus, we have ˜α1 = 0
981
+ θ2 = β2 = 0; HaCa = core system in (Hadeler and Castillo-Chavez 1995); KZVH = (Kribs-Zaleta and
982
+ Velasco-Hernandez 2000); LM = (J. Li and Ma 2002); AABH = (Avram, Adenane, Bianchin, et al. 2022).
983
+ BuDr and AABH come in two versions, the subscript 1 refers to βS > βR and 2 to βS < βR.
984
+ 10 The bulk of results in Section 5 and 6 of (Avram, Adenane, Bianchin, et al. 2022) assumes µ1 = µ2.
985
+ 11To be comparable Eq. (2.22) refers to the sub-case µ1 = µ2 in (Busenberg and Driessche 1990), so
986
+ dim ABuDr = 4. Allowing also an excess mortality µ2 − µ1 > 0 gives #eff = 5 in Table 2.
987
+
988
+ 14
989
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
990
+ and ˜γ1 = δ as in SIRI1, but ˜α2 = α2 + δ becomes independent. If, for comparison, we
991
+ restrict to µ1 = µ2 = µ then β2 = 0 implies ˜β2 = −∆µI ≤ 0.
992
+
993
+ If q1 > 0 then one of the three parameters (γ1, α2, δ) always becomes redundant.
994
+ So the models of (Hadeler and Castillo-Chavez 1995) and (Kribs-Zaleta and Velasco-
995
+ Hernandez 2000) are isomorphic, in spite of the latter containing the additional im-
996
+ munity waning rate α2. Also, they both satisfy ˜β2 = β2 > 0.
997
+
998
+ Putting q2 = 1 in the SIS-type model of (J. Li and Ma 2002) the mapping (α1, α2, γ1, δ) �→
999
+ (˜αi, ˜γi) is bijective. Also, the authors have defined µi = f(N) and µI = f(N) + ∆µI.
1000
+ Hence, the only restrictions in this model are ˜β2 = −∆µI < 0 and ˜γ1 > 0.
1001
+ In summary we get the following conclusions, which apparently have not yet been realized
1002
+ in the literature.
1003
+ Corollary 2.9. Assume µ1 = µ2 =: µ and put ∆µI := µI − µ.
1004
+ i)
1005
+ For β1 > β2 = ∆µI the SIRI model of (Derrick and Driessche 1993) is isomorphic to
1006
+ Hethcote’s classic endemic model.
1007
+ Moreover, restricting to ˜γ1 > 0 and β2 ̸= ∆µI we have
1008
+ ii) The SIRS-type model of (Busenberg and Driessche 1990) reduces to a sub-case of the
1009
+ SIS-type model of (J. Li and Ma 2002), which in turn covers Sectors III-VII of the
1010
+ SSISS model at θi = 0.
1011
+ iii) The models of (Hadeler and Castillo-Chavez 1995) and (Kribs-Zaleta and Velasco-
1012
+ Hernandez 2000) are isomorphic and cover Sector I of the SSISS model at θi = 0.
1013
+ iv) The models of (J. Li and Ma 2002) and (Hadeler and Castillo-Chavez 1995; Kribs-
1014
+ Zaleta and Velasco-Hernandez 2000) only differ by the sign of ˜β2.
1015
+ v) Their disjoint union covers the SIRI model of (Derrick and Driessche 1993) and co-
1016
+ incides with the model of (Avram, Adenane, Bianchin, et al. 2022).
1017
+ An equivalent formulation of Corollary 2.9 based on normalized parameters and vari-
1018
+ ables is given in Corollary 3.19 in Section 3.4.
1019
+ 2.5. Absence of periodic solutions. In this subsection I will specify parameter ranges
1020
+ guaranteeing absence of periodic solutions by optimizing methods from (Busenberg and
1021
+ Driessche 1990) (see also (Busenberg and Driessche 1991; Derrick and Driessche 1993)) for
1022
+ the present situation, including θi ̸= 0. To start with, the Busenberg-Driessche version of
1023
+ the classical Bendixson–Dulac Theorem may be given the following alternative formulation
1024
+ Lemma 2.10. (Busenberg and Driessche 1990) Let F : R3 → R3 be smooth in a neigh-
1025
+ borhood of Tphys and assume Tphys forward invariant under the flow of ˙y = F(y). Assume
1026
+ there exists a smooth scalar function u(y) defined in a neighborhood of Tphys such that
1027
+ Ψ(y) := ∇ · (uF)(y) − (y · ∇)(u
1028
+
1029
+ i
1030
+ Fi)(y) ≤ 0 ,
1031
+ ∀y ∈ Tphys
1032
+ (2.27)
1033
+ and Ψ(y) < 0 for some y ∈ Tphys. Then in Tphys \ ∂Tphys there exist no periodic solutions,
1034
+ homoclinic loops or oriented phase polygons of the dynamical system ˙y = F(y).
1035
+ Proof. Put 1 := (1, 1, 1) and g := y × uF. Then g · F = 0 and ⟨1 | ∇ × g⟩|Tphys = Ψ|Tphys,
1036
+ where the second identity easily follows from ⟨1 | F⟩|Tphys = 0. Now the claim follows by
1037
+ Stoke’s Theorem as in the proof of Theorem 4.1 of (Busenberg and Driessche 1990).
1038
+
1039
+ Remark 2.11. In Lemma A.1 in Appendix A it is shown that for models with constant
1040
+ per capita birth and death rates one may always replace F by ˜F obeying F|Tphys = ˜F|Tphys
1041
+
1042
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1043
+ 15
1044
+ and ⟨1 | ˜F⟩ = 0 also outside Tphys. So in this case the second term in (2.27) vanishes and
1045
+ the condition ∇(u˜F) ≤ 0 looks like in the classical Bendixson-Dulac theorem.
1046
+ As in (Busenberg and Driessche 1990) putting y = (S1, S2, I) and u = 1/(S1S2I) we now
1047
+ apply this to the dynamical system Eqs. (2.8)-(2.10). We have uF(y) = uMy + uf(y)
1048
+ where
1049
+ M =
1050
+
1051
+
1052
+ −˜α1
1053
+ ˜α2
1054
+ ˜γ1
1055
+ ˜α1
1056
+ −˜α2
1057
+ ˜γ2
1058
+ 0
1059
+ 0
1060
+ −1
1061
+
1062
+ � ,
1063
+ (uf)(y) =
1064
+
1065
+
1066
+ −( ˜β1 + θ1)/S2 + θ2/S1 + (ν1 − ν2)/I
1067
+ −( ˜β2 + θ2)/S1 + θ1/S2 + (ν2 − ν1)/I
1068
+ ˜β1/S2 + ˜β2/S1
1069
+
1070
+ � .
1071
+ (2.28)
1072
+ Here the time scale normalization ˜γ1 + ˜γ2 = 1 is understood.
1073
+ Theorem 2.12. Under the following conditions there exist no periodic solutions, homo-
1074
+ clinic loops or oriented phase polygons of the SSISS system (2.8)-(2.10) in Tphys.
1075
+ i) (˜αi, ˜γi, θi) ∈ R6
1076
+ ≥0.
1077
+ ii) (˜αi, ˜γi, θi) ∈ Cbio and ν1 = ν2.
1078
+ Proof. First note that ˜γ1 + ˜γ2 = 1 implies that the boundary lines {S1 = 0} and {S2 = 0}
1079
+ cannot both be forward invariant. Hence, ∂Tphys cannot be a phase polygon. Next, the
1080
+ second term in (2.27) vanishes, because we have ⟨1 | F⟩ = 0 also outside of Tphys. We are
1081
+ left to compute ∇·(u(y)My) = − �
1082
+ i̸=j Mi,jyj/yi < 0 and ∇·f = −θ2/S2
1083
+ 1 −θ1/S2
1084
+ 2. Part i)
1085
+ follows by Lemma 2.5 and Lemma 2.10. The proof of part ii) relies on the normalization
1086
+ formalism of Section 3 and follows from Corollary 3.17.
1087
+
1088
+ Remark 2.13. Note that Theorem 2.12ii) doesn’t follow directly from Theorem 2.6, because
1089
+ there the equivalence transformation Tg need not preserve Tphys, see also Remark 2.7.
1090
+ Remark 2.14. Usually in the literature on models with constant per capita birth and death
1091
+ rates the vector field F appears in the form F = FM + f, where FM = My − ⟨1 | My⟩y,
1092
+ the second term being nonzero. This makes computations more involved but still yields
1093
+ ΨM|Tphys ≡ ∇ · (uFM)|Tphys − (y · ∇)⟨1 | uFM⟩|Tphys = − �
1094
+ i̸=j Mi,jyj/yi, see Eq. (3.8)
1095
+ in (Derrick and Driessche 1993). The fact that M may be chosen to satisfy ⟨1|M = 0
1096
+ (Lemma A.1 in Appendix A, see also remark 2.11) is rarely noticed in the literature.
1097
+ 3. Normalization
1098
+ 3.1. Phase space. From now on we drop again the tilde above parameters and also
1099
+ require ν1 = ν2. To proceed one has to choose suitable coordinates (X, Y ) on a phase space
1100
+ P ⊃ Tphys. Let’s first do some linear algebra. Put V = R2 and consider S ≡ |S⟩ =
1101
+ � S1
1102
+ S2
1103
+
1104
+ ,
1105
+ α ≡ |α⟩ = ( α1
1106
+ α2 ), γ ≡ |γ⟩ = ( γ1
1107
+ γ2 ), θ ≡ |θ⟩ =
1108
+ � θ1
1109
+ θ2
1110
+
1111
+ as elements of V (“ket-” or “column-”
1112
+ vectors). Denote
1113
+ e ≡ ⟨e| := (1, 1) ,
1114
+ β ≡ ⟨β| := (β1, β2)
1115
+ (3.1)
1116
+ as a basis in the dual space V ∗ (“bra-” or “row-” vectors). Putting L(β, θ) := D(β)+E(θ)
1117
+ we then have
1118
+ ⟨e|E(α) = 0,
1119
+ ⟨e|L(β, θ) = ⟨β|,
1120
+ ⟨e | γ⟩ = 1
1121
+ (3.2)
1122
+ where ⟨· | ·⟩ denotes the dual pairing V ∗ ⊗ V → R. Generalizing this setting, pick (e, β)
1123
+ any oriented12 basis in V ∗ and γ ∈ V satisfying ⟨e | γ⟩ = 1. Denote E ⊂ End V the right
1124
+ 12The requirement of being oriented (with respect to a given orientation in V ) is a coordinate free
1125
+ version of the condition β2 < β1.
1126
+
1127
+ 16
1128
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1129
+ ideal anihilated by ⟨e| and L := {L ∈ End V | ⟨e|L = ⟨β|}. On V × R = R3 consider the
1130
+ dynamical system
1131
+ ˙S = − [E + IL] S + Iγ,
1132
+ S ∈ V, E ∈ E, L ∈ L,
1133
+ (3.3)
1134
+ ˙I = (X − 1)I,
1135
+ I ∈ R, X := ⟨β | S⟩.
1136
+ (3.4)
1137
+ Fixing e and varying (β, E, L, γ) under the above constraints defines a 7-parameter dy-
1138
+ namical system which in fact provides a coordinate free reformulation of the SSISS model
1139
+ (2.4). Note that the conditions imply ⟨e | ˙S⟩ + ˙I ≡ ˙S1 + ˙S2 + ˙I = 0, so the dynamics
1140
+ (3.3)-(3.4) leaves the cosets {⟨e | S⟩ + I = const.} ⊂ R3 invariant. Since I = 0 implies
1141
+ ˙I = 0 also the half spaces {I ∈ R±} as well as the plane {I = 0} stay invariant.
1142
+ Definition 3.1. The dynamical system (3.3)-(3.4) on phase space P = {(S, I) ∈ V ×R≥0 |
1143
+ ⟨e | S⟩ + I = 1} with parameter space A = C × B is called the extended SSISS model.
1144
+ Remark 3.2. The extension to negative values of variables Si and parameters a is needed
1145
+ to construct the symmetry operation of GS in Theorem 2.6.
1146
+ 3.2. Canonical coordinates. Putting I := 1 − ⟨e | S⟩ and using S as independent
1147
+ coordinates on P Eq. (3.4) becomes redundant and we end up with a two-dimensional
1148
+ system. However, based on the coordinate free formulation (3.3)-(3.4), there is another
1149
+ natural set of canonical coordinates for this system. Put
1150
+ X := ⟨β | S⟩,
1151
+ Y := ⟨e | S⟩,
1152
+ (3.5)
1153
+ or equivalently choose the basis dual to (3.1) in V
1154
+ e⊥ ≡ |e⊥⟩ :=
1155
+ 1
1156
+ β1 − β2
1157
+
1158
+ 1
1159
+ −1
1160
+
1161
+ ,
1162
+ β⊥ ≡ |β⊥⟩ :=
1163
+ 1
1164
+ β1 − β2
1165
+
1166
+ −β2
1167
+ β1
1168
+
1169
+ (3.6)
1170
+ Hence we have X ≡ Xrep, Y ≡ S1 + S2 and
1171
+ S = Xe⊥ + Y β⊥.
1172
+ (3.7)
1173
+ Lemma 3.3. In canonical coordinates the extended SSISS model becomes
1174
+ ˙X
1175
+ =
1176
+ (−aX + b) + (−cX + d)I − ϵI2 ,
1177
+ (3.8)
1178
+ ˙Y
1179
+ =
1180
+ (1 − X)I = − ˙I ,
1181
+ (3.9)
1182
+ where I = 1 − Y and where the new parameters are given by
1183
+ a := α1 + α2
1184
+ (3.10)
1185
+ b := α2β1 + α1β2
1186
+ (3.11)
1187
+ c := β1 + β2 + θ1 + θ2
1188
+ (3.12)
1189
+ d := γ1β1 + γ2β2 − b + ϵ
1190
+ (3.13)
1191
+ ϵ := β1β2 + β1θ2 + β2θ1 .
1192
+ (3.14)
1193
+ Proof. In canonical coordinates the matrices E(α) and L(β, θ) := D(β) + E(θ) take the
1194
+ normal form
1195
+ E(α) =
1196
+
1197
+ a
1198
+ −b
1199
+ 0
1200
+ 0
1201
+
1202
+ ,
1203
+ L(β, θ) =
1204
+
1205
+ c
1206
+ −ϵ
1207
+ 1
1208
+ 0
1209
+
1210
+ .
1211
+ (3.15)
1212
+ Using |γ⟩ = (β1γ1 +β2γ2)|e⊥⟩+|β⊥⟩ the claim follows by straightforward calculation.
1213
+
1214
+
1215
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1216
+ 17
1217
+ The canonical form of the SSISS dynamical system (3.8)-(3.9) will also be called the RN-
1218
+ dynamical system (RN = replacement number). Beware that unless β2 ≥ 0 the “would-be”
1219
+ replacement number X may take negative values even for Si ≥ 0. In fact, in canonical
1220
+ coordinates the physical triangle takes the form
1221
+ Tphys(β) = {(X, Y ) ∈ R × [0, 1] | β2Y ≤ X ≤ β1Y }
1222
+ = {(X, I) ∈ R × [0, 1] | β2(1 − I) ≤ X ≤ β1(1 − I)}.
1223
+ (3.16)
1224
+ So in (X, I)-space Tphys is given by the corners T< = (β2, 0), T> = (β1, 0) and T∧ = (0, 1).
1225
+ To stay with epidemiological conventions, from now on I will use X ≡ Xrep and I ≡ 1−Y
1226
+ as independent variables, in terms of which phase space is now given by
1227
+ P = {(X, I) ∈ R × R≥0}.
1228
+ Also note that in canonical coordinates the dynamics is reduced from seven to five pa-
1229
+ rameters, i.e. the system no longer depends on β. So, the role of β is reduced to fixing
1230
+ the image of physical triangles Tphys in canonical coordinates. Equivalently this means
1231
+ that fixing x = (a, b, c, d, ϵ) and varying β ∈ B we get an equivalence class of isomorphic
1232
+ dynamical systems, albeit physical triangles are not mapped onto each other under these
1233
+ isomorphisms.
1234
+ Proposition 3.4. For a, a′ ∈ A, a = (α, β, γ, θ) and a′ = (α′, β′, γ′, θ′), assume x(a) =
1235
+ x(a′). Following Eq. (3.7) put
1236
+ S := Xe⊥(β) + (1 − I)β⊥ ,
1237
+ S′ := Xe⊥(β′) + (1 − I)β′⊥ .
1238
+ (3.17)
1239
+ Then S1 + S2 = S′
1240
+ 1 + S′
1241
+ 2 = 1 − I and S = gS′ where g ∈ GL+(R2) is uniquely defined by
1242
+ g = |β⊥⟩⟨e| + |e⊥(β)⟩⟨β′| =
1243
+ 1
1244
+ β1 − β2
1245
+
1246
+ β′
1247
+ 1 − β2
1248
+ β′
1249
+ 2 − β2
1250
+ β1 − β′
1251
+ 1
1252
+ β1 − β′
1253
+ 2 ,
1254
+
1255
+ (3.18)
1256
+ implying det g = (β′
1257
+ 1 − β′
1258
+ 2)/(β1 − β2) > 0. Moreover, (S, I) satisfies the SSISS dynamics
1259
+ (3.3)-(3.4) at parameter values a iff (S′, I) satisfies it at parameter values a′.
1260
+ Proof. Eq. (3.17) implies ⟨e|S⟩ = ⟨e|S′⟩ = 1−I and ⟨β|S⟩ = ⟨β′|S′⟩ = X. Hence, g must
1261
+ satisfy ⟨e|g = ⟨e| and ⟨β|g = ⟨β′| with unique solution (3.18).
1262
+
1263
+ Remark 3.5. Apparently we have g ∈ GS := {g ∈ GL+(R2) | ⟨e|g = ⟨e|} and by Eq.
1264
+ (3.18) β �→ βg defines a transitive and free right action of GS on B13. In Corollary 3.7
1265
+ below this action will be transported to a free GS-action on A, thus proving parts i)-iv)
1266
+ of Theorem 2.6.
1267
+ 3.3. Main results. In this subsection we study the constraints on the new parameters
1268
+ x := (a, b, c, d, ϵ) and admissible ranges of β - or equivalently Tphys(β) - for given values
1269
+ of x, which will finally lead to a proof of Theorems 2.6 and 2.12. Recalling A ≡ C × B
1270
+ denote
1271
+ φ : A ∋ a �→ (x(a), β) ∈ D × B,
1272
+ D := R+ × R4
1273
+ (3.19)
1274
+ where x(a) is given by (3.10)-(3.14). The proof of the following Lemma is by straight
1275
+ forward calculation and hence omitted.
1276
+ 13Note that dim GS = 2.
1277
+ The parametrization of g in (3.18) is redundant by invariance under
1278
+ (β1, β2) �→ (β1 + λ, β2 + λ) and (β1, β2) �→ (χβ1, χβ2), (λ, χ) ∈ R × R+.
1279
+
1280
+ 18
1281
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1282
+ Lemma 3.6. The map φ : A → D × B provides a diffeomorphism with φ−1 given by
1283
+ αi = b − aβi
1284
+ βj − βi
1285
+ ,
1286
+ γi = d + b − ϵ − βj
1287
+ βi − βj
1288
+ ,
1289
+ θi = β2
1290
+ i − cβi + ϵ
1291
+ βj − βi
1292
+ ,
1293
+ j ̸= i
1294
+ (3.20)
1295
+
1296
+ Corollary 3.7. Consider D×B as a trivial principal GS-bundle with fiber B and GS right
1297
+ action (x, β) ◁ g := (x, βg), see Remark 3.5. Defining a ◁ g := φ−1(x(a), βg) we get an
1298
+ isomorphic GS-bundle structure on A. Putting y := (S1, S2, I) and writing the dynamical
1299
+ system (3.3)-(3.4) with parameters a ∈ A as ˙y = Fa(y), Proposition 3.4 becomes
1300
+ Fa ◦ Tg = Tg ◦ Fa◁ g ,
1301
+ Tg := g ⊕ id ,
1302
+ g ∈ GS.
1303
+ This proves parts i), iii) and iv) of Theorem 2.6.
1304
+
1305
+ The remaining transformation rules in part ii) of Theorem 2.6 now boil down to an exercise
1306
+ in linear algebra.
1307
+ Lemma 3.8. Let D(β) and E(α) be given as in Eq. (2.3) and ϑ(β, β′) as in part ii) of
1308
+ Theorem 2.6. Then for all g ∈ GS, α ∈ R2 and β′ = βg ∈ B
1309
+ E(¯gα)g = gE(α),
1310
+ D(β)g = g [D(β′) + E(ϑ(β, β′))]
1311
+ Applying these identities to the dynamical system (3.3)-(3.4) proves Theorem 2.6ii).
1312
+
1313
+ Remark 3.9. Beware that the transformation matrix g preserves S1+S2 but not necessarily
1314
+ R2
1315
+ ≥0.
1316
+ Also, if a ∈ Aphys (or Abio) and x(a) = x(a′) then it depends on β′ whether
1317
+ a′ ∈ Aphys (or Abio), see Proposition 3.15 below. Hence, the above equivalencies may
1318
+ produce scenarios where a ∈ Aphys and a′ = a ◁ g ̸∈ Aphys and T−1
1319
+ g Tphys ̸∈ R3
1320
+ ≥0 but still
1321
+ T−1
1322
+ g Tphys is forward invariant under the flow of Fa′.
1323
+ Next, on D define the functions
1324
+ R0(x) := b/a
1325
+ ≡ α2β1 + α1β2
1326
+ α1 + α2
1327
+ ,
1328
+ (3.21)
1329
+ R1(x) := d + b − ϵ ≡ γ1β1 + γ2β2 .
1330
+ (3.22)
1331
+ Obviously we may also use x ≡ (a, R0, R1, c, ϵ) ∈ R+ × R4 as independent parameters in
1332
+ D. Moreover we clearly have
1333
+ φ(A+) = {(x, β) ∈ D × B | β2 ≤ Ri ≤ β1 , i = 1, 2} ,
1334
+ (3.23)
1335
+ i.e. on A+ the functions Ri may be interpreted as two kinds of mean values of β1 and β2.
1336
+ Again beware that for β2 < 0 we may have Ri < 0 even on A+. To explain the meaning
1337
+ of R0 note that for a > 0 the value of the replacement number X at the disease-free
1338
+ equilibrium (DFE) of the RN-dynamical system (3.8)-(3.9) is precisely given by X∗
1339
+ 0 = R0.
1340
+ Following results of (Driessche and Watmough 2002) this leads to
1341
+ Definition 3.10. R0 is called the reduced reproduction number.
1342
+ Remark 3.11. As has been shown by (Driessche and Watmough 2002, 2008), in models
1343
+ with just one infectious compartment the more general notion of R0 as the spectral
1344
+ radius of the next generation matrix ((Diekmann, Heesterbeek, and Metz 1990), see also
1345
+ (Diekmann and Heesterbeek 2000)) reduces to the above definition. Denoting the values
1346
+
1347
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1348
+ 19
1349
+ of Si at the DFE by S∗
1350
+ i we have R0 = β1S∗
1351
+ 1 + β2S∗
1352
+ 2, which is the usual formula, see e.g.
1353
+ (Kribs-Zaleta and Velasco-Hernandez 2000) or (Avram, Adenane, Bianchin, et al. 2022).
1354
+ Remark 3.12. Mostly in the literature R0 is called the basic reproduction number. But in
1355
+ case β2 = 0 this terminology is already occupied by r0 := β1/γ as the expected number
1356
+ of secondary cases produced by a typical infectious individual in a totally susceptible
1357
+ population. So to avoid confusion I prefer to call R0 the reduced reproduction number.
1358
+ Next put Dx := πD(φ(Ax)), x = phys or x = bio, where πD : D × B → D denotes
1359
+ the canonical projection. We look for suitable coordinates describing Dx and then derive
1360
+ additional bounds on β to describe φ(Ax). Consider the following functions on D.
1361
+ A±(x) := 1
1362
+ 2
1363
+
1364
+ a + c ±
1365
+
1366
+ (a + c)2 − 4(b + ϵ)
1367
+
1368
+ ,
1369
+ (3.24)
1370
+ B±(x) := 1
1371
+ 2
1372
+
1373
+ c ±
1374
+
1375
+ c2 − 4ϵ
1376
+
1377
+ .
1378
+ (3.25)
1379
+ Then by (3.15) and the trace-det formula A± and B± provide the eigenvalues of E + L
1380
+ and L, respectively. The meaning of these eigenvalues becomes clear by looking at (3.20)
1381
+ β1 = A+ ⇔ α1 + θ1 = 0
1382
+ β1 = B+ ⇔ θ1 = 0
1383
+ (3.26)
1384
+ β2 = A− ⇔ α2 + θ2 = 0
1385
+ β2 = B− ⇔ θ2 = 0
1386
+ (3.27)
1387
+ βi = R0 ⇔ αi = 0
1388
+ βi = R1 ⇔ γj = 0, j ̸= i
1389
+ (3.28)
1390
+ More generally from (3.20) we get
1391
+ θi = (βi − B−)(βi − B+)
1392
+ βj − βi
1393
+ ,
1394
+ j ̸= i ,
1395
+ (3.29)
1396
+ αi + θi = (βi − A−)(βi − A+)
1397
+ βj − βi
1398
+ ,
1399
+ j ̸= i .
1400
+ (3.30)
1401
+ Hence A± will serve to fix the constraints on (x, β) ∈ φ(Aphys) and B± (B ≡ “bio”) to fix
1402
+ constraints on (x, β) ∈ φ(Abio). First we gather some trivial identities.
1403
+ c = B+ + B− = A+ + A− − a ,
1404
+ ϵ = B+B− = A+A− − b ,
1405
+ (3.31)
1406
+ a = A+ + A− − B+ − B− ,
1407
+ aR0 ≡ b = A+A− − B+B−
1408
+ (3.32)
1409
+ From these one immediately computes
1410
+ a(A± − R0) = (A± − B+)(A± − B−) = A2
1411
+ ± − cA± + ϵ
1412
+ (3.33)
1413
+ a(R0 − B±) = (B± − A+)(B± − A−) = B2
1414
+ ± − (a + c)B± + (b + ϵ)
1415
+ (3.34)
1416
+ Now let’s introduce the notation
1417
+ DA := D ∩ {A± ∈ R}
1418
+ (3.35)
1419
+ DB := D ∩ {B± ∈ R ∧ B− < B+}
1420
+ (3.36)
1421
+ DAB := DA ∩ DB ∩ {B− ≤ A− ≤ B+ ≤ A+}.
1422
+ (3.37)
1423
+ Lemma 3.13. The following identities hold
1424
+ DAB = {x ∈ DA | A− ≤ R0 ≤ A+ ∧ c2 ̸= 4ϵ} = {x ∈ DB | B− ≤ R0 ≤ B+}
1425
+ (3.38)
1426
+ Hence in DAB we always have the additional bound
1427
+ B− ≤ A− ≤ R0 ≤ B+ ≤ A+ .
1428
+ (3.39)
1429
+
1430
+ 20
1431
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1432
+ Proof. By Eqs. (3.33) and (3.34) on DAB we always have A− ≤ R0 ≤ A+ and B− ≤
1433
+ R0 ≤ B+.
1434
+ Conversely, if A± ∈ R, c2 ̸= ϵ and A− ≤ R0 ≤ A+ then (3.33) implies
1435
+ A2
1436
+ − − cA− + ϵ ≤ 0 ≤ A2
1437
+ + − cA+ + ϵ and therefore c2 > 4ϵ. Hence B− < B+ ∈ R and again
1438
+ by (3.33) B− ≤ A− ≤ B+ ≤ A+. The second identity follows analogously.
1439
+
1440
+ Lemma 3.14. Denoting iB : DB ∋ x �→ (x, B+, B−) ∈ DB × B the following identities
1441
+ hold
1442
+ φ(Asplit) = {(x, β) ∈ DB × B | B− ≤ β2 < β1 ≤ B+} ,
1443
+ (3.40)
1444
+ φ(Aθ+α≥0) = {(x, β) ∈ DA × B | β2 ≤ A− ≤ β1 ≤ A+} .
1445
+ (3.41)
1446
+ φ(Aθ=0) = iB(DB)
1447
+ (3.42)
1448
+ φ(Aθ=0 ∩ Aα≥0) = iB(DAB)
1449
+ (3.43)
1450
+ Proof. We have B± ∈ R iff there exists β ∈ R such that β2 − cβ + ϵ ≤ 0. Hence, by
1451
+ (3.20), if θ1 ≥ 0 and θ2 ≤ 0 then B± ∈ R and B− ≤ β2 < β1 ≤ B+, proving the “⊂”-part
1452
+ in (3.40). The opposite direction follows from (3.29). Similarly, A± ∈ R iff there exists
1453
+ β ∈ R such that β2 − (a + c)β + b + ϵ ≤ 0. Hence, by (3.20), if α1 + θ1 ≥ 0 then A± ∈ R
1454
+ and A− ≤ β1 ≤ A+. If in addition α2 + θ2 ≥ 0 then also β2 ≤ A−, proving the “⊂”-part
1455
+ in (3.41). The opposite direction follows from (3.30). Eq. (3.42) follows since in Aθ=0 we
1456
+ have β1 = B+ and β2 = B−. If in addition αi ≥ 0 then (3.30) implies Eq. (3.43).
1457
+
1458
+ We are now in the position to summarize the constraints describing φ(Aphys) and φ(Abio).
1459
+ Proposition 3.15. For Ax = Cx × B as defined in (2.14) - (2.15) we have
1460
+ φ(Aphys) ≡ φ(A+) ∩ φ(Aθ+α≥0)
1461
+ = (DA × B) ∩ {β2 ≤ {A−, R0, R1} ≤ β1 ≤ A+} ,
1462
+ (3.44)
1463
+ Dphys = DA ∩ {R0,1 ≤ A+} ,
1464
+ (3.45)
1465
+ φ(Abio) ≡ φ(Asplit) ∩ φ(Aphys)
1466
+ = (DAB × B) ∩ {B− ≤ β2 ≤ A− ≤ R0 ≤ β1 ≤ B+} ∩ {R1 ∈ [β2, β1]} ,
1467
+ (3.46)
1468
+ Dbio = DAB ∩ {R1 ∈ [B−, B+]} ⊂ Dphys .
1469
+ (3.47)
1470
+ Proof. This is a summary of Eq. (3.23) and Lemmas 3.13 - 3.14.
1471
+
1472
+ Proposition 3.15 motivates the following notation and definition
1473
+ Definition 3.16. For x ∈ Dbio put
1474
+ βmax
1475
+ 2
1476
+ (x) := min{A−, R1},
1477
+ βmin
1478
+ 1
1479
+ (x) := max{R0, R1}.
1480
+ (3.48)
1481
+ Then β ∈ B is called bio-compatible with x if B− ≤ β2 ≤ βmax
1482
+ 2
1483
+ and βmin
1484
+ 1
1485
+ ≤ β1 ≤ B+,
1486
+ equivalently if φ−1(x, β) ∈ Abio.
1487
+ Similarly, β is called compatible if β2 ≤ βmax
1488
+ 2
1489
+ and
1490
+ βmin
1491
+ 1
1492
+ ≤ β1 ≤ A+, equivalently if φ−1(x, β) ∈ Aphys. A physical triangle Tphys(β) is called
1493
+ (bio)-compatible, if β is (bio)-compatible.
1494
+ Hence, (bio-)compatible physical triangles are always forward invariant under the RN-
1495
+ dynamics (3.8)-(3.9) and the smallest one is just Tphys(βmin
1496
+ 1
1497
+ , βmax
1498
+ 2
1499
+ ). The following Corollary
1500
+ also proves part ii) of Theorem 2.12.
1501
+
1502
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1503
+ 21
1504
+ Corollary 3.17. Let x ∈ Dbio and let β ∈ B be compatible with x. Then there exist
1505
+ no periodic solutions, homoclinic loops or oriented phase polygons of the RN-dynamical
1506
+ system (3.8)-(3.9) in Tphys(β).
1507
+ Proof. Let Z ⊂ Tphys(β) be a solution cycle (image of a periodic solution, a homoclinic
1508
+ loop or an oriented phase polygon). As argued in the proof of Theorem 2.12, we must
1509
+ have Z ̸= ∂Tphys(β′) for all β′ ∈ B. Hence, by forward invariance, Z must lie inside the
1510
+ smallest compatible triangle, Z ⊂ Tphys(βmin
1511
+ 1
1512
+ , βmax
1513
+ 2
1514
+ ) ⊂ Tphys(B+, B−). But, by Proposition
1515
+ 3.15 and Eq.
1516
+ (3.43), φ−1(x, B+, B−) ∈ Abio ∩ Aθ=0 and we get a contradiction with
1517
+ Theorem 2.12i).
1518
+
1519
+ Finally, to prove Theorem 2.6v), note that Lemma 3.14 and Proposition 3.15 in particular
1520
+ imply (use that GS acts transitively on B)
1521
+ Aθ=0 ◁ GS = Asplit ◁ GS = φ−1(DB × B)
1522
+ (3.49)
1523
+ Aθ+α≥0 ◁ GS = φ−1(DA × B)
1524
+ (3.50)
1525
+ (Aθ=0 ∩ Aα≥0) ◁ GS = φ−1(DAB × B)
1526
+ (3.51)
1527
+ Aphys ◁ GS = φ−1(Dphys × B)
1528
+ (3.52)
1529
+ Abio ◁ GS = φ−1(Dbio × B)
1530
+ (3.53)
1531
+ Aθ=0 ∩ Abio ◁ GS ⊂ Abio
1532
+ (3.54)
1533
+ where the last equation follows from (Dbio × B) ∩ iB(DB) = iB(Dbio) ⊂ φ(Abio). Part v)
1534
+ of Theorem 2.6 now follows from Eqs. (3.49), (3.54) and Lemma 3.18 below.
1535
+ Lemma 3.18. Put AB := φ−1(DB × B) = Aθ=0 ◁ GS, then
1536
+ AB ⊃ A ∩ {θ1 ≥ θ2 ∨ θ1θ2 > 0} ⊃ Asplit ⊃ Abio.
1537
+ Proof. The second and third inclusions are obvious from the definitions (2.13) and (2.15)
1538
+ and the first inclusion follows from DB = D ∩ {c2 > 4ϵ} and
1539
+ c2 − 4ϵ = (β1 − β2)2 + (θ1 + θ2)2 + 2(β1 − β2)(θ1 − θ2)
1540
+ = (β1 − β2 + θ1 − θ2)2 + 4θ1θ2.
1541
+
1542
+ Table 3. Sector classification in Abio generalizing Table 1.
1543
+ Sector
1544
+ c = B− + B+
1545
+ ϵ = B−B+
1546
+ Interval [B−, B+]
1547
+ I
1548
+ +
1549
+ +
1550
+ 0 < B− < B+
1551
+ II (SIRS)
1552
+ +
1553
+ 0
1554
+ 0 = B− < B+
1555
+ III
1556
+ +
1557
+
1558
+ 0 < −B− < B+
1559
+ IV
1560
+ 0
1561
+
1562
+ 0 < −B− = B+
1563
+ V
1564
+
1565
+
1566
+ B− < −B+ < 0
1567
+ VI
1568
+
1569
+ 0
1570
+ B− < B+ = 0
1571
+ VII
1572
+
1573
+ +
1574
+ B− < B+ < 0
1575
+
1576
+ 22
1577
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1578
+ Let me close by mentioning that the parametrizations (3.31) can now be used to generalize
1579
+ the Sector classification of Table 1 from the special case Aθ=0 to all of Abio (more generally
1580
+ to AB := φ−1(DB × B) ⊃ Abio) as shown in Table 3.
1581
+ 3.4. Examples revisited. For completeness let us revisit the examples in Section 2.4
1582
+ within the present setting. Eqs. (2.20)-(2.26) translate into14
1583
+ DHeth = Dbio ∩ {R0 = B+ ∧ a < 1 ∧ d = B− = 0}
1584
+ (3.55)
1585
+ DSIRI1,2 = Dbio ∩ {R0 = B± ∧ a < 1 ∧ d = B∓(B± + 1 − a)}
1586
+ (3.56)
1587
+ DBuDr = Dbio ∩ {R1 < R0 = B+ ∧ B− < 0}15
1588
+ (3.57)
1589
+ DSIRS = Dbio ∩ {B− = 0}
1590
+ (3.58)
1591
+ DLM = Dbio ∩ {B− < min{0, R1}}
1592
+ (3.59)
1593
+ DKZVH = Dbio ∩ {B− > 0} = DHaCa
1594
+ (3.60)
1595
+ DAABH1 = Dbio ∩ {B− ≤ R1 < B+}15
1596
+ (3.61)
1597
+ DAABH2 = Dbio ∩ {B− < R1 ≤ B+}15
1598
+ (3.62)
1599
+ Note that all models except SI(R)S already satisfy θi = 0 whence ˜β1 = B+, ˜β2 = B−
1600
+ by Eqs. (3.26)-(3.28). In the SI(R)S model we have instead 0 = ˜β2 = B− < ˜β1 ≤ B+.
1601
+ Corollary 2.9 may now be reformulated as follows
1602
+ Corollary 3.19. Referring to the sub-cases µ1 = µ2 in (Avram, Adenane, Bianchin, et al.
1603
+ 2022; Busenberg and Driessche 1990) and putting DAABH := DAABH1 ∪ DAABH2 we have
1604
+ DHeth = DSIRI1 ∩ {B− = 0}
1605
+ (3.63)
1606
+ = DSIRS ∩ {a < 1 ∧ R0 = c ∧ d = 0}
1607
+ (3.64)
1608
+ DLM ⊃ DBuDr ∩ {B− ̸= R1}
1609
+ (3.65)
1610
+ DLM = DAABH2 ∩ {B− < 0}
1611
+ (3.66)
1612
+ DKZVH = DAABH ∩ {B− > 0}
1613
+ (3.67)
1614
+ Finally, we are now in the position to generalize the scaling symmetry for SI(R)S models
1615
+ of (Nill 2022) to the present setting. First note that having started from the 10-parameter
1616
+ extended SI(R)S model we now have arrived at dim DSIRS = 4. Also, dim DHeth = 2 with
1617
+ independent parameters a ∈ (0, 1) and c = R0 = B+ > 0. In particular, if x ∈ DHeth
1618
+ then putting (u, v) := (X, cI) the RN-dynamical system (3.8)-(3.9) reduces to the classic
1619
+ endemic model in Eq. (1.1). In a second normalization step the number of parameters in
1620
+ the SI(R)S case may now be reduced again by two. In this way, for c > d,16 the normalized
1621
+ SI(R)S model also looks like the classic endemic model
1622
+ ˙u = −uv − c1u + c2 ,
1623
+ ˙v = uv − v ,
1624
+ (3.68)
1625
+ the difference being that coming from DHeth we have c1 = a ∈ (0, 1) and c2 = aR0 ≥ 0,
1626
+ whereas coming from DSIRS gives (c1, c2) ∈ R+ × R17. However, since endemic bifurcation
1627
+ 14Heth = (Hethcote 1974, 1976, 1989); SIRI = (Derrick and Driessche 1993); BuDr = (Busenberg
1628
+ and Driessche 1990); SIRS = 10-parameter mixed SIRS/SIS model with constant population size and
1629
+ θ2 = β2 = 0; HaCa = core system in (Hadeler and Castillo-Chavez 1995); KZVH = (Kribs-Zaleta and
1630
+ Velasco-Hernandez 2000); LM = (J. Li and Ma 2002); AABH = (Avram, Adenane, Bianchin, et al. 2022).
1631
+ BuDr and AABH come in two versions, the subscript 1 refers to βS > βR and 2 to βS < βR.
1632
+ 15Referring to the sub-case µ1 = µ2 in these models, see Footnotes 10 and 11.
1633
+ 16Note that in DSIRS we have c = B+ ≥ R1 − aR0 = d where equality implies R0 = 0 and R1 = B+.
1634
+ 17In case a = 0 we would get c1 = c2 = 0.
1635
+
1636
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1637
+ 23
1638
+ in the model (3.68) occurs at R0 = c2/c1 = 1, extending this model to the SI(R)S case
1639
+ by including also values c2 < 0 and c1 ≥ 1 doesn’t change its characteristic behavior.
1640
+ In particular, various proofs in the literature on variants of constant population SI(R)S
1641
+ models with standard incidence become obsolete, it’s all contained in Hethcote’s work.
1642
+ Eq. (3.68) is proven in Appendix B. In principle, the proof relies on the same structure
1643
+ as in Theorem 2.6, with the symmetry group GS acting on A replaced by a dilatation
1644
+ group Gdil = R2
1645
+ + acting on D. Since these dilatations may blow up physical triangles to
1646
+ arbitrary size, we also get the following
1647
+ Lemma 3.20. For x ∈ DSIRS the forward flow of the RN-dynamical system (3.8)-(3.9)
1648
+ stays bounded for all initial conditions (X0, I0) ∈ R × R≥0.
1649
+ This result may be used to prove, that SI(R)S models as above are always Hamiltonian
1650
+ (Nill n.d.[a]). Lemma 3.20 is also proven in Appendix B.
1651
+ 4. Summary and outlook
1652
+ In summary we have seen, that in canonical coordinates the 14-parameter SSISS model,
1653
+ constraint by ν1 = ν2, effectively depends on at most five parameters x = (a, b, c, d, ϵ). De-
1654
+ pending on natural model restrictions like “phys” or “bio” these parameters obey various
1655
+ relations which can be encoded by further reparametrizations like x = (a, R0, R1, B+, B−),
1656
+ see Eqs. (3.21), (3.22), (3.31) and Proposition 3.15. The incidence rates βi have disap-
1657
+ peared from the equations of motion. Their role is reduced to fixing physical triangles
1658
+ Tphys(β) in (X, I)-space, see Eq.
1659
+ (3.16).
1660
+ If x ∈ Dbio, then for all compatible values
1661
+ β = (β1, β2) the triangles Tphys(β) stay forward invariant under the RN-dynamics (3.8)-
1662
+ (3.9). Independence of β also means that SSISS models at parameter values φ−1(x, β)
1663
+ for fixed x ∈ D and varying β ∈ B are all isomorphic to each other18 (Proposition 3.4).
1664
+ The isomorphisms are provided by a parameter symmetry group GS ⊂ GL+(R2) acting
1665
+ simultaneously on phase space P and parameter space A (Theorem 2.6i-iv). If x ∈ DB
1666
+ then a representative in A of the equivalence class x may always be chosen by putting
1667
+ β1 = B+ and β2 = B− and hence θi = 0 (Theorem 2.6v). In combination with methods
1668
+ from (Busenberg and Driessche 1990) this also leads to a proof of absence of periodic
1669
+ solutions for all a ∈ Abio (Theorem 2.12).
1670
+ In part III of this work it will be shown, that the model also admits an additional scaling
1671
+ symmetry leading to a second normalization step, similar as described for the SI(R)S
1672
+ model in Appendix B, see also (Nill 2022). In this way the number of essential parameters
1673
+ will further reduce from five to three (respectively two in Sectors II and VI).
1674
+ Part II of this work will reanalyze equilibrium points and their stability properties in
1675
+ all Sectors of Abio, thereby recovering and extending the results of (Avram, Adenane,
1676
+ Bianchin, et al. 2022; Hadeler and Castillo-Chavez 1995; Kribs-Zaleta and Velasco-Hernandez
1677
+ 2000; J. Li and Ma 2002), which had been obtained for θi = 0 and some more parameter
1678
+ restrictions, see Table 2 and Corollary 2.9/3.19. This approach will differ from previ-
1679
+ ous papers by relying on the normalization formalism and sector classification of the
1680
+ present work. In this way the search for endemic equilibria (X∗, I∗) simplifies consider-
1681
+ ably, since always X∗ = 1. So one is left with analyzing roots of the quadratic equation
1682
+ h(I∗) := ˙X(X∗ = 1, I∗) = 0. This will also uncover an exceptional scenario in Sectors
1683
+ III-V, which apparently has been overlooked in the literature so far.
1684
+ 18By Remark 3.9, physical triangles are not mapped onto each other under these isomorphisms.
1685
+
1686
+ 24
1687
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1688
+ Appendix A. Normalizing linear vital dynamics
1689
+ This Appendix gives a normalization prescription for the dynamics of fractional vari-
1690
+ ables in an n-compartment model with linear vital dynamics. Let the vectorfield V :
1691
+ Rn → Rn be homogeneous of degree one and assume there exists ν = (ν1, · · · , νn) such
1692
+ that ⟨1|V(Y)⟩ ≡ �
1693
+ i Vi(Y) = ⟨ν|Y⟩ for all Y ∈ Rn, where 1 := (1, · · · , 1).
1694
+ Call
1695
+ N(Y) := ⟨1|Y⟩ the total population and y := N−1Y the fractional compartment vari-
1696
+ ables, then the dynamical system ˙Y = V(Y) implies
1697
+ ˙y = V(y) − ⟨ν | y⟩y =: F(y).
1698
+ Denote S := {y ∈ Rn | ⟨1|y⟩ = 1}, then clearly ⟨1|F⟩|S = 0. The aim is to substitute F
1699
+ by ˜F such that F|S = ˜F|S and ⟨1|˜F⟩ = 0 holds as an identity on all of Rn. The following
1700
+ Lemma holds by straight forward calculation.
1701
+ Lemma A.1. Put Λijk := (δij − δik)(νk − νj) and Λi(y) := �
1702
+ j,k Λijkyjyk.
1703
+ i)
1704
+ For all y ∈ Rn and i = 1, · · · , n we have
1705
+ 1
1706
+ 2Λi(y) =
1707
+
1708
+ k
1709
+ (νk − νi)yiyk ≡ yi⟨ν|y⟩ − νiyi⟨1|y⟩.
1710
+ (A.1)
1711
+ ii) Put
1712
+ ˜F := V − diag(ν) − 1
1713
+ 2Λ.
1714
+ (A.2)
1715
+ Then F|S = ˜F|S and ⟨1|˜F⟩ = 0 as an identity on Rn.
1716
+ By this method we also get conditions guaranteeing that constant per capita birth and
1717
+ death rates become redundant as in Eq. (2.7).
1718
+ Lemma A.2. Let V(Y) be of the form
1719
+ Vi(Y) =
1720
+
1721
+ j
1722
+ MijYj + 1
1723
+ 2
1724
+
1725
+ j,k
1726
+ ΓijkYjYk/N +
1727
+
1728
+ j
1729
+ LijYj
1730
+ where without loss Γijk = Γikj and where �
1731
+ i Mij = �
1732
+ i Γijk = 0. Hence, all vital dynamics
1733
+ parameters are encoded in (Lij) and νj := �
1734
+ i Lij satisfies ⟨1|V = ⟨ν|. If in this case
1735
+ Lij ̸= νiδij ⇒ Mij ̸= 0 and νj ̸= νk ⇒ (Γjjk ̸= 0 ∧ Γkkj ̸= 0), then for the dynamics of
1736
+ fractional variables all parameters Lij are redundant.
1737
+ Proof. Applying (A.2) we have ˜Fi(y) = �
1738
+ j ˜
1739
+ Mijyj+ 1
1740
+ 2 ˜Γijkyjyk, where ˜
1741
+ Mij = Mij+Lij−νiδij
1742
+ and ˜Γijk = Γijk − Λijk. The claim follows since Λijk = Λikj, Λjjk = −Λkkj and Λijk = 0 if
1743
+ νj = νk or if j ̸= i ̸= k, which also yields �
1744
+ i Λijk = 0 .
1745
+
1746
+ Appendix B. Scaling the SI(R)S model
1747
+ In this appendix we extend the dilatation symmetry as proposed for a 6-parameter
1748
+ SI(R)S model in (Nill 2022) to the 10-parameter extended SI(R)S model as classified in
1749
+ this paper. Denote Sector II in DB by DII := DB ∩ {B− = 0} and DSIRS := DII ∩ Dbio.
1750
+ Recall that in DII we have c = B+ > 0 and in DSIRS we have 0 ≤ Ri ≤ B+ and hence
1751
+ d − c = R1 − aR0 − B+ ≤ 0, where equality implies R0 = 0 and R1 = B+. Hence the
1752
+ following Lemma in particular includes Lemma 3.20.
1753
+
1754
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1755
+ 25
1756
+ Lemma B.1. Consider the RN-dynamical system (3.8) - (3.9) on phase space P ≡ R×R≥0
1757
+ for parameter values x = (a, b, c, d, ϵ = 0) ∈ DII ∩ {d ≤ c ∧ d = c ⇒ R0 < 1} ⊃ DSIRS.
1758
+ Let T ⊂ P be a rectangular triangle with corners T◁ = (X◁, 0), T▷ = (X▷, 0) and T△ =
1759
+ (X◁, I△), where X◁ < X▷. Call T compatible with x if
1760
+ I△ = (X▷ − X◁)/c
1761
+ X◁ ≤ min{R0, d/c}
1762
+ R0 − X◁ ≤ I△ min{c, (c − d)/a}
1763
+ i)
1764
+ Then every x-compatible triangle T is forward invariant.
1765
+ ii) The forward flow for arbitrary initial conditions (X0, I0) ∈ P stays bounded.
1766
+ Proof. To prove part i), the upper bounds on X◁ imply ˙X > 0 on the line {X = X◁}. We
1767
+ are left to show ˙X + c ˙I ≤ 0 on the hypotenuse X(I) = X◁ + c(I△ − I), 0 ≤ I ≤ I△.
1768
+ ˙X + c ˙I = a(R0 − X(I)) + (d − c)I
1769
+ = a(R0 − X◁ − c(I△ − I)) + (d − c)I
1770
+ ≤ I△ min{ac, c − d} − ac(I△ − I) + (d − c)I
1771
+ ≤ 0
1772
+ Part ii) follows since for d < c we may always choose X◁ < X0 and X▷ large enough,
1773
+ such T is x-compatible and (X0, I0) ∈ T . For d = c and R0 < 1 x-compatibility requires
1774
+ X◁ = R0. If in this case X0 < R0 glue the rectangle R = [X0, R0]×[0, I△] to the left of T .
1775
+ Then (X0, I0) ∈ R ∪ T for X▷ large enough and R ∪ T is forward invariant, since ˙I < 0
1776
+ and ˙X > 0 for (X, I) ∈ R.
1777
+
1778
+ Given x ∈ DII ∩ {d ≤ c ∧ d = c ⇒ R0 < 1} as above and T compatible with x
1779
+ we now show that the RN-dynamical system (3.8) - (3.9) may always be rescaled to an
1780
+ isomorphic system with parameters x′ ∈ DSIRS such that T maps to the physical triangle
1781
+ Tphys(B′
1782
+ +, 0) of the SI(R)S system. Following (Nill 2022) the dilatation symmetry group
1783
+ Gdil ≡ GX × GI ≡ R2
1784
+ + is defined by rescaling (X, I) variables according to
1785
+ X(ξ,λ)(t) − 1 := ξ(X(ξt) − 1),
1786
+ I(ξ,λ)(t) := λI(ξt),
1787
+ (ξ, λ) ∈ R2
1788
+ +
1789
+ The following Lemma is easily verified by straightforward calculation.
1790
+ Lemma B.2. Let the group action ▷ : Gdil × D ∋ (ξ, λ, x) �→ (ξ, λ) ▷ x ∈ D be given by
1791
+ (ξ, λ) ▷ (a, R0 − 1, c, d − c, ϵ) := (ξa, ξ(R0 − 1), ξc/λ, ξ2(d − c)/λ, ξ2ϵ/λ2)
1792
+ (B.1)
1793
+ and for x ∈ D let fx(X, I) denote the vector field of the system (3.8) - (3.9). Then
1794
+ ( ˙X, ˙I) = fx(X, I) ⇐⇒ ( ˙X(ξ,λ), ˙I(ξ,λ)) = fx′(X(ξ,λ), I(ξ,λ)),
1795
+ x′ = (ξ, λ) ▷ x.
1796
+
1797
+ Note that this action leaves all Sectors in DB invariant, but in general not Dbio ⊂ DB. We
1798
+ now determine Gdil ▷ DSIRS, thereby also providing an alternative proof of Lemma B.1i).
1799
+ Proposition B.3.
1800
+ i)
1801
+ Let T be compatible with x ∈ DII ∩ {d ≤ c ∧ d = c ⇒ R0 < 1} in the sense of Lemma
1802
+ B.1.
1803
+ Then there exists a unique dilatation transformation (ξ, λ) ∈ Gdil such that
1804
+ x′ := (ξ, λ) ▷ x ∈ Dbio and such that the rescaled triangle satisfies T(ξ,λ) = Tphys(B′
1805
+ +, 0).
1806
+ ii) Gdil ▷ DSIRS = DII ∩ {d ≤ c ∧ d = c ⇒ R0 < 1}.
1807
+
1808
+ 26
1809
+ SYMMETRIES AND NORMALIZATION IN 3-COMPARTMENT EPIDEMIC MODELS I
1810
+ Proof. To prove part i) denote transformed quantities by a prime.
1811
+ The requirements
1812
+ T′
1813
+ ◁ = (0, 0) and T′
1814
+ △ = (0, 1) fix ξ = (1 − X◁)−1 and λ = I−1
1815
+ △ .
1816
+ Hence X▷ maps to
1817
+ ξcI△ = c′ = B′
1818
+ + and therefore T(ξ,λ) = Tphys(B′
1819
+ +, 0). To show 0 ≤ R′
1820
+ i ≤ B′
1821
+ + use R′
1822
+ 0 =
1823
+ ξ(R0 − 1) + 1 = ξ(R0 − X◁) and therefore
1824
+ 0 ≤ R′
1825
+ 0 ≤ ξ
1826
+ λ min{c, (c − d)/a} = min{c′, (c′ − d′)/a′} ≤ B′
1827
+ +
1828
+ By the above we also have R′
1829
+ 1 = a′R′
1830
+ 0 + d′ ≤ c′ = B′
1831
+ + and we are left to show R′
1832
+ 1 ≥ 0.
1833
+ Sufficient is d′ ≥ 0 which follows from 1−d′/c′ = ξ(1−d/c) ≤ ξ(1−X◁) = 1. This proves
1834
+ part i) and therefore also the “⊃”-direction of part ii). To prove the “⊂”-direction use that
1835
+ the action of Gdil on D preserves the sign of d − c and in case d = c we have R0 = 0 and
1836
+ therefore R′
1837
+ 0 = ξ(R0 − 1) + 1 = 1 − ξ < 1.
1838
+
1839
+ As in (Nill 2022), the above dilatation symmetry leads to a second normalization step
1840
+ for the SIRS-Sector, thus further reducing its number of essential parameters from four
1841
+ to two. Equivalently this means, that equivalence classes of Gdil-isomorphic systems with
1842
+ parameters in Gdil ▷ DSIRS are naturally parametrized by KSIRS := (Gdil ▷ DSIRS)/Gdil. A
1843
+ convenient realization of the normalized system on phase space P = {(q, p) ∈ R × R≥0} is
1844
+ given by putting
1845
+ q(t) := 1
1846
+ a(X(t/a) − 1) ,
1847
+ p(t) := c
1848
+ aI(t/a)
1849
+ (B.2)
1850
+ In terms of these variables the RN-dynamical system (3.8) - (3.9) becomes
1851
+ ˙q = −q(p + 1) + κ0 − κ1p ,
1852
+ ˙p = qp ,
1853
+ (B.3)
1854
+ where the new Gdil-invariant parameters are given by
1855
+ κ0 := R0 − 1
1856
+ a
1857
+ ,
1858
+ κ1 := c − d
1859
+ ac
1860
+ .
1861
+ (B.4)
1862
+ The only remaining constraint on the reduced parameter space says
1863
+ KSIRS = {(κ0, κ1) ∈ R × R≥0 | κ1 = 0 ⇒ κ0 < 0} .
1864
+ (B.5)
1865
+ Thus, after normalization the whole SIRS Sector just looks like Hethcote’s classic endemic
1866
+ model except for a somewhat less restricted parameter space. In fact, by Eq. (3.55),
1867
+ DHeth ⊂ DSIRS is already two-dimensional with independent parameters a ∈ (0, 1) and
1868
+ c = R0 = B+ > 0. These map injectively to KSIRS via κ0 = (c − 1)/a and κ1 = 1/a,
1869
+ whence
1870
+ DHeth ∼= KHeth = KSIRS ∩ {κ1 > 1 ∧ κ0 + κ1 > 0}
1871
+ (B.6)
1872
+ The normalization convention in Eq. (3.68) is obtained under the restriction c > d or
1873
+ equivalently κ1 > 0. In this case one may alternatively use
1874
+ u(t) − 1 :=
1875
+ c
1876
+ c − d(X(ct/(c − d)) − 1) = 1
1877
+ κ1
1878
+ q(t/κ1) ,
1879
+ (B.7)
1880
+ v(t) :=
1881
+ c2
1882
+ c − dI(ct/(c − d)) = 1
1883
+ κ1
1884
+ p(t/κ1) .
1885
+ (B.8)
1886
+ In terms of these variables we recover the normalization convention (1.1), (3.68)
1887
+ ˙u = −uv − c1u + c2 ,
1888
+ ˙v = uv − v ,
1889
+ (B.9)
1890
+ where c1 = 1/κ1 and c2 = 1/κ1 + κ0/κ2
1891
+ 1, which is also the version given in (Nill 2022).
1892
+ In part III of this work the above normalization step will be generalized to all Sectors of
1893
+ Dbio. In this way the equation for ˙q in (B.3) gets an additional term −κ2p2, and so our
1894
+
1895
+ REFERENCES
1896
+ 27
1897
+ initial 14-parameter19 SSISS model boils down to a much simpler 3-parameter dynamical
1898
+ system.
1899
+ Appendix C. The case α1 = α2 = 0
1900
+ This Appendix shortly discusses the border case α1 = α2 = 020. In this case define
1901
+ parameter spaces C0
1902
+ x as in Eqs. (2.11)-(2.15) with αi = 0 and A0
1903
+ x := C0
1904
+ x ×B. In particular,
1905
+ in A0
1906
+ bio we have θ1 ≥ 0, θ2 = 0, γi ≥ 0 and γ1 + γ2 = 1. Lemma 3.3 still holds with
1907
+ a = b = 0 and d = R1 + ϵ, i.e. the replacement number dynamics becomes
1908
+ ˙X = (d − cX)I − ϵI2 ,
1909
+ ˙I = (X − 1)I .
1910
+ (C.1)
1911
+ In this case R0 is undefined and there is a continuum of disease free equilibria at I = 0,
1912
+ which are locally stable for X < 1 and unstable for X > 1. Proposition 3.4 remains
1913
+ unchanged provided a = a′ ∈ A0. Putting D0 = {(c, d, ϵ) ∈ R3} Lemma 3.6 still holds
1914
+ with A replaced by A0 and D replaced by D0. Moreover, in A0
1915
+ bio we get A+ = B+ = β1+θ1,
1916
+ A− = B− = β2, c = β1+β2+θ1, ϵ = β2(β1+θ1) and putting D0
1917
+ A = D0
1918
+ B = D0
1919
+ AB := D0∩{c2 >
1920
+ 4ϵ} Proposition 3.15 becomes
1921
+ φ(A0
1922
+ phys) = (D0
1923
+ B × B) ∩ {β2 ≤ {B−, R1} ≤ β1 ≤ B+} ,
1924
+ (C.2)
1925
+ D0
1926
+ phys = D0
1927
+ B ∩ {R1 ≤ B+} ,
1928
+ (C.3)
1929
+ φ(A0
1930
+ bio) = (D0
1931
+ B × B) ∩ {B− = β2 ≤ R1 ≤ β1 ≤ B+} ,
1932
+ (C.4)
1933
+ D0
1934
+ bio = D0
1935
+ B ∩ {B− ≤ R1 ≤ B+} ⊂ D0
1936
+ phys .
1937
+ (C.5)
1938
+ So, for x ∈ D0
1939
+ phys physical triangles Tphys(β1, β2) are forward invariant provided (β1, β2)
1940
+ satisfy the bounds C.2. Finally, Eq. (3.42) becomes φ−1(iB(D0
1941
+ B)) = φ(Aθ=0 ∩ Aα=0) and
1942
+ Theorem 2.12, Theorem 2.6 and Corollary 3.17 stay valid also for α = 0.
1943
+ References
1944
+ Arino, J., C.C. Mccluskey, and P. van den Driessche (2003). “Global results for an epidemic
1945
+ model with vaccination that exhibits backwad bifurcation.” In: SIAM J. Appl. Math.
1946
+ 64, pp. 260–276. doi: 10.1137/S0036139902413829.
1947
+ Avram, F., R. Adenane, L. Basnarkov, et al. (Dec. 2021). “On matrix-SIR Arino models
1948
+ with linear birth rate, loss of immunity, disease and vaccination fatalities, and their
1949
+ approximations.” In: arXiv preprint. url: http://arxiv.org/abs/2112.03436.
1950
+ Avram, F., R. Adenane, G. Bianchin, et al. (2022). “Stability analysis of an eight parameter
1951
+ SIR- type model including loss of immunity, and disease and vaccination fatalities.” In:
1952
+ Mathematics 10.3, p. 402. doi: 10.3390/math10030402.
1953
+ Batistela, C.M. et al. (2021). “Vaccination and social distance to prevent Covid-19.” In:
1954
+ IFAC PapersOnLine 54-15, pp. 151–156.
1955
+ Busenberg, S. N. and P. van den Driessche (1990). “Analysis of a disease transmission
1956
+ model in a population with varying size.” In: J. Math. Biol. 28, pp. 257–270.
1957
+ Busenberg, S. N. and P. van den Driessche (1991). “Nonexistence of periodic solutions
1958
+ for a class of epidemiological models.” In: Biology, Epidemiology, and Ecology. Ed. by
1959
+ S.N. Busenberg and M. Martelli. Vol. 92. Lecture Notes in Biomath. Berlin Heidelberg
1960
+ New York: Springer, pp. 70–79.
1961
+ 19i.e. constraint by ν1 = ν2.
1962
+ 20Here, for simplicity of notation, the tilde is still omitted.
1963
+ So beware that truly this appendix
1964
+ addresses the cases αi = νi = νI = 0 (constant population (2.2)) or αi = δi = 0 and µ1 = µ2 (time
1965
+ varying population (2.7)).
1966
+
1967
+ 28
1968
+ REFERENCES
1969
+ Chauhan, S., O.P. Misra, and J. Dhar (2014). “Stability Analysis of Sir Model with Vac-
1970
+ cination.” In: American J. Comp. Appl. Math. 2014.4(1), pp. 17–23. doi: 10.5923/j.
1971
+ ajcam.20140401.03.
1972
+ Derrick, W.R. and P. van den Driessche (1993). “A disease transmission model in a non-
1973
+ constant population.” In: J Math Biol 31.5, pp. 495–512. doi: 10.1007/BF00173889.
1974
+ Diagne, M.L. et al. (2021). “A Mathematical Model of COVID-19 with Vaccination and
1975
+ Treatment.” In: Computational and Mathematical Methods in Medicine 2021, p. 1250129.
1976
+ doi: 10.1155/2021/1250129.
1977
+ Diekmann, O. and J.A.P. Heesterbeek (2000). Mathematical epidemiology of in-fectious
1978
+ diseases. Wiley series in mathematical and computational biology. West Sussex, Eng-
1979
+ land: John Wiley & Sons.
1980
+ Diekmann, O., J.A.P. Heesterbeek, and J.A.J. Metz (1990). “On the definition and the
1981
+ computation of the basic reproduction ratio R0 in models for infectious diseases in
1982
+ heterogeneous populations.” In: J. Math. Biol. 28, p. 365.
1983
+ Driessche, P. van den and J. Watmough (2002). “Reproduction numbers and sub-threshold
1984
+ endemic equilibria for compartmental models of disease transmission.” In: Math.Biosci.
1985
+ 180, pp. 29–48.
1986
+ Driessche, P. van den and J. Watmough (2008). “Further notes on the basic reproduction
1987
+ number.” In: Mathematical Epidemiology. Ed. by F. Bauer, P. van den Driessche, and
1988
+ J. Wu. Vol. 1945. Lecture Notes in Mathematics, pp. 159–178. isbn: 978-3-540-78910-9.
1989
+ doi: 10.1007/978-3-540-78911-6.
1990
+ Greenhalgh, D. (1997). “Hopf bifurcation in epidemic models with a latent period and
1991
+ nonpermanent immunity.” In: Mathematical and Computer Modelling 25.2, pp. 85–107.
1992
+ Hadeler, K. P. and C. Castillo-Chavez (1995). “A Core Group Model for Disease Trans-
1993
+ mission.” In: Math.Biosci. 128, pp. 41–55. url: https://www.researchgate.net/
1994
+ publication/216242267.
1995
+ Hadeler, K. P. and P. van den Driessche (1997). “Backward Bifurcation in epidemic Con-
1996
+ trol.” In: Math.Biosci. 146, pp. 15–35.
1997
+ Hethcote, H.W. (1974). “Asymptotic behavior and stability in epidemic models.” In: Math-
1998
+ ematical Problems in Biology. Victoria Conference 1973. Ed. by P. van den Driessche.
1999
+ Vol. 2. Lecture Notes in Biomathematics. Springer Verlag, pp. 83–92. doi: 10.1007/
2000
+ 978-3-642-45455-4_10.
2001
+ Hethcote, H.W. (1976). “Qualitative analysis for communicable disease models.” In: Math.
2002
+ Biosci. 28, pp. 335–356.
2003
+ Hethcote, H.W. (1989). “Three basic epidemiological models.” In: Applied Mathematical
2004
+ Ecology. Ed. by L. Gross, T.G. Hallam, and S.A. Levin. Vol. 18. Biomathematics. Berlin
2005
+ Heidelberg New York: Springer Verlag, pp. 119–144.
2006
+ Hethcote, H.W. (2000). “The Mathematics of Infectious Diseases.” In: SIAM Rev. 42,
2007
+ p. 599.
2008
+ Kermack, W. O. and A. G. McKendrick (1927). “Contribution to the mathematical theory
2009
+ of epidemics, part I.” In: Proc. Roy. Soc. Lond A 115, pp. 700–721.
2010
+ Korobeinikov, A. and G.C. Wake (2002). “Lyapunov Functions and Global Stability for
2011
+ SIR, SIRS, and SIS Epidemiological Models.” In: Appl. Math. Lett. 15, pp. 955–960.
2012
+ Kribs-Zaleta, C.M. and J.X. Velasco-Hernandez (2000). “A simple vaccination model with
2013
+ multiple endemic states.” In: Mathematical Biosciences 164, pp. 183–201. doi: 10.
2014
+ 1007/s00285-021-01629-8.
2015
+
2016
+ REFERENCES
2017
+ 29
2018
+ Li, Jianquan and Zhien Ma (2002). “Qualitative analyses of SIS epidemic model with vac-
2019
+ cination and varying total population size.” In: Mathematical and Computer Modelling
2020
+ 35, pp. 1235–1243.
2021
+ Li, Michael Y et al. (1999). “Global dynamics of a SEIR model with varying total popu-
2022
+ lation size.” In: Mathematical biosciences 160.2, pp. 191–21325.
2023
+ Lu, Guichen and Zhengyi Lu (2018). “Global asymptotic stability for the seirs models
2024
+ with varying total population size.” In: Mathematical biosciences 296, pp. 17–25.
2025
+ Mena-Lorca, J. and H.W. Hethcote (1992). “Dynamic models of infectious diseases as
2026
+ regulators of population sizes.” In: J. Math. Biol 30, pp. 693–716.
2027
+ Nadim, S.S. and J. Chattopadhyay (2020). “Occurrence of backward bifurcation and pre-
2028
+ diction of disease transmission with imperfect lockdown: A case study on COVID-19.”
2029
+ In: Chaos, Solitons and Fractals 140, p. 110163.
2030
+ Nill, Florian (Mar. 13, 2022). “Endemic oscillations for SARS-COV-2 Omicron - A SIRS
2031
+ model analysis.” In: doi: https://doi.org/10.48550/arXiv.2211.09005. eprint:
2032
+ arXiv:2211.09005[q-bio.PE].
2033
+ Nill, Florian (n.d.[a]). “Hamiltonian structure in epidemic SIRS models.” paper to be
2034
+ written up.
2035
+ Nill, Florian (n.d.[b]). “Symmetries and normalization in 3-compartment epidemic models.
2036
+ II: Equilibria and stability.” paper to be written up.
2037
+ Nill, Florian (n.d.[c]). “Symmetries and normalization in 3-compartment epidemic models.
2038
+ III: The dilatation symmetry.” paper to be written up.
2039
+ O’Regan, S.M. et al. (2010). “Lyapunov functions for SIR and SIRS epidemic models.”
2040
+ In: Applied Mathematics Letters 23, pp. 446–448. doi: 10.1016/j.aml.2009.11.014.
2041
+ Razvan, M.R. (2001). “Multiple Equilibria for an SIRS Epidemiological System.” In: arXiv
2042
+ preprint arXiv:math/0101051v1. doi: 10.48550/arXiv.math/0101051.
2043
+ Sun, Chengjun and Ying-Hen Hsieh (2010). “Global analysis of an SEIR model with vary-
2044
+ ing population size and vaccination.” In: Applied Mathematical Modelling 34, pp. 2685–
2045
+ 2697.
2046
+ Yang, Wei, Chengjun Sun, and Julien Arino (2010). “Global analysis for a general epidemi-
2047
+ ological model with vaccination and varying population.” In: Journal of Mathematical
2048
+ Analysis and Applications 372.1, pp. 208–223.
2049
+
WtAyT4oBgHgl3EQfWPdu/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
XtFJT4oBgHgl3EQf5y0l/content/tmp_files/2301.11671v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
XtFJT4oBgHgl3EQf5y0l/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
_NE1T4oBgHgl3EQfUwPr/content/2301.03095v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0bf4634f16876e894f65fb62f424b6111195673005af2726eaf3d571531c2889
3
+ size 1123336
b9AyT4oBgHgl3EQfXPer/content/tmp_files/2301.00180v1.pdf.txt ADDED
@@ -0,0 +1,700 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.00180v1 [math.DG] 31 Dec 2022
2
+ A blow-up formula for stationary quaternionic maps ∗†
3
+ Jiayu Li‡
4
+ Chaona Zhu§
5
+ Abstract
6
+ Let (M, Jα, α = 1, 2, 3) and (N, J α, α = 1, 2, 3) be Hyperk¨ahler manifolds. Suppose
7
+ that uk is a sequence of stationary quaternionic maps and converges weakly to u in
8
+ H1,2(M, N), we derive a blow-up formula for limk→∞ d(u∗
9
+ kJ α), for α = 1, 2, 3, in the
10
+ weak sense. As a corollary, we show that the maps constructed by Chen-Li [CL2] and by
11
+ Foscolo [F] can not be tangent maps (c.f [LT], Theorem 3.1) of a stationary quaternionic
12
+ map satisfing d(u∗J α) = 0.
13
+ 1
14
+ Introduction and the main result
15
+ A hyperk¨ahler manifold is a Riemannian manifold (M, g) with three parallel com-
16
+ plex structures {J1, J2, J3} compatible with the metric g such that (J1)2 = (J2)2 =
17
+ (J3)2 = J1J2J3 = −id. The simplest hyperk¨ahler manifold is the Euclidean space
18
+ R4m. It is well-known that the only compact hyperk¨ahler manifolds of dimension 4 are
19
+ K3 surfaces and complex tori. Let (M, g, Jα, α = 1, 2, 3) and (N, h, J α, α = 1, 2, 3) be
20
+ hyperk¨ahler manifolds. Let ωα(·, ·) = g(·, Jα·) and Ωα(·, ·) = h(·, J α·), (α = 1, 2, 3)
21
+ be the K¨ahler forms on M and N respectively. A smooth map u : M → N is called a
22
+ quaternionic map (triholomorphic map) if
23
+ AαβJ β ◦ du ◦ Jα = du
24
+ (1)
25
+ where Aαβ denote the entries of a matrix A in SO(3). For simplicity, we choose
26
+ Aαβ = δαβ.
27
+ The quaternionic maps (triholomorphic maps) between Hyperk¨ahler manifolds
28
+ has been studied by many aothors (cf. [BT], [Ch], [CL1, [CL2], [FKS], [W]). Quater-
29
+ nionic maps automatically minimize the energy functional in their homotopy classes
30
+ (cf. [Ch], [CL1] and [FKS]) and hence they are harmonic. It can be verified that
31
+ holomorphic and anti-holomorphic maps with respect to some complex structures on
32
+ M and N are quaternionic maps. However, Chen-Li constructed quaternionic maps
33
+ which are not holomorphic with respect to any complex structures on M and N (cf.
34
+ [CL1]).
35
+ ∗This work is supported by NSF grant 11721101.
36
+ †MSC (2000): 53C26, 53C43, 58E12, 58E20. Keywords: Stationary harmonic maps, quaternionic maps,
37
+ blow-up formula.
38
39
40
+ 1
41
+
42
+ Definition 1.1 A map u from M to N is called a stationary quaternionic map if it
43
+ is a stationary harmonic map and it is a quaternionic map outside its singular set.
44
+ It is clear that (c.f. [BT]), if u satisfies (1) almost everywhere, and
45
+ d(u∗J α) = 0,
46
+ for α = 1, 2, 3,
47
+ (2)
48
+ then u is a stationary quaternionic map.
49
+ Chen-Li ([CL2]) proved that, if there is a harmonic sphere φ : S2 → N which
50
+ satisfies
51
+ dφ JS2 = −
52
+ 3
53
+
54
+ k=1
55
+ akJ k dφ,
56
+ (3)
57
+ where ⃗a = (a1, a2, a3) : S2 → S2, and
58
+
59
+ S2 xi|∇φ|2dσ = 0, i = 1, 2, 3, (x1, x2, x3) ∈ S2,
60
+ (4)
61
+ then
62
+ u(x, x4) = φ( x
63
+ |x|) for any x ∈ R3\{0}
64
+ is a stationary quaternionic map with the x4-axis as its singular set.
65
+ Chen-Li ([CL2]) showed that there does exist a complete noncompact hyperk¨ahler
66
+ manifold, into which there is a harmonic S2 which satisfies (3) and (4). Recently,
67
+ Foscolo [F] showed that there exists a compact K3 surface with the above property.
68
+ However, the map u constructed by Chen-Li or by Foscolo does not satisfy (2). Now
69
+ the question is whether the maps constructed by Chen-Li or by Foscolo could be a
70
+ tangent map of a stationary quaternionic map with identity (2), if not the singular
71
+ set of a stationary quaternionic map with identity (2) might be of codimensional 4
72
+ (Remark 1.2 in [BT]).
73
+ Suppose that uk is a sequence of stationary quaternionic maps with bounded
74
+ energies E(uk) ≤ Λ. The blow-up set of uk can be defined as
75
+ Σ = ∩r>0{x ∈ M| lim inf
76
+ k→∞ r2−m
77
+
78
+ Br(x)
79
+ | ▽ uk|2dy ≥ ǫ0}.
80
+ We can always assume that uk ⇀ u weakly in W 1,2(M, N) and that
81
+ | ▽ uk|2dx ⇀ | ▽ u|2dx + ν
82
+ in the sense of measure as k → ∞. Here ν is a nonnegative Radon measure on M
83
+ with support in Σ. It is known that Σ is a Hm−2-rectifiable set, and we may write
84
+ ν = θ(x)Hm−2⌊Σ. It is clear that strongly convergence in H1,2(M, N) preserves the
85
+ identity (2). In this paper we mainly prove the following blow-up formula for weakly
86
+ convergence sequence of stationary quaternionic maps.
87
+ 2
88
+
89
+ Theorem 1.2 Let uk be a sequence of stationary quaternionic map with E(uk) ≤ Λ.
90
+ Assume that uk → u weakly in H1(M, N). Then there exist
91
+ (a1, a2, a3) ∈ R3 with
92
+ �3
93
+ α=1(aα)2 = 1 such that, for any smooth (m − 3)-form η with compact support in
94
+ M,
95
+ lim
96
+ k→∞
97
+ 3
98
+
99
+ α=1
100
+
101
+
102
+ M
103
+ dη ∧ u∗
104
+ kJ α =
105
+ 3
106
+
107
+ α=1
108
+
109
+
110
+ M
111
+ dη ∧ u∗J α +
112
+
113
+ Σ
114
+ θdη|Σ
115
+ (5)
116
+ and for any (b1, b2, b3) ⊥ (a1, a2, a3), there holds
117
+ lim
118
+ k→∞
119
+ 3
120
+
121
+ α=1
122
+
123
+
124
+ M
125
+ dη ∧ u∗
126
+ kJ α =
127
+ 3
128
+
129
+ α=1
130
+
131
+
132
+ M
133
+ dη ∧ u∗J α.
134
+ As a corollary of the theorem, the maps constructed by Chen-Li [CL2] and by Fos-
135
+ colo [F] can not be tangent maps (c.f [LT], Theorem 3.1) of a stationary quaternionic
136
+ map satisfing d(u∗J α) = 0.
137
+ 2
138
+ The proof of the blow-up formula
139
+ If u is a strong limit of a sequence of stationary quaternionic maps in H1,2(M, N),
140
+ then it’s easy to see that u satisfies (2). If u is just a weak limit, i.e. there exists a
141
+ sequence of stationary quaternionic maps uk satisfying uk → u weakly in H1,2(M, N)
142
+ and |∇uk|2dV → |∇u|2dV +θHm−2|Σ in the sense of measure, we prove in this section
143
+ a formula for the blow-up set θHm−2|Σ and the limiting map u.
144
+ Without loss of generality, we may assume that m = 4. Because Σ is a Hm−2-
145
+ rectifiable set, so we may assume that Σ = ∪∞
146
+ i=0Σi, Σi ∩Σi′ = φ if i ̸= i′, Hm−2(Σ0) =
147
+ 0, Σi ⊂ Ni and Ni (i = 1, 2, · · ·) is an (m − 2)-dimensional embedded C1 submanifold
148
+ of M. It is important that (see p. 61 in [Si]) TxΣ = TxNi for Hm−2-a.e. x ∈ Σi.
149
+ It is known that ν = θ(x)Hm−2⌊Σ, where θ(x) is upper semi-continuous with
150
+ ǫ0 ≤ θ(x) ≤ C(l1) for Hm−2-a.e. x ∈ Σ, C(l1) is a positive constant depending only
151
+ on M and l1 (cf. [Lin], Lemma 1.6). Since Hm−2(Σ) < +∞, for any 1. > 0, there
152
+ exist Σ1. ⊂ Σ and i0 such that Hm−2(Σ1. ) < 1., Σc
153
+ 1. = Σ\Σ1. = ∪i0
154
+ i=1Σ1.
155
+ i where Σ1.
156
+ i ⊂ Σi
157
+ (i = 1, · · ·, i0) is a bounded closed set. We choose a covering {Brn|n = 1, 2, · · ·} of
158
+ Σ1. such that �
159
+ n rm−2
160
+ n
161
+ < C1.. Here and in the sequel, C always denotes a uniform
162
+ constant depending only on M and N.
163
+ Suppose that (x1, ..., x4) is a local normal coordinate system in Bǫ(Σδ
164
+ i), and that
165
+ (x3, x4) is the corresponding coordinate system in Σi, and the matrix expressions of
166
+ the complex structures are given by (6), (7) and (8).
167
+ J1 =
168
+
169
+
170
+
171
+
172
+ 0
173
+ 0
174
+ 0
175
+ −1
176
+ 0
177
+ 0
178
+ 1
179
+ 0
180
+ 0
181
+ −1
182
+ 0
183
+ 0
184
+ 1
185
+ 0
186
+ 0
187
+ 0
188
+
189
+
190
+
191
+  ,
192
+ A1βJ β =
193
+
194
+
195
+
196
+
197
+ J1
198
+ ·
199
+ ·
200
+ J1
201
+
202
+
203
+
204
+
205
+ (6)
206
+ 3
207
+
208
+ J2 =
209
+
210
+
211
+
212
+
213
+ 0
214
+ −1
215
+ 0
216
+ 0
217
+ 1
218
+ 0
219
+ 0
220
+ 0
221
+ 0
222
+ 0
223
+ 0
224
+ 1
225
+ 0
226
+ 0
227
+ −1
228
+ 0
229
+
230
+
231
+
232
+  ,
233
+ A2βJ β =
234
+
235
+
236
+
237
+
238
+ J2
239
+ ·
240
+ ·
241
+ J2
242
+
243
+
244
+
245
+
246
+ (7)
247
+ J3 =
248
+
249
+
250
+
251
+
252
+ 0
253
+ 0
254
+ 1
255
+ 0
256
+ 0
257
+ 0
258
+ 0
259
+ 1
260
+ −1
261
+ 0
262
+ 0
263
+ 0
264
+ 0
265
+ −1
266
+ 0
267
+ 0
268
+
269
+
270
+
271
+  ,
272
+ A3βJ β =
273
+
274
+
275
+
276
+
277
+ J3
278
+ ·
279
+ ·
280
+ J3
281
+
282
+
283
+
284
+
285
+ (8)
286
+ where AαβJ β are 4n×4n-matrices, Aαβ are the entries of a matrix A in SO(3). Then
287
+ the quaternionic equation is
288
+
289
+
290
+
291
+
292
+
293
+
294
+
295
+
296
+
297
+
298
+
299
+
300
+
301
+
302
+
303
+
304
+
305
+
306
+
307
+
308
+
309
+
310
+
311
+
312
+
313
+
314
+
315
+
316
+
317
+
318
+
319
+
320
+
321
+
322
+
323
+
324
+
325
+
326
+
327
+
328
+
329
+ u1
330
+ 1 + u2
331
+ 2 + u3
332
+ 3 + u4
333
+ 4
334
+ =
335
+ 0
336
+ u2
337
+ 1 − u1
338
+ 2 + u4
339
+ 3 − u3
340
+ 4
341
+ =
342
+ 0
343
+ u3
344
+ 1 − u1
345
+ 3 − u4
346
+ 2 + u2
347
+ 4
348
+ =
349
+ 0
350
+ u4
351
+ 1 − u1
352
+ 4 − u2
353
+ 3 + u3
354
+ 2
355
+ =
356
+ 0
357
+ u5
358
+ 1 + u6
359
+ 2 + u7
360
+ 3 + u8
361
+ 4
362
+ =
363
+ 0
364
+ u6
365
+ 1 − u5
366
+ 2 + u8
367
+ 3 − u7
368
+ 4
369
+ =
370
+ 0
371
+ u7
372
+ 1 − u5
373
+ 3 − u8
374
+ 2 + u6
375
+ 4
376
+ =
377
+ 0
378
+ u8
379
+ 1 − u5
380
+ 4 − u6
381
+ 3 + u7
382
+ 2
383
+ =
384
+ 0
385
+ · · ·
386
+ u4n−3
387
+ 1
388
+ + u4n−2
389
+ 2
390
+ + u4n−1
391
+ 3
392
+ + u4n
393
+ 4
394
+ =
395
+ 0
396
+ u4n−2
397
+ 1
398
+ − u4n−3
399
+ 2
400
+ + u4n
401
+ 3 − u4n−1
402
+ 4
403
+ =
404
+ 0
405
+ u4n−1
406
+ 1
407
+ − u4n−3
408
+ 3
409
+ − u4n
410
+ 2 + u4n−2
411
+ 4
412
+ =
413
+ 0
414
+ u4n
415
+ 1 − u4n−3
416
+ 4
417
+ − u4n−2
418
+ 3
419
+ + u4n−1
420
+ 2
421
+ =
422
+ 0.
423
+ (9)
424
+ Theorem 2.1 For any smooth (m − 3)-form η with compact support in M, we have
425
+ lim
426
+ k→∞
427
+ 3
428
+
429
+ α=1
430
+ Aαβ
431
+
432
+ M
433
+ dη ∧ u∗
434
+ kJ β =
435
+ 3
436
+
437
+ α=1
438
+ Aαβ
439
+
440
+ M
441
+ dη ∧ u∗J β +
442
+
443
+ Σ
444
+ θdη|Σ
445
+ and
446
+ lim
447
+ k→∞ A1β
448
+
449
+ M
450
+ dη ∧ u∗
451
+ kJ β = A1β
452
+
453
+ M
454
+ dη ∧ u∗J β,
455
+ lim
456
+ k→∞ A3β
457
+
458
+ M
459
+ dη ∧ u∗
460
+ kJ β = A3β
461
+
462
+ M
463
+ dη ∧ u∗J β,
464
+ Proof. Assume that η = �
465
+ I ηIdxI. We have
466
+ lim
467
+ k→∞
468
+
469
+ M
470
+ dη ∧ u∗
471
+ k(AαβJ β) =
472
+
473
+ M
474
+ dη ∧ u∗(AαβJ β)
475
+ +
476
+ lim
477
+ δ→0 lim
478
+ ǫ→0 lim
479
+ k→∞
480
+
481
+ Bǫ(∪i0
482
+ i=1Σδ
483
+ i )
484
+ dη ∧ u∗
485
+ k(AαβJ β)
486
+ +
487
+ lim
488
+ δ→0 lim
489
+ ǫ→0 lim
490
+ k→∞
491
+
492
+ ∪nBrn\Bǫ(∪i0
493
+ i=1Σδ
494
+ i )
495
+ dη ∧ u∗
496
+ k(AαβJ β).
497
+ (10)
498
+ 4
499
+
500
+ It’s easy to see that
501
+ lim
502
+ δ→0 lim
503
+ ǫ→0 lim
504
+ k→∞
505
+
506
+ ∪nBrn
507
+ dη ∧ u∗
508
+ k(J β) = 0
509
+ (11)
510
+ By Lemma 2.2 in [LT], we get
511
+ lim
512
+ δ→0 lim
513
+ ǫ→0 lim
514
+ k→∞
515
+
516
+ Bǫ(Σδ
517
+ i )
518
+ dη ∧ u∗
519
+ k(AαβJ β)
520
+ =
521
+ lim
522
+ δ→0 lim
523
+ ǫ→0 lim
524
+ k→∞
525
+
526
+ Bǫ(Σδ
527
+ i )
528
+ 2∂ηI
529
+ ∂xl
530
+ ∂uσ
531
+ k
532
+ ∂x1 (AαβJ β)σγ
533
+ ∂uγ
534
+ k
535
+ ∂x2 dxl ∧ dxI ∧ dx1 ∧ dx2
536
+ (12)
537
+ Substituting (9) to (12) and applying Lemma 2.2 in [LT], we have
538
+ lim
539
+ δ→0 lim
540
+ ǫ→0 lim
541
+ k→∞
542
+
543
+ Bǫ(Σδ
544
+ i )
545
+ dη ∧ u∗
546
+ k(A1βJ β) = lim
547
+ δ→0 lim
548
+ ǫ→0 lim
549
+ k→∞
550
+
551
+ Bǫ(Σδ
552
+ i )
553
+ dη ∧ u∗
554
+ k(A3βJ β) = 0
555
+ and
556
+ lim
557
+ δ→0 lim
558
+ ǫ→0 lim
559
+ k→∞
560
+
561
+ Bǫ(Σδ
562
+ i )
563
+ dη ∧ u∗
564
+ k(A2βJ β) = lim
565
+ δ→0 lim
566
+ ǫ→0 lim
567
+ k→∞
568
+
569
+ Bǫ(Σδ
570
+ i )
571
+ |∇uk|2dη ∧ dx1 ∧ dx2
572
+ =
573
+ lim
574
+ δ→0 lim
575
+ ǫ→0(
576
+
577
+ Bǫ(Σδ
578
+ i )
579
+ |∇u|2dη ∧ dx1 ∧ dx2 +
580
+
581
+ Bǫ(Σδ
582
+ i )∩Σ
583
+ θdη|Σ) =
584
+
585
+ Σi
586
+ θdη|Σ
587
+ (13)
588
+ Then the proof of the theorem is completed.
589
+ Q.E.D.
590
+ Remark 2.2 From this theorem, we see that if uk satisfies (2), the weak limit u still
591
+ satisfies (2) if and only if θ = constant.
592
+ As a corollary, we can derive that θ(x) is locally constant. Precisely,
593
+ Corollary 2.3 Under the assumption of Theorem 1.2, and assume that there is an
594
+ open ball Bm ⊂ M \ Singu with Hm−2(Σ ∩ Bm) > 0. We have θ(x) is constant on
595
+ Σ ∩ Bm.
596
+ Proof.
597
+ In (5), we choose cutoff function η such that suppη ⊂ Bm.
598
+ Since Bm ⊂
599
+ M \ Singu, we have u is smooth on Bm. Then du∗J β = 0 on Bm for β = 1, 2, 3. In
600
+ view of (5), we conclude that θ is constant on Σ ∩ Bm.
601
+ Q.E.D.
602
+ Let φ : S2 → N be a nonconstant smooth map satisfying (3) and (4). Set
603
+ u(x, x4) = φ( x
604
+ |x|) for any x ∈ R3\{0} x4 ∈ Rm−3
605
+ (14)
606
+ as Chen-Li ([CL2]) did. Then we have
607
+ 5
608
+
609
+ Proposition 2.4 For any smooth (m − 3)-form η with compact support in Rm, we
610
+ have
611
+
612
+ Rm dη ∧ u∗J α = −Eα
613
+ T (φ)
614
+
615
+ Rm−3 η(0, x4),
616
+ (15)
617
+ where
618
+ ET(φ) =
619
+
620
+ S2⟨Jα
621
+ S2, u∗J α⟩dσ.
622
+ Proof. We choose a spherical coordinate system (r, ϕ, θ) in R3, because u is smooth
623
+ for any r > 0, we have
624
+
625
+ Rm dη ∧ u∗J α
626
+ =
627
+
628
+ Rm−3
629
+ � ∞
630
+ 0
631
+ ∂ηI
632
+ ∂r dr ∧ dxI
633
+
634
+ S2 φ∗J α
635
+ =
636
+
637
+
638
+ Rm−3 η(0, x4)
639
+
640
+ S2 φ∗J α
641
+ =
642
+ −Eα
643
+ T (φ)
644
+
645
+ Rm−3 η(0, x4)
646
+ Q.E.D.
647
+ By Theorem 2.1 and Proposition 2.4, we have the following corollary.
648
+ Corollary 2.5 The map u defined in (14) can not be a tangent map (c.f [LT], The-
649
+ orem 3.1) of a stationary quaternionic map with the property (2) at a singular point.
650
+ Proof. Suppose that u is defined as in (14). If it is a tangent map, then we have
651
+ by Theorem 2.1,
652
+ 3
653
+
654
+ α=1
655
+ Aαβ
656
+
657
+ M
658
+ dη ∧ u∗J β +
659
+
660
+ Σ
661
+ θdη|Σ = 0.
662
+ By Proposition 2.4, we obtain
663
+ 3
664
+
665
+ α=1
666
+ AαβEβ
667
+ T(φ)
668
+
669
+ Rm−3 η(0, x4) =
670
+
671
+ Σ
672
+ θdη|Σ.
673
+ Since u is stationary, by the blow-up formula of Li-Tian [LT], we have Σ is station-
674
+ ary. Using the constancy theorem (Theorem 41.1 in [Si]), it follows that the density
675
+ function θ is constant in every connected component of Σ, which implies that φ is
676
+ homotopy to a constant map. We therefore get a contradiction.
677
+ Q.E.D.
678
+ REFERENCES
679
+ [BT] C. Bellettini and G. Tian, Compactness results for triholomorphic maps, J. Eur. Math. Soc., 2(2019),
680
+ 1271-1317.
681
+ 6
682
+
683
+ [Ch] J. Chen, Complex anti-self-dual connections on product of Calabi-Yau surfaces and triholomorphic
684
+ curves, Commun. Math. Phys. 201(1999), 201-247.
685
+ [CL1] J. Chen and J. Li, Quaternionic maps between Hyperk¨ahler manifolds, J. Diff. Geom. 55(2000), no.
686
+ 2, 355-384.
687
+ [CL2] J. Chen and J. Li, Quarternionic maps and minimal surfaces, Ann. Sc. Norm. Super. Pisa Cl. Sci.
688
+ 4 (2005), no. 3, 375-388.
689
+ [FKS] J.M. Figuroa-O’Farrill, C. K¨ohl and B. Spence, Supersymmetric Yang-Mills, octonionic instantons
690
+ and triholomorphic curves, Nucl. Phys. B 521 (1998) no. 3, 419-443.
691
+ [F] L. Foscolo, ALF gravitational instantons and collapsing Ricci-flat metrics on the K3 surface, J. Diff.
692
+ Geom., 112(2019), 79-120.
693
+ [LT] J. Li, and G. Tian, A blow-up formula for stationary harmonic maps, IMRN, 14(1998), 735-755.
694
+ [Lin] F.-H. Lin, Gradient estimates and blow-up analysis for stationary harmonic maps I, Ann. of Math.
695
+ 149(1999), 785-829.
696
+ [Si] L. Simon, Lectures on Geometric Measure Theory, Proc. Center Math. Anal. 3(1983), Australian
697
+ National Univ. Press.
698
+ [W] C. Wang, Energy quantization for triholomorphic maps, Calc. Var. PDE 18(2003), 145-158.
699
+ 7
700
+
b9AyT4oBgHgl3EQfXPer/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,337 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf,len=336
2
+ page_content='arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
3
+ page_content='00180v1 [math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
4
+ page_content='DG] 31 Dec 2022 A blow-up formula for stationary quaternionic maps ∗† Jiayu Li‡ Chaona Zhu§ Abstract Let (M, Jα, α = 1, 2, 3) and (N, J α, α = 1, 2, 3) be Hyperk¨ahler manifolds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
5
+ page_content=' Suppose that uk is a sequence of stationary quaternionic maps and converges weakly to u in H1,2(M, N), we derive a blow-up formula for limk→∞ d(u∗ kJ α), for α = 1, 2, 3, in the weak sense.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
6
+ page_content=' As a corollary, we show that the maps constructed by Chen-Li [CL2] and by Foscolo [F] can not be tangent maps (c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
7
+ page_content='f [LT], Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
8
+ page_content='1) of a stationary quaternionic map satisfing d(u∗J α) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
9
+ page_content=' 1 Introduction and the main result A hyperk¨ahler manifold is a Riemannian manifold (M, g) with three parallel com- plex structures {J1, J2, J3} compatible with the metric g such that (J1)2 = (J2)2 = (J3)2 = J1J2J3 = −id.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
10
+ page_content=' The simplest hyperk¨ahler manifold is the Euclidean space R4m.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
11
+ page_content=' It is well-known that the only compact hyperk¨ahler manifolds of dimension 4 are K3 surfaces and complex tori.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
12
+ page_content=' Let (M, g, Jα, α = 1, 2, 3) and (N, h, J α, α = 1, 2, 3) be hyperk¨ahler manifolds.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
13
+ page_content=' Let ωα(·, ·) = g(·, Jα·) and Ωα(·, ·) = h(·, J α·), (α = 1, 2, 3) be the K¨ahler forms on M and N respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
14
+ page_content=' A smooth map u : M → N is called a quaternionic map (triholomorphic map) if AαβJ β ◦ du ◦ Jα = du (1) where Aαβ denote the entries of a matrix A in SO(3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
15
+ page_content=' For simplicity, we choose Aαβ = δαβ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
16
+ page_content=' The quaternionic maps (triholomorphic maps) between Hyperk¨ahler manifolds has been studied by many aothors (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
17
+ page_content=' [BT], [Ch], [CL1, [CL2], [FKS], [W]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
18
+ page_content=' Quater- nionic maps automatically minimize the energy functional in their homotopy classes (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
19
+ page_content=' [Ch], [CL1] and [FKS]) and hence they are harmonic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
20
+ page_content=' It can be verified that holomorphic and anti-holomorphic maps with respect to some complex structures on M and N are quaternionic maps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
21
+ page_content=' However, Chen-Li constructed quaternionic maps which are not holomorphic with respect to any complex structures on M and N (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
22
+ page_content=' [CL1]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
23
+ page_content=' ∗This work is supported by NSF grant 11721101.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
24
+ page_content=' †MSC (2000): 53C26, 53C43, 58E12, 58E20.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
25
+ page_content=' Keywords: Stationary harmonic maps, quaternionic maps, blow-up formula.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
26
+ page_content=' ‡jiayuli@ustc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
27
+ page_content='edu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
28
+ page_content='cn §zcn1991@mail.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
29
+ page_content='ustc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
30
+ page_content='edu.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
31
+ page_content='cn 1 Definition 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
32
+ page_content='1 A map u from M to N is called a stationary quaternionic map if it is a stationary harmonic map and it is a quaternionic map outside its singular set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
33
+ page_content=' It is clear that (c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
34
+ page_content='f.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
35
+ page_content=' [BT]), if u satisfies (1) almost everywhere, and d(u∗J α) = 0, for α = 1, 2, 3, (2) then u is a stationary quaternionic map.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
36
+ page_content=' Chen-Li ([CL2]) proved that, if there is a harmonic sphere φ : S2 → N which satisfies dφ JS2 = − 3 � k=1 akJ k dφ, (3) where ⃗a = (a1, a2, a3) : S2 → S2, and � S2 xi|∇φ|2dσ = 0, i = 1, 2, 3, (x1, x2, x3) ∈ S2, (4) then u(x, x4) = φ( x |x|) for any x ∈ R3\\{0} is a stationary quaternionic map with the x4-axis as its singular set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
37
+ page_content=' Chen-Li ([CL2]) showed that there does exist a complete noncompact hyperk¨ahler manifold, into which there is a harmonic S2 which satisfies (3) and (4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
38
+ page_content=' Recently, Foscolo [F] showed that there exists a compact K3 surface with the above property.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
39
+ page_content=' However, the map u constructed by Chen-Li or by Foscolo does not satisfy (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
40
+ page_content=' Now the question is whether the maps constructed by Chen-Li or by Foscolo could be a tangent map of a stationary quaternionic map with identity (2), if not the singular set of a stationary quaternionic map with identity (2) might be of codimensional 4 (Remark 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
41
+ page_content='2 in [BT]).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
42
+ page_content=' Suppose that uk is a sequence of stationary quaternionic maps with bounded energies E(uk) ≤ Λ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
43
+ page_content=' The blow-up set of uk can be defined as Σ = ∩r>0{x ∈ M| lim inf k→∞ r2−m � Br(x) | ▽ uk|2dy ≥ ǫ0}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
44
+ page_content=' We can always assume that uk ⇀ u weakly in W 1,2(M, N) and that | ▽ uk|2dx ⇀ | ▽ u|2dx + ν in the sense of measure as k → ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
45
+ page_content=' Here ν is a nonnegative Radon measure on M with support in Σ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
46
+ page_content=' It is known that Σ is a Hm−2-rectifiable set, and we may write ν = θ(x)Hm−2⌊Σ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
47
+ page_content=' It is clear that strongly convergence in H1,2(M, N) preserves the identity (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
48
+ page_content=' In this paper we mainly prove the following blow-up formula for weakly convergence sequence of stationary quaternionic maps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
49
+ page_content=' 2 Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
50
+ page_content='2 Let uk be a sequence of stationary quaternionic map with E(uk) ≤ Λ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
51
+ page_content=' Assume that uk → u weakly in H1(M, N).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
52
+ page_content=' Then there exist (a1, a2, a3) ∈ R3 with �3 α=1(aα)2 = 1 such that, for any smooth (m − 3)-form η with compact support in M, lim k→∞ 3 � α=1 aα � M dη ∧ u∗ kJ α = 3 � α=1 aα � M dη ∧ u∗J α + � Σ θdη|Σ (5) and for any (b1, b2, b3) ⊥ (a1, a2, a3), there holds lim k→∞ 3 � α=1 bα � M dη ∧ u∗ kJ α = 3 � α=1 bα � M dη ∧ u∗J α.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
53
+ page_content=' As a corollary of the theorem, the maps constructed by Chen-Li [CL2] and by Fos- colo [F] can not be tangent maps (c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
54
+ page_content='f [LT], Theorem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
55
+ page_content='1) of a stationary quaternionic map satisfing d(u∗J α) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
56
+ page_content=' 2 The proof of the blow-up formula If u is a strong limit of a sequence of stationary quaternionic maps in H1,2(M, N), then it’s easy to see that u satisfies (2).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
57
+ page_content=' If u is just a weak limit, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
58
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
59
+ page_content=' there exists a sequence of stationary quaternionic maps uk satisfying uk → u weakly in H1,2(M, N) and |∇uk|2dV → |∇u|2dV +θHm−2|Σ in the sense of measure, we prove in this section a formula for the blow-up set θHm−2|Σ and the limiting map u.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
60
+ page_content=' Without loss of generality, we may assume that m = 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
61
+ page_content=' Because Σ is a Hm−2- rectifiable set, so we may assume that Σ = ∪∞ i=0Σi, Σi ∩Σi′ = φ if i ̸= i′, Hm−2(Σ0) = 0, Σi ⊂ Ni and Ni (i = 1, 2, · · ·) is an (m − 2)-dimensional embedded C1 submanifold of M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
62
+ page_content=' It is important that (see p.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
63
+ page_content=' 61 in [Si]) TxΣ = TxNi for Hm−2-a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
64
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
65
+ page_content=' x ∈ Σi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
66
+ page_content=' It is known that ν = θ(x)Hm−2⌊Σ, where θ(x) is upper semi-continuous with ǫ0 ≤ θ(x) ≤ C(l1) for Hm−2-a.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
67
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
68
+ page_content=' x ∈ Σ, C(l1) is a positive constant depending only on M and l1 (cf.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
69
+ page_content=' [Lin], Lemma 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
70
+ page_content='6).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
71
+ page_content=' Since Hm−2(Σ) < +∞, for any 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
72
+ page_content=' > 0, there exist Σ1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
73
+ page_content=' ⊂ Σ and i0 such that Hm−2(Σ1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
74
+ page_content=' ) < 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
75
+ page_content=', Σc 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
76
+ page_content=' = Σ\\Σ1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
77
+ page_content=' = ∪i0 i=1Σ1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
78
+ page_content=' i where Σ1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
79
+ page_content=' i ⊂ Σi (i = 1, · · ·, i0) is a bounded closed set.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
80
+ page_content=' We choose a covering {Brn|n = 1, 2, · · ·} of Σ1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
81
+ page_content=' such that � n rm−2 n < C1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
82
+ page_content='. Here and in the sequel, C always denotes a uniform constant depending only on M and N.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
83
+ page_content=' Suppose that (x1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
84
+ page_content='..' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
85
+ page_content=', x4) is a local normal coordinate system in Bǫ(Σδ i), and that (x3, x4) is the corresponding coordinate system in Σi, and the matrix expressions of the complex structures are given by (6), (7) and (8).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
86
+ page_content=' J1 = \uf8eb \uf8ec \uf8ec \uf8ed 0 0 0 −1 0 0 1 0 0 −1 0 0 1 0 0 0 \uf8f6 \uf8f7 \uf8f7 \uf8f8 , A1βJ β = \uf8eb \uf8ec \uf8ec \uf8ed J1 J1 \uf8f6 \uf8f7 \uf8f7 \uf8f8 (6) 3 J2 = \uf8eb \uf8ec \uf8ec \uf8ed 0 −1 0 0 1 0 0 0 0 0 0 1 0 0 −1 0 \uf8f6 \uf8f7 \uf8f7 \uf8f8 , A2βJ β = \uf8eb \uf8ec \uf8ec \uf8ed J2 J2 \uf8f6 \uf8f7 \uf8f7 \uf8f8 (7) J3 = \uf8eb \uf8ec \uf8ec \uf8ed 0 0 1 0 0 0 0 1 −1 0 0 0 0 −1 0 0 \uf8f6 \uf8f7 \uf8f7 \uf8f8 , A3βJ β = \uf8eb \uf8ec \uf8ec \uf8ed J3 J3 \uf8f6 \uf8f7 \uf8f7 \uf8f8 (8) where AαβJ β are 4n×4n-matrices, Aαβ are the entries of a matrix A in SO(3).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
87
+ page_content=' Then ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
88
+ page_content='the quaternionic equation is ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
89
+ page_content='\uf8f1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
90
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
91
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
92
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
93
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
94
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
95
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
96
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
97
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
98
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
99
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
100
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
101
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
102
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
103
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
104
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
105
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
106
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
107
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
108
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
109
+ page_content='\uf8f2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
110
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
111
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
112
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
113
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
114
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
115
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
116
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
117
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
118
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
119
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
120
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
121
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
122
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
123
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
124
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
125
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
126
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
127
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
128
+ page_content='\uf8f4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
129
+ page_content='\uf8f3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
130
+ page_content='u1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
131
+ page_content='1 + u2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
132
+ page_content='2 + u3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
133
+ page_content='3 + u4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
134
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
135
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
136
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
137
+ page_content='u2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
138
+ page_content='1 − u1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
139
+ page_content='2 + u4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
140
+ page_content='3 − u3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
141
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
142
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
143
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
144
+ page_content='u3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
145
+ page_content='1 − u1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
146
+ page_content='3 − u4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
147
+ page_content='2 + u2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
148
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
149
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
150
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
151
+ page_content='u4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
152
+ page_content='1 − u1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
153
+ page_content='4 − u2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
154
+ page_content='3 + u3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
155
+ page_content='2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
156
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
157
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
158
+ page_content='u5 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
159
+ page_content='1 + u6 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
160
+ page_content='2 + u7 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
161
+ page_content='3 + u8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
162
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
163
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
164
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
165
+ page_content='u6 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
166
+ page_content='1 − u5 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
167
+ page_content='2 + u8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
168
+ page_content='3 − u7 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
169
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
170
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
171
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
172
+ page_content='u7 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
173
+ page_content='1 − u5 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
174
+ page_content='3 − u8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
175
+ page_content='2 + u6 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
176
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
177
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
178
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
179
+ page_content='u8 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
180
+ page_content='1 − u5 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
181
+ page_content='4 − u6 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
182
+ page_content='3 + u7 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
183
+ page_content='2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
184
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
185
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
186
+ page_content='· · ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
187
+ page_content='u4n−3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
188
+ page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
189
+ page_content='+ u4n−2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
190
+ page_content='2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
191
+ page_content='+ u4n−1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
192
+ page_content='3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
193
+ page_content='+ u4n ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
194
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
195
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
196
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
197
+ page_content='u4n−2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
198
+ page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
199
+ page_content='− u4n−3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
200
+ page_content='2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
201
+ page_content='+ u4n ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
202
+ page_content='3 − u4n−1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
203
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
204
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
205
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
206
+ page_content='u4n−1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
207
+ page_content='1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
208
+ page_content='− u4n−3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
209
+ page_content='3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
210
+ page_content='− u4n ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
211
+ page_content='2 + u4n−2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
212
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
213
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
214
+ page_content='0 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
215
+ page_content='u4n ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
216
+ page_content='1 − u4n−3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
217
+ page_content='4 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
218
+ page_content='− u4n−2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
219
+ page_content='3 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
220
+ page_content='+ u4n−1 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
221
+ page_content='2 ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
222
+ page_content='= ' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
223
+ page_content='0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
224
+ page_content=' (9) Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
225
+ page_content='1 For any smooth (m − 3)-form η with compact support in M, we have lim k→∞ 3 � α=1 Aαβ � M dη ∧ u∗ kJ β = 3 � α=1 Aαβ � M dη ∧ u∗J β + � Σ θdη|Σ and lim k→∞ A1β � M dη ∧ u∗ kJ β = A1β � M dη ∧ u∗J β, lim k→∞ A3β � M dη ∧ u∗ kJ β = A3β � M dη ∧ u∗J β, Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
226
+ page_content=' Assume that η = � I ηIdxI.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
227
+ page_content=' We have lim k→∞ � M dη ∧ u∗ k(AαβJ β) = � M dη ∧ u∗(AαβJ β) + lim δ→0 lim ǫ→0 lim k→∞ � Bǫ(∪i0 i=1Σδ i ) dη ∧ u∗ k(AαβJ β) + lim δ→0 lim ǫ→0 lim k→∞ � ∪nBrn\\Bǫ(∪i0 i=1Σδ i ) dη ∧ u∗ k(AαβJ β).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
228
+ page_content=' (10) 4 It’s easy to see that lim δ→0 lim ǫ→0 lim k→∞ � ∪nBrn dη ∧ u∗ k(J β) = 0 (11) By Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
229
+ page_content='2 in [LT], we get lim δ→0 lim ǫ→0 lim k→∞ � Bǫ(Σδ i ) dη ∧ u∗ k(AαβJ β) = lim δ→0 lim ǫ→0 lim k→∞ � Bǫ(Σδ i ) 2∂ηI ∂xl ∂uσ k ∂x1 (AαβJ β)σγ ∂uγ k ∂x2 dxl ∧ dxI ∧ dx1 ∧ dx2 (12) Substituting (9) to (12) and applying Lemma 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
230
+ page_content='2 in [LT], we have lim δ→0 lim ǫ→0 lim k→∞ � Bǫ(Σδ i ) dη ∧ u∗ k(A1βJ β) = lim δ→0 lim ǫ→0 lim k→∞ � Bǫ(Σδ i ) dη ∧ u∗ k(A3βJ β) = 0 and lim δ→0 lim ǫ→0 lim k→∞ � Bǫ(Σδ i ) dη ∧ u∗ k(A2βJ β) = lim δ→0 lim ǫ→0 lim k→∞ � Bǫ(Σδ i ) |∇uk|2dη ∧ dx1 ∧ dx2 = lim δ→0 lim ǫ→0( � Bǫ(Σδ i ) |∇u|2dη ∧ dx1 ∧ dx2 + � Bǫ(Σδ i )∩Σ θdη|Σ) = � Σi θdη|Σ (13) Then the proof of the theorem is completed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
231
+ page_content=' Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
232
+ page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
233
+ page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
234
+ page_content=' Remark 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
235
+ page_content='2 From this theorem, we see that if uk satisfies (2), the weak limit u still satisfies (2) if and only if θ = constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
236
+ page_content=' As a corollary, we can derive that θ(x) is locally constant.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
237
+ page_content=' Precisely, Corollary 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
238
+ page_content='3 Under the assumption of Theorem 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
239
+ page_content='2, and assume that there is an open ball Bm ⊂ M \\ Singu with Hm−2(Σ ∩ Bm) > 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
240
+ page_content=' We have θ(x) is constant on Σ ∩ Bm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
241
+ page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
242
+ page_content=' In (5), we choose cutoff function η such that suppη ⊂ Bm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
243
+ page_content=' Since Bm ⊂ M \\ Singu, we have u is smooth on Bm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
244
+ page_content=' Then du∗J β = 0 on Bm for β = 1, 2, 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
245
+ page_content=' In view of (5), we conclude that θ is constant on Σ ∩ Bm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
246
+ page_content=' Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
247
+ page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
248
+ page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
249
+ page_content=' Let φ : S2 → N be a nonconstant smooth map satisfying (3) and (4).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
250
+ page_content=' Set u(x, x4) = φ( x |x|) for any x ∈ R3\\{0} x4 ∈ Rm−3 (14) as Chen-Li ([CL2]) did.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
251
+ page_content=' Then we have 5 Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
252
+ page_content='4 For any smooth (m − 3)-form η with compact support in Rm, we have � Rm dη ∧ u∗J α = −Eα T (φ) � Rm−3 η(0, x4), (15) where ET(φ) = � S2⟨Jα S2, u∗J α⟩dσ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
253
+ page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
254
+ page_content=' We choose a spherical coordinate system (r, ϕ, θ) in R3, because u is smooth for any r > 0, we have � Rm dη ∧ u∗J α = � Rm−3 � ∞ 0 ∂ηI ∂r dr ∧ dxI � S2 φ∗J α = − � Rm−3 η(0, x4) � S2 φ∗J α = −Eα T (φ) � Rm−3 η(0, x4) Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
255
+ page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
256
+ page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
257
+ page_content=' By Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
258
+ page_content='1 and Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
259
+ page_content='4, we have the following corollary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
260
+ page_content=' Corollary 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
261
+ page_content='5 The map u defined in (14) can not be a tangent map (c.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
262
+ page_content='f [LT], The- orem 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
263
+ page_content='1) of a stationary quaternionic map with the property (2) at a singular point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
264
+ page_content=' Proof.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
265
+ page_content=' Suppose that u is defined as in (14).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
266
+ page_content=' If it is a tangent map, then we have by Theorem 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
267
+ page_content='1, 3 � α=1 Aαβ � M dη ∧ u∗J β + � Σ θdη|Σ = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
268
+ page_content=' By Proposition 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
269
+ page_content='4, we obtain 3 � α=1 AαβEβ T(φ) � Rm−3 η(0, x4) = � Σ θdη|Σ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
270
+ page_content=' Since u is stationary, by the blow-up formula of Li-Tian [LT], we have Σ is station- ary.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
271
+ page_content=' Using the constancy theorem (Theorem 41.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
272
+ page_content='1 in [Si]), it follows that the density function θ is constant in every connected component of Σ, which implies that φ is homotopy to a constant map.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
273
+ page_content=' We therefore get a contradiction.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
274
+ page_content=' Q.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
275
+ page_content='E.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
276
+ page_content='D.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
277
+ page_content=' REFERENCES [BT] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
278
+ page_content=' Bellettini and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
279
+ page_content=' Tian, Compactness results for triholomorphic maps, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
280
+ page_content=' Eur.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
281
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
282
+ page_content=' Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
283
+ page_content=', 2(2019), 1271-1317.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
284
+ page_content=' 6 [Ch] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
285
+ page_content=' Chen, Complex anti-self-dual connections on product of Calabi-Yau surfaces and triholomorphic curves, Commun.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
286
+ page_content=' Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
287
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
288
+ page_content=' 201(1999), 201-247.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
289
+ page_content=' [CL1] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
290
+ page_content=' Chen and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
291
+ page_content=' Li, Quaternionic maps between Hyperk¨ahler manifolds, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
292
+ page_content=' Diff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
293
+ page_content=' Geom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
294
+ page_content=' 55(2000), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
295
+ page_content=' 2, 355-384.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
296
+ page_content=' [CL2] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
297
+ page_content=' Chen and J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
298
+ page_content=' Li, Quarternionic maps and minimal surfaces, Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
299
+ page_content=' Sc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
300
+ page_content=' Norm.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
301
+ page_content=' Super.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
302
+ page_content=' Pisa Cl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
303
+ page_content=' Sci.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
304
+ page_content=' 4 (2005), no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
305
+ page_content=' 3, 375-388.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
306
+ page_content=' [FKS] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
307
+ page_content='M.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
308
+ page_content=' Figuroa-O’Farrill, C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
309
+ page_content=' K¨ohl and B.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
310
+ page_content=' Spence, Supersymmetric Yang-Mills, octonionic instantons and triholomorphic curves, Nucl.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
311
+ page_content=' Phys.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
312
+ page_content=' B 521 (1998) no.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
313
+ page_content=' 3, 419-443.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
314
+ page_content=' [F] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
315
+ page_content=' Foscolo, ALF gravitational instantons and collapsing Ricci-flat metrics on the K3 surface, J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
316
+ page_content=' Diff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
317
+ page_content=' Geom.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
318
+ page_content=', 112(2019), 79-120.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
319
+ page_content=' [LT] J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
320
+ page_content=' Li, and G.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
321
+ page_content=' Tian, A blow-up formula for stationary harmonic maps, IMRN, 14(1998), 735-755.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
322
+ page_content=' [Lin] F.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
323
+ page_content='-H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
324
+ page_content=' Lin, Gradient estimates and blow-up analysis for stationary harmonic maps I, Ann.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
325
+ page_content=' of Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
326
+ page_content=' 149(1999), 785-829.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
327
+ page_content=' [Si] L.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
328
+ page_content=' Simon, Lectures on Geometric Measure Theory, Proc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
329
+ page_content=' Center Math.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
330
+ page_content=' Anal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
331
+ page_content=' 3(1983), Australian National Univ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
332
+ page_content=' Press.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
333
+ page_content=' [W] C.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
334
+ page_content=' Wang, Energy quantization for triholomorphic maps, Calc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
335
+ page_content=' Var.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
336
+ page_content=' PDE 18(2003), 145-158.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
337
+ page_content=' 7' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/b9AyT4oBgHgl3EQfXPer/content/2301.00180v1.pdf'}
bdE_T4oBgHgl3EQfzRz6/content/2301.08323v1.pdf ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af89757008b63f81da7e860e1f5f4fae789e1f4c8e26f106925a1e3d52a51d53
3
+ size 2347832
d9FIT4oBgHgl3EQfoiuQ/content/tmp_files/2301.11319v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
d9FIT4oBgHgl3EQfoiuQ/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
dtE0T4oBgHgl3EQfoQFr/content/tmp_files/2301.02523v1.pdf.txt ADDED
@@ -0,0 +1,1043 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+
3
+ Many-body Hybrid Excitons with strong molecular orientation
4
+ dependence
5
+ in
6
+ Organic-Inorganic
7
+ van
8
+ der
9
+ Waals
10
+ Heterostructures
11
+ Shaohua Fu,1,2,4 # Jianwei Ding3 #, Haifeng Lv,5 Shuangyan Liu,1 Kun Zhao1, Zhiying
12
+ Bai1, Dawei He,1 Rui Wang,6 Jimin Zhao,4 Xiaojun Wu,5 Dongsheng Tang,2 * Xiaohui
13
+ Qiu,3 * Yongsheng Wang1, Xiaoxian Zhang,1 *
14
+ 1Key Laboratory of Luminescence and Optical Information, Ministry of Education,
15
+ Institute of Optoelectronic Technology, Beijing Jiaotong University, Beijing 100044,
16
+ China
17
+ 2Synergetic Innovation Center for Quantum Effects an Application, Key Laboratory of
18
+ Low-dimensional Quantum Structures and Quantum Control of Ministry of Education,
19
+ School of Physics and Electronics, Hunan Normal University, Changsha 410081, China
20
+ 3CAS Key Laboratory of Standardization and Measurement for Nanotechnology, CAS
21
+ Center for Excellence in Nanoscience, National Center for Nanoscience and
22
+ Technology, Beijing 100190, P. R. China.
23
+ 4Beijing National Laboratory for Condensed Matter Physics, Institute of Physics,
24
+ Chinese Academy of Sciences, Beijing 100190, China
25
+ 5Hefei National Laboratory for Physical Sciences at the Microscale, CAS Key
26
+ Laboratory of Materials for Energy Conversion, Synergetic Innovation of Quantum
27
+ Information & Quantum Technology, School of Chemistry and Materials Sciences, and
28
+ CAS Center for Excellence in Nanoscience, University of Science and Technology of
29
+ China, Hefei, Anhui 230026, P.R. China
30
+
31
+ 2
32
+
33
+ 6Beijing Information technology college, Beijing 100015, P. R. China
34
+ #These authors contributed equally
35
36
37
+
38
+
39
+
40
+
41
+
42
+
43
+
44
+
45
+
46
+
47
+
48
+
49
+
50
+
51
+
52
+
53
+ 3
54
+
55
+ Abstract
56
+ The coherent many-body interaction at the organic-inorganic interface can give rise to
57
+ intriguing hybrid excitons that combine the advantages of the Wannier-Mott and
58
+ Frenkel excitons simultaneously. Unlike the 2D inorganic heterostructures that suffer
59
+ from moment mismatch, the hybrid excitons formed at the organic-inorganic interface
60
+ have a momentum-direct nature, which have yet to be explored. Here, we report hybrid
61
+ excitons at the copper phthalocyanine/molybdenum diselenide (CuPc/MoSe2) interface
62
+ with
63
+ strong
64
+ molecular
65
+ orientation
66
+ dependence
67
+ using
68
+ low-temperature
69
+ photoluminescence spectroscopy. The new emission peaks observed in the
70
+ CuPc/MoSe2 heterostructure indicate the formation of interfacial hybrid excitons. The
71
+ density functional theory (DFT) calculation confirms the strong hybridization between
72
+ the lowest unoccupied molecular orbital (LUMO) of CuPc and the conduction band
73
+ minimum (CBM) of MoSe2, suggesting that the hybrid excitons consist of electrons
74
+ extended in both layers and holes confined in individual layers. The temperature-
75
+ dependent measurements show that the hybrid excitons can gain the signatures of the
76
+ Frenkel excitons of CuPc and the Wannier-Mott excitons of MoSe2 simultaneously. The
77
+ out-of-plane molecular orientation is used to tailor the interfacial hybrid exciton states.
78
+ Our results reveal the hybrid excitons at the CuPc/MoSe2 interface with tunability by
79
+ molecular orientation, which suggests that the emerging organic-inorganic
80
+ heterostructure can be a promising platform for many-body exciton physics.
81
+
82
+
83
+
84
+ 4
85
+
86
+ Introduction
87
+ Hybrid excitons are many-body exciton states that originate from the hybridization
88
+ of electronic states at interfaces1,2, which have been realized in distinct systems,
89
+ including quantum dots coupled to a Fermi sea1,3, coupled quantum-dot molecules4,5,
90
+ and emerging van der Waals heterostructures2,6-8, displaying great potential in kondo
91
+ physics1,8, quantum optics4,7, and strongly correlated electronic physics2. Transition
92
+ metal dichalcogenides (TMDs) have become a promising building block for hybrid
93
+ excitons due to their strong light-matter interaction9,10 and rich exciton physics11-14,
94
+ such as valley polarized excitons15-17, long-lived interlayer excitons18 and moiré
95
+ excitons19-21. However, in TMD heterostructures, the momentum-mismatch problem
96
+ severely restricts the formation of prominent hybrid excitons, which is moment-direct
97
+ only at a small twist angle, requiring a precise control of the interlayer angle alignment
98
+ during fabrication2,6,7.
99
+ Unlike the momentum-mismatch issue encountered in inorganic heterostructures,
100
+ the hybrid excitons formed at the organic-inorganic heterostructures have a momentum-
101
+ direct nature22, which could simplify the fabrication process and maintain the novel
102
+ exciton physics at the same time. In addition, theoretical calculations have predicted
103
+ that the hybrid excitons at the organic-inorganic interfaces can gain the signature of the
104
+ Wannier-Mott excitons of inorganics and the Frenkel excitons of organics
105
+ simultaneously23,24. Nevertheless, the coupling at the organic-inorganic interfaces is
106
+ generally weak25, and an ordered structure of the organic materials is required to
107
+ achieve strong electronic coupling24. TMDs can be an ideal building brick for realizing
108
+
109
+ 5
110
+
111
+ hybrid excitons at the organic-inorganic interfaces, because they not only contain rich
112
+ exciton physics but can also serve as a suitable template for the growth of well-ordered
113
+ organic films through the van der Waals epitaxial method26-31. Moreover, the short-
114
+ range interactions, such as ultrafast charge transfer31-34 and interfacial spin orbital
115
+ coupling35, have been observed at organic-TMD interfaces, which suggest that the
116
+ coherent superposition of the electron wavefunctions and the formation of hybrid
117
+ excitons are possible at such interfaces. However, experimental evidence for hybrid
118
+ Wannier-Mott-Frenkel excitons at organic-TMD interfaces remains to be explored.
119
+ It is well-known that the interfacial hybridization strength depends sensitively on
120
+ the interlayer twist angle and stacking configuration in inorganic heterostructures6,7.
121
+ Similarly, the molecular orientation at the organic-inorganic interface can be used to
122
+ tune the interfacial distance, i.e., the hybridization strength, so the interfacial hybrid
123
+ exciton behavior could be effectively tailored, which is beneficial for designing the
124
+ organic-inorganic interface functionality.
125
+ In this article, we report the formation of hybrid excitons at the CuPc/MoSe2
126
+ interface and their further modulation by molecular orientation. The new emission
127
+ peaks observed in photoluminescence spectroscopy and the further DFT calculations
128
+ confirm the emergence of new hybrid excitons. The temperature-dependent
129
+ measurements reveal that the hybrid excitons combine the signature of both Wannier-
130
+ Mott and Frenkel exciton species. The out-of-plane molecular orientation is also
131
+ applied to tailor the interfacial hybrid excitons. Our results suggest that the organic-
132
+ inorganic heterostructure is a promising platform to explore many-body exciton physics.
133
+
134
+ 6
135
+
136
+ Results
137
+
138
+ Fig. 1 | Sample configuration and basic optical characterization. a Schematic illustration of the
139
+ different molecular orientations (face-on and edge-on) at CuPc/MoSe2 heterostructure interface. b
140
+ AFM topographic image of a typical CuPc/MoSe2 heterostructure on Si/SiO2 substrate. c PL spectra
141
+ of MoSe2 and CuPc/MoSe2 heterostructure at 298 K. d Time-resolved PL spectra of MoSe2 and
142
+ CuPc/MoSe2 heterostructure at 298 K. The solid curves are the fitted results.
143
+ Sample configuration and basic optical characterization
144
+ Figure 1a shows a schematic of the CuPc/MoSe2 heterostructure configuration.
145
+ We consider two different molecular orientations, i.e., face-on and edge-on, at the
146
+ CuPc/MoSe2 interface, which will sensitively influence the interfacial coupling strength.
147
+ The CuPc/MoSe2 heterostructure is prepared by directly evaporating CuPc molecules
148
+ on top of a monolayer MoSe2 surface in vacuum (see details in methods). A film
149
+ thickness of ~5 nm is determined by AFM (Fig. 1b and Supplementary Fig. 1c). The
150
+
151
+ a
152
+ b
153
+ 10 nm
154
+ edge-on
155
+ Cu
156
+ CuPc
157
+ N
158
+ face-on
159
+ CuPc
160
+ CuPclMoSe
161
+ C
162
+ H
163
+ Mo
164
+ Se
165
+ MoSe2
166
+ 600nm
167
+ 0
168
+ c
169
+ d
170
+ ×20
171
+ PL intensity (a.u.)
172
+ -MoSe,
173
+ Normalized PL
174
+ - CuPc/MoSe2
175
+ 0.1
176
+ 1.4
177
+ 1.5
178
+ 1.6
179
+ 1.7
180
+ 0
181
+ 1
182
+ 2
183
+ Photon energy (eV)
184
+ Delay time (ns)7
185
+
186
+ optimal molecular orientation at interface of the as-grown sample is the face-on
187
+ orientation, which has been reported in similar systems34,36 and revealed by our
188
+ theoretical calculation. The edge-on orientation is introduced by using the CuPc single
189
+ crystal later. At the face-on orientation, the planar conjugated structure of the CuPc
190
+ molecule37 and the atomic flat surface of monolayer MoSe210 without dangling bonds
191
+ facilitate interfacial coupling between them (Fig. 1a). The photoluminescence (PL)
192
+ spectra of MoSe2 and CuPc/MoSe2 heterostructure acquired at room temperature are
193
+ shown in Fig. 1c and Supplementary Fig.1d. The MoSe2 exhibits a pronounced PL peak
194
+ located at ~1.58 eV from the A excitonic transition38. In contrast, a remarkable redshift
195
+ of ~20 meV in the PL peak energy and a strong quenching in the PL peak intensity are
196
+ observed in CuPc/MoSe2 heterostructure. The pure CuPc film shows no detectable PL
197
+ signal (Supplementary Fig. 2a) due to the weak absorption at approximately 514 nm34
198
+ and its strong intersystem crossing39. Control experiments are further performed to
199
+ examine the underlying possibilities of the observed phenomena. By changing the
200
+ thickness of CuPc thin film, it is found that the PL quenching ratio remains almost
201
+ unchanged (Supplementary Fig. 2c), indicating that absorption of the CuPc film is not
202
+ the main reason and the phenomena may stem from interfacial interaction. We also
203
+ adopt another heterostructure configuration by dry transferring MoSe2 on top of the
204
+ CuPc film (Supplementary Fig.3) and observe similar phenomena, which indicates that
205
+ the dielectric environment change has negligible influence here40,41. Therefore, the
206
+ observed phenomena should originate from the intrinsic interfacial coupling between
207
+ CuPc and MoSe2. Time-resolved PL measurements are performed to compare the PL
208
+
209
+ 8
210
+
211
+ lifetimes of MoSe2 and CuPc/MoSe2 heterostructure (Fig. 1c). The PL decays of both
212
+ MoSe2 and CuPc/MoSe2 heterostructure can be well fitted with a biexponential function,
213
+ thus, two processes can be derived from both decay curves. For MoSe2, the fast decay
214
+ constituent has a lifetime of ~85 ps, which is consistent with the lifetime of the A
215
+ exciton42. The slow decay component has a lifetime of ~1495 ps and is likely to
216
+ originate from defect-bound excitons43,44. The PL decay of heterostructure exhibits
217
+ much shorter lifetime than the A exciton of MoSe2, which is not consistent with the
218
+ behavior of interfacial charge transfer exciton because it usually has a longer lifetime
219
+ due to the spatial indirect nature31,45.
220
+
221
+ Fig. 2 | Hybrid excitons in CuPc/MoSe2 heterostructure. a PL spectra of the MoSe2 and
222
+ CuPc/MoSe2 heterostructure at 78 K. b The PL peak intensity of MoSe2 and CuPc/MoSe2
223
+ heterostructure as a function of the excitation power. c PL spectra of the MoSe2 and CuPc/MoSe2
224
+
225
+ a
226
+ b
227
+ 1.0
228
+ x
229
+ hX2
230
+ X
231
+ MoSe2
232
+ 105
233
+ T
234
+ 78 K
235
+ 78 K
236
+
237
+ hx
238
+ ~p1.13
239
+ CuPc/MoSe2
240
+ intensity (a.u.)
241
+ hx,
242
+ PL
243
+ 104
244
+ hX
245
+ P1.13
246
+ -P1.19
247
+ 103
248
+ P1.26
249
+ Normal
250
+ PLi
251
+ 102
252
+ P1.09
253
+ hX?
254
+ T
255
+ hX4
256
+ 101
257
+ ~P1.26
258
+ 0.0
259
+ 1.50
260
+ 1.65
261
+ 1.80
262
+ 1.95
263
+ 1
264
+ 10
265
+ 100
266
+ Photon energy (eV)
267
+ Excitation power(μW)
268
+ C 1.0
269
+ d
270
+ hX2
271
+ T
272
+ 4 K
273
+ MoSe2
274
+ 10
275
+ 000
276
+ Ix/lT
277
+ hX,
278
+ Peak intensity ratio
279
+ CuPc/MoSe2
280
+ PL
281
+ Normalized I
282
+ Q00
283
+ 00.0000
284
+ 0.5
285
+ Ihx1/lhx2
286
+ X
287
+ hX3
288
+ 00
289
+ hX4
290
+ 0.1
291
+ 0.0
292
+ 1.50
293
+ 1.65
294
+ 1.80
295
+ 1.95
296
+ 0
297
+ 50
298
+ 100
299
+ 150
300
+ 200
301
+ Photon energy (eV)
302
+ Temperature (K)9
303
+
304
+ heterostructure at 4 K. d The PL peak intensity ratio of exciton versus trion in MoSe2 (IX/IT) and hx1
305
+ versus hx2 in the CuPc/MoSe2 heterostructure (IhX1/IhX2) as a function of temperature.
306
+ Emergence of interfacial hybrid excitons
307
+ Low-temperature PL spectra under 514 nm excitation are obtained at 78 K to
308
+ further reveal the possible mechanism. We still observe no detectable PL signal in the
309
+ pure CuPc thin film (Supplementary Fig. 2b). As displayed in Fig. 2a, two emission
310
+ peaks located at ~1.648 eV and 1.618 eV are observed in the PL spectrum of MoSe2,
311
+ which can be ascribed to the emission from the A exciton (X) and trion (T) of MoSe238.
312
+ A striking contrast is observed in the PL spectrum of heterostructure, with four new
313
+ emission peaks located at ~1.630 eV, ~1.606 eV, ~1.727 eV and ~1.848 eV emerging,
314
+ which are labeled hX1, hX2, hX3, and hX4, respectively. The hX1 and hX2 show a clear
315
+ redshift compared with the A exciton of MoSe2, and the hX4 peak displays an obvious
316
+ redshift with respect to the B exciton of MoSe2 (Supplementary Fig. 4). The hX3 peak
317
+ is a totally new PL peak that are not observed in pure MoSe2 and CuPc films. Charge
318
+ transfer exciton31 or dark exciton46 of MoSe2 is also excluded because it has a much
319
+ higher energy (~79 meV) than the A exciton of MoSe2. In addition, the PL peaks of
320
+ heterostructure show clear broadening compared with those of MoSe2. We ascribe the
321
+ observed PL peak redshift and broadening to the signature of interfacial hybridization
322
+ as reported in similar MoSe2/WS2 heterostructure6. Power-dependent PL spectra are
323
+ further obtained to examine the origin of the new peaks in heterostructure (Fig. 2b and
324
+ Supplementary Fig. 5). It is obvious that the peak intensity is enhanced with increasing
325
+ power. The relationship between excitation power and PL intensity can be expressed
326
+ as47 𝐼 ∝ 𝑃𝛼, in which 𝐼 represents the PL intensity and 𝑃 represents the excitation
327
+
328
+ 10
329
+
330
+ power. The intensities of X and T peaks in MoSe2 show a linear relationship with
331
+ excitation power with a slope of ~1.13, which indicates recombination from excitons48.
332
+ Interestingly, all the new peaks in heterostructure also show the linear relationship with
333
+ similar slopes, which suggests similar exciton behavior with no biexciton49 or defect
334
+ effect50. Since we have observed new PL peaks with exciton behavior and the signature
335
+ of hybridization, it is possible that new hybrid excitons are formed in CuPc/MoSe2
336
+ heterostructure due to interfacial hybridization.
337
+ We further perform PL measurements at 4 K to examine the influence of interfacial
338
+ hybridization. As illustrated in Fig. 2c, the PL spectrum of MoSe2 at 4 K is dominated
339
+ by trion rather than A exciton due to the enhanced trion localization, in great contrast
340
+ to that at 78 K42. On the contrary, the PL spectrum of heterostructure is still dominated
341
+ by hX1 and hX2, similar to that at 78 K. Figure 2d displays the evolution of X/T from 4
342
+ K to 200 K, it is clear that the ratio of X/T has increased from 0.09 to 10 when the
343
+ temperature is increased from 4 K to 200 K due to the increased thermal perturbance to
344
+ trion formation. Nevertheless, the radio of hX1/hX2 shows a weak temperature
345
+ dependence from 4 K to 200 K, differing from the pure exciton and trion behavior in
346
+ MoSe2, which indicates that the interfacial hybridization effect has changed the exciton
347
+ behavior in heterostructure. In addition, the PL spectrum of heterostructure displays a
348
+ highly asymmetric line-shape with prominent low energy tail (Fig. 2a, c), which can be
349
+ ascribed to an energy shakeup process during the recombination of hybrid excitons 8.
350
+
351
+
352
+ 11
353
+
354
+ Fig. 3 | Theoretical calculation of electronic structure in CuPc/MoSe2 heterostructure. a Top
355
+ -view and side-view for the optimized structure of CuPc/MoSe2 heterostructure. b Calculated band
356
+ structure of CuPc/MoSe2 heterostructure. c Conduction bands (CB), CB+1, CB+2 and CB+3 in the
357
+ energy range of 0.72 to 0.80 eV, which correspond to the blue square in b. d Projected charge
358
+ density for bands in b. ΔE (23 meV) is defined as the energy difference between CB+1 and CB+3,
359
+ which is mostly contributed by CuPc and MoSe2, respectively. e Schematic illustration of the
360
+ formation of interfacial hybrid excitons due to the hybridization between LUMO of CuPc and CBM
361
+ of MoSe2.
362
+ The above results indicate that the interfacial hybridization effect could lead to the
363
+ formation of hybrid excitons and change the exciton behavior at the CuPc/MoSe2
364
+ interface. First-principles calculations are further performed to confirm this. The
365
+ heterostructure is built by adsorbing a CuPc molecule on a 5×3√3 supercell of MoSe2
366
+ and the optimal molecular orientation is the face-on orientation (Fig. 3a). The calculated
367
+ electronic structure of CuPc/MoSe2 heterostructure with face-on orientation is shown
368
+
369
+ a
370
+ Top-view
371
+ Side-view
372
+ e
373
+ CuPc
374
+ 3.36A
375
+ hX
376
+ CuPc
377
+ h2o
378
+ MoSe2
379
+ MoSe2
380
+ b
381
+ 1.0
382
+ C 0.80
383
+ CB
384
+ (eV)
385
+ CB+1
386
+ 0.5
387
+
388
+ AE
389
+ W-0.5
390
+ CB+2
391
+ CB+3
392
+ CB
393
+ CB+1
394
+ CB+2
395
+ CB+3
396
+ -1.0
397
+ 0.72
398
+ x
399
+ Y12
400
+
401
+ in Fig. 3b. We could recognize two nearly flat bands near 0.5 and -0.5 eV, which
402
+ correspond to the singly occupied and unoccupied molecular orbitals (SOMO and
403
+ SUMO) of the CuPc molecule. Then, we concentrate on the bands in the energy range
404
+ of 0.72 to 0.80 eV (Fig. 3c), which are denoted as conduction band CB, CB+1, CB+2
405
+ and CB+3. As shown in Fig. 3d, the projected charge density shows that CB and CB+1
406
+ are mostly contributed by CuPc, and CB+3 is mostly contributed by MoSe2. Notably,
407
+ CB+2 is contributed both by CuPc and MoSe2, which could be regarded as the
408
+ hybridization between the LUMO of CuPc and the conduction band of MoSe2. The
409
+ energy difference between CB+3 and CB+1 in the same spin channel is approximately
410
+ 23 meV, leading to strong hybridization between CuPc and MoSe2 at the face-on
411
+ orientation, which can explain the observed new hybrid excitons at CuPc/MoSe2
412
+ interface. The calculation also reveals that the formed hybrid excitons consist of
413
+ electrons extended in both layers and holes confined in individual layers (Fig. 3e),
414
+ which can be used to achieve novel quantum control at organic-inorganic interfaces7.
415
+
416
+
417
+
418
+ 13
419
+
420
+
421
+ Fig. 4 | Temperature dependence of the hybrid excitons. a Two-dimensional PL spectrum of
422
+ CuPc/MoSe2 heterostructure as a function of temperature. b PL spectra of CuPc/MoSe2
423
+ heterostructure in the energy range of 1.4 - 1.95 eV at the temperatures of 78 K, 98 K, 138 K, 178
424
+ K, and 208 K, respectively. c PL spectra of CuPc/MoSe2 heterostructure in the energy range of 1.68-
425
+ 1.95 eV at the temperatures of 78 K, 98 K, 138 K, 178 K, and 208 K, respectively. d The peak
426
+ energy of hX1 (black), hX2 (red), and hX3 (purple) as a function of temperature.
427
+ Temperature-dependent behavior of interfacial hybrid excitons
428
+ The temperature-dependent behavior of the observed hybrid excitons is carefully
429
+ examined from 78 K to 298 K. For CuPc /MoSe2 heterostructure (Fig. 4a), we clearly
430
+ observed a remarkable increase in the whole PL intensity when cooling from room
431
+ temperature (298 K) to low temperature (78 K), which can be explained by the
432
+ suppression of nonradiative recombination51. When the temperature is higher than 178
433
+ K, hX2 becomes undistinguishable and hX1 dominates the PL spectra (Fig. 4b). For
434
+ MoSe2, the trion peak disappears at 98 K and the A exciton becomes dominant
435
+
436
+ a
437
+ 78
438
+ b
439
+ hX
440
+ C
441
+ 6
442
+ hX,
443
+ hX4
444
+ hX3
445
+ 2
446
+ 208 K
447
+ hX3
448
+ 208 K
449
+ max
450
+ 98
451
+ hX4
452
+ 1
453
+ 3
454
+ hX3hX4
455
+ 118
456
+ 0
457
+ 0
458
+ hX
459
+ 148
460
+ 178K
461
+ 178 K
462
+ 2
463
+ 4
464
+ hX4
465
+ 1
466
+ 188
467
+ hX3 hX4
468
+ 2
469
+ (×103)
470
+ 0
471
+ 0
472
+ 228
473
+ 138K
474
+ hx
475
+ 138 K
476
+ min
477
+ intensity
478
+ intensity
479
+ 4
480
+ 278
481
+ 2
482
+ hX
483
+ 1.4
484
+ 1.6
485
+ 1.8
486
+ 2.0
487
+ hX3 hX4
488
+ 2
489
+ Photon energy (eV)
490
+ 0
491
+ d 1.75
492
+ μ20
493
+ xyaxy
494
+ PL
495
+ 0
496
+ 98K
497
+ hX.
498
+ 98 K
499
+ 14
500
+ 1.70
501
+ 10
502
+ hX
503
+ 7
504
+ 4
505
+ (eV)
506
+ hX3
507
+ hX
508
+ hX4
509
+ hX
510
+ 0
511
+ hX2/
512
+ 0
513
+ 40
514
+ hX,
515
+ 78 K
516
+ 78 K
517
+ 18
518
+ 1.60
519
+ ...
520
+ 20
521
+ hX3
522
+ hX3 hX4
523
+ 9
524
+ 0
525
+ 1.55
526
+ 0
527
+ 100
528
+ 150
529
+ 200
530
+ 250
531
+ 300
532
+ 1.4
533
+ 1.6
534
+ 1.8
535
+ 1.71
536
+ 1.80
537
+ 1.89
538
+ Temperature (K)
539
+ Photon energy (eV)
540
+ Photonenergy (eV)14
541
+
542
+ (Supplementary Fig. 6). To our surprise, the peak energy of the hybrid excitons shows
543
+ different temperature dependence (Fig. 4a, d). We first focus on the hX1, hX2, and hX4
544
+ peaks, of which the peak energy shows obvious redshift with increasing temperature
545
+ (Fig. 4b, d) due to the increased electron-phonon interactions51, similar to the
546
+ temperature-dependent behavior of exciton and trion in MoSe2 (Supplementary Fig. 6).
547
+ By fitting the peak energy with the standard semiconductor bandgap model52:𝐸𝑔(0) =
548
+ 𝐸𝑔(𝑇) − 𝑆ℏ𝜔 [𝑐𝑜𝑡ℎ (
549
+ ℏ𝜔
550
+ 2𝑘𝐵𝑇) −1] , where 𝐸𝑔 represents the bandgap, ℏ���� represents
551
+ the phonon energy, 𝑆 represents the electron-phonon coupling strength, and 𝑇
552
+ represents the temperature, we obtain a similar phonon energy for the CuPc/MoSe2
553
+ heterostructure and MoSe2 (Supplementary Fig. 7a, and Supplementary Table 1),
554
+ suggesting that these hybrid excitons are also influenced by the phonons of MoSe2.
555
+ The hX3 peak located at ~1.72 eV is more unique among the four hybrid excitons.
556
+ The peak energy shows a rather weak temperature dependence (Fig 4c, d), which is
557
+ totally different from the other three peaks. Such weak temperature-dependent behavior
558
+ of excitons has been observed in organic molecules, in which the effects of thermal
559
+ expansion and exciton-phonon coupling almost cancel out53, indicating that the hX3
560
+ peak displays the signature of Frenkel excitons in CuPc. However, this peak cannot be
561
+ simply assigned to the emission of the CuPc molecules since no PL signals of CuPc
562
+ film were observed at 78 K (Supplementary Fig. 2b). Furthermore, we can even observe
563
+ this peak in the heterostructure region when we dry transferred monolayer MoSe2 on
564
+ top of the CuPc thin film immediately without any further treatment (Supplementary
565
+ Fig. 8), which can exclude the influence of the CuPc film morphology. This also
566
+
567
+ 15
568
+
569
+ indicates that the coupling at the CuPc/MoSe2 interface is very robust and the hybrid
570
+ excitons can be formed immediately once they are in contact without any further
571
+ treatment. On the other hand, it also combines the character of the Wannier-Mott
572
+ exciton in MoSe2. For example, the peak intensity of hX3 shows similar temperature-
573
+ dependent behavior with the A exciton of MoSe2 (Supplementary Fig. 7b), which
574
+ suggests that it gains a large oscillator strength from MoSe2 and shows a detectable PL
575
+ signal compared with the pure CuPc film. Therefore, the peak energy of hX3 shows the
576
+ signature of the Frenkel excitons in the organic CuPc film and its emission properties
577
+ display the character of Wannier-Mott excitons in the inorganic MoSe2 monolayer,
578
+ which unambiguously reveal the formation of hybrid Frenkel-Wannier-Mott excitons
579
+ at the CuPc/MoSe2 interface.
580
+
581
+
582
+ a
583
+ b
584
+ hX,
585
+ 100°C
586
+ ?200°C
587
+ X
588
+ 100°℃
589
+ 200°C
590
+ anneal
591
+ hX2
592
+ MOSe
593
+ CL
594
+ UPC
595
+ MoSe,/cuPc
596
+ (crvstal
597
+ crystal)
598
+ Normalized PL
599
+ Edge-on
600
+ MoSe,/CuPccrystal region
601
+ C
602
+ hX3
603
+ hX4
604
+ hx
605
+ mixed Face-on
606
+ hX3
607
+ 3mixedFace-on
608
+ and Edge-on
609
+ andEdge-on
610
+ hX
611
+ hX
612
+ hx.
613
+ hx4
614
+ MoSe, region
615
+ Face-on
616
+ Face-on
617
+ hX3
618
+ hX4
619
+ 1.5
620
+ 1.6
621
+ 1.7
622
+ 1.8
623
+ 1.91.5
624
+ 1.6
625
+ 1.7
626
+ 1.8
627
+ 1.9
628
+ 1.7
629
+ 1.8
630
+ 1.9
631
+ Photon energy (ev)
632
+ Photon energy (ev)
633
+ Photonenergy (eV)
634
+ d
635
+ e
636
+ HS(film)
637
+ Face-on
638
+ ('n'e)
639
+ CuPc
640
+ hybridization
641
+ MoSe,
642
+ Raman intensity
643
+ ?
644
+ h
645
+ Hybrid Exciton
646
+ HybridExciton
647
+ CuPc
648
+ 200°℃
649
+ Mixed
650
+ Face-on
651
+ (h)
652
+ Intralayer Exciton
653
+ Interlayer
654
+ HS(crystal)Edge-on
655
+ MoSe,region
656
+ 100℃
657
+ @h
658
+ 200240
659
+ 1380
660
+ 1610
661
+ 100°C
662
+ 200°C
663
+ Ramanshift(cm-1)
664
+ Annealingtemperature16
665
+
666
+ Fig. 5 | Molecular orientation-dependent hybrid excitons. a The PL spectra in the MoSe2 region
667
+ and MoSe2/CuPc crystal region for the same MoSe2/CuPc crystal heterostructure after annealing at
668
+ 100°C and 200°C. After annealing at 200°C, the CuPc crystal partially decomposes and the CuPc
669
+ molecules can migrate on MoSe2, which lead to the formation hybrid excitons in both regions. b
670
+ AFM topographic image of the MoSe2/CuPc crystal heterostructure after annealing at 100℃ and
671
+ 200℃. c Enlarged view of the image in the gray dotted box in a. d Raman spectra of the MoSe2
672
+ region in the MoSe2/CuPc crystal heterostructure sample after annealing at 100°C and 200°C. e
673
+ Schematic illustration of the relationship between molecular orientation (face-on, edge-on) of CuPc
674
+ and interlayer hybridization.
675
+ Tailoring the hybrid exciton using molecular orientation
676
+ The molecular orientation is introduced as a new degree of freedom to modulate the
677
+ interfacial hybridization strength, which will further tailor the interfacial hybrid
678
+ excitons. In general, the CuPc molecule tends to adopt a face-on orientation on the
679
+ MoSe2 surface that allows efficient interfacial hybridization, as revealed by our
680
+ theoretical calculation. In contrast, the edge-on orientation will experience insufficient
681
+ interfacial hybridization due to the larger interfacial distance. To demonstrate that the
682
+ molecular orientation can be used to tailor the interfacial hybrid exciton states, we
683
+ carefully prepared MoSe2/CuPc film heterostructure and MoSe2/CuPc crystal
684
+ heterostructure simultaneously by the dry transfer method. Since the CuPc molecule
685
+ stacks randomly in the CuPc film, it easily adopts a face-on orientation on the MoSe2
686
+ surface. However, because the CuPc molecule shows a herringbone stacking in the
687
+ crystal54, it can only adopt an edge-on orientation on MoSe2 surface before crystal
688
+ decomposition. Because the monolayer MoSe2 partially covers the CuPc crystal, we
689
+ could compare the measurements from the MoSe2 region and MoSe2/CuPc crystal
690
+
691
+ 17
692
+
693
+ region in the same sample (Supplementary Fig. 9a). After annealing simultaneously at
694
+ 100°C, the PL spectra of MoSe2/CuPc film heterostructure and MoSe2/CuPc crystal
695
+ heterostructure display great contrast as expected. The PL of crystal heterostructure
696
+ shows similar spectral features and slight quenching compared with monolayer MoSe2
697
+ (Fig. 5a and Supplementary Fig. 9b), indicating a weak interfacial hybridization
698
+ strength. However, the PL of film heterostructure presents obvious quenching and the
699
+ formation of hybrid excitons (Supplementary Fig. 9b, c). The influence of the
700
+ morphology of CuPc can be excluded since the surface of CuPc crystal is flatter than
701
+ that of the CuPc film (Supplementary Fig. 10). Therefore, the above results suggest that
702
+ the molecular orientation can be used to tune the interlayer hybridization strength and
703
+ further tailor the interfacial hybrid excitons.
704
+ To confirm this deduction, the MoSe2/CuPc crystal heterostructure is further
705
+ annealed at 200oC to decompose the CuPc crystal. When the CuPc crystal decomposes,
706
+ the CuPc molecules can easily adopt a face-on orientation on the MoSe2 surface, thus,
707
+ the interfacial hybrid excitons should also be observed. After annealing at 200°C, the
708
+ PL spectra display obvious quenching in both the MoSe2/CuPc crystal region and
709
+ MoSe2 region (Supplementary Fig. 9d), and clearly shows the formation of hybrid
710
+ excitons (Fig. 5a, c), which indicates that the molecular orientation is changed from
711
+ edge-on to face-on after CuPc crystal decomposition. The decomposition of the CuPc
712
+ crystal is confirmed by AFM, as shown in Fig. 5b. It is obvious that the CuPc crystal
713
+ partially decomposes after annealing at 200°C, as evidenced by the change in the AFM
714
+ height profile. Note that we can even observe hybrid excitons in the MoSe2 region
715
+
716
+ 18
717
+
718
+ because the CuPc molecules can migrate on MoSe2 surface after the decomposition of
719
+ CuPc crystal. The Raman spectra further confirms this, as the Raman peaks of both
720
+ MoSe2 and CuPc molecules appear in the MoSe2 region after annealing at 200°C (Fig.
721
+ 5d). These results unambiguously show that we can successfully tailor the interfacial
722
+ hybrid excitons by changing the molecular orientation (Fig. 5e). The theoretical
723
+ calculation also supports our results. As shown in the electronic structure of
724
+ heterostructure at the edge-on orientation (Supplementary Fig. 11), we observe no
725
+ obvious interfacial hybridization, which coincides with the observed phenomena in
726
+ MoSe2/CuPc crystal heterostructure.
727
+ Discussion
728
+ We have demonstrated the formation of interfacial hybrid excitons in CuPc/MoSe2
729
+ heterostructure due to the hybridization between CuPc and MoSe2. The observed
730
+ phenomenon is unusual as the coupling at the organic-inorganic interface is generally
731
+ weak24. The first principles calculations rationalize the results as the LUMO of CuPc
732
+ strongly hybridized with the CBM of MoSe2, which leads to the emergence of new
733
+ eigenstates. The new hybrid excitons consist of electrons delocalized in both layers and
734
+ holes confined in individual layer, enabling simultaneous large optical and electrical
735
+ dipoles7. The temperature-dependent behavior suggests that the hybrid excitons
736
+ simultaneously gain the signature of the Wannier-Mott excitons in MoSe2 and the
737
+ Frenkel excitons in CuPc. The excellent agreement between the theoretical and
738
+ experimental results not only validates the observed strong coupling phenomenon, but
739
+ also provides a basis for manipulating hybrid excitons at the organic-inorganic interface.
740
+
741
+ 19
742
+
743
+ For instance, the large electrical dipole in the out-of-plane direction can be used to
744
+ achieve novel electrical control of the hybrid excitons55,56. Our result is also of great
745
+ importance for realizing tunable interlayer hybridization strength by changing the
746
+ molecular orientation, which can be used to tailor the exciton states at the organic-
747
+ inorganic interface. In conclusion, we report the formation of interfacial hybrid excitons
748
+ with strong molecular orientation dependence that originate from the hybridization
749
+ between CuPc and MoSe2, which is meaningful for many-body exciton physics at the
750
+ organic-inorganic interface.
751
+
752
+ Methods
753
+ Sample preparation
754
+ (1) Construction of the CuPc film /MoSe2 heterostructure
755
+ Monolayer MoSe2 was mechanically exfoliated on a SiO2/Si substrate from bulk
756
+ crystals and further annealed in vacuum at 200 °C to remove surface contaminants. The
757
+ thickness of MoSe2 was confirmed by optical contrast, atomic force microscopy (AFM),
758
+ and Raman measurements (Supplementary Fig. 1). To construct the CuPc (film)/MoSe2
759
+ heterostructure, CuPc thin film was directly deposited on top of monolayer MoSe2 using
760
+ thermal evaporation in vacuum (home-built evaporator). The heating current was
761
+ maintained at 5 amperes and the average evaporation speed was 0.25 nm/min.
762
+ (2) Construction of the MoSe2/CuPc film heterostructure
763
+ The CuPc film was firstly thermally evaporated on a SiO2/Si substrate using the same
764
+ conditions as (1). Then, monolayer MoSe2 was mechanically exfoliated on the PDMS
765
+
766
+ 20
767
+
768
+ substrate from bulk crystals, and further transferred on top of the CuPc film using the
769
+ dry transfer method.
770
+ (3) Construction of the MoSe2/CuPc crystal heterostructure
771
+ The single crystals of CuPc were grown by the physical vapor deposition (PVT) method
772
+ in a quartz tube with a hot zone temperature of 400°C. To construct the MoSe2/CuPc
773
+ (crystal) heterostructure, monolayer MoSe2 was mechanically exfoliated on a PDMS
774
+ substrate, and further transferred on top of a CuPc single crystal using the dry transfer
775
+ method.
776
+ Low temperature PL Measurements.
777
+ The measurements at 78 K were conducted in a temperature-controlled cryostat
778
+ (THMS600, Linkam) with a diffraction-limited excitation beam diameter of 1µm.The
779
+ signal was collected using a 50X long-working distance objective and detected on a
780
+ commercial Renishaw inVia spectrometer. The excitation power was selected to be
781
+ below 200 µW to avoid heating damage to the sample. The measurements at 4 K were
782
+ conducted in a temperature-controlled cryostat (Montana Instruments) with an
783
+ excitation beam diameter of 1µm. The signal was collected using a 100X objective and
784
+ detected on a commercial Ocean Optics spectrometer.
785
+ Theoretical calculation. First-principles calculations were carried out based on the
786
+ density functional theory (DFT) framework by utilizing the Vienna Ab initio Simulation
787
+ Package (VASP) 5.4.4 package57,58. Pseudopotentials were used to describe the
788
+ electron-ion interactions within the PAW approach and generalized gradient
789
+ approximations (GGA) of Perdew-Burke-Ernzerhof (PBE) were adopted for the
790
+
791
+ 21
792
+
793
+ exchange-correlation potential59-61. To better describe the interlayer van der Waals
794
+ (vdW) interactions, we adopt optB88-vdW corrections for the optimization of
795
+ structures62. The electron wave functions are expanded on a plane-wave basis set with
796
+ an energy cutoff of 520 eV. The atomic coordinates of all structures were allowed to
797
+ relax until the forces acting on the ions were less than 0.01 eV Å-1. The convergence
798
+ criterion for the electronic self-consistent cycle is fixed at 1×10-5 eV. The integrations
799
+ in the reduced Brillouin zone are performed on a 3×3×1 Monkhorst-Pack special k-
800
+ points for optimization and self-consistent calculations63,64. A vacuum slab above 15 Å
801
+ was used in all calculations to avoid interlayer interactions. The CuPc/MoSe2
802
+ heterostructure is modeled by adsorbing one CuPc molecule on a 5×3√3 supercell of
803
+ MoSe2, which can be written as Mo30Se60C32N8H16Cu. The lattice parameters of the
804
+ CuPc/MoSe2 heterostructure were calculated to be a = 16.62 Å, b = 17.27 Å, and
805
+ α=β=γ=90°. The interlayer distance between CuPc and MoSe2 substrate is
806
+ approximately 3.36 Å.×
807
+
808
+ Data availability
809
+ The data that support the findings of this study are available from the corresponding
810
+ authors upon reasonable request.
811
+
812
+ References
813
+ 1.
814
+ Kleemans, N.A.J.M., et al. Many-body exciton states in self-assembled quantum dots coupled
815
+ to a Fermi sea. Nat. Phys. 6, 534-538 (2010).
816
+ 2.
817
+ Shimazaki, Y., Schwartz, I., Watanabe, K., Taniguchi, T., Kroner, M. & Imamoglu, A. Strongly
818
+ correlated electrons and hybrid excitons in a moire heterostructure. Nature 580, 472-477 (2020).
819
+ 3.
820
+ Dalgarno, P.A., et al. Optically induced hybridization of a quantum dot state with a filled
821
+
822
+ 22
823
+
824
+ continuum. Phys. Rev. Lett. 100, 176801 (2008).
825
+ 4.
826
+ Stinaff, E.A., et al. Optical signatures of coupled quantum dots. Science 311, 636-639 (2006).
827
+ 5.
828
+ Krenner, H.J., et al. Optically probing spin and charge interactions in a tunable artificial
829
+ molecule. Phys. Rev. Lett. 97, 076403 (2006).
830
+ 6.
831
+ Alexeev, E.M., et al. Resonantly hybridized excitons in moire superlattices in van der Waals
832
+ heterostructures. Nature 567, 81-86 (2019).
833
+ 7.
834
+ Hsu, W.T., et al. Tailoring excitonic states of van der Waals bilayers through stacking
835
+ configuration, band alignment, and valley spin. Sci .Adv. 5, eaax7407 (2019).
836
+ 8.
837
+ Brotons-Gisbert, M., et al. Coulomb blockade in an atomically thin quantum dot coupled to a
838
+ tunable Fermi reservoir. Nat. Nanotechnol. 14, 442-446 (2019).
839
+ 9.
840
+ Mak, K.F., Lee, C., Hone, J., Shan, J. & Heinz, T.F. Atomically thin MoS2: a new direct-gap
841
+ semiconductor. Phys. Rev. Lett. 105, 136805 (2010).
842
+ 10.
843
+ Ugeda, M.M., et al. Giant bandgap renormalization and excitonic effects in a monolayer
844
+ transition metal dichalcogenide semiconductor. Nat. Mater. 13, 1091-1095 (2014).
845
+ 11.
846
+ Wang, G., et al. Colloquium: Excitons in atomically thin transition metal dichalcogenides. Rev.
847
+ Mod. Phys. 90 (2): 021001 (2018).
848
+ 12.
849
+ Regan, E.C., et al. Emerging exciton physics in transition metal dichalcogenide heterobilayers.
850
+ Nat. Rev. Mater. 1-18 (2022).
851
+ 13.
852
+ Huang, D., Choi, J., Shih, C.K. & Li, X. Excitons in semiconductor moire superlattices. Nat.
853
+ Nanotechnol. 17, 227-238 (2022).
854
+ 14.
855
+ Wilson, N.P., Yao, W., Shan, J. & Xu, X. Excitons and emergent quantum phenomena in stacked
856
+ 2D semiconductors. Nature 599, 383-392 (2021).
857
+ 15.
858
+ Xiao, D., Liu, G.B., Feng, W., Xu, X. & Yao, W. Coupled spin and valley physics in monolayers
859
+ of MoS2 and other group-VI dichalcogenides. Phys. Rev. Lett. 108, 196802 (2012).
860
+ 16.
861
+ Rivera, P., et al. Valley-polarized exciton dynamics in a 2D semiconductor heterostructure.
862
+ Science 351, 688-691 (2016).
863
+ 17.
864
+ Rivera, P., Yu, H., Seyler, K.L., Wilson, N.P., Yao, W. & Xu X. Interlayer valley excitons in
865
+ heterobilayers of transition metal dichalcogenides. Nat. Nanotechnol. 13, 1004-1015 (2018).
866
+ 18.
867
+ Rivera, P., et al. Observation of long-lived interlayer excitons in monolayer MoSe2-WSe2
868
+ heterostructures. Nat. Commun. 6, 6242 (2015).
869
+ 19.
870
+ Jin, C., et al. Observation of moire excitons in WSe2/WS2 heterostructure superlattices. Nature
871
+ 567, 76-80 (2019).
872
+ 20.
873
+ Seyler, K.L., et al. Signatures of moiré-trapped valley excitons in MoSe2/WSe2 heterobilayers.
874
+ Nature 567, 66-70 (2019).
875
+ 21.
876
+ Tran, K., et al. Evidence for moiré excitons in van der Waals heterostructures. Nature, 567, 71-
877
+ 75 (2019).
878
+ 22.
879
+ Ulman, K. & Quek, S.Y. Organic-2D material heterostructures: a promising platform for exciton
880
+ Condensation and multiplication. Nano Lett. 21, 8888-8894 (2021).
881
+ 23.
882
+ Agranovich, V.M., Basko, D.M., Rocca, G.C.L. & Bassani, F. Excitons and optical
883
+ nonlinearities in hybrid organic-inorganic nanostructures. J. Phys.: Condens. Matter 10, 9369-
884
+ 9400 (1998).
885
+ 24.
886
+ Agranovich, V.M., Gartstein, Y.N. & Litinskaya, M. Hybrid resonant organic-inorganic
887
+ nanostructures for optoelectronic applications. Chem. Rev. 111, 5179-5214 (2011).
888
+ 25.
889
+ Blumstengel, S., Sadofev, S., Xu, C., Puls, J. & Henneberger, F. Converting Wannier into
890
+
891
+ 23
892
+
893
+ Frenkel excitons in an inorganic/organic hybrid semiconductor nanostructure. Phys. Rev. Lett.
894
+ 97, 237401 (2006).
895
+ 26.
896
+ Gobbi, M., Orgiu, E. & Samori, P. When 2D materials meet molecules: opportunities and
897
+ challenges of hybrid organic/inorganic van der Waals Heterostructures. Adv. Mater. 30,
898
+ e1706103 (2018).
899
+ 27.
900
+ Huang, Y.L., Zheng, Y.J., Song, Z., Chi, D., Wee, A.T.S. & Quek, S.Y. The organic-2D transition
901
+ metal dichalcogenide heterointerface. Chem. Soc. Rev. 47, 3241-3264 (2018).
902
+ 28.
903
+ Cho, K., Pak, J., Chung, S. & Lee, T. Recent advances in interface engineering of transition-
904
+ metal dichalcogenides with organic molecules and polymers. ACS Nano 13, 9713-9734 (2019).
905
+ 29.
906
+ He, D., et al. Two-dimensional quasi-freestanding molecular crystals for high-performance
907
+ organic field-effect transistors. Nat. Commun. 5, 5162 (2014).
908
+ 30.
909
+ Wang, S., et al. A MoS2 /PTCDA hybrid heterojunction synapse with efficient photoelectric dual
910
+ modulation and versatility. Adv. Mater. 31, e1806227 (2019).
911
+ 31.
912
+ Ding, J., Fu, S., Hu, K., Zhang, G., Liu, M., Zhang, X., Wang, R. & Qiu, X. Efficient hot electron
913
+ capture in CuPc/MoSe2 heterostructure assisted by intersystem crossinng. Nano Lett. 22, 8463-
914
+ 8469 (2022).
915
+ 32.
916
+ Homan, S.B., Sangwan, V.K., Balla, I., Bergeron, H., Weiss, E.A. & Hersam, M.C. Ultrafast
917
+ exciton dissociation and long-lived charge separation in a photovoltaic pentacene-MoS2 van der
918
+ Waals heterojunction. Nano Lett. 17, 164-169 (2017).
919
+ 33.
920
+ Zhong, C., Sangwan, V.K., Wang, C., Bergeron, H., Hersam, M.C. & Weiss, E.A. Mechanisms
921
+ of ultrafast charge separation in a PTB7/monolayer MoS2 van der Waals Heterojunction. J. Phys.
922
+ Chem. Lett. 9, 2484-2491 (2018).
923
+ 34.
924
+ Padgaonkar, S., et al. Molecular-orientation-dependent interfacial charge transfer in
925
+ phthalocyanine/MoS2 mixed-dimensional heterojunctions. J. Phys. Chem. C 123, 13337-13343
926
+ (2019).
927
+ 35.
928
+ Kafle, T.R., Kattel, B., Lane, S.D., Wang, T., Zhao, H. & Chan, W.L. Charge transfer exciton
929
+ and spin flipping at organic-transition-metal dichalcogenide interfaces. ACS Nano 11, 10184-
930
+ 10192 (2017).
931
+ 36.
932
+ Choi, J., Zhang, H. & Choi, J.H. Modulating optoelectronic properties of two-dimensional
933
+ transition metal dichalcogenide semiconductors by photoinduced charge transfer. ACS Nano 10,
934
+ 1671-1680 (2016).
935
+ 37.
936
+ Aristov, V.Y., et al. Electronic structure of the organic semiconductor copper phthalocyanine:
937
+ experiment and theory. J. Chem. Phys. 128, 034703 (2008).
938
+ 38.
939
+ Ross, J.S., et al. Electrical control of neutral and charged excitons in a monolayer semiconductor.
940
+ Nat. Commun. 4, 1474 (2013).
941
+ 39.
942
+ Caplins, B.W., Mullenbach, T.K., Holmes, R.J. & Blank, D.A. Femtosecond to nanosecond
943
+ excited state dynamics of vapor deposited copper phthalocyanine thin films. Phys. Chem. Chem.
944
+ Phys. 18, 11454-11459 (2016).
945
+ 40.
946
+ Shi, N. & Ramprasad, R. Dielectric properties of Cu-phthalocyanine systems from first
947
+ principles. Appl. Phys. Lett. 89, 102904 (2006).
948
+ 41.
949
+ Robertson, J. High dielectric constant oxides. Eur.Phys.J-Appl. Phys. 28, 265-291 (2004).
950
+ 42.
951
+ Godde, T., et al. Exciton and trion dynamics in atomically thin MoSe2 and WSe2: effect of
952
+ localization. Phys. Rev. B 94, 165301 (2016).
953
+ 43.
954
+ Anghel, S., Passmann, F., Ruppert, C., Bristow, A.D. & Betz, M. Coupled exciton-trion spin
955
+
956
+ 24
957
+
958
+ dynamics in a MoSe2 monolayer. 2D Mater. 5, 045024 (2018).
959
+ 44.
960
+ Ruppert, C., Chernikov, A., Hill, H.M., Rigosi, A.F. & Heinz, T.F. The role of electronic and
961
+ phononic excitation in the optical response of monolayer WS2 after ultrafast excitation. Nano
962
+ Lett. 17, 644-651 (2017).
963
+ 45.
964
+ Zhu, T., et al. Highly mobile charge-transfer excitons in two-dimensional WS2/tetracene
965
+ heterostructures. Sci. Adv. 4, eaao3104 (2018).
966
+ 46.
967
+ Robert, C., et al. Measurement of the spin-forbidden dark excitons in MoS2 and MoSe2
968
+ monolayers. Nat. Commun. 11, 4037 (2020).
969
+ 47.
970
+ Schmidt, T., Lischka, K. & Zulehner, W. Excitation-power dependence of the near-band-edge
971
+ photoluminescence of semiconductors. Phys. Rev. B Condens. Matter 45, 8989-8994 (1992).
972
+ 48.
973
+ Li, Z., et al. Revealing the biexciton and trion-exciton complexes in BN encapsulated WSe2.
974
+ Nat. Commun. 9, 3719 (2018).
975
+ 49.
976
+ Kylänpää, I. & Komsa, H.P. Binding energies of exciton complexes in transition metal
977
+ dichalcogenide monolayers and effect of dielectric environment. Phys. Rev. B 92, 205418 (2015).
978
+ 50.
979
+ Tongay, S., et al. Defects activated photoluminescence in two-dimensional semiconductors:
980
+ interplay between bound, charged, and free excitons. Sci. Rep. 3, 2657 (2013).
981
+ 51.
982
+ Li, D., et al. Large‐area 2D/3D MoS2–MoO2 heterostructures with thermally stable exciton and
983
+ intriguing electrical transport behaviors. Adv. Electron. Mater. 3, 1600335 (2017).
984
+ 52.
985
+ O’Donnell, K.P. & Chen, X. Temperature dependence of semiconductor band gaps. Appl. Phys.
986
+ Lett. 58, 2924-2926 (1991).
987
+ 53.
988
+ Alvertis, A.M., et al. Impact of exciton delocalization on exciton-vibration interactions in
989
+ organic semiconductors. Phys. Rev. B 102, 081122 (2020).
990
+ 54.
991
+ Zou, T., et al. Controllable molecular packing motif and overlap type in organic nanomaterials
992
+ for advanced optical properties. Crystals 8, 22 (2018).
993
+ 55.
994
+ Peimyoo, N., et al. Electrical tuning of optically active interlayer excitons in bilayer MoS2. Nat.
995
+ Nanotechnol. 16, 888-893 (2021).
996
+ 56.
997
+ Leisgang, N., et al. Giant Stark splitting of an exciton in bilayer MoS2. Nat. Nanotechnol. 15,
998
+ 901-907 (2020).
999
+ 57.
1000
+ Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio total-energy calculations
1001
+ using a plane-wave basis set. Phys. Rev. B 54, 11169-11186 (1996).
1002
+ 58.
1003
+ Kresse, G., & Hafner, J. Ab initio molecular dynamics for liquid metals. Phys. Rev. B 47, 558-
1004
+ 561 (1993).
1005
+ 59.
1006
+ Perdew, J.P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys.
1007
+ Rev. Lett. 77, 3865-3868 (1996).
1008
+ 60.
1009
+ Blöchl, P.E. Projector augmented-wave method. Phys. Rev. B 50, 17953-17979 (1994).
1010
+ 61.
1011
+ Kresse, G., & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave
1012
+ method. Phys. Rev. B 59, 1758-1775 (1999).
1013
+ 62.
1014
+ Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. A consistent and accurate ab initio
1015
+ parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu.
1016
+ J. Chem. Phys. 132, 154104 (2010).
1017
+ 63.
1018
+ Monkhorst, H.J. & Pack, J.D. Special points for Brillouin-zone integrations. Phys Rev B 13,
1019
+ 5188-5192 (1976).
1020
+ 64.
1021
+ Methfessel, M. & Paxton, A.T. High-precision sampling for Brillouin-zone integration in metals.
1022
+ Phys. Rev. B 40, 3616-3621 (1989).
1023
+
1024
+ 25
1025
+
1026
+
1027
+ Acknowledgements
1028
+ This work was supported by the National Nature Science Foundation of China (Grant
1029
+ Nos. 11974088, 12074116, 21790353, 61875236, 61975007), the National Key
1030
+ Research and Development Program of China (Grant Nos.2016YFA0202302,
1031
+ 2017YFA0205000, 2021YFA1400201), the Strategic Priority Research Program of
1032
+ CAS (Grant No. XDB30000000), the CAS Project for Young Scientists in Basic
1033
+ Research (Grant No. YSBR-059).
1034
+ Author contributions
1035
+ X.Z., X.Q. and D.T. conceived the idea; S.F. and J.D. prepared the samples and
1036
+ conducted all the optical measurements and the corresponding data analysis; X.W. and
1037
+ H.L. performed the DFT calculations; This manuscript was prepared primarily by X.Z.,
1038
+ S.F. and J.D., and all authors contributed to discussing and commenting on the paper.
1039
+ Competing interests
1040
+ The authors declare no competing interests
1041
+ Additional information
1042
+ Correspondence and requests for materials should be addressed to Xiaoxian Zhang.
1043
+
dtE0T4oBgHgl3EQfoQFr/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
dtE1T4oBgHgl3EQfyAVc/content/tmp_files/2301.03428v1.pdf.txt ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Regularized Optimal Mass Transport with Nonlinear Diffusion
2
+ Kaiming Xu, Xinan Chen, Helene Benveniste, Allen Tannenbaum ∗†‡§
3
+ January 10, 2023
4
+ Abstract
5
+ In this paper, we combine nonlinear diffusion with the regularized optimal mass
6
+ transport (rOMT) model. As we will demonstrate, this new approach provides further
7
+ insights into certain applications of fluid flow analysis in the brain. From the point
8
+ of view of image processing, the anisotropic diffusion method, based on Perona-Malik,
9
+ explicitly considers edge information. Applied to rOMT analysis of glymphatic trans-
10
+ port based on dynamic contrast-enhanced magnetic resonance imaging data, this new
11
+ framework appears to capture a larger advection-dominant volume.
12
+ 1
13
+ Introduction
14
+ The theory of optimal mass transport(OMT) was first proposed by Gaspard Monge in 1781
15
+ and has since evolved into a unique scientific field which has had significant impact on
16
+ research in many disciplines [22, 23]. Mass transport theory has been applied to diverse
17
+ fields including physics, biology, economics and engineering. OMT defines a distance called
18
+ the Wasserstein distance, and thus creates a natural geometry on the space of probability
19
+ distributions.
20
+ Our study is based on a fluid dynamics reformulation of OMT [1] which
21
+ allows us to calculate the flow fields between two density distributions.
22
+ Regularized optimal mass transport (rOMT), an extension of fluid dynamics reformulation
23
+ of OMT, is a tool to study temporal flow fields as a physically inspired model of optical flow.
24
+ It has the ability to capture the flow dynamics, handle noise and simulate diffusion [3, 5, 9].
25
+ rOMT utilizes an advection-diffusion equation as its flow-driven partial different equation
26
+ and is endpoint free. A source term may be added to rOMT in which case the total mass
27
+ preservation condition can be circumvented. This line of research will be pursued in other
28
+ work.
29
+ Anisotropic diffusion, a major tool for image segmentation, edge detection and image de-
30
+ noising, was first proposed by Perona and Malik [17]. Notably, instead of using a constant
31
+ diffusion coefficient, Perona and Malik considered a nonnegative function (conductivity
32
+ ∗K. Xu is with the Department of Applied Mathematics & Statistics, Stony Brook University, NY; email:
33
34
+ †X. Chen is with the Department of Medical Physics, Memorial Sloan Kettering Cancer Center, NY
35
+ ‡H. Benveniste is with the Department of Anesthesiology, Yale School of Medicine, CT
36
+ §A. Tannenbaum is with the Departments of Computer Science and Applied Mathematics & Statistics,
37
+ Stony Brook University, NY; email: [email protected]
38
+ 1
39
+ arXiv:2301.03428v1 [physics.flu-dyn] 3 Dec 2022
40
+
41
+ coefficient) of the magnitude of the local density gradient; see equation (8). The authors
42
+ suggested two possible conductivity coefficients (see (9) and (10)), wherein the diffusion will
43
+ be very small near the edges, i.e. reflecting the fact that near edges images tend to have
44
+ very large intensity gradients. In this work, we show that anisotropic diffusion enhances
45
+ the interpretation of glymphatic dynamic contrast-enhanced magnetic resonance imaging
46
+ (DCE-MRI) flow data and may be used in conjunction with the constant diffusion coefficient
47
+ approach [3]. The anisotropic diffusion equation may be derived via the steepest descend
48
+ method for solving an energy minimization problem [25].
49
+ The glymphatic system is involved in transporting waste products from the brain to the
50
+ meningeal lymphatic system which connects to the cervical lymph nodes [14]. The function-
51
+ ing of the glymphatic and lymphatic systems decrease with age and have been implicated in
52
+ the pathophysiology of a wide range of neurodegenerative diseases including cerebral amy-
53
+ loid angiopathy [3, 24] and Alzheimer’s disease [4, 10, 13, 16]. We study glymphatic trans-
54
+ port using a temporal series of DCE-MRI data acquired from the rodent brain [6, 11, 12].
55
+ Since the data are acquired at discrete time points, our work is motivated by the need to
56
+ find a dynamic physically based model of the transport. Several different versions of OMT
57
+ [18] and rOMT [3, 5, 9] have been used to model the glymphatic flow.
58
+ In the present work, we propose a new version of rOMT. Specifically, we replace the lin-
59
+ ear diffusion in rOMT [3, 5, 9] with the Perona-Malik based anisotropic diffusion. Here,
60
+ we argue that this gives us enhanced flexibility to study image-based flows inherent to
61
+ glymphatic transport. Notably, many diffusion processes in fluids are better captured by
62
+ nonlinear models, e.g., axisymmetric surface diffusion [2] and thin fluid films [7, 8]. We
63
+ utilize Lagrangian coordinates for visualizing the glymphatic transport pathlines. Several
64
+ properties of solute particle movement are computed along the pathlines such as speed and
65
+ the P´eclet number. Here we compare various parameters of the anisotropic diffusion coef-
66
+ ficient, and observe the impact of different values on several data metrics including P´eclet
67
+ plots which can map diffusion dominated versus advection dominated regions of the brain.
68
+ We briefly summarize the contents of the present paper. In Section 2, we review the theory
69
+ of OMT, rOMT and nonlinear diffusion. Section 3 introduces the algorithm and numerical
70
+ methods we employ for our current work. In Section 4, we explicate the application of the
71
+ model to glymphatic DCE-MRI data and analyze the experimental results and we conclude
72
+ our paper in Section 5.
73
+ 2
74
+ Model
75
+ 2.1
76
+ OMT
77
+ In this section, we introduce OMT and its fluid dynamics formulation. All the technical
78
+ details as well as a complete set of references may be found in [22, 23].
79
+ The original
80
+ formulation of OMT was given by Gaspard Monge and may be expressed as
81
+ inf
82
+ T {
83
+
84
+
85
+ c(x, T(x))ρ0(x)dx | T#ρ0 = ρ1},
86
+ (1)
87
+ where c(x, y) is the cost function of moving the unit mass from x to y, ρ0 and ρ1 are two
88
+ probability distributions in the domain Ω ⊆ Rd, T is the transport map, and T# is the
89
+ 2
90
+
91
+ push-forward of T. This formulation assumes that ρ0 and ρ1 have the same total mass, i.e.
92
+
93
+ Ω ρ0(x)dx =
94
+
95
+ Ω ρ1(x)dx and then seeks for the optimal transport map T to minimize the
96
+ total cost, the integral in equation (1), subject to the push-forward constraint.
97
+ Later, Leonid Kantorovich formulated a relaxed version of OMT as follows:
98
+ inf
99
+ π∈Π(ρ0,ρ1)
100
+
101
+ Ω×Ω
102
+ c(x, y)π(dx, dy),
103
+ (2)
104
+ where Π(ρ0, ρ1) denotes the set of all couplings (joint distributions) between the marginals
105
+ ρ0 and ρ1. From here on, the cost function c will be taken as the square of the Euclidean
106
+ distance c(x, y) = ∥x − y∥2.
107
+ Benemou and Brenier [1] proved that for c(x, y) = ∥x − y∥2, the specific infimum of Monge-
108
+ Kantorovich formulation is equal to the result in following fluid dynamics formulation for
109
+ density/probability distributions with compact support:
110
+ inf
111
+ ρ,v
112
+ � 1
113
+ 0
114
+
115
+
116
+ ρ(t, x)|v(t, x)|2dxdt,
117
+ (3)
118
+ ∂ρ
119
+ ∂t + ∇ · (ρv) = 0,
120
+ (4)
121
+ ρ(0, x) = ρ0(x),
122
+ ρ(1, x) = ρ1(x),
123
+ (5)
124
+ where ρ : [0, 1]×Ω → R≥0 is the family of density/probability distributions defining geodesic
125
+ path from ρ0 to ρ1, and v : [0, 1] × Ω → Rd is the velocity vector field.
126
+ 2.2
127
+ rOMT
128
+ The regularized OMT model (rOMT) [5, 9] adds two assumptions: 1. the image data we
129
+ use are noisy observations and thus we do not want to make the final density we calculate
130
+ coincide with the MR images; and 2. the flow is driven by an advection-diffusion equation.
131
+ Based on these two assumptions, the rOMT formulation may be written as:
132
+ inf
133
+ ρ,v
134
+ � 1
135
+ 0
136
+
137
+
138
+ ρ(t, x)|v(t, x)|2dxdt + β
139
+
140
+
141
+ (ρ(1, x) − ρ1(x))2dx,
142
+ (6)
143
+ ∂ρ
144
+ ∂t + ∇ · (ρv) = ∇ · (σ0∇ρ),
145
+ (7)
146
+ ρ(0, x) = ρ0(x).
147
+ In this formulation, the final marginal condition is removed and a penalty of the error
148
+ between final density and ground truth is added in the objective function (6), where β is
149
+ the penalty parameter. Equation (7) is an advection-diffusion equation with a constant σ0
150
+ denoting the diffusion coefficient.
151
+ 2.3
152
+ Nonlinear diffusion
153
+ Instead of using linear diffusion in which σ0 is a constant, nonlinear diffusion seems to have
154
+ certain advantages that we will now describe. Perona and Malik proposed an anisotropic
155
+ 3
156
+
157
+ diffusion [17], which is a useful tool for image segmentation, edge detection and image
158
+ denoising. The anisotropic diffusion equation is
159
+ ∂ρ
160
+ ∂t = ∇ · (σ(|∇ρ|)∇ρ),
161
+ (8)
162
+ where σ(·) is a nonnegative strictly decreasing function. If we consider a 3D problem, then
163
+ |∇ρ| =
164
+
165
+ ρ2x + ρ2y + ρ2z. The proper diffusion should be large in smooth homogeneous areas
166
+ and become smaller near edges, the places where |∇ρ| is large.
167
+ Perona and Malik [17]
168
+ suggested two versions of the diffusion (conductivity) coefficient:
169
+ σ(x) = σ0
170
+ 1
171
+ 1 + ( x
172
+ K )2 ,
173
+ (9)
174
+ σ(x) = σ0e−( x
175
+ K )2.
176
+ (10)
177
+ Both are 0 when x approaches ∞ and attend upper bound σ0 while x = 0. K is a constant
178
+ and controls the sensitivity to edges and can be tuned for different applications.
179
+ Following [25], we may derive the anisotropic diffusion equation (8) via the steepest descent
180
+ from an energy minimization problem. More precisely, considering the following minimiza-
181
+ tion problem:
182
+ min
183
+
184
+
185
+ f(|∇ρ|)dΩ,
186
+ (11)
187
+ then the steepest descend equation may be computed to be
188
+ ∂ρ
189
+ ∂t = ∇ · (f′(|∇ρ| ∇ρ
190
+ |∇ρ|)).
191
+ (12)
192
+ Obviously, (12) is identical to (8) if
193
+ f′(x) = xσ(x).
194
+ (13)
195
+ For example, the corresponding f function of σ function (9) is
196
+ f(x) = σ0K2
197
+ 2
198
+ ln[1 + ( x
199
+ K )2]
200
+ (14)
201
+ 2.4
202
+ rOMT with nonlinear diffusion
203
+ In this section, we present our new rOMT formulation. We replace the diffusion in (7) by
204
+ anisotropic diffusion in (8) and obtain the following formulation:
205
+ inf
206
+ ρ,v
207
+ � 1
208
+ 0
209
+
210
+
211
+ ρ(t, x)|v(t, x)|2dxdt + β
212
+
213
+
214
+ (ρ(1, x) − ρ1(x))2dx,
215
+ ∂ρ
216
+ ∂t + ∇ · (ρv) = ∇ · (σ(|∇ρ|)∇ρ),
217
+ (15)
218
+ ρ(0, x) = ρ0(x).
219
+ One may employ various versions of the σ function and in this work, we choose the function
220
+ given in (9). Note that, there are two parameters σ0 and K which may be tuned based on
221
+ the data we use.
222
+ 4
223
+
224
+ Equation (15) may be written in conservation form as
225
+ ∂ρ
226
+ ∂t + ∇ · (ρ(v − σ(|∇ρ|)∇ log ρ)) = 0,
227
+ and after defining an augmented velocity
228
+ vaug = v − σ(|∇ρ|)∇ log ρ,
229
+ we derive a simple conservation form of equation (15)
230
+ ∂ρ
231
+ ∂t + ∇ · (ρvaug) = 0.
232
+ The Lagrangian representation X = X(x, t) of the optimal trajectory for this rOMT with
233
+ nonlinear diffusion model is given by
234
+ X(x, 0) = x,
235
+ ∂X(x, t)
236
+ ∂t
237
+ = vaug
238
+ opt (X(x, t), t),
239
+ (16)
240
+ where
241
+ vaug
242
+ opt = vopt − σ(|∇ρopt|)∇ log ρopt,
243
+ (17)
244
+ and vopt and ρopt denote the optimal solution of the rOMT with nonlinear diffusion model.
245
+ In Section 4, we exhibit the pathlines in Figure 2 and Figure 3 derived from the Lagrangian
246
+ coordinates (16).
247
+ 3
248
+ Numerical scheme
249
+ In this section, we focus on the numerical solution of the nonlinear diffusive rOMT model.
250
+ The pipeline that comes from [5, 9] is based on the Gauss-Newton method:
251
+ 1. Give initial guess of v at each time and spatial point.
252
+ 2. Use v, ρ0 and the advection-diffusion equation (15) to calculate ρ at each subsequent
253
+ time step.
254
+ 3. Calculate the objective function (6), which we will denote with Γ(v) as the discrete
255
+ form.
256
+ 4. Calculate the gradient g(v) and the Hessian matrix H(v) of Γ(v) with respect to v.
257
+ 5. Solve the descent direction s by solving H(v)s = −g(v).
258
+ 6. Do line search to find l and update v by setting v = v + ls.
259
+ 7. Repeat step 2-6 until the results attain the final condition.
260
+ Space is discretized into a cell-center grid of size nx × ny × nz with a total number of N
261
+ cells, each with width ∆x, height ∆y and depth ∆z. Time is divided into m intervals of
262
+ length ∆t with m + 1 time steps. Moreover, the superscript 0 corresponds to initial time
263
+ t = 0, M corresponds to final time t = 1 and dt × m = 1. We use ρ = [(ρ0)T , . . . , (ρm)T ]T
264
+ and v = [(v1)T , . . . , (vm)T ]T to represent temporal density and velocity, respectively. Note
265
+ that the velocity vi describes the velocity field from (i − 1)th time step to ith time step.
266
+ 5
267
+
268
+ 3.1
269
+ Advection-diffusion equation
270
+ Here we describe the numerical scheme for equation (15).
271
+ The discrete form of equation (15) between time tn and tn+1 is
272
+ ρn+1 − ρn
273
+ ∆t
274
+ + A(ρ, v) = D(ρ),
275
+ (18)
276
+ where A and D are discretizations of advective and diffusive terms, respectively. We will
277
+ describe these in greater detail below. Following the work of Steklova and Haber [21], we
278
+ split equation (18) into two parts,
279
+ ρadv − ρn
280
+ ∆t
281
+ + A(ρ, v) = 0,
282
+ (19)
283
+ ρn+1 − ρadv
284
+ ∆t
285
+ = D(ρ),
286
+ (20)
287
+ where ρadv is an auxiliary variable. Simply by adding (19) and (20), we obtain the equation
288
+ (18).
289
+ So far we have not chosen the time step of ρ in the advective part A(ρ, v) and
290
+ diffusive part D(ρ). We use a standard forward scheme, i.e. ρ = ρn in our implementation.
291
+ Summarizing up to this point, to solve for the next time step density ρn+1, we first calculate
292
+ ρadv by solving equation (19) and use ρadv and ρn to calculate ρn+1 following equation
293
+ (20).
294
+ For the advective part A(ρ, v), we utilize a particle-in-cell method which is also how Steklova
295
+ and Haber[21] dealt with their advective part to solve equation (19):
296
+ ρadv = S(v)ρ.
297
+ (21)
298
+ S(v) is the averaging matrix with respect to v.
299
+ The basic idea of particle-in-cell method is moving density the ρi in the cell center to
300
+ the target ρnew
301
+ i
302
+ according to its velocity vi and using its nearest neighbor cell centers to
303
+ interpolate.
304
+ The numerical techniques of solving equation (20) are based on hyperbolic conservation
305
+ laws and the theory of viscosity solutions [15, 19, 20], and we explicitly write D in the next
306
+ section.
307
+ 3.2
308
+ Anisotropic diffusion
309
+ From now on, we explore in 3-dimension (d = 3), following [19, 25] and discretize the
310
+ anisotropic diffusion as follows:
311
+ D(ρi,j,k) = ∆x
312
+ −{σ[
313
+
314
+ (∆x
315
+ +ρi,j,k)2 + m2(∆y
316
+ +ρi,j,k, ∆y
317
+ −ρi,j,k) + m2(∆z
318
+ +ρi,j,k, ∆z
319
+ −ρi,j,k)]∆x
320
+ +ρi,j,k}
321
+ + ∆y
322
+ −{σ[
323
+
324
+ (∆y
325
+ +ρi,j,k)2 + m2(∆x
326
+ +ρi,j,k, ∆x
327
+ −ρi,j,k) + m2(∆z
328
+ +ρi,j,k, ∆z
329
+ −ρi,j,k)]∆y
330
+ +ρi,j,k}
331
+ + ∆z
332
+ −{σ[
333
+
334
+ (∆z
335
+ +ρi,j,k)2 + m2(∆x
336
+ +ρi,j,k, ∆x
337
+ −ρi,j,k) + m2(∆y
338
+ +ρi,j,k, ∆y
339
+ −ρi,j,k)]∆z
340
+ +ρi,j,k}.
341
+ (22)
342
+ 6
343
+
344
+ Here,
345
+ ∆x
346
+ −ai,j,k = ai,j,k − ai−1,j,k
347
+ ∆x
348
+ ,
349
+ ∆x
350
+ +ai,j,k = ai+1,j,k − ai,j,k
351
+ ∆x
352
+ ,
353
+ ∆y
354
+ −ai,j,k = ai,j,k − ai,j−1,k
355
+ ∆y
356
+ ,
357
+ ∆y
358
+ +ai,j,k = ai,j+1,k − ai,j,k
359
+ ∆y
360
+ ,
361
+ ∆z
362
+ −ai,j,k = ai,j,k − ai,j,k−1
363
+ ∆z
364
+ ,
365
+ ∆z
366
+ +ai,j,k = ai,j,k+1 − ai,j,k
367
+ ∆z
368
+ ,
369
+ m(a, b) = median(a, b, 0).
370
+ We note that the solution of equation (18) may be written recursively:
371
+ ρn+1 = S(vn)ρn + ∆tD(ρn).
372
+ (23)
373
+ 3.3
374
+ Objective function Γ(v)
375
+ A straightforward way [5, 9] to discretize the objective function Γ(v) in (6) is
376
+ hd ∗ ∆t ∗ ρT (Im ⊗ [IN|IN|IN])(v ⊙ v) + β|ρm − ρT |2.
377
+ (24)
378
+ Here hd = ∆x∗∆y∗∆z, ρ, v are column vectors, ⊗ is Kronecker product and ⊙ is Hadamard
379
+ product.
380
+ 3.4
381
+ Gradient, hessian and sensitivity
382
+ In order to apply the Gauss-Newton minimization procedure such as described in Steklova
383
+ and Haber [21], we need expressions for the gradient g(v) and the Hessian H(v). Taking
384
+ the gradient of (24) with respect to v, we find
385
+ g(v) = ∂Γ(v)
386
+ ∂v
387
+ = hd ∗ ∆t ∗ [2ρT Mdiag(v) + (M(v ⊙ v))T J] + β(ρm − ρ1)T ∂ρm
388
+ ∂v ,
389
+ (25)
390
+ where M = Im ⊗ [IN|IN|IN], matrix J = (Jk
391
+ j ). Here Jk
392
+ j = ∂ρk
393
+ ∂vj , k = 1, . . . , m and j =
394
+ 0, . . . , m − 1.
395
+ The Hessian matrix is
396
+ H(v) = ∂g
397
+ ∂v = hd∗∆t∗[2ρT ∇(Mdiag(v))+2∇(ρ)Mdiag(v)+M(v⊙v)∇J +∇[M(v⊙v)]J]
398
+ + β[(∂ρm
399
+ ∂v )T (∂ρm
400
+ ∂v ) + (ρm − ρ1)∂2ρm
401
+ ∂v2 ].
402
+ (26)
403
+ Numerically we approximate the Hessian by
404
+ H(v) = 2hd ∗ ∆t ∗ ρT ∇(Mdiag(v)) + β(∂ρm
405
+ ∂v )T (∂ρm
406
+ ∂v )
407
+ = 2hd ∗ ∆t ∗ diag(ρT M) + β(∂ρm
408
+ ∂v )T (∂ρm
409
+ ∂v ).
410
+ (27)
411
+ In the formulae for the gradient (25) and Hessian (27), we still need to know the sensitivity
412
+ of the density ρ with respect to the velocity v. We recall equation (23)
413
+ ρn+1 = S(vn)ρn + ∆tD(ρn).
414
+ 7
415
+
416
+ From that, the sensitivity can be calculated as below:
417
+ ∂ρk
418
+ ∂vj =
419
+
420
+
421
+
422
+ S(vk−1) ∂ρk−1
423
+ ∂vj
424
+ + ∆tD′(ρk−1) ∂ρk−1
425
+ ∂vj
426
+ k ≥ j + 2
427
+
428
+ ∂vj (S(vj)ρj)
429
+ k = j + 1
430
+ 0
431
+ k ≤ j
432
+ (28)
433
+ 4
434
+ Experimental results
435
+ In this section, we test our proposed methodology on 3D DCE-MRI data derived from
436
+ [3]. In this dataset, rats were anesthetized, and a Gd-tagged tracer was injected into the
437
+ cerebrospinal fluid (CSF). The rat underwent dynamic 3D MRI scanning every 5 minutes
438
+ to collect a total 29 3D brain images with a voxel size of 100×106×100. Post-processing of
439
+ the DCE-MRI data included head motion correction, intensity normalization, and voxel-by-
440
+ voxel conversion to percentage of baseline signal. In our experiment, we chose a 12-month-
441
+ old wild type rat for demonstrating the results.
442
+ The new algorithm was run for data covering a 100-minute time period (60 minutes to
443
+ 160 minutes) which includes 23 frames, and we used every other image as inputs to reduce
444
+ runtime, leaving 12 frames for the numerical experiment.
445
+ We use In, n = 1, . . . , 12 to
446
+ represent these frames. To derive the interpolations, we applied our model between each of
447
+ two consecutive frames, i.e. Ik and Ik+1. To ensure continuity, (except for the first step),
448
+ the initial density originates from the previous step. For example, if we are considering
449
+ the problem between I2 and I3, and we will use the final density I′
450
+ 2 calculated between I1
451
+ and I2 as the new initial density here and apply our model between I′
452
+ 2 and I3. One of the
453
+ metrics that can measure the model accuracy is the error between the final density I′
454
+ k and
455
+ the ground truth Ik at each step.
456
+ Here we are using σ function (9) with σ0 = 0.002. The choice of σ0 follows [3]. We tested
457
+ rOMT on the 3D DCE-MRI data set with σ0 = 0.00002, 0.0002, 0.002, 0.02, 0.2. The speed
458
+ maps in Figure 4 show a stable trend between σ0 = 0.00002 and σ0 = 0.002 and among
459
+ these three σ0 (0.00002,0.0002 and 0.002), 0.002 has the minimal interpolation error (see
460
+ Figure 5).
461
+ We computed pathlines based on Lagrangian coordinates (16). We compared different K’s
462
+ and the results are shown in Figures 1-3. Figure 1 shows the relative error
463
+ e = |I′ − I|2
464
+ |I|2
465
+ on each frame with different K’s. The x-axis represents the indices of frames and the y-
466
+ axis is the relative error. From Figure 1, we observe that rOMT with anisotropic diffusion
467
+ has similar accuracy as the original rOMT model. Figure 2 compares the P´eclet number
468
+ along pathlines in the right lateral view plane for different K’s. Further, Figure 3 shows
469
+ the ventral surface of the brain. Red color represents larger P´eclet numbers (advection
470
+ dominant) and blue represents smaller P´eclet numbers (diffusion dominant). As shown in
471
+ Figure 2 and Figure 3, a smaller K value results in more advection dominated transport
472
+ in ‘surface’ areas of the brain which corresponds to the CSF compartment. When we set
473
+ K = ∞, then clearly σ(x) = σ0, since
474
+ lim
475
+ K→∞ σ0
476
+ 1
477
+ 1 + ( x
478
+ K )2 = σ0.
479
+ 8
480
+
481
+ Figure 1: Relative interpolation error plot for different parameter K’s.
482
+ Original means
483
+ constant diffusion coefficient, i.e. K = ∞.
484
+ 9
485
+
486
+ relativeinterpolationerror
487
+ 0.25
488
+ 0.2
489
+ 0.15
490
+ original
491
+ 0.1
492
+ K=10
493
+ K=100
494
+ K=1000
495
+ K=10000
496
+ K=100000
497
+ 0.05
498
+ K=1000000
499
+ 0
500
+ 1
501
+ 2
502
+ 3
503
+ 4
504
+ 5
505
+ 6
506
+ 7
507
+ 8
508
+ 9
509
+ 10
510
+ 11Figure 2: Pathlines endowed with P´eclet Number shown in the lateral view plane. Parameter
511
+ K = 10, 100, 1000, 10000, 100000, ∞. The maximal limit of color bar is 300. When K is
512
+ small, the advective (red) pathline dominates in CSF rich areas.
513
+ 10
514
+
515
+ K=10
516
+ Pseudocolor
517
+ Var: PathPoint
518
+ 300.0
519
+ 225.0
520
+ 150.0
521
+ 75.00
522
+ 0.000
523
+ Max: 2.210e+13
524
+ Min: 0.000K=100
525
+ Pseudocolor
526
+ Var: PathPoint
527
+ 300.0
528
+ 225.0
529
+ 150.0
530
+ 75.00
531
+ 0.000
532
+ Max: 6.551e+13
533
+ Min: 0.000K= 1000
534
+ Pseudocolor
535
+ Var: PathPoint
536
+ 300.0
537
+ 225.0
538
+ 150.0
539
+ 75.00
540
+ 0.000
541
+ Max: 7.892e+13
542
+ Min: 0.000K=10000
543
+ Pseudocolor
544
+ Var: PathPoint
545
+ 300.0
546
+ 225.0
547
+ 150.0
548
+ 75.00
549
+ 0.000
550
+ Max:1.703e+14
551
+ Min: 0.000K=100000
552
+ Pseudocolor
553
+ Var: PathPoint
554
+ 300.0
555
+ 225.0
556
+ 150.0
557
+ 75.00
558
+ 0.000
559
+ Max:1.770e+14
560
+ Min: 0.000K=Infinity
561
+ Pseudocolor
562
+ Var: PathPoint
563
+ 300.0
564
+ 225.0
565
+ 150.0
566
+ 75.00
567
+ 0.000
568
+ Max:1.839e+13
569
+ Min: 0.000Figure 3: P´eclet number endowed pathlines shown in ventral view plane. Parameter K =
570
+ 10, 100, 1000, 10000, 100000, ∞. The maximal limit of the color bar is 300.
571
+ 11
572
+
573
+ K=1000
574
+ Pseudocolor
575
+ Var: PathPoint
576
+ 300.0
577
+ 225.0
578
+ 150.0
579
+ 75.00
580
+ 0.000
581
+ Max:7.892e+13
582
+ Min: 0.000K= 10000
583
+ Pseudocolor
584
+ Var: PathPoint
585
+ 300.0
586
+ 225.0
587
+ 150.0
588
+ 75.00
589
+ 0.000
590
+ Max: 1.703e+1
591
+ Min: 0.000K=100000
592
+ Pseudocolor
593
+ Var: PathPoint
594
+ 300.0
595
+ 225.0
596
+ 150.0
597
+ 75.00
598
+ 0.000
599
+ Max:1.770e+14
600
+ Min: 0.000K=Infinity
601
+ Pseudocolor
602
+ Var: PathPoint
603
+ 300.0
604
+ 225.0
605
+ 150.0
606
+ 75.00
607
+ 0.000
608
+ Max:1.839e+13
609
+ Min: 0.000K=10
610
+ Pseudocolor
611
+ Var: PathPoint
612
+ 300.0
613
+ 225.0
614
+ 150.0
615
+ 75.00
616
+ 0.000
617
+ Max: 2.210e+13
618
+ Min: 0.000K=100
619
+ Pseudocolor
620
+ Var: PathPoint
621
+ 300.0
622
+ 225.0
623
+ 150.0
624
+ 75.00
625
+ 0.000
626
+ Max: 6.551e+13
627
+ Min: 0.000Figure 4: Speed map for different σ0’s. The maximal limit of the color bar is 0.6. The first
628
+ three speed maps exhibit a stable trend. The last two speed maps with higher values of
629
+ diffusion dramatically (and erroneously) increase speed suggesting that σ0 is too large.
630
+ Figure 5: Mean speed (blue line) and interpolation error (orange line) of different σ0’s. The
631
+ interpolation error is the relative error between interpolated frames and data image of the
632
+ last frame. The interpolation error reflects the closeness between interpolations from rOMT
633
+ and the data image. Lower interpolation error means more accurate the rOMT is fitting
634
+ the real data. This figure shows larger σ0 has better interpolation error but when σ0 goes
635
+ to 0.2, the mean speed accelerates dramatically, which is unrealistic given previous data of
636
+ the expected magnitude of solute transport in brain tissue.
637
+ 12
638
+
639
+ 。=0.00020.=0.002.=0.020.6
640
+ 0。=0.2
641
+ 0.5
642
+ 0.4
643
+ 0.3
644
+ 0.2
645
+ 0.10.25
646
+ 0.2
647
+ meanspeed
648
+ interpolation error
649
+ 0.15
650
+ 0.1
651
+ 0.05
652
+ e
653
+ 0.00002
654
+ 0.0002
655
+ 0.002
656
+ 0.02
657
+ 0.2
658
+ do5
659
+ Discussion
660
+ In this paper, we proposed a novel extension of the rOMT model. Specifically, we replaced
661
+ the linear diffusion term in the advection-diffusion equation by a nonlinear diffusion term
662
+ based on the Perona-Malik anisotropic diffusion approach. The updated model was tested
663
+ on glymphatic DCE-MRI data comparing different parameter K’s in the conductivity co-
664
+ efficient (σ) function and we observed that smaller K yields increased number of advective
665
+ pathlines in CSF rich areas. More uniform advective solutes flow in the CSF compartment
666
+ including at the level of the basal cisterns, ambient cistern and subarachnoid space above
667
+ the cerebellum may be more biologically realistic.
668
+ This paper only applied the model on glymphatic DCE-MRI data, but it can be generally
669
+ applied to other types of biological imaging data.
670
+ In the future, we plan to apply our
671
+ approach to tumor vasculature imagery also derived from DCE-MRI, since the mass (tracer)
672
+ is injected and may leak, we also plan to explore an unbalanced version of rOMT with
673
+ nonlinear diffusion.
674
+ Acknowledgments
675
+ This research was funded in part by AFOSR grant FA9550-20-1-0029, NIH grant R01-
676
+ AG048769, a grant from Breast Cancer Research Foundation BCRF-17-193, Army Research
677
+ Office grant W911NF2210292, and a grant from the Cure Alzheimer’s Foundation.
678
+ References
679
+ [1] Jean-David Benamou and Yann Brenier. A computational fluid mechanics solution to
680
+ the monge-kantorovich mass transfer problem. Numerische Mathematik, 84(3):375–393,
681
+ 2000.
682
+ [2] Andrew J Bernoff, Andrea L Bertozzi, and Thomas P Witelski. Axisymmetric surface
683
+ diffusion: dynamics and stability of self-similar pinchoff. Journal of statistical physics,
684
+ 93(3):725–776, 1998.
685
+ [3] Xinan Chen, Xiaodan Liu, Sunil Koundal, Rena Elkin, Xiaoyue Zhu, Brittany Monte,
686
+ Feng Xu, Feng Dai, Maysam Pedram, Hedok Lee, et al. Cerebral amyloid angiopathy
687
+ is associated with glymphatic transport reduction and time-delayed solute drainage
688
+ along the neck arteries. Nature Aging, 2(3):214–223, 2022.
689
+ [4] Sandro Da Mesquita, Antoine Louveau, Andrea Vaccari, Igor Smirnov, R Chase Cor-
690
+ nelison, Kathryn M Kingsmore, Christian Contarino, Suna Onengut-Gumuscu, Emily
691
+ Farber, Daniel Raper, et al. Functional aspects of meningeal lymphatics in ageing and
692
+ alzheimer’s disease. Nature, 560(7717):185–191, 2018.
693
+ [5] Rena Elkin, Saad Nadeem, Eldad Haber, Klara Steklova, Hedok Lee, Helene Ben-
694
+ veniste, and Allen Tannenbaum. Glymphvis: visualizing glymphatic transport path-
695
+ ways using regularized optimal transport. In International Conference on Medical Im-
696
+ age Computing and Computer-Assisted Intervention, pages 844–852. Springer, 2018.
697
+ 13
698
+
699
+ [6] Jeffrey J Iliff, Hedok Lee, Mei Yu, Tian Feng, Jean Logan, Maiken Nedergaard, He-
700
+ lene Benveniste, et al. Brain-wide pathway for waste clearance captured by contrast-
701
+ enhanced mri. The Journal of clinical investigation, 123(3):1299–1309, 2013.
702
+ [7] JR King. Emerging areas of mathematical modelling. Philosophical Transactions of the
703
+ Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences,
704
+ 358(1765):3–19, 2000.
705
+ [8] JR King. Two generalisations of the thin film equation. Mathematical and Computer
706
+ modelling, 34(7-8):737–756, 2001.
707
+ [9] Sunil Koundal, Rena Elkin, Saad Nadeem, Yuechuan Xue, Stefan Constantinou, Simon
708
+ Sanggaard, Xiaodan Liu, Brittany Monte, Feng Xu, William Van Nostrand, et al.
709
+ Optimal mass transport with lagrangian workflow reveals advective and diffusion driven
710
+ solute transport in the glymphatic system. Scientific reports, 10(1):1–18, 2020.
711
+ [10] Benjamin T Kress, Jeffrey J Iliff, Maosheng Xia, Minghuan Wang, Helen S Wei, Dou-
712
+ glas Zeppenfeld, Lulu Xie, Hongyi Kang, Qiwu Xu, Jason A Liew, et al. Impairment
713
+ of paravascular clearance pathways in the aging brain. Annals of neurology, 76(6):845–
714
+ 861, 2014.
715
+ [11] Hedok Lee, Kristian Mortensen, Simon Sanggaard, Palle Koch, Hans Brunner, Bjørn
716
+ Quistorff, Maiken Nedergaard, and Helene Benveniste. Quantitative gd-dota uptake
717
+ from cerebrospinal fluid into rat brain using 3d vfa-spgr at 9.4 t. Magnetic resonance
718
+ in medicine, 79(3):1568–1578, 2018.
719
+ [12] Hedok Lee, Lulu Xie, Mei Yu, Hongyi Kang, Tian Feng, Rashid Deane, Jean Logan,
720
+ Maiken Nedergaard, and Helene Benveniste.
721
+ The effect of body posture on brain
722
+ glymphatic transport. Journal of Neuroscience, 35(31):11034–11044, 2015.
723
+ [13] Qiaoli Ma, Benjamin V Ineichen, Michael Detmar, and Steven T Proulx. Outflow of
724
+ cerebrospinal fluid is predominantly through lymphatic vessels and is reduced in aged
725
+ mice. Nature communications, 8(1):1–13, 2017.
726
+ [14] Maiken Nedergaard. Garbage truck of the brain. Science, 340(6140):1529–1530, 2013.
727
+ [15] Stanley Osher and James A Sethian. Fronts propagating with curvature-dependent
728
+ speed: Algorithms based on hamilton-jacobi formulations. Journal of computational
729
+ physics, 79(1):12–49, 1988.
730
+ [16] Weiguo Peng, Thiyagarajan M Achariyar, Baoman Li, Yonghong Liao, Humberto
731
+ Mestre, Emi Hitomi, Sean Regan, Tristan Kasper, Sisi Peng, Fengfei Ding, et al.
732
+ Suppression of glymphatic fluid transport in a mouse model of alzheimer’s disease.
733
+ Neurobiology of disease, 93:215–225, 2016.
734
+ [17] Pietro Perona and Jitendra Malik. Scale-space and edge detection using anisotropic
735
+ diffusion. IEEE Transactions on pattern analysis and machine intelligence, 12(7):629–
736
+ 639, 1990.
737
+ [18] Vadim Ratner, Yi Gao, Hedok Lee, Rena Elkin, Maiken Nedergaard, Helene Ben-
738
+ veniste, and Allen Tannenbaum. Cerebrospinal and interstitial fluid transport via the
739
+ glymphatic pathway modeled by optimal mass transport. Neuroimage, 152:530–537,
740
+ 2017.
741
+ 14
742
+
743
+ [19] Leonid I Rudin, Stanley Osher, and Emad Fatemi. Nonlinear total variation based
744
+ noise removal algorithms. Physica D: nonlinear phenomena, 60(1-4):259–268, 1992.
745
+ [20] James A Sethian. A review of recent numerical algorithms for hypersurfaces moving
746
+ with curvature dependent speed. J. Differential Geometry, 31(31):131–161, 1989.
747
+ [21] Klara Steklova and Eldad Haber. Joint hydrogeophysical inversion: state estimation
748
+ for seawater intrusion models in 3d. Computational Geosciences, 21(1):75–94, 2017.
749
+ [22] C´edric Villani. Optimal transport: old and new, volume 338. Springer, 2009.
750
+ [23] C´edric Villani. Topics in optimal transportation, volume 58. American Mathematical
751
+ Soc., 2021.
752
+ [24] Jiajie Xu, Ya Su, Jiayu Fu, Xiaoxiao Wang, Benedictor Alexander Nguchu, Bensheng
753
+ Qiu, Qiang Dong, and Xin Cheng. Glymphatic dysfunction correlates with severity of
754
+ small vessel disease and cognitive impairment in cerebral amyloid angiopathy. European
755
+ Journal of Neurology, 29(10):2895–2904, 2022.
756
+ [25] Yu-Li You, Wenyuan Xu, Allen Tannenbaum, and Mostafa Kaveh. Behavioral analysis
757
+ of anisotropic diffusion in image processing. IEEE Transactions on Image Processing,
758
+ 5(11):1539–1553, 1996.
759
+ 15
760
+
dtE1T4oBgHgl3EQfyAVc/content/tmp_files/load_file.txt ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ filepath=/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf,len=430
2
+ page_content='Regularized Optimal Mass Transport with Nonlinear Diffusion Kaiming Xu, Xinan Chen, Helene Benveniste, Allen Tannenbaum ∗†‡§ January 10, 2023 Abstract In this paper, we combine nonlinear diffusion with the regularized optimal mass transport (rOMT) model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
3
+ page_content=' As we will demonstrate, this new approach provides further insights into certain applications of fluid flow analysis in the brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
4
+ page_content=' From the point of view of image processing, the anisotropic diffusion method, based on Perona-Malik, explicitly considers edge information.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
5
+ page_content=' Applied to rOMT analysis of glymphatic trans- port based on dynamic contrast-enhanced magnetic resonance imaging data, this new framework appears to capture a larger advection-dominant volume.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
6
+ page_content=' 1 Introduction The theory of optimal mass transport(OMT) was first proposed by Gaspard Monge in 1781 and has since evolved into a unique scientific field which has had significant impact on research in many disciplines [22, 23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
7
+ page_content=' Mass transport theory has been applied to diverse fields including physics, biology, economics and engineering.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
8
+ page_content=' OMT defines a distance called the Wasserstein distance, and thus creates a natural geometry on the space of probability distributions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
9
+ page_content=' Our study is based on a fluid dynamics reformulation of OMT [1] which allows us to calculate the flow fields between two density distributions.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
10
+ page_content=' Regularized optimal mass transport (rOMT), an extension of fluid dynamics reformulation of OMT, is a tool to study temporal flow fields as a physically inspired model of optical flow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
11
+ page_content=' It has the ability to capture the flow dynamics, handle noise and simulate diffusion [3, 5, 9].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
12
+ page_content=' rOMT utilizes an advection-diffusion equation as its flow-driven partial different equation and is endpoint free.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
13
+ page_content=' A source term may be added to rOMT in which case the total mass preservation condition can be circumvented.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
14
+ page_content=' This line of research will be pursued in other work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
15
+ page_content=' Anisotropic diffusion, a major tool for image segmentation, edge detection and image de- noising, was first proposed by Perona and Malik [17].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
16
+ page_content=' Notably, instead of using a constant diffusion coefficient, Perona and Malik considered a nonnegative function (conductivity ∗K.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
17
+ page_content=' Xu is with the Department of Applied Mathematics & Statistics, Stony Brook University, NY;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
18
+ page_content=' email: kaiming.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
19
+ page_content='xu@stonybrook.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
20
+ page_content='edu †X.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
21
+ page_content=' Chen is with the Department of Medical Physics, Memorial Sloan Kettering Cancer Center, NY ‡H.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
22
+ page_content=' Benveniste is with the Department of Anesthesiology, Yale School of Medicine, CT §A.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
23
+ page_content=' Tannenbaum is with the Departments of Computer Science and Applied Mathematics & Statistics, Stony Brook University, NY;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
24
+ page_content=' email: allen.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
25
+ page_content='tannenbaum@stonybrook.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
26
+ page_content='edu 1 arXiv:2301.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
27
+ page_content='03428v1 [physics.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
28
+ page_content='flu-dyn] 3 Dec 2022 coefficient) of the magnitude of the local density gradient;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
29
+ page_content=' see equation (8).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
30
+ page_content=' The authors suggested two possible conductivity coefficients (see (9) and (10)), wherein the diffusion will be very small near the edges, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
31
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
32
+ page_content=' reflecting the fact that near edges images tend to have very large intensity gradients.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
33
+ page_content=' In this work, we show that anisotropic diffusion enhances the interpretation of glymphatic dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) flow data and may be used in conjunction with the constant diffusion coefficient approach [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
34
+ page_content=' The anisotropic diffusion equation may be derived via the steepest descend method for solving an energy minimization problem [25].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
35
+ page_content=' The glymphatic system is involved in transporting waste products from the brain to the meningeal lymphatic system which connects to the cervical lymph nodes [14].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
36
+ page_content=' The function- ing of the glymphatic and lymphatic systems decrease with age and have been implicated in the pathophysiology of a wide range of neurodegenerative diseases including cerebral amy- loid angiopathy [3, 24] and Alzheimer’s disease [4, 10, 13, 16].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
37
+ page_content=' We study glymphatic trans- port using a temporal series of DCE-MRI data acquired from the rodent brain [6, 11, 12].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
38
+ page_content=' Since the data are acquired at discrete time points, our work is motivated by the need to find a dynamic physically based model of the transport.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
39
+ page_content=' Several different versions of OMT [18] and rOMT [3, 5, 9] have been used to model the glymphatic flow.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
40
+ page_content=' In the present work, we propose a new version of rOMT.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
41
+ page_content=' Specifically, we replace the lin- ear diffusion in rOMT [3, 5, 9] with the Perona-Malik based anisotropic diffusion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
42
+ page_content=' Here, we argue that this gives us enhanced flexibility to study image-based flows inherent to glymphatic transport.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
43
+ page_content=' Notably, many diffusion processes in fluids are better captured by nonlinear models, e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
44
+ page_content='g.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
45
+ page_content=', axisymmetric surface diffusion [2] and thin fluid films [7, 8].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
46
+ page_content=' We utilize Lagrangian coordinates for visualizing the glymphatic transport pathlines.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
47
+ page_content=' Several properties of solute particle movement are computed along the pathlines such as speed and the P´eclet number.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
48
+ page_content=' Here we compare various parameters of the anisotropic diffusion coef- ficient, and observe the impact of different values on several data metrics including P´eclet plots which can map diffusion dominated versus advection dominated regions of the brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
49
+ page_content=' We briefly summarize the contents of the present paper.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
50
+ page_content=' In Section 2, we review the theory of OMT, rOMT and nonlinear diffusion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
51
+ page_content=' Section 3 introduces the algorithm and numerical methods we employ for our current work.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
52
+ page_content=' In Section 4, we explicate the application of the model to glymphatic DCE-MRI data and analyze the experimental results and we conclude our paper in Section 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
53
+ page_content=' 2 Model 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
54
+ page_content='1 OMT In this section, we introduce OMT and its fluid dynamics formulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
55
+ page_content=' All the technical details as well as a complete set of references may be found in [22, 23].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
56
+ page_content=' The original formulation of OMT was given by Gaspard Monge and may be expressed as inf T { � Ω c(x, T(x))ρ0(x)dx | T#ρ0 = ρ1}, (1) where c(x, y) is the cost function of moving the unit mass from x to y, ρ0 and ρ1 are two probability distributions in the domain Ω ⊆ Rd, T is the transport map, and T# is the 2 push-forward of T.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
57
+ page_content=' This formulation assumes that ρ0 and ρ1 have the same total mass, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
58
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
59
+ page_content=' � Ω ρ0(x)dx = � Ω ρ1(x)dx and then seeks for the optimal transport map T to minimize the total cost, the integral in equation (1), subject to the push-forward constraint.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
60
+ page_content=' Later, Leonid Kantorovich formulated a relaxed version of OMT as follows: inf π∈Π(ρ0,ρ1) � Ω×Ω c(x, y)π(dx, dy), (2) where Π(ρ0, ρ1) denotes the set of all couplings (joint distributions) between the marginals ρ0 and ρ1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
61
+ page_content=' From here on, the cost function c will be taken as the square of the Euclidean distance c(x, y) = ∥x − y∥2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
62
+ page_content=' Benemou and Brenier [1] proved that for c(x,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
63
+ page_content=' y) = ∥x − y∥2,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
64
+ page_content=' the specific infimum of Monge- Kantorovich formulation is equal to the result in following fluid dynamics formulation for density/probability distributions with compact support: inf ρ,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
65
+ page_content='v � 1 0 � Ω ρ(t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
66
+ page_content=' x)|v(t,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
67
+ page_content=' x)|2dxdt,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
68
+ page_content=' (3) ∂ρ ∂t + ∇ · (ρv) = 0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
69
+ page_content=' (4) ρ(0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
70
+ page_content=' x) = ρ0(x),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
71
+ page_content=' ρ(1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
72
+ page_content=' x) = ρ1(x),' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
73
+ page_content=' (5) where ρ : [0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
74
+ page_content=' 1]×Ω → R≥0 is the family of density/probability distributions defining geodesic path from ρ0 to ρ1,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
75
+ page_content=' and v : [0,' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
76
+ page_content=' 1] × Ω → Rd is the velocity vector field.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
77
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
78
+ page_content='2 rOMT The regularized OMT model (rOMT) [5, 9] adds two assumptions: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
79
+ page_content=' the image data we use are noisy observations and thus we do not want to make the final density we calculate coincide with the MR images;' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
80
+ page_content=' and 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
81
+ page_content=' the flow is driven by an advection-diffusion equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
82
+ page_content=' Based on these two assumptions, the rOMT formulation may be written as: inf ρ,v � 1 0 � Ω ρ(t, x)|v(t, x)|2dxdt + β � Ω (ρ(1, x) − ρ1(x))2dx, (6) ∂ρ ∂t + ∇ · (ρv) = ∇ · (σ0∇ρ), (7) ρ(0, x) = ρ0(x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
83
+ page_content=' In this formulation, the final marginal condition is removed and a penalty of the error between final density and ground truth is added in the objective function (6), where β is the penalty parameter.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
84
+ page_content=' Equation (7) is an advection-diffusion equation with a constant σ0 denoting the diffusion coefficient.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
85
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
86
+ page_content='3 Nonlinear diffusion Instead of using linear diffusion in which σ0 is a constant, nonlinear diffusion seems to have certain advantages that we will now describe.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
87
+ page_content=' Perona and Malik proposed an anisotropic 3 diffusion [17], which is a useful tool for image segmentation, edge detection and image denoising.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
88
+ page_content=' The anisotropic diffusion equation is ∂ρ ∂t = ∇ · (σ(|∇ρ|)∇ρ), (8) where σ(·) is a nonnegative strictly decreasing function.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
89
+ page_content=' If we consider a 3D problem, then |∇ρ| = � ρ2x + ρ2y + ρ2z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
90
+ page_content=' The proper diffusion should be large in smooth homogeneous areas and become smaller near edges, the places where |∇ρ| is large.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
91
+ page_content=' Perona and Malik [17] suggested two versions of the diffusion (conductivity) coefficient: σ(x) = σ0 1 1 + ( x K )2 , (9) σ(x) = σ0e−( x K )2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
92
+ page_content=' (10) Both are 0 when x approaches ∞ and attend upper bound σ0 while x = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
93
+ page_content=' K is a constant and controls the sensitivity to edges and can be tuned for different applications.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
94
+ page_content=' Following [25], we may derive the anisotropic diffusion equation (8) via the steepest descent from an energy minimization problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
95
+ page_content=' More precisely, considering the following minimiza- tion problem: min � Ω f(|∇ρ|)dΩ, (11) then the steepest descend equation may be computed to be ∂ρ ∂t = ∇ · (f′(|∇ρ| ∇ρ |∇ρ|)).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
96
+ page_content=' (12) Obviously, (12) is identical to (8) if f′(x) = xσ(x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
97
+ page_content=' (13) For example, the corresponding f function of σ function (9) is f(x) = σ0K2 2 ln[1 + ( x K )2] (14) 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
98
+ page_content='4 rOMT with nonlinear diffusion In this section, we present our new rOMT formulation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
99
+ page_content=' We replace the diffusion in (7) by anisotropic diffusion in (8) and obtain the following formulation: inf ρ,v � 1 0 � Ω ρ(t, x)|v(t, x)|2dxdt + β � Ω (ρ(1, x) − ρ1(x))2dx, ∂ρ ∂t + ∇ · (ρv) = ∇ · (σ(|∇ρ|)∇ρ), (15) ρ(0, x) = ρ0(x).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
100
+ page_content=' One may employ various versions of the σ function and in this work, we choose the function given in (9).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
101
+ page_content=' Note that, there are two parameters σ0 and K which may be tuned based on the data we use.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
102
+ page_content=' 4 Equation (15) may be written in conservation form as ∂ρ ∂t + ∇ · (ρ(v − σ(|∇ρ|)∇ log ρ)) = 0, and after defining an augmented velocity vaug = v − σ(|∇ρ|)∇ log ρ, we derive a simple conservation form of equation (15) ∂ρ ∂t + ∇ · (ρvaug) = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
103
+ page_content=' The Lagrangian representation X = X(x, t) of the optimal trajectory for this rOMT with nonlinear diffusion model is given by X(x, 0) = x, ∂X(x, t) ∂t = vaug opt (X(x, t), t), (16) where vaug opt = vopt − σ(|∇ρopt|)∇ log ρopt, (17) and vopt and ρopt denote the optimal solution of the rOMT with nonlinear diffusion model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
104
+ page_content=' In Section 4, we exhibit the pathlines in Figure 2 and Figure 3 derived from the Lagrangian coordinates (16).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
105
+ page_content=' 3 Numerical scheme In this section, we focus on the numerical solution of the nonlinear diffusive rOMT model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
106
+ page_content=' The pipeline that comes from [5, 9] is based on the Gauss-Newton method: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
107
+ page_content=' Give initial guess of v at each time and spatial point.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
108
+ page_content=' 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
109
+ page_content=' Use v, ρ0 and the advection-diffusion equation (15) to calculate ρ at each subsequent time step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
110
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
111
+ page_content=' Calculate the objective function (6), which we will denote with Γ(v) as the discrete form.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
112
+ page_content=' 4.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
113
+ page_content=' Calculate the gradient g(v) and the Hessian matrix H(v) of Γ(v) with respect to v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
114
+ page_content=' 5.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
115
+ page_content=' Solve the descent direction s by solving H(v)s = −g(v).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
116
+ page_content=' 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
117
+ page_content=' Do line search to find l and update v by setting v = v + ls.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
118
+ page_content=' 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
119
+ page_content=' Repeat step 2-6 until the results attain the final condition.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
120
+ page_content=' Space is discretized into a cell-center grid of size nx × ny × nz with a total number of N cells, each with width ∆x, height ∆y and depth ∆z.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
121
+ page_content=' Time is divided into m intervals of length ∆t with m + 1 time steps.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
122
+ page_content=' Moreover, the superscript 0 corresponds to initial time t = 0, M corresponds to final time t = 1 and dt × m = 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
123
+ page_content=' We use ρ = [(ρ0)T , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
124
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
125
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
126
+ page_content=' , (ρm)T ]T and v = [(v1)T , .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
127
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
128
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
129
+ page_content=' , (vm)T ]T to represent temporal density and velocity, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
130
+ page_content=' Note that the velocity vi describes the velocity field from (i − 1)th time step to ith time step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
131
+ page_content=' 5 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
132
+ page_content='1 Advection-diffusion equation Here we describe the numerical scheme for equation (15).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
133
+ page_content=' The discrete form of equation (15) between time tn and tn+1 is ρn+1 − ρn ∆t + A(ρ, v) = D(ρ), (18) where A and D are discretizations of advective and diffusive terms, respectively.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
134
+ page_content=' We will describe these in greater detail below.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
135
+ page_content=' Following the work of Steklova and Haber [21], we split equation (18) into two parts, ρadv − ρn ∆t + A(ρ, v) = 0, (19) ρn+1 − ρadv ∆t = D(ρ), (20) where ρadv is an auxiliary variable.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
136
+ page_content=' Simply by adding (19) and (20), we obtain the equation (18).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
137
+ page_content=' So far we have not chosen the time step of ρ in the advective part A(ρ, v) and diffusive part D(ρ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
138
+ page_content=' We use a standard forward scheme, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
139
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
140
+ page_content=' ρ = ρn in our implementation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
141
+ page_content=' Summarizing up to this point, to solve for the next time step density ρn+1, we first calculate ρadv by solving equation (19) and use ρadv and ρn to calculate ρn+1 following equation (20).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
142
+ page_content=' For the advective part A(ρ, v), we utilize a particle-in-cell method which is also how Steklova and Haber[21] dealt with their advective part to solve equation (19): ρadv = S(v)ρ.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
143
+ page_content=' (21) S(v) is the averaging matrix with respect to v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
144
+ page_content=' The basic idea of particle-in-cell method is moving density the ρi in the cell center to the target ρnew i according to its velocity vi and using its nearest neighbor cell centers to interpolate.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
145
+ page_content=' The numerical techniques of solving equation (20) are based on hyperbolic conservation laws and the theory of viscosity solutions [15, 19, 20], and we explicitly write D in the next section.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
146
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
147
+ page_content='2 Anisotropic diffusion From now on, we explore in 3-dimension (d = 3), following [19, 25] and discretize the anisotropic diffusion as follows: D(ρi,j,k) = ∆x −{σ[ � (∆x +ρi,j,k)2 + m2(∆y +ρi,j,k, ∆y −ρi,j,k) + m2(∆z +ρi,j,k, ∆z −ρi,j,k)]∆x +ρi,j,k} + ∆y −{σ[ � (∆y +ρi,j,k)2 + m2(∆x +ρi,j,k, ∆x −ρi,j,k) + m2(∆z +ρi,j,k, ∆z −ρi,j,k)]∆y +ρi,j,k} + ∆z −{σ[ � (∆z +ρi,j,k)2 + m2(∆x +ρi,j,k, ∆x −ρi,j,k) + m2(∆y +ρi,j,k, ∆y −ρi,j,k)]∆z +ρi,j,k}.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
148
+ page_content=' (22) 6 Here, ∆x −ai,j,k = ai,j,k − ai−1,j,k ∆x , ∆x +ai,j,k = ai+1,j,k − ai,j,k ∆x , ∆y −ai,j,k = ai,j,k − ai,j−1,k ∆y , ∆y +ai,j,k = ai,j+1,k − ai,j,k ∆y , ∆z −ai,j,k = ai,j,k − ai,j,k−1 ∆z , ∆z +ai,j,k = ai,j,k+1 − ai,j,k ∆z , m(a, b) = median(a, b, 0).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
149
+ page_content=' We note that the solution of equation (18) may be written recursively: ρn+1 = S(vn)ρn + ∆tD(ρn).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
150
+ page_content=' (23) 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
151
+ page_content='3 Objective function Γ(v) A straightforward way [5, 9] to discretize the objective function Γ(v) in (6) is hd ∗ ∆t ∗ ρT (Im ⊗ [IN|IN|IN])(v ⊙ v) + β|ρm − ρT |2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
152
+ page_content=' (24) Here hd = ∆x∗∆y∗∆z, ρ, v are column vectors, ⊗ is Kronecker product and ⊙ is Hadamard product.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
153
+ page_content=' 3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
154
+ page_content='4 Gradient, hessian and sensitivity In order to apply the Gauss-Newton minimization procedure such as described in Steklova and Haber [21], we need expressions for the gradient g(v) and the Hessian H(v).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
155
+ page_content=' Taking the gradient of (24) with respect to v, we find g(v) = ∂Γ(v) ∂v = hd ∗ ∆t ∗ [2ρT Mdiag(v) + (M(v ⊙ v))T J] + β(ρm − ρ1)T ∂ρm ∂v , (25) where M = Im ⊗ [IN|IN|IN], matrix J = (Jk j ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
156
+ page_content=' Here Jk j = ∂ρk ∂vj , k = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
157
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
158
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
159
+ page_content=' , m and j = 0, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
160
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
161
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
162
+ page_content=' , m − 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
163
+ page_content=' The Hessian matrix is H(v) = ∂g ∂v = hd∗∆t∗[2ρT ∇(Mdiag(v))+2∇(ρ)Mdiag(v)+M(v⊙v)∇J +∇[M(v⊙v)]J] + β[(∂ρm ∂v )T (∂ρm ∂v ) + (ρm − ρ1)∂2ρm ∂v2 ].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
164
+ page_content=' (26) Numerically we approximate the Hessian by H(v) = 2hd ∗ ∆t ∗ ρT ∇(Mdiag(v)) + β(∂ρm ∂v )T (∂ρm ∂v ) = 2hd ∗ ∆t ∗ diag(ρT M) + β(∂ρm ∂v )T (∂ρm ∂v ).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
165
+ page_content=' (27) In the formulae for the gradient (25) and Hessian (27), we still need to know the sensitivity of the density ρ with respect to the velocity v.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
166
+ page_content=' We recall equation (23) ρn+1 = S(vn)ρn + ∆tD(ρn).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
167
+ page_content=' 7 From that, the sensitivity can be calculated as below: ∂ρk ∂vj = � � � S(vk−1) ∂ρk−1 ∂vj + ∆tD′(ρk−1) ∂ρk−1 ∂vj k ≥ j + 2 ∂ ∂vj (S(vj)ρj) k = j + 1 0 k ≤ j (28) 4 Experimental results In this section, we test our proposed methodology on 3D DCE-MRI data derived from [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
168
+ page_content=' In this dataset, rats were anesthetized, and a Gd-tagged tracer was injected into the cerebrospinal fluid (CSF).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
169
+ page_content=' The rat underwent dynamic 3D MRI scanning every 5 minutes to collect a total 29 3D brain images with a voxel size of 100×106×100.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
170
+ page_content=' Post-processing of the DCE-MRI data included head motion correction, intensity normalization, and voxel-by- voxel conversion to percentage of baseline signal.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
171
+ page_content=' In our experiment, we chose a 12-month- old wild type rat for demonstrating the results.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
172
+ page_content=' The new algorithm was run for data covering a 100-minute time period (60 minutes to 160 minutes) which includes 23 frames, and we used every other image as inputs to reduce runtime, leaving 12 frames for the numerical experiment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
173
+ page_content=' We use In, n = 1, .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
174
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
175
+ page_content=' .' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
176
+ page_content=' , 12 to represent these frames.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
177
+ page_content=' To derive the interpolations, we applied our model between each of two consecutive frames, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
178
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
179
+ page_content=' Ik and Ik+1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
180
+ page_content=' To ensure continuity, (except for the first step), the initial density originates from the previous step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
181
+ page_content=' For example, if we are considering the problem between I2 and I3, and we will use the final density I′ 2 calculated between I1 and I2 as the new initial density here and apply our model between I′ 2 and I3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
182
+ page_content=' One of the metrics that can measure the model accuracy is the error between the final density I′ k and the ground truth Ik at each step.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
183
+ page_content=' Here we are using σ function (9) with σ0 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
184
+ page_content='002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
185
+ page_content=' The choice of σ0 follows [3].' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
186
+ page_content=' We tested rOMT on the 3D DCE-MRI data set with σ0 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
187
+ page_content='00002, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
188
+ page_content='0002, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
189
+ page_content='002, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
190
+ page_content='02, 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
191
+ page_content='2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
192
+ page_content=' The speed maps in Figure 4 show a stable trend between σ0 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
193
+ page_content='00002 and σ0 = 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
194
+ page_content='002 and among these three σ0 (0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
195
+ page_content='00002,0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
196
+ page_content='0002 and 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
197
+ page_content='002), 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
198
+ page_content='002 has the minimal interpolation error (see Figure 5).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
199
+ page_content=' We computed pathlines based on Lagrangian coordinates (16).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
200
+ page_content=' We compared different K’s and the results are shown in Figures 1-3.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
201
+ page_content=' Figure 1 shows the relative error e = |I′ − I|2 |I|2 on each frame with different K’s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
202
+ page_content=' The x-axis represents the indices of frames and the y- axis is the relative error.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
203
+ page_content=' From Figure 1, we observe that rOMT with anisotropic diffusion has similar accuracy as the original rOMT model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
204
+ page_content=' Figure 2 compares the P´eclet number along pathlines in the right lateral view plane for different K’s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
205
+ page_content=' Further, Figure 3 shows the ventral surface of the brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
206
+ page_content=' Red color represents larger P´eclet numbers (advection dominant) and blue represents smaller P´eclet numbers (diffusion dominant).' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
207
+ page_content=' As shown in Figure 2 and Figure 3, a smaller K value results in more advection dominated transport in ‘surface’ areas of the brain which corresponds to the CSF compartment.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
208
+ page_content=' When we set K = ∞, then clearly σ(x) = σ0, since lim K→∞ σ0 1 1 + ( x K )2 = σ0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
209
+ page_content=' 8 Figure 1: Relative interpolation error plot for different parameter K’s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
210
+ page_content=' Original means constant diffusion coefficient, i.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
211
+ page_content='e.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
212
+ page_content=' K = ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
213
+ page_content=' 9 relativeinterpolationerror 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
214
+ page_content='25 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
215
+ page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
216
+ page_content='15 original 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
217
+ page_content='1 K=10 K=100 K=1000 K=10000 K=100000 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
218
+ page_content='05 K=1000000 0 1 2 3 4 5 6 7 8 9 10 11Figure 2: Pathlines endowed with P´eclet Number shown in the lateral view plane.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
219
+ page_content=' Parameter K = 10, 100, 1000, 10000, 100000, ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
220
+ page_content=' The maximal limit of color bar is 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
221
+ page_content=' When K is small, the advective (red) pathline dominates in CSF rich areas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
222
+ page_content=' 10 K=10 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
223
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
224
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
225
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
226
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
227
+ page_content='000 Max: 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
228
+ page_content='210e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
229
+ page_content='000K=100 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
230
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
231
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
232
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
233
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
234
+ page_content='000 Max: 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
235
+ page_content='551e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
236
+ page_content='000K= 1000 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
237
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
238
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
239
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
240
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
241
+ page_content='000 Max: 7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
242
+ page_content='892e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
243
+ page_content='000K=10000 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
244
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
245
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
246
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
247
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
248
+ page_content='000 Max:1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
249
+ page_content='703e+14 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
250
+ page_content='000K=100000 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
251
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
252
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
253
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
254
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
255
+ page_content='000 Max:1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
256
+ page_content='770e+14 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
257
+ page_content='000K=Infinity Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
258
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
259
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
260
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
261
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
262
+ page_content='000 Max:1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
263
+ page_content='839e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
264
+ page_content='000Figure 3: P´eclet number endowed pathlines shown in ventral view plane.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
265
+ page_content=' Parameter K = 10, 100, 1000, 10000, 100000, ∞.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
266
+ page_content=' The maximal limit of the color bar is 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
267
+ page_content=' 11 K=1000 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
268
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
269
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
270
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
271
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
272
+ page_content='000 Max:7.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
273
+ page_content='892e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
274
+ page_content='000K= 10000 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
275
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
276
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
277
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
278
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
279
+ page_content='000 Max: 1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
280
+ page_content='703e+1 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
281
+ page_content='000K=100000 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
282
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
283
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
284
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
285
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
286
+ page_content='000 Max:1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
287
+ page_content='770e+14 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
288
+ page_content='000K=Infinity Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
289
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
290
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
291
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
292
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
293
+ page_content='000 Max:1.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
294
+ page_content='839e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
295
+ page_content='000K=10 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
296
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
297
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
298
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
299
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
300
+ page_content='000 Max: 2.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
301
+ page_content='210e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
302
+ page_content='000K=100 Pseudocolor Var: PathPoint 300.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
303
+ page_content='0 225.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
304
+ page_content='0 150.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
305
+ page_content='0 75.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
306
+ page_content='00 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
307
+ page_content='000 Max: 6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
308
+ page_content='551e+13 Min: 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
309
+ page_content='000Figure 4: Speed map for different σ0’s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
310
+ page_content=' The maximal limit of the color bar is 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
311
+ page_content='6.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
312
+ page_content=' The first three speed maps exhibit a stable trend.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
313
+ page_content=' The last two speed maps with higher values of diffusion dramatically (and erroneously) increase speed suggesting that σ0 is too large.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
314
+ page_content=' Figure 5: Mean speed (blue line) and interpolation error (orange line) of different σ0’s.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
315
+ page_content=' The interpolation error is the relative error between interpolated frames and data image of the last frame.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
316
+ page_content=' The interpolation error reflects the closeness between interpolations from rOMT and the data image.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
317
+ page_content=' Lower interpolation error means more accurate the rOMT is fitting the real data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
318
+ page_content=' This figure shows larger σ0 has better interpolation error but when σ0 goes to 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
319
+ page_content='2, the mean speed accelerates dramatically, which is unrealistic given previous data of the expected magnitude of solute transport in brain tissue.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
320
+ page_content=' 12 。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
321
+ page_content='=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
322
+ page_content='00020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
323
+ page_content='=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
324
+ page_content='002.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
325
+ page_content='=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
326
+ page_content='020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
327
+ page_content='6 0。' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
328
+ page_content='=0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
329
+ page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
330
+ page_content='5 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
331
+ page_content='4 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
332
+ page_content='3 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
333
+ page_content='2 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
334
+ page_content='10.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
335
+ page_content='25 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
336
+ page_content='2 meanspeed interpolation error 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
337
+ page_content='15 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
338
+ page_content='1 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
339
+ page_content='05 e 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
340
+ page_content='00002 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
341
+ page_content='0002 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
342
+ page_content='002 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
343
+ page_content='02 0.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
344
+ page_content='2 do5 Discussion In this paper, we proposed a novel extension of the rOMT model.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
345
+ page_content=' Specifically, we replaced the linear diffusion term in the advection-diffusion equation by a nonlinear diffusion term based on the Perona-Malik anisotropic diffusion approach.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
346
+ page_content=' The updated model was tested on glymphatic DCE-MRI data comparing different parameter K’s in the conductivity co- efficient (σ) function and we observed that smaller K yields increased number of advective pathlines in CSF rich areas.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
347
+ page_content=' More uniform advective solutes flow in the CSF compartment including at the level of the basal cisterns, ambient cistern and subarachnoid space above the cerebellum may be more biologically realistic.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
348
+ page_content=' This paper only applied the model on glymphatic DCE-MRI data, but it can be generally applied to other types of biological imaging data.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
349
+ page_content=' In the future, we plan to apply our approach to tumor vasculature imagery also derived from DCE-MRI, since the mass (tracer) is injected and may leak, we also plan to explore an unbalanced version of rOMT with nonlinear diffusion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
350
+ page_content=' Acknowledgments This research was funded in part by AFOSR grant FA9550-20-1-0029, NIH grant R01- AG048769, a grant from Breast Cancer Research Foundation BCRF-17-193, Army Research Office grant W911NF2210292, and a grant from the Cure Alzheimer’s Foundation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
351
+ page_content=' References [1] Jean-David Benamou and Yann Brenier.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
352
+ page_content=' A computational fluid mechanics solution to the monge-kantorovich mass transfer problem.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
353
+ page_content=' Numerische Mathematik, 84(3):375–393, 2000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
354
+ page_content=' [2] Andrew J Bernoff, Andrea L Bertozzi, and Thomas P Witelski.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
355
+ page_content=' Axisymmetric surface diffusion: dynamics and stability of self-similar pinchoff.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
356
+ page_content=' Journal of statistical physics, 93(3):725–776, 1998.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
357
+ page_content=' [3] Xinan Chen, Xiaodan Liu, Sunil Koundal, Rena Elkin, Xiaoyue Zhu, Brittany Monte, Feng Xu, Feng Dai, Maysam Pedram, Hedok Lee, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
358
+ page_content=' Cerebral amyloid angiopathy is associated with glymphatic transport reduction and time-delayed solute drainage along the neck arteries.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
359
+ page_content=' Nature Aging, 2(3):214–223, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
360
+ page_content=' [4] Sandro Da Mesquita, Antoine Louveau, Andrea Vaccari, Igor Smirnov, R Chase Cor- nelison, Kathryn M Kingsmore, Christian Contarino, Suna Onengut-Gumuscu, Emily Farber, Daniel Raper, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
361
+ page_content=' Functional aspects of meningeal lymphatics in ageing and alzheimer’s disease.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
362
+ page_content=' Nature, 560(7717):185–191, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
363
+ page_content=' [5] Rena Elkin, Saad Nadeem, Eldad Haber, Klara Steklova, Hedok Lee, Helene Ben- veniste, and Allen Tannenbaum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
364
+ page_content=' Glymphvis: visualizing glymphatic transport path- ways using regularized optimal transport.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
365
+ page_content=' In International Conference on Medical Im- age Computing and Computer-Assisted Intervention, pages 844–852.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
366
+ page_content=' Springer, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
367
+ page_content=' 13 [6] Jeffrey J Iliff, Hedok Lee, Mei Yu, Tian Feng, Jean Logan, Maiken Nedergaard, He- lene Benveniste, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
368
+ page_content=' Brain-wide pathway for waste clearance captured by contrast- enhanced mri.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
369
+ page_content=' The Journal of clinical investigation, 123(3):1299–1309, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
370
+ page_content=' [7] JR King.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
371
+ page_content=' Emerging areas of mathematical modelling.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
372
+ page_content=' Philosophical Transactions of the Royal Society of London.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
373
+ page_content=' Series A: Mathematical, Physical and Engineering Sciences, 358(1765):3–19, 2000.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
374
+ page_content=' [8] JR King.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
375
+ page_content=' Two generalisations of the thin film equation.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
376
+ page_content=' Mathematical and Computer modelling, 34(7-8):737–756, 2001.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
377
+ page_content=' [9] Sunil Koundal, Rena Elkin, Saad Nadeem, Yuechuan Xue, Stefan Constantinou, Simon Sanggaard, Xiaodan Liu, Brittany Monte, Feng Xu, William Van Nostrand, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
378
+ page_content=' Optimal mass transport with lagrangian workflow reveals advective and diffusion driven solute transport in the glymphatic system.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
379
+ page_content=' Scientific reports, 10(1):1–18, 2020.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
380
+ page_content=' [10] Benjamin T Kress, Jeffrey J Iliff, Maosheng Xia, Minghuan Wang, Helen S Wei, Dou- glas Zeppenfeld, Lulu Xie, Hongyi Kang, Qiwu Xu, Jason A Liew, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
381
+ page_content=' Impairment of paravascular clearance pathways in the aging brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
382
+ page_content=' Annals of neurology, 76(6):845– 861, 2014.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
383
+ page_content=' [11] Hedok Lee, Kristian Mortensen, Simon Sanggaard, Palle Koch, Hans Brunner, Bjørn Quistorff, Maiken Nedergaard, and Helene Benveniste.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
384
+ page_content=' Quantitative gd-dota uptake from cerebrospinal fluid into rat brain using 3d vfa-spgr at 9.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
385
+ page_content='4 t.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
386
+ page_content=' Magnetic resonance in medicine, 79(3):1568–1578, 2018.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
387
+ page_content=' [12] Hedok Lee, Lulu Xie, Mei Yu, Hongyi Kang, Tian Feng, Rashid Deane, Jean Logan, Maiken Nedergaard, and Helene Benveniste.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
388
+ page_content=' The effect of body posture on brain glymphatic transport.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
389
+ page_content=' Journal of Neuroscience, 35(31):11034–11044, 2015.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
390
+ page_content=' [13] Qiaoli Ma, Benjamin V Ineichen, Michael Detmar, and Steven T Proulx.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
391
+ page_content=' Outflow of cerebrospinal fluid is predominantly through lymphatic vessels and is reduced in aged mice.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
392
+ page_content=' Nature communications, 8(1):1–13, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
393
+ page_content=' [14] Maiken Nedergaard.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
394
+ page_content=' Garbage truck of the brain.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
395
+ page_content=' Science, 340(6140):1529–1530, 2013.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
396
+ page_content=' [15] Stanley Osher and James A Sethian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
397
+ page_content=' Fronts propagating with curvature-dependent speed: Algorithms based on hamilton-jacobi formulations.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
398
+ page_content=' Journal of computational physics, 79(1):12–49, 1988.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
399
+ page_content=' [16] Weiguo Peng, Thiyagarajan M Achariyar, Baoman Li, Yonghong Liao, Humberto Mestre, Emi Hitomi, Sean Regan, Tristan Kasper, Sisi Peng, Fengfei Ding, et al.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
400
+ page_content=' Suppression of glymphatic fluid transport in a mouse model of alzheimer’s disease.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
401
+ page_content=' Neurobiology of disease, 93:215–225, 2016.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
402
+ page_content=' [17] Pietro Perona and Jitendra Malik.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
403
+ page_content=' Scale-space and edge detection using anisotropic diffusion.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
404
+ page_content=' IEEE Transactions on pattern analysis and machine intelligence, 12(7):629– 639, 1990.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
405
+ page_content=' [18] Vadim Ratner, Yi Gao, Hedok Lee, Rena Elkin, Maiken Nedergaard, Helene Ben- veniste, and Allen Tannenbaum.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
406
+ page_content=' Cerebrospinal and interstitial fluid transport via the glymphatic pathway modeled by optimal mass transport.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
407
+ page_content=' Neuroimage, 152:530–537, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
408
+ page_content=' 14 [19] Leonid I Rudin, Stanley Osher, and Emad Fatemi.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
409
+ page_content=' Nonlinear total variation based noise removal algorithms.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
410
+ page_content=' Physica D: nonlinear phenomena, 60(1-4):259–268, 1992.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
411
+ page_content=' [20] James A Sethian.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
412
+ page_content=' A review of recent numerical algorithms for hypersurfaces moving with curvature dependent speed.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
413
+ page_content=' J.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
414
+ page_content=' Differential Geometry, 31(31):131–161, 1989.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
415
+ page_content=' [21] Klara Steklova and Eldad Haber.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
416
+ page_content=' Joint hydrogeophysical inversion: state estimation for seawater intrusion models in 3d.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
417
+ page_content=' Computational Geosciences, 21(1):75–94, 2017.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
418
+ page_content=' [22] C´edric Villani.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
419
+ page_content=' Optimal transport: old and new, volume 338.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
420
+ page_content=' Springer, 2009.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
421
+ page_content=' [23] C´edric Villani.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
422
+ page_content=' Topics in optimal transportation, volume 58.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
423
+ page_content=' American Mathematical Soc.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
424
+ page_content=', 2021.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
425
+ page_content=' [24] Jiajie Xu, Ya Su, Jiayu Fu, Xiaoxiao Wang, Benedictor Alexander Nguchu, Bensheng Qiu, Qiang Dong, and Xin Cheng.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
426
+ page_content=' Glymphatic dysfunction correlates with severity of small vessel disease and cognitive impairment in cerebral amyloid angiopathy.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
427
+ page_content=' European Journal of Neurology, 29(10):2895–2904, 2022.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
428
+ page_content=' [25] Yu-Li You, Wenyuan Xu, Allen Tannenbaum, and Mostafa Kaveh.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
429
+ page_content=' Behavioral analysis of anisotropic diffusion in image processing.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
430
+ page_content=' IEEE Transactions on Image Processing, 5(11):1539–1553, 1996.' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
431
+ page_content=' 15' metadata={'source': '/home/zjlab/wf/langchain-ChatGLM/knowledge_base/dtE1T4oBgHgl3EQfyAVc/content/2301.03428v1.pdf'}
e9E2T4oBgHgl3EQfGgbf/content/tmp_files/2301.03659v1.pdf.txt ADDED
@@ -0,0 +1,1307 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 1
2
+
3
+ Multifunctional Fiber-based Optoacoustic Emitter for Non-genetic
4
+ Bidirectional Neural Communication
5
+ Author Information
6
+ Nan Zheng1, Ying Jiang2, Shan Jiang3, Jongwoon Kim3, Yueming Li4, Ji-Xin Cheng2, 6, Xiaoting Jia3 *
7
+ and Chen Yang5, 6 *
8
+ Affiliations
9
+ 1 Division of Materials Science and Engineering, Boston University, Boston, MA, USA
10
+ 2 Department of Biomedical Engineering, Boston University, Boston, MA, USA
11
+ 3 Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, VA,
12
+ USA
13
+ 4 Department of Mechanical Engineering, Boston University, Boston, MA, USA
14
+ 5 Department of Chemistry, Boston University, Boston, MA, USA
15
+ 6 Department of Electrical and Computer Engineering, Boston University, Boston, MA, USA
16
+ Contributions
17
+ C.Y. conceived the project. N.Z. and S.J. performed fabrication and characterization of materials. N.Z.
18
+ and Y.J. performed the stimulation and recording experiments in vitro and in vivo. N.Z. and Y.L.
19
+ performed the in vivo biocompatibility evaluations. X.J. provided guidance on the multifunctional fiber
20
+ system. J.X.C. provided guidance on the design of fiber optoacoustic emitter. J.K. provided guidance on
21
+ optimization of recording and data analysis. The manuscript was written through contributions of all
22
+ authors. All authors have given approval to the final version of the manuscript.
23
+ Corresponding author
24
+ Correspondence to: Chen Yang ([email protected]) and Xiaoting Jia ([email protected])
25
+
26
+
27
+
28
+ 2
29
+
30
+ Abstract
31
+ A bidirectional brain interface with both “write” and “read” functions can be an important tool for
32
+ fundamental studies and potential clinical treatments for neurological diseases. Here we report a
33
+ miniaturized multifunctional fiber based optoacoustic emitter (mFOE) that first integrates simultaneous
34
+ non-genetic optoacoustic stimulation for “write” and electrophysiology recording of neural circuits for
35
+ “read”. The non-genetic feature addresses the challenges of the viral transfection required by optogenetics
36
+ in primates and human. The orthogonality between optoacoustic waves and electrical field provides a
37
+ solution to avoid the interference between electrical stimulation and recording. We first validated the non-
38
+ genetic stimulation function of the mFOE in rat cultured neurons using calcium imaging. In vivo
39
+ application of mFOE for successful simultaneous optoacoustic stimulation and electrical recording of
40
+ brain activities was confirmed in mouse hippocampus in both acute and chronical applications up to 1
41
+ month. Minimal brain tissue damage has been confirmed after these applications. The capability of non-
42
+ genetic neural stimulation and recording enabled by mFOE opens up new possibilities for the
43
+ investigation of neural circuits and brings new insights into the study of ultrasound neurostimulation.
44
+ Introduction
45
+ Bidirectional communication with dynamic local circuits inside the brain of individual behaving animals
46
+ or humans has been an invaluable approach for fundamental studies of neural circuits and for effective
47
+ clinical treatment of neurological diseases, like epilepsy, Parkinson’ s disease, and depression1, 2.
48
+ Additionally, bidirectional neural interface paves the way for the closed-loop control, as it could enable
49
+ more sophisticated, real-time control over neural dynamics3, behaviors4 and achieve effective therapeutic
50
+ effect in neurological disease5, 6. To achieve real time assessment of the stimulated outcome, neural
51
+ interfaces with ability to simultaneously manipulate and directly monitor the neural activities are
52
+ preferred. Among the technologies developed in past decades, electrical stimulation and
53
+ electrophysiology recording have been widely used and forms the basis of current implantable devices,
54
+ which has been applied to clinical applications7. For example, to restore both the motor and sensory
55
+
56
+ 3
57
+
58
+ modalities, electric stimulation of the cortical surface is often associated with electrophysiology
59
+ recording8, 9, like electrocorticography (ECoG). Also, the bidirectional electrical stimulation has
60
+ demonstrated promising treatment effect in neurological diseases, such as epilepsy. The responsive focal
61
+ cortical stimulation (RNS), leveraging ECoG recording as the trigger to provide stimulation, showed a
62
+ statistically significantly greater reduction in seizure frequency and the benefits increased over time in a
63
+ two-year study10, 11. However, electrical stimulation has a limited spatial resolution due to current spread.
64
+ It also interferes with the electrical signals used for recording, leading to “contamination” in
65
+ electrophysiology recording2, 12. Although researchers are improving its performance through
66
+ technologies such as current steering13, novel electrode design14, and artifacts cancellation15, considering
67
+ the intrinsic physical properties of brain tissue16, the current spread, root cause of above-mentioned issues
68
+ is hard to be fully eliminated. Therefore, electrical stimulation for the bidirectional communication of
69
+ brain may not be the ideal candidate.
70
+ Being orthogonal with electrical recording, optical stimuli not only avoids the interference but
71
+ also enables a high spatial resolution. To take this advantage, early efforts developed so-called
72
+ optoelectrodes by simply assembling the optical fibers for optogenetics stimulation with the electrodes,
73
+ such as Utah arrays17-19, Michigan probes20, 21 and microwires22. Semiconductor fabrication techniques
74
+ and multiple material processing methods have recently been applied to improve the integration of those
75
+ bidirectional devices. New processing techniques not only make the device more compact but also
76
+ strengthen its functionality and biocompatibility. For example, monolithically integrated micro-light-
77
+ emitting-diodes (µLEDs) were used to reduce the complexity of light-guide structures and significantly
78
+ boosted the number of stimulation sites and stimulation resolution 23, 24. Alternatively, a high-throughput
79
+ thermal drawing method has been used to integrate the function components, for example, electrodes,
80
+ microfluidic channels, and optical waveguides, to the flexible multifunctional polymer fiber 25, 26.
81
+ Through this approach, the flexible fiber probes showed low bending-stiffness and enabled
82
+ multifunctionalities, including optogenetics, electrical recording and drug delivery 27-29. Since
83
+
84
+ 4
85
+
86
+ optogenetics relies on the expression of light-sensitive opsins in neurons through gene modification26, it is
87
+ challenging to apply optogenetics to non-human primates and human effectively and safely30.
88
+ Recently, our team showed non-genetic optoacoustic neural stimulation with a high spatial
89
+ resolution up to single neuron level31, 32. In an optoacoustic process, the pulsed light is illuminated on an
90
+ absorber, causing transient heating and thermal expansion, and generating broadband acoustic pulses at
91
+ ultrasonic frequencies33, 34. As a light mediated neural modulation method, optoacoustic is an ideal
92
+ candidate to work with electrical recording for bidirectional neural communication. Compared with
93
+ existing technologies, it exhibited the advantages as a light mediated method, including a high spatial
94
+ resolution and minimal crosstalk noise with electrical recording. Importantly, the non-genetic
95
+ optoacoustic neurostimulation alleviates the challenges and safety concern in optogenetics since no viral
96
+ transfection is required.
97
+ Here, we developed a multifunctional fiber-based optoacoustic emitter (mFOE) as a miniaturized
98
+ bidirectional brain interface performing simultaneously non-genetic neural stimulation and electrical
99
+ recording of the neural activities. Through a thermal drawing process,25, 35 fabrication of mFOE integrated
100
+ an optical waveguide and multiple electrodes within a single fiber with a total diameter of 300 µm,
101
+ compatible to the typical size of silica fibers used in optogenetic studies. An optoacoustic coating was
102
+ selectively deposited to the tip of the core optical waveguide in the mFOE through a controlled micro-
103
+ injection process. Upon nanosecond pulse laser delivered to the photoacoustic coating, the mFOE
104
+ generates a peak-to-peak pressure greater than 1 MPa, confirmed by the hydrophone measurement, which
105
+ is sufficient for successful neural stimulation in vitro and in vivo. By calcium imaging, the optoacoustic
106
+ stimulation function of the mFOE was validated in Oregon green-loaded rat primary neurons.
107
+ Importantly, we demonstrated the reliable functions of the chronic implanted mFOE for simultaneously
108
+ stimulating and recording neurons in mouse hippocampus. Chronic recording also demonstrated that the
109
+ embedded electrodes could provide long-term neural monitoring with a single-unit resolution. The
110
+ histological evaluation of the brain tissue response confirmed that our flexible mFOE established a stable
111
+
112
+ 5
113
+
114
+ and biocompatible multifunctional neural interface. mFOE is the first device integrated both optoacoustic
115
+ stimulation with electrical recording for bidirectional neural communication. With the bidirectional
116
+ capabilities and excellent biocompatibility, it offers a non-generic tools probing brain circuits, alternative
117
+ to the optoelectrode devices, with improved feasibility in non-human primates and human. It also opens
118
+ up potentials for closed-loop neural stimulation and brain machine interface.
119
+ Results
120
+ Design, fabrication and characterization of mFOE
121
+ Towards bidirectional neural communication, we have designed the mFOE to utilize the optoacoustic
122
+ stimulation as “writing” and electrophysiological recording as “reading” of the neural interface (Fig. 1a).
123
+ Previously, fiber based optoacoustic emitters have been developed as a miniature invasive ultrasound
124
+ transducer for the biomedical applications, such as intravascular imaging and interventional cardiology36,
125
+ 37. Recently, our work showed that fiber based optoacoustic emitters can also be applied to neural
126
+ stimulation in vitro and in vivo, with single neuron resolution and dual site capability32, 38. In these
127
+ studies, typically commercial silica fibers were used, together with optoacoustic coating. However, the
128
+ silica fiber, with Young’s modulus of ~70 GPa, is mismatched with mechanical properties of native
129
+ neural tissue (kilo- to mega pascals)2 and not easy to integrate with miniaturized electrodes for recording.
130
+ In this study, we took advantage of the fiber fabrication method developed by Anikeeva and Yoel25, and
131
+ utilized the polymer multifunctional fiber design as the base for the mFOE to delivering nanosecond laser
132
+ to the optoacoustic coating and to record electrical signals. Specifically, a multifunctional fiber with a
133
+ core optical waveguide and miniaturized electrodes was fabricated using the thermal drawing process
134
+ (TDP) as previously reported27 (Fig. 1b). The waveguide is made of polycarbonate core (PC, refractive
135
+ index nPC = 1.586, diameter = 150 µm) and polyvinylidene difluoride cladding (PVDF, refractive index
136
+ nPVDF = 1.426, thickness = 50 µm) as the core and the shell, respectively (Fig. 1c). BiSn alloy is used in
137
+ surrounding electrodes with diameters of 35 µm because of its conductivity and compatibility with TDP
138
+
139
+ 6
140
+
141
+ (Fig. 1c).This multifunctional fiber showed broadband transmission across the visible range to near
142
+ infrared region and sub-megaohm impedance when it has been prepared into two centimetres long27, 39.
143
+ To integrate the optoacoustic converter to the multifunctional fiber, the optoacoustic coating,
144
+ composed of light absorbers and thermal expansion matrix, is needed to be selectively coated on the core
145
+ waveguide distal end while keeping the surrounding electrodes exposed and conducting. Compared to
146
+ previously reported FOE fabrication, here we took several innovative steps. First, a pressure-driven pico-
147
+ litter injector was used to precisely deposit the optoacoustic materials to the core waveguide distal end.
148
+ The coating area was controlled through varying the injection volume (0.1 – 0.5 nL), which is controlled
149
+ by the regulated pressure (2-4 psi) over a set period of time (1-2 s, Supplementary Fig. S1) as described in
150
+ equation (1),
151
+ ������������ = ������������ ∙ ������������������������������������������������������������������������
152
+ 3
153
+ ∙ ������������ ∙ ������������
154
+
155
+
156
+
157
+
158
+
159
+ (1)
160
+ where ������������ is the injection volume, ������������ is a constant attributed to the unit conversion factors, effects of liquid
161
+ viscosity and the taper angle of micropipette, ������������������������������������������������������������������������ is the inner diameter of the pico-litter injector, ������������ is
162
+ the pressure, and ������������ is the deposition time. Two 3D translational stages with stereo microscopes were used
163
+ to precisely control the deposition localization. Second, instead of using carbon nanotubes (CNT), we
164
+ used carbon black (CB) embedded polydimethylsiloxane (PDMS) as the composite optoacoustic material.
165
+ CB exhibited similar wideband light absorption40, assuring the sufficient photoacoustic conversion for
166
+ neural stimulation. Importantly, due to its relative low viscosity 41, 42, CB/PDMS composite shows much
167
+ higher injectability compared with CNT/PDMS, therefore more comparable to the pico-liter deposition
168
+ process. Through these steps, we successfully coated 10-20 µm thick 10% w/w CB/PDMS composite
169
+ onto the 150 µm diameter core waveguide distal end while electrodes were still exposed as shown in Fig.
170
+ 1e. Collectively mFOE with the photoacoustic emitter and multiple electrodes has been successfully
171
+ fabricated.
172
+ To characterize the optoacoustic performance of mFOE, a Q-switched 1030 nm pulsed
173
+ nanosecond laser was applied with pulse energies of 16.6 µJ, 27.3 µJ and 41.8 µJ, respectively. The
174
+
175
+ 7
176
+
177
+ generated acoustic waves were measured by a 40 µm needle hydrophone placed at about 100 µm away
178
+ from the fiber tip. Representative pulse acoustic pulse with a width of approximately 0.08 µs was
179
+ generated by a single laser pulse as shown in Fig. 1f. Higher input laser pulse energy led to larger acoustic
180
+ pressure. A peak-to-peak pressure of 1.0, 1.6 and 2.3 MPa were measured with the pulse energy of 16.6,
181
+ 27.3 and 41.8 µJ, respectively. The frequency spectrum shows the broadband characteristic of typical
182
+ optoacoustic waves34, and the peak frequencies are around 12.5 MHz (Fig. 1g). Based on previous work,
183
+ we expected that such pressure and frequency is capable to successfully stimulate neurons in vitro and in
184
+ vivo. We also calculated the mechanical index (MI), a commonly used matrix, to evaluate the probability
185
+ of mechanical damage due to ultrasound generated. The MI of acoustic waves generated by 2.3 MPa is
186
+ 0.198, lower than 1.9, the safety threshold suggested by the Food and Drug Administration (FDA) safety
187
+ guidelines.
188
+
189
+ 8
190
+
191
+ Figure. 1 Design, fabrication and characterization of mFOE
192
+ a. Schematic of mFOE for bidirectional communication with neurons. Input laser pulse (red) is used to
193
+ generate optoacoustic waves (black) by the converter and the neural activities are recorded by embedded
194
+ electrodes as the output electrical signal (blue). b. Illustration of the thermal drawing
195
+ process. c. Components of the multifunctional fiber, including a PC/PVDF waveguide, BiSn alloy
196
+ electrodes and PC sacrifice layer. d. The selective deposition process for integrating the optoacoustic
197
+ converter to the core wave guide in the multifunctional fiber. A pressure-driven micro-injector is used to
198
+ control the volume of CS/PDMS deposited. 3D translation stages and microscope are used to control the
199
+ deposition location. Zoom-in: The micro pipette was aligned to the center of the fiber under the microscope.
200
+ e. Top view microscope image of the mFOE. Scale bar: 100 µm. f. Representative acoustic waveforms
201
+ under different laser pulse energy recorded by a needle hydrophone. g. Frequency spectrum of acoustic
202
+ waveforms shown in f.
203
+ mFOE stimulation of cultured primary neurons
204
+
205
+ a
206
+ ptoacousticwave
207
+ Pulsed laser
208
+ Electrical signal
209
+ q
210
+ C
211
+ d
212
+ Injector on 3D stage
213
+ Wave guide
214
+ Micro pipette
215
+ (PC/PVDF)
216
+ Heat
217
+ Fiber
218
+ Sacrificelayer(PC)
219
+ Vdrawing speed
220
+ Electrode (BiSn alloy)
221
+ Fiber holder
222
+ on 3D stage
223
+ e
224
+ Optoacoustic
225
+ g
226
+ 0.4
227
+ 41.8 μJ
228
+ 41.8 μJ
229
+ converter
230
+ 1.5
231
+ 27.3 μJ
232
+ 27.3 μJ
233
+ Pressure (MPa)
234
+ 16.6 μJ
235
+ 0.3
236
+ 16.6μJ
237
+ (a.u
238
+ Magnitude
239
+ 0.5
240
+ 0.2
241
+ 0
242
+ 0.1
243
+ 0.5
244
+ Electrode
245
+ 0
246
+ 0
247
+ 0.1
248
+ 0.2
249
+ 0.3
250
+ 0.4
251
+ 0
252
+ 20
253
+ 40
254
+ 60
255
+ 80
256
+ 100
257
+ Time (μs)
258
+ Frequency (MHz)9
259
+
260
+ To investigate mFOE can directly trigger the neuronal activity, we examined the response of cultured
261
+ primary neurons under mFOE stimulation. Because of the presence of calcium channels in neuronal
262
+ membrane and their activation during the depolarization, calcium imaging has been widely used to
263
+ monitor the neuronal activities43, 44. Here, we cultured and loaded the rat cortical neurons (days in vitro
264
+ 10-14) with a calcium indicator, Oregon Green™ 488 BAPTA-1 dextran (OGD-1)45 , and performed the
265
+ calcium imaging with an inverted wide-field fluorescence microscope (Supplementary Fig. S2). To
266
+ perform the optoacoustic stimulation, mFOE was placed approximately 50 µm above the in-focus target
267
+ neurons (Fig. 2a) by a micromanipulator under the microscope. 1030 nm 3 ns pulsed laser with a
268
+ repetition rate of 1.7 kHz was delivered to the mFOE through an optical fiber. The energy of laser pulse
269
+ was 41.8 µJ, corresponding to a peak-to-peak pressure of 2.3 MPa generated. Lower energy was tested
270
+ but did not induce calcium transient. The stimulation duration determined by each laser burst was 100 ms,
271
+ corresponding to 170 pulses (Supplementary Fig. S3). By applying 5 bursts of laser pulses with interval
272
+ of 1s, we investigated the reproducibility of the stimulation.
273
+ Using calcium imaging, we monitored the activities of all neurons in the field of view and divided
274
+ them into two groups: groups within the converter area (Fig. 2b) and outside the converter area (Fig. 2c).
275
+ For neurons within the converter area, i.e. the 100 µm from the center of the mFOE, Fig. 2b shows that 8
276
+ of 10 neurons showed successful and repeatable calcium transient (ΔF/F > 1%, the baseline standard
277
+ deviation) corresponding to each stimulation. Calcium transients are also repeatable for each burst applied
278
+ over the 1 s period, indicating the evoked neuronal activities and confirming the reliability and biosafety
279
+ of mFOE stimulation. For neurons outside the converter area, only 2 of 10 neurons responded. This result
280
+ also suggested the mFOE with the 150 µm center waveguide with photoacoustic coating provided a
281
+ spatial precision of ~200 µm for stimulation in vitro. This observation is consistent with that fiber based
282
+ optoacoustic converters generate a confined ultrasound fields with sizes comparable with the radius of
283
+ converter31.
284
+ Next, to investigate the threshold of mFOE stimulation, we varied the stimulation duration from 5
285
+ ms, 50 ms, 100 ms to 200 ms on neurons in different cultures (N = 15) under the same laser pulse energy
286
+
287
+ 10
288
+
289
+ of 41.8 µJ and the same repetition rate of 1.7 kHz. mFOE stimulation with duration of 5 ms did not
290
+ evoked any observable fluorescence change (n.s., p > 0.05) (Fig. 2g). Only when the duration was 50 ms
291
+ or longer, the mFOE successfully produced neural activation (ΔF/F > 1%, p < 0.01) as shown in Fig. 2d-f,
292
+ and Fig. 2h. Longer pulse durations leads to larger peak fluorescence changes, from 2.9 ± 1.1%, 6.0 ±
293
+ 2.8% to 7.8 ± 1.3% corresponding to 50 ms, 100 ms and 200 ms, respectively. For the longest stimulation
294
+ duration of 200 ms tested, no obvious change on morphology or elevation of baseline fluorescence
295
+ intensity was detected in neurons after multiple stimulations (Supplementary Fig. 4), indicating the safety
296
+ of stimulation.
297
+ Laser only control experiment was also performed. Laser light with same pulse energy of 41.8 µJ
298
+ and duration (200 ms, 100 ms and 50 ms) was delivered to OGD-1 loaded neurons through
299
+ multifunctional fiber without optoacoustic coating. None of neuron culture showed detectable calcium
300
+ response, distinct from the observed in mFOE stimulated neurons (Supplementary Fig. 5).
301
+ To evaluate the photothermal effect of the mFOE stimulation and its potential impact on neurons,
302
+ we also characterized the thermal profile of the mFOE in PBS during the acoustic generation.
303
+ Temperature was measured by an ultrafast thermal sensor with a sampling rate of 2000 Hz placed in
304
+ contact with mFOE optoacoustic coating under the microscope. The laser conditions were consistent with
305
+ neural stimulation test, i.e., the pulse energy was maintained at 41.8 µJ and the burst duration was varied
306
+ from 50 ms, 100 ms to 200 ms. The temperature increase on the mFOE surface was found to be 1.23 ±
307
+ 0.09 °C, 1.07 ± 0.08 °C, 0.96 ± 0.08 °C for 200, 100, 50 ms laser durations, respectively (Supplementary
308
+ Fig. 6). Such temperature increase is far below the previously reported threshold of thermal-induced
309
+ neural stimulation (ΔT > 5 °C)46, 47. Taken together, we conclude that activation of neurons was due to the
310
+ mFOE optoacoustic stimulation.
311
+
312
+ 11
313
+
314
+
315
+ Figure 2. Calcium transients induced by mFOE in cultured primary neurons.
316
+ a. Calcium image of primary cultured neurons loaded with OGD-1. Twenty neurons within (orange) and
317
+ outside (blue) the optoacoustic converter area are circled and labelled. Scale bar: 100 µm. Solid circle:
318
+ area outside the converter area; dashed line circle: area within the optoacoustic converter area. b-c.
319
+ Calcium traces of neurons undergone repeated mFOE stimulations with a laser pulse train duration of 100
320
+ ms (red dots). Each pulse train was repeated 5 times. Colors and numbers of the traces are corresponding
321
+ to the neurons labelled in a. d-g. Average calcium traces of neurons triggered by mFOE stimulation with
322
+ durations of 200 ms (d), 100 ms (e), 50 ms (f) and 5 ms (g), respectively. Shaded area: the standard
323
+ deviation (SD). N=15 h. Average maximum ΔF/F of neurons stimulated by mFOE. N = 15. (n.s.: non-
324
+ significant, p > 0.05; *: p < 0.05; **: p < 0.01; ***: p < 0.001, One-Way ANOVA and Tukey’s mean
325
+ comparison test)
326
+ In vivo simultaneous optoacoustic stimulation and electrophysiological recording
327
+ Since the animal experiment is a significant part of the study in neuroscience and neurological diseases,
328
+ we further investigated the performance of mFOE in the wild type C57BL/6J mice. In vivo optoacoustic
329
+ stimulation was performed by delivering pulsed laser to the implanted mFOE, and the optoacoustically
330
+ stimulated neuronal activities were recorded through electrodes in the mFOE (Fig. 3a). Experimentally,
331
+ we implanted the mFOE into the hippocampus of mice (N =5). The chronically implanted mFOE allows
332
+
333
+ a
334
+ b
335
+ 10
336
+ hhyeh10
337
+ C
338
+ 9
339
+ 9
340
+ 9
341
+ 8
342
+ 8
343
+ Me
344
+ 40
345
+ 8
346
+ 6
347
+ 6
348
+ 5
349
+ 4
350
+ 3
351
+ 2
352
+ 2
353
+ 10%
354
+ AA
355
+ 5s'
356
+ d
357
+ e
358
+ +
359
+ g
360
+ h
361
+ 0.12
362
+ 200 ms
363
+ 100 ms
364
+ 50 ms
365
+ 5 ms
366
+ 0.1
367
+ 0.1
368
+ 0.1
369
+ 0.1
370
+ 0.1
371
+ T
372
+ 0.08
373
+ F 0.05
374
+ 0.05
375
+ 0.05
376
+ 0.05
377
+ F
378
+ 0.06
379
+ A
380
+ F
381
+ 0.04
382
+ n.s.
383
+ 0.02
384
+ 0
385
+ A
386
+ 0
387
+ 0
388
+ 0
389
+ 0
390
+ 0
391
+ 1
392
+ 2
393
+ 3
394
+ 0
395
+ 1
396
+ 2
397
+ 3
398
+ 0
399
+ 1
400
+ 2
401
+ 3
402
+ 0
403
+ 1
404
+ 2
405
+ 3
406
+ -0.02
407
+ Time (s)
408
+ Time (s)
409
+ Time (s)
410
+ Time (s)
411
+ 200ms100ms50ms5ms12
412
+
413
+ mice to move freely after surgery (Fig 3b). During stimulation and recording tests, the mFOE was
414
+ coupled with the laser source and electrophysiological recording headstage through the standard ferrule
415
+ and pin connector, respectively. The stimulation and recording were conducted in the mice under
416
+ continuous anesthesia induced and maintained by isoflurane. Based on the threshold of optoacoustic
417
+ stimulation obtained in in vitro studies, 50 ms bursts of laser pulses with a pulse energy of 41.8 µJ were
418
+ delivered to the mFOE at 1Hz during the 5 second treatment period. The simultaneous
419
+ electrophysiological recording by mFOE electrodes was bandpass filtered to examine the local field
420
+ potential (LFP, 0.5-300 Hz). Simultaneous optoacoustic stimulation and electrophysiological recording
421
+ were performed at multiple time points, including 3 days, 7 days, 2 weeks and 1 month (Fig. 3c-f). Three
422
+ out of five mice tested showed successful simultaneous stimulation and recording functions for testing
423
+ periods of 3 days to one month.
424
+ The evoked brain activities corresponding to the optoacoustic stimulation were confirmed by
425
+ monitoring the LFP response. LFP response at two weeks after implantation was detected with latency of
426
+ 7.19 ± 2.29 ms (N = 15, from three mice). The amplitude of LFP response varied at four time points. The
427
+ largest and smallest responses occurred at 2 weeks and 1 month, respectively. A possible reason for this
428
+ observation may be the brain tissue injury and healing after the implantation. These results collectively
429
+ demonstrate the reliability of the optoacoustic stimulation and recording functions of the implanted
430
+ mFOE in the animals.
431
+ To eliminate the possibility that LFP response was induced by electrical noise or laser artifacts,
432
+ we also conducted two sham control experiments. In the light only control group, we implanted a
433
+ multifunctional fiber without optoacoustic coating to the mouse hippocampus and delivered the laser light
434
+ with the same condition. The LFP recorded didn’t correlate to the laser pulse train, indicating the
435
+ spontaneous brain activities were recorded and light only did not invoke the LFP response
436
+ (Supplementary Fig. 7a). In the dead brain control group, we tested the optoacoustic stimulation through
437
+ mFOE implanted to the euthanized mouse and did not observe the corresponding LFP response
438
+
439
+ 13
440
+
441
+ (Supplementary Fig. 7b). These results collectively confirm the signals we detected from mFOE
442
+ stimulation were not artifacts.
443
+ We further evaluated the recording performance of implanted mFOE. To evaluate the ability of
444
+ mFOE for single unit recording, the electrophysiological signals recorded were bandpass filtered for spike
445
+ activity (0.5-3 kHz, Fig. 3g). Through a principal-component analysis (PCA) based spike sorting
446
+ algorithm, two spike clusters can be isolated from an endogenous neural recording (Fig. 3j). The cluster
447
+ quality was assessed by two common measures48, Lratio and isolation distance. Lratio is 0.0017 and isolation
448
+ distance has the value of 99.37. The first averaged spike shape (Fig. 3h) showed a narrower and larger
449
+ depolarization than that of the second spike shape (Fig. 3i). The different spike waveform and the cluster
450
+ analysis suggested that the action potentials were recorded from at least two different groups of neurons49,
451
+ 50. Thus, the successfully spike sorted neural activities from CA3 confirmed the ability of mFOE
452
+ electrodes for the single-unit recording.
453
+ To examine the sensitivity of LFP recording, at one month after implantation we altered the
454
+ anesthesia level via adjusting the induced isoflurane concentration during the recording to see if the
455
+ characteristic anesthesia dosage-dependent changes can be observed (Fig. 3k). Initially, a low level of
456
+ anesthesia was maintained at 0.5% v/v isoflurane, and recorded LFP showed that spontaneous brain
457
+ activities occurred continuously (i in Fig. 3h. and Fig. 3l). Then a higher-level anesthesia (3% v/v
458
+ isoflurane) was applied for 3 minutes. After increased the isoflurane level, some spontaneous brain
459
+ activities were suppressed and a hyperexcitable brain state was induced, where the voltage alternation
460
+ (bursts) and isoelectric quiescence (suppression) appeared quasiperiodically27, 51 (ii in Fig. 3h and Fig.
461
+ 3m). With maintaining 3% v/v isoflurane, a deep anesthesia state was induced in the animal. At the same
462
+ time, both respiration rate and responsiveness to toe pinch decreased due to the higher anesthetic level.
463
+
464
+ Less voltage alternation occurred and for the most of time the LFP signal was a flat line
465
+ (suppression, iii in Fig. 3h and Fig. 3n). Compared with initial stage, γ band LFP activity in 30-100 Hz
466
+ was decreased due to the higher concentration of isoflurane as shown in the power spectrum52 (Fig. 3n).
467
+ Later, when the concentration of isoflurane was reduced to 0.5% v/v again, the LFP activity returned to a
468
+
469
+ 14
470
+
471
+ similar level as measured in the initial stage. Taken together, this isoflurane dosage-dependent
472
+ characteristic confirmed the accuracy of LFP recording by mFOE.
473
+
474
+ Figure. 3 Simultaneous optoacoustic stimulation and electrophysiological recording by implanted
475
+ mFOE in mouse hippocampus.
476
+ a. Illustration of the mFOE enabled bidirectional neural communication using laser signal as input and
477
+ electrical signal as readout. b. mFOE was implanted into hippocampus of a wild type C57BL/6J mouse.
478
+ c-f. Simultaneous optoacoustic stimulation and electrophysiological recording performed at 3 days (c), 7
479
+ days (d), two weeks (e) and one month (f) after implantation. Blue dots the laser pulse trains. For each
480
+ laser train: 50 ms burst of pulses, pulse energy of 41.8 µJ, laser repetition rate 1.7 kHz. g. Part of the
481
+ filtered spontaneous activity containing two separable units recorded by mFOE electrode at one month
482
+ after implantation. h-i. Spike shapes of two separable units in g. j. Principal-components analysis (PCA)
483
+ of the two units. k. Local field potential (LFP) recorded by mFOE one month after implantation with an
484
+ alternating anaesthesia level (0.5-3% v/v isoflurane). l-n. different LFP responses induced by varying the
485
+ concentration of isoflurane: l corresponds to the initial stage (0.5% of isoflurane level); m corresponds to
486
+ the burst/suppression transition stage (after increasing the isoflurane level to 3%); n corresponds to the
487
+ suppression stage (the isoflurane level was maintained at 3% and took effect).
488
+
489
+ a
490
+ c
491
+ d
492
+ Optical
493
+ Electrical
494
+ 0.5
495
+ 3 days
496
+ 0.5
497
+ 7 days
498
+ input
499
+ readout
500
+ (mV)
501
+ 0.0
502
+ Voltage (
503
+ 0.0
504
+ -0.5
505
+ -1.0-
506
+ -1.0-
507
+ 0
508
+ 2
509
+ 4
510
+ 6
511
+ 10
512
+ 0
513
+ 4
514
+ 6
515
+ 8
516
+ 10
517
+ Time (s)
518
+ Time (s)
519
+ b
520
+ e
521
+ (mV)
522
+ 0.5
523
+ 2 weeks
524
+ 0.5-
525
+ 1 month
526
+ (mV)
527
+ Voltage (
528
+ 0.0
529
+ Voltage
530
+ 0.0
531
+ -0.5
532
+ -0.5
533
+ -1.0
534
+ -1.0
535
+ 0
536
+ 2
537
+ 4
538
+ 6
539
+ 8
540
+ 10
541
+ 0
542
+ 2
543
+ 4
544
+ 6
545
+ 8
546
+ 10
547
+ Time (s)
548
+ Time (s)
549
+ g
550
+ h
551
+ 100
552
+ 40
553
+ 40
554
+ 20
555
+ 20
556
+ Yoltage (μV)
557
+ (Λ)
558
+ 50
559
+ 0
560
+ 0
561
+ -20
562
+ PC2
563
+ 40
564
+ 40
565
+ -50
566
+ 60
567
+ -60
568
+ 0
569
+ 0.5
570
+ 1
571
+ 1.5
572
+ 2
573
+ 0
574
+ 2 s
575
+ 0.5
576
+ 1
577
+ 1.5
578
+ 2
579
+ -100
580
+ Time (ms)
581
+ Time(ms)
582
+ -200
583
+ -150
584
+ -100
585
+ -50
586
+ 0
587
+ 50
588
+ 100
589
+ PC1
590
+ k
591
+ 3 %
592
+ i Initial stage
593
+ m
594
+ ii Burst/suppression
595
+ n
596
+ ii Suppression
597
+ 0.5 %
598
+ M
599
+ 2 s
600
+ N
601
+ (ZH)
602
+ 100
603
+ 100
604
+ Frequency
605
+ (dB)
606
+ Frequency
607
+ (p)
608
+ Frequency
609
+ -50
610
+ 50
611
+ -50
612
+ 50
613
+ Power
614
+ 50
615
+ 100
616
+ -100
617
+ 50
618
+ -100
619
+ 150
620
+ ii
621
+ ili
622
+ 2
623
+ 4
624
+ 68101214
625
+ 2
626
+ 468101214
627
+ 2
628
+ 68101214
629
+ 50 s
630
+ Time (s)
631
+ Time (s)
632
+ Time (s)15
633
+
634
+ Foreign body response comparison between mFOE and standard optical fiber using
635
+ immunohistochemistry
636
+ Foreign body response is a critical property of implantable neural interface to assure their usage in a safe
637
+ and chronic way, since the physical insertion into brain tissue commonly initiates a progressive
638
+ inflammatory tissue response53. To evaluate the biocompatibility of mFOE, we compared the foreign
639
+ body response of mouse brain to mFOE with the similar size standard silica optical fibers (diameter = 300
640
+ µm), which is widely used in optogenetic technologies54, 55. The immunohistochemistry analysis of
641
+ surrounding brain tissue was performed from mice (N = 3) implanted with the mFOE and a conventional
642
+ silica fiber 3 days and 1 month after implantation (Fig. 4a). The damage to surrounding neurons from
643
+ implant was assessed through evaluating neuronal density using the neuronal nuclei (NeuN) markers (Fig.
644
+ 4b). Number of neurons was calculated by counting the NeuN-positive cells per field of view (650 × 650
645
+ µm). The presence of ionized calcium-binding adaptor molecule 1 (Iba1, Fig. 4c) and glial fibrillary
646
+ acidic protein (GFAP, Fig. 4d) were used as the markers for activated microglia and astrocytic response,
647
+ respectively.
648
+ Compared with the silica fiber, mFOE induced significantly less microglial response (p < 0.01,
649
+ Fig. 4c, f) and astrocyte reactivity (p < 0.001, Fig. 4d, g), but no significant difference was observed on
650
+ the neuronal density (Fig. 4b, e) 3 days after implantation. A decrease in foreign body response,
651
+ specifically, higher neuronal density and lower microglia and astrocytic response (Fig. 4e-g), was
652
+ observed from 3 days to 1 month after implantation of both mFOE and silica fiber and no significant
653
+ difference was observed between mFOE and silica fiber 1 month after implantation. Taken together, the
654
+ immunohistochemistry analysis confirmed that mFOE yielded less foreign body response in the short
655
+ period, i.e., 3 days, after implantation and showed similar biocompatibilty with silica fiber at longer
656
+ implantation time, i.e., 1 month.
657
+
658
+ 16
659
+
660
+
661
+ Figure. 4 Foreign body response comparison of mFOE and silica fiber using
662
+ immunohistochemistry.
663
+ a-d. Immunohistochemistry images of mouse brains implanted with mFOE and silica fiber one month
664
+ after implantation (N = 3). Scale bar: 100 µm. Brain slices were labelled with the neuron-specific protein
665
+ (NeuN, cyan), ionized calcium-binding adaptor molecule 1 (Iba1, red) and glial fibrillary acidic protein
666
+ (GFAP, green). e. Number of neurons in the field of view, calculated by counting the NeuN-positive cells
667
+ for mFOE and silica fiber at 3 days and 1 mon after implantation. f. Microglial reactivity, assessed by
668
+ counting the Iba-1 labelled area, for mFOE and silica fiber at 3 days and 1 mon after implantation. g.
669
+ Astrocyte reactivity, assessed by counting the GFAP labelled area, for mFOE and silica fiber at 3 days
670
+ and 1 mon after implantation. For each experimental group, two to four brain slices were used from each
671
+
672
+ a
673
+ mFOE
674
+ Silica Fiber
675
+ 800
676
+ e
677
+ n.s.
678
+ 700
679
+ neurons
680
+ Composite
681
+ 600
682
+ 500
683
+ Number
684
+ n.s.
685
+ 400
686
+ 300
687
+ mFOE
688
+ b
689
+ Silica fiber
690
+ 200
691
+ Day 3
692
+ Day30
693
+ TimePoint
694
+ NeuN
695
+ **
696
+ mFOE
697
+ 3×104
698
+ Silica fiber
699
+ n.s.
700
+ 2 ×10
701
+ C
702
+ 10
703
+ 6×103
704
+ Iba1
705
+ Day 3
706
+ Day30
707
+ TimePoint
708
+ g
709
+ 3×10
710
+ mFOE
711
+ Silica fiber
712
+ d
713
+ 2 × 10
714
+ GFAP area (μm"
715
+ n.s.
716
+ GFAP
717
+ 10
718
+ 6 ×10°
719
+ Day 3
720
+ Day30
721
+ TimePoint17
722
+
723
+ mouse (N= 3). (n.s.: non-significant, p > 0.05; *: p < 0.05; **: p < 0.01; ***: p < 0.001, One-Way
724
+ ANOVA and Tukey’s mean comparison test)
725
+
726
+ Discussion
727
+ In this study, we designed and developed a miniaturized fiber-based device, i.e. mFOE, for
728
+ bidirectional neural communication. mFOE performs the “write” function, i.e. non-genetic optoacoustic
729
+ stimulation and the “read” function, i.e. simultaneous electrophysiological recording. The broadband
730
+ acoustic wave with a broadband ultrasound pulse with pulse width about 0.1 µs and a center frequency at
731
+ 12.5 MHz and a peak pressure of 2.3 MPa with pulse numbers >85 generated by mFOE successfully
732
+ stimulate neurons with a spatial resolution of approximately 200 µm in primary rat cortical neuron
733
+ culture. By implanting mFOE into mouse hippocampus, we demonstrated its ability for simultaneous
734
+ optoacoustic stimulation and electrophysiological recording and superior biocompatibility as a chronic
735
+ bidirectional neural interface. Reliable stimulation and LFP recording have been achieved up to one
736
+ month post implantation. Recording quality has been demonstrated by single unit recording.
737
+ For the first time, combining this pico-liter deposition and thermal fiber pulling, we successfully
738
+ integrated an optoacoustic converter to the polymer multifunctional fiber. Different from the conventional
739
+ dip-coating method36, 56, the selective deposition through micro-injection allows the easy fabrication of
740
+ optoacoustic emitter in a volume and position-controlled way. Through the selective deposition, the
741
+ dimension of optoacoustic emitter is no longer limited by the tip sizes of optical fibers. Our choice of
742
+ CB/PDMS composite as the optoacoustic material is also essential as it is comparable with this deposition
743
+ process with a fine volume control at pico liter level. Besides the application in neural interface, such
744
+ design and fabrication method can also be applied to optical ultrasound probes used in imaging37, 57, for
745
+ example, in the tip engineering and the integration to photonics crystal fibers.
746
+ We introduced the optoacoustic stimulation as a new strategy for “writing” in the bidirectional
747
+ neural interface. Compared with previous optoelectrode devices based on optogenetics24, 25, 27 and
748
+
749
+ 18
750
+
751
+ photothermal58, 59, the non-genetic optoacoustic stimulation enabled by mFOE reduces the barrier of
752
+ transgenic techniques for applications in primate and potentially human, and avoids the thermal toxicity.
753
+ At the same time, it offers the spatial precision benefit from the confined ultrasound field. It is orthogonal
754
+ to electrical recording, therefore minimizing crosstalk with electrical recording. As an emerging
755
+ neuromodulation method, the mechanism of optoacoustic stimulation is still not fully understood but
756
+ more studies indicated that mechanosensitive ion channels are responsible for the activation of neurons60,
757
+ 61.
758
+ Bidirectional brain interfaces are important research tools to understand brain circuits, potential
759
+ treatments for neurological disease and bridges to brain computer interface for broad applications. New
760
+ features of mFOE compared to the previous fiber based interface, such as non-genetic and non-electrical
761
+ stimulation are critical to advance these applications. For example, closed-loop neuromodulation has been
762
+ demonstrated to be superior to the conventional open-loop system, as it can achieve more responsive and
763
+ real-time control over neural dynamics. In neurological diseases treatment, combining the detection and
764
+ in situ intervention improves the treatment effectiveness and safety. Because of its bidirectional
765
+ capabilities, mFOE has the potential to be used as a new brain interface with closed-loop capability.
766
+ Using epilepsy as an example, by implanting the mFOE into seizure foci, the continuous LFP recording
767
+ can guide the localized optoacoustic stimulation and intervene can be triggered at the early stage before
768
+ seizure progresses into a generalized seizure. The unique orthogonal non-electrical optoacoustic
769
+ stimulation and electrical recording prevents “contamination” of the recording signals, potentially
770
+ offering a more effective closed-loop strategy.
771
+ In comparison of the optoelectrodes fabricated through semiconductor fabrication process, the
772
+ recording and stimulation sites of the current mFOE design is fixed at the core waveguide and the number
773
+ of channels is limited because of the nature of multifunctional fiber. Some post processing methods have
774
+ been proposed to tackle this challenge, like the laser micromachining technique27. In addition, it is
775
+ possible to further engineer the fiber to offer multiple and selective stimulation sites62. With the further
776
+
777
+ 19
778
+
779
+ development of multifunctional fiber strategy, we believe the bandwidth of mFOE would be improved
780
+ and open more opportunities in the research of neuroscience and neurological diseases.
781
+
782
+ Methods
783
+ Multifunctional fiber fabrication and optoacoustic emitter integration
784
+ Multifunctional fibers were fabricated from a preform fiber and then drawn into thin fibers through TDP
785
+ in a customized furnace. For the preform fiber, PVDF film (Mcmaster) and PC film (laminated plastics)
786
+ were rolled onto a PC rod (Mcmaster) and followed by a consolidation process in vacuum at 200 °C.
787
+ Next, four rectangular grooves (2 mm × 2 mm) were machined on the solid PC layer and inserted with the
788
+ BiSn (Indium Corporate) electrodes. Then, another PVDF layer was rolled over the rod to form an
789
+ insulation layer for the electrodes and followed by an additional PC as the sacrifice layer for the
790
+ convenience of TDP. The detailed fabrication process was discussed in the previous paper27.
791
+ A composite of 10% carbon black (diameter < 500 nm, Sigma Aldrich) and 90%
792
+ polydimethylsiloxane (PDMS, Sylgard 184, Dow Corning Corporation, USA) were used as the
793
+ optoacoustic material. The mixture was sonicated for 1 hour followed by degassing in vacuum for 30
794
+ minutes. The mixture was then filled in the glass micropipette (Inner diameter = 30 µm, TIP30TW1,
795
+ World Precision Instruments, USA) connected to the pico-liter injector (PLI-100A, Warner Instruments,
796
+ USA). Under the microscope, the glass micropipette was aligned with the core waveguide of
797
+ multifunctional fiber and the mixture was deposited to the surface of the core waveguide by controlling
798
+ the injection pressure and time. The deposited fiber was then cured vertically at room temperature for 2
799
+ days.
800
+ Before use, mFOE was further prepared for the optical coupling and electrodes connection. For
801
+ the optical coupling, a ceramic ferrule (Thorlabs, USA) was added and affixed to the end of the fiber by
802
+ the 5-min epoxy (Devcon, ITW Performance Polymers, USA). Then the end surface was polished by
803
+
804
+ 20
805
+
806
+ optical polishing papers to reduce roughness from 30 µm to 1 µm. For the connection to electrodes
807
+ embedded in the multifunctional fiber, the electrodes were exposed manually along the side wall of the
808
+ fiber by using a blade and silver paint (SPI Supplies, USA). Then copper wires were wrapped around the
809
+ fiber at each exposure locations along the fiber and the silver paint were applied for the fixation and lower
810
+ resistance. The copper wires connected to fiber electrodes were soldered to the pin connector while a
811
+ stainless-steel wire was also soldered as the ground wire for later extracellular recording. In addition, the
812
+ 5-min epoxy (Devcon, ITW Performance Polymers, USA) was applied to the connection interface for
813
+ strengthening affixation and better electrical insulation.
814
+ Optoacoustic wave characterization
815
+ To generate the optoacoustic signal, a compact Q-switched diode-pumped solid-state laser (1030 nm, 3
816
+ ns, 100 μJ, repetition rate of 1.7 kHz, RPMC Lasers Inc., USA) was used as the excitation laser source.
817
+ The laser was first connected to an optical fiber through a 200 µm fiber coupling module and then
818
+ connected to the mFOE with a SubMiniature version A (SMA) connector. The pulse energy was adjusted
819
+ through a fiber optic attenuator (varied gap SMA Connector, Thorlabs, Inc., USA). The acoustic signal
820
+ was measured through a homebuilt system including a needle hydrophone (ID. 40 µm; OD, 300 µm) with
821
+ a frequency range of 1–30 MHz (NH0040, Precision Acoustics Inc., Dorchester, UK), an amplifier and an
822
+ oscilloscope. The mFOE tip and hydrophone tip were both immersed in degassed water. The pressure
823
+ values were calculated based on the calibration factor provided by the hydrophone manufacturer. The
824
+ frequency data was obtained through a fast Fourier transform (FFT) calculation using the OriginPro 2019.
825
+ Embryonic neuron culture
826
+ All experimental procedures complied with all relevant guidelines and ethical regulations for animal
827
+ testing and research established and approved by Institutional Animal Care and Use Committee (IACUC)
828
+ of Boston University (PROTO201800534). Primary cortical neurons were isolated from embryonic day
829
+ 15 (E15) Sprague−Dawley rat embryos of either sex (Charles River Laboratories, MA, USA). Cortices
830
+
831
+ 21
832
+
833
+ were isolated and digested in TrypLE Express (ThermoFisher Scientific, USA). Then the neurons were
834
+ plated on poly-D-lysine (50 μgmL−1, ThermoFisher Scientific, USA)-coated glass bottom dish (P35G-
835
+ 1.5-14-C, MatTek Corporation, USA). Neurons were first cultured with a seeding medium composed of
836
+ 90% Dulbecco’s modified Eagle medium (ThermoFisher Scientific, USA) and 10% fetal bovine serum
837
+ (ThermoFisher Scientific, USA) and 1% GlutaMAX (ThermoFisher Scientific, USA), which was then
838
+ replaced 24 h later by a growth medium composed of Neurobasal Media (ThermoFisher Scientific, USA)
839
+ supplemented with 1× B27 (ThermoFisher Scientific, USA), 1× N2 (ThermoFisher Scientific, USA), and
840
+ 1× GlutaMAX (ThermoFisher Scientific, USA). Half of the medium was replaced with fresh growth
841
+ medium every 3 or 4 days. Cells cultured in vitro for 10−14 days were used for Oregon Green labelling
842
+ and PA stimulation experiments.
843
+ In vitro neurostimulation and calcium imaging
844
+ Oregon Green™ 488 BAPTA-1 dextran (OGD-1) (ThermoFisher Scientific, USA) was dissolved in 20%
845
+ Pluronic F-127 in dimethyl sulfoxide (DMSO) at a concentration of 1 mM as stock solution. Before
846
+ imaging, neurons were incubated with 2 µM OGD-1 for 30 min, followed by incubation with normal
847
+ medium for 30 min. Q-switched 1030 nm nanosecond laser was used to generate light and delivered to
848
+ mFOE. The pulse energy was adjusted through a fiber optic attenuator (varied gap SMA Connector,
849
+ Thorlabs, Inc., USA). Notably, 1030 nm is far from the excitation peak of Oregon Green (494 nm) and
850
+ pass band of emission filter (500-540 nm), therefore assuring no effect from direct excitation of OGD by
851
+ any light leak from the fiber. A 3D translational stage was used to position the mFOE approaching the
852
+ target neurons.
853
+ Calcium fluorescence imaging was performed on a lab-built wide-field fluorescence microscope
854
+ based on an Olympus IX71 microscope frame with a 20× air objective (UPLSAPO20X, 0.75NA,
855
+ Olympus, USA), illuminated by a 470 nm LED (M470L2, Thorlabs, USA), an emission filter (FBH520-
856
+ 40, Thorlabs, USA), an excitation filter (MF469-35, Thorlabs) and a dichroic mirror (DMLP505R,
857
+ Thorlabs, USA). Image sequences were acquired with a scientific CMOS camera (Zyla 5.5, Andor,
858
+
859
+ 22
860
+
861
+ Oxfords Instruments, UK) at 20 frames per second. The fluorescence intensities, data analysis, and
862
+ exponential curve fitting were analyzed using ImageJ (Fiji) and MATLAB 2022.
863
+ Implantation surgery procedure
864
+ All surgery procedures complied with all relevant guidelines and ethical regulations for animal testing and
865
+ research established and approved by Institutional Animal Care and Use Committee (IACUC) of Boston
866
+ University (PROTO201800534). Eight to ten weeks old male wildtype C57BL/6-E mice (Charles River
867
+ Laboratories, US) were received and allowed to acclimate for at least 3 days before enrolling them in
868
+ experiments. All mice in experiments had access to food and water ad libitum and were kept in the BU
869
+ animal facility maintained for 12-h light/dark cycle. During the implantation surgery, mice were
870
+ anesthetized by isoflurane (5% for induction, 1-3.5% during the procedure) and positioned on a
871
+ stereotaxic apparatus (51500D, Stoelting Co., USA). After hair removal, a small incision was made by
872
+ sterile surgery scalpel at the target region and then a small craniotomy was made by using a dental drill.
873
+ Assembled mFOE was inserted into mice hippocampus (−2.0 mm AP, 1.5 mm ML, 2 mm DV) using the
874
+ manipulator with respect to the Mouse Brain Atlas. The ground stainless steel wire was soldered to a
875
+ miniaturized screw (J.I. Morris) on the skull. Finally, the whole exposed skull area was fully covered by a
876
+ layer of Metabond (C&B METABOND, Parkell, USA) and dental cement (51458, Stoelting Co., USA).
877
+ Buprenorphine SR was used to provide long effective analgesia after the surgery.
878
+ In vivo electrophysiology recording and optoacoustic stimulation
879
+ Extracellular recording was performed through an electrophysiology system (Molecular Devices, LLC,
880
+ USA). mFOE electrodes were connected to the amplifier (Multiclamp 700B, Molecular Devices, LLC,
881
+ USA) through the pin connector and headstages after the animals recovered from surgeries. The amplified
882
+ analog signal was then converted and recorded by the digitizer (Digidata 1550, Molecular Devices, LLC,
883
+ USA).
884
+
885
+ 23
886
+
887
+ Q-switched 1030 nm nanosecond laser was used to generate light and delivered to mFOE. During the
888
+ extracellular electrophysiological recording, the preset trigger signal was generated by the digitizer and
889
+ used to trigger the Q-switch laser for optoacoustic stimulation. The pulse energy was adjusted through a
890
+ fiber optic attenuator (varied gap SMA Connector, Thorlabs, Inc., USA).
891
+ Data analysis was performed with Matlab and OriginPro and custom scripts were used to analyse the local
892
+ field potential and spike sorting. The raw extracellular recordings were first band filtered for local field
893
+ potential results (LFP, 0.5 – 300 Hz) and spike results (300 – 5000 Hz). A custom Matlab script was used
894
+ to create spectrograms to visually support the analysis of the LFPs in both the time domain and the
895
+ frequency domain. The spike sorting algorithm was implemented through several steps: first, individual
896
+ spike signals with length of 3 ms were picked up from the full recording through a standard amplitude
897
+ threshold method; then the dimensionality of each spike signal was reduced via the principal component
898
+ analysis (PCA) and unsupervised learning algorithms (K-means clustering) was used to separate out the
899
+ clusters.
900
+ Foreign body response assessment via immunohistochemistry
901
+ To compare the tissue response, animals were implanted with a silica optical fiber (diameter = 300 µm,
902
+ FT300EMT, Thorlabs, Inc, USA) and mFOE for 3 days or 4 weeks. Then at target timepoints, animals
903
+ were euthanized and transcardially perfused with phosphate-buffered saline (PBS, ThermoFisher
904
+ Scientific, USA) followed by 4% paraformaldehyde (PFA, ThermoFisher Scientific, USA) in PBS. The
905
+ fiber probes were carefully extracted before the extraction and then the brains were kept in 4% PFA
906
+ solution for one day at 4 °C. Brains were sectioned in the horizontal plane at 75 µm on a vibrating blade
907
+ vibratome. Free-floating brain slices were washed in PBS and blocked for 1 hour at room temperature in a
908
+ blocking solution consisting of 0.3% Triton X-100 (vol/vol) and 2.5% goat serum (vol/vol) in PBS. After
909
+ blocking, brain slices were incubated with the primary antibodies in the PBS solution with 2.5% goat
910
+ serum (vol/vol) for 24 hours at 4 °C. Primary antibodies used included rat anti-GFAP (Abcam Cat. #
911
+
912
+ 24
913
+
914
+ ab279291, 1:500), chicken anti-NeuN (Millipore Cat. # ABN91, 1:500), and rabbit anti-Iba1 (Abcam Cat.
915
+ # ab178846, 1:500). Following primary incubation, slices were washed three times with PBS for 10 min
916
+ at room temperature. The brain slices were then incubated with secondary antibodies in the PBS solution
917
+ with 2.5% goat serum (vol/vol) for 2 hours at room temperature. Secondary antibodies used included goat
918
+ anti-rat Alexa Fluor 488 (Abcam Cat. # ab150157, 1:1000), goat anti-rabbit Alexa Fluor 568 (Abcam Cat.
919
+ # ab175471, 1:1000) and goat anti-chicken Alexa Fluor 647 (Abcam Cat. # ab150171, 1:1000). Slices
920
+ were then washed three times with PBS for 10 min at room temperature. Before imaging, slices were
921
+ stained with DAPI solution (1 µg/ml, Millipore, USA) for 15 minutes at room temperature. All
922
+ fluorescent images were acquired with a laser scanning confocal microscope (Olympus FV3000) with an
923
+ air 20× objective with a numerical aperture NA = 0.75 unless otherwise noted. Neuron density was then
924
+ calculated within the normalized area by counting NeuN labeled cell bodies using the cell counter plugin
925
+ (ImageJ). Area analysis of Iba1 and GFAP labeled cells was performed by creating binary layers of the
926
+ fluorescence images using the threshold function and quantified using the measurement tool (ImageJ).
927
+ Statistical information
928
+ Data shown are mean ± standard deviation. For the comparison on peak fluorescence change of in vitro
929
+ optoacoustic stimulation, one-way ANOVA and Tukey’s mean comparison test were conducted by using
930
+ OriginLab. 15 stimulation events were compared for each condition. For the comparison of foreign body
931
+ response between silica fiber and mFOE, N > 8 brain slices from 3 animals were analysed using one-way
932
+ ANOVA and Tukey’s mean comparison test. The p values were determined as n.s.: nonsignificant, p >
933
+ 0.05; *: p < 0.05; **: p < 0.01; ***: p < 0.001. Statistic analysis were conducted using OriginPro.
934
+ Data Availability
935
+ The raw data that support the findings of this study are available from the corresponding author upon
936
+ request.
937
+ Code Availability
938
+
939
+ 25
940
+
941
+ The MATLAB scripts for analysis are available from the corresponding author upon request.
942
+ Acknowledgements
943
+ This work was supported by National Institute of Health Brain Initiative R01 NS109794 to J-XC and CY.
944
+ Research reported in this publication was supported by the Boston University Micro and Nano Imaging
945
+ Facility and the Office of the Director, National Institutes of Health of the National Institutes of Health
946
+ under award Number S10OD024993
947
+ References
948
+ 1.
949
+ Wilbrecht, L. & Shohamy, D. Neural Circuits can Bridge Systems and Cognitive Neuroscience.
950
+ Front Hum Neurosci 3, 81 (2010).
951
+ 2.
952
+ Chen, R., Canales, A. & Anikeeva, P. Neural Recording and Modulation Technologies. Nat Rev
953
+ Mater 2 (2017).
954
+ 3.
955
+ Sohal, V.S., Zhang, F., Yizhar, O. & Deisseroth, K. Parvalbumin neurons and gamma rhythms
956
+ enhance cortical circuit performance. Nature 459, 698-702 (2009).
957
+ 4.
958
+ Paulk, A.C., Kirszenblat, L., Zhou, Y. & van Swinderen, B. Closed-Loop Behavioral Control
959
+ Increases Coherence in the Fly Brain. J Neurosci 35, 10304-10315 (2015).
960
+ 5.
961
+ Rosin, B. et al. Closed-loop deep brain stimulation is superior in ameliorating parkinsonism.
962
+ Neuron 72, 370-384 (2011).
963
+ 6.
964
+ Paz, J.T. et al. Closed-loop optogenetic control of thalamus as a tool for interrupting seizures
965
+ after cortical injury. Nat Neurosci 16, 64-70 (2013).
966
+ 7.
967
+ Harmsen, I.E. et al. Clinical trials for deep brain stimulation: Current state of affairs. Brain Stimul
968
+ 13, 378-385 (2020).
969
+ 8.
970
+ Hughes, C., Herrera, A., Gaunt, R. & Collinger, J. Bidirectional brain-computer interfaces. Handb
971
+ Clin Neurol 168, 163-181 (2020).
972
+ 9.
973
+ Zhou, A., Johnson, B.C. & Muller, R. Toward true closed-loop neuromodulation: artifact-free
974
+ recording during stimulation. Curr Opin Neurobiol 50, 119-127 (2018).
975
+ 10.
976
+ Heck, C.N. et al. Two-year seizure reduction in adults with medically intractable partial onset
977
+ epilepsy treated with responsive neurostimulation: final results of the RNS System Pivotal trial.
978
+ Epilepsia 55, 432-441 (2014).
979
+ 11.
980
+ Sun, F.T. & Morrell, M.J. The RNS System: responsive cortical stimulation for the treatment of
981
+ refractory partial epilepsy. Expert Rev Med Devices 11, 563-572 (2014).
982
+ 12.
983
+ Miocinovic, S. et al. Experimental and theoretical characterization of the voltage distribution
984
+ generated by deep brain stimulation. Exp Neurol 216, 166-176 (2009).
985
+ 13.
986
+ Paff, M., Loh, A., Sarica, C., Lozano, A.M. & Fasano, A. Update on Current Technologies for Deep
987
+ Brain Stimulation in Parkinson's Disease. J Mov Disord 13, 185-198 (2020).
988
+ 14.
989
+ Buhlmann, J., Hofmann, L., Tass, P.A. & Hauptmann, C. Modeling of a segmented electrode for
990
+ desynchronizing deep brain stimulation. Front Neuroeng 4, 15 (2011).
991
+ 15.
992
+ Qian, X. et al. A Method for Removal of Deep Brain Stimulation Artifact From Local Field
993
+ Potentials. IEEE Trans Neural Syst Rehabil Eng 25, 2217-2226 (2017).
994
+
995
+ 26
996
+
997
+ 16.
998
+ Koessler, L. et al. In-vivo measurements of human brain tissue conductivity using focal electrical
999
+ current injection through intracerebral multicontact electrodes. Hum Brain Mapp 38, 974-986
1000
+ (2017).
1001
+ 17.
1002
+ Zhang, J. et al. Integrated device for optical stimulation and spatiotemporal electrical recording
1003
+ of neural activity in light-sensitized brain tissue. J Neural Eng 6, 055007 (2009).
1004
+ 18.
1005
+ Abaya, T.V., Blair, S., Tathireddy, P., Rieth, L. & Solzbacher, F. A 3D glass optrode array for optical
1006
+ neural stimulation. Biomed Opt Express 3, 3087-3104 (2012).
1007
+ 19.
1008
+ Wang, J. et al. Integrated device for combined optical neuromodulation and electrical recording
1009
+ for chronic in vivo applications. J Neural Eng 9, 016001 (2012).
1010
+ 20.
1011
+ Royer, S. et al. Multi-array silicon probes with integrated optical fibers: light-assisted
1012
+ perturbation and recording of local neural circuits in the behaving animal. Eur J Neurosci 31,
1013
+ 2279-2291 (2010).
1014
+ 21.
1015
+ He, Z. et al. A sapphire based monolithic integrated optrode. Annu Int Conf IEEE Eng Med Biol
1016
+ Soc 2016, 6186-6189 (2016).
1017
+ 22.
1018
+ Kravitz, A.V., Owen, S.F. & Kreitzer, A.C. Optogenetic identification of striatal projection neuron
1019
+ subtypes during in vivo recordings. Brain Res 1511, 21-32 (2013).
1020
+ 23.
1021
+ Wu, F. et al. Monolithically Integrated muLEDs on Silicon Neural Probes for High-Resolution
1022
+ Optogenetic Studies in Behaving Animals. Neuron 88, 1136-1148 (2015).
1023
+ 24.
1024
+ Voroslakos, M. et al. HectoSTAR muLED Optoelectrodes for Large-Scale, High-Precision In Vivo
1025
+ Opto-Electrophysiology. Adv Sci (Weinh) 9, e2105414 (2022).
1026
+ 25.
1027
+ Canales, A. et al. Multifunctional fibers for simultaneous optical, electrical and chemical
1028
+ interrogation of neural circuits in vivo. Nat Biotechnol 33, 277-284 (2015).
1029
+ 26.
1030
+ Park, S. et al. One-step optogenetics with multifunctional flexible polymer fibers. Nat Neurosci
1031
+ 20, 612-619 (2017).
1032
+ 27.
1033
+ Jiang, S. et al. Spatially expandable fiber-based probes as a multifunctional deep brain interface.
1034
+ Nat Commun 11, 6115 (2020).
1035
+ 28.
1036
+ Antonini, M.J. et al. Customizing MRI-Compatible Multifunctional Neural Interfaces through
1037
+ Fiber Drawing. Adv Funct Mater 31 (2021).
1038
+ 29.
1039
+ Park, S. et al. Adaptive and multifunctional hydrogel hybrid probes for long-term sensing and
1040
+ modulation of neural activity. Nat Commun 12, 3435 (2021).
1041
+ 30.
1042
+ Bansal, A., Shikha, S. & Zhang, Y. Towards translational optogenetics. Nat Biomed Eng (2022).
1043
+ 31.
1044
+ Jiang, Y. et al. Optoacoustic brain stimulation at sub-millimeter spatial precision. Nature
1045
+ Communications (2020).
1046
+ 32.
1047
+ Shi, L. et al. Non-genetic photoacoustic stimulation of single neurons by a tapered fiber
1048
+ optoacoustic emitter. Light Sci Appl 10, 143 (2021).
1049
+ 33.
1050
+ Wang, L.V. & Hu, S. Photoacoustic tomography: in vivo imaging from organelles to organs.
1051
+ Science 335, 1458-1462 (2012).
1052
+ 34.
1053
+ Lee, T., Baac, H.W., Li, Q. & Guo, L.J. Efficient Photoacoustic Conversion in Optical Nanomaterials
1054
+ and Composites. Advanced Optical Materials 6 (2018).
1055
+ 35.
1056
+ Abouraddy, A.F. et al. Towards multimaterial multifunctional fibres that see, hear, sense and
1057
+ communicate. Nat Mater 6, 336-347 (2007).
1058
+ 36.
1059
+ Colchester, R.J. et al. Laser-generated ultrasound with optical fibres using functionalised carbon
1060
+ nanotube composite coatings. Applied Physics Letters 104 (2014).
1061
+ 37.
1062
+ Noimark, S. et al. Carbon-Nanotube-PDMS Composite Coatings on Optical Fibers for All-Optical
1063
+ Ultrasound Imaging. Advanced Functional Materials 26, 8390-8396 (2016).
1064
+ 38.
1065
+ Chen, G. et al. High-precision neural stimulation by a highly efficient candle soot fiber
1066
+ optoacoustic emitter. Front Neurosci 16, 1005810 (2022).
1067
+
1068
+ 27
1069
+
1070
+ 39.
1071
+ Guo, Y. et al. Polymer-fiber-coupled field-effect sensors for label-free deep brain recordings.
1072
+ PLoS One 15, e0228076 (2020).
1073
+ 40.
1074
+ Han, D., Meng, Z., Wu, D., Zhang, C. & Zhu, H. Thermal properties of carbon black aqueous
1075
+ nanofluids for solar absorption. Nanoscale Res Lett 6, 457 (2011).
1076
+ 41.
1077
+ Hilarius, K. et al. Influence of shear deformation on the electrical and rheological properties of
1078
+ combined filler networks in polymer melts: Carbon nanotubes and carbon black in
1079
+ polycarbonate. Polymer 54, 5865-5874 (2013).
1080
+ 42.
1081
+ Brigandi, P.J., Cogen, J.M., Reffner, J.R., Wolf, C.A. & Pearson, R.A. Influence of carbon black and
1082
+ carbon nanotubes on the conductivity, morphology, and rheology of conductive ternary polymer
1083
+ blends. Polymer Engineering & Science 57, 1329-1339 (2017).
1084
+ 43.
1085
+ Smetters, D., Majewska, A. & Yuste, R. Detecting action potentials in neuronal populations with
1086
+ calcium imaging. Methods 18, 215-221 (1999).
1087
+ 44.
1088
+ Yang, W. & Yuste, R. In vivo imaging of neural activity. Nat Methods 14, 349-359 (2017).
1089
+ 45.
1090
+ Nagayama, S. et al. In vivo simultaneous tracing and Ca(2+) imaging of local neuronal circuits.
1091
+ Neuron 53, 789-803 (2007).
1092
+ 46.
1093
+ Lyu, Y., Xie, C., Chechetka, S.A., Miyako, E. & Pu, K. Semiconducting Polymer Nanobioconjugates
1094
+ for Targeted Photothermal Activation of Neurons. J Am Chem Soc 138, 9049-9052 (2016).
1095
+ 47.
1096
+ Shapiro, M.G., Homma, K., Villarreal, S., Richter, C.P. & Bezanilla, F. Infrared light excites cells by
1097
+ changing their electrical capacitance. Nat Commun 3, 736 (2012).
1098
+ 48.
1099
+ Schmitzer-Torbert, N., Jackson, J., Henze, D., Harris, K. & Redish, A.D. Quantitative measures of
1100
+ cluster quality for use in extracellular recordings. Neuroscience 131, 1-11 (2005).
1101
+ 49.
1102
+ Heinricher, M.M. Principles of extracellular single-unit recording. Microelectrode Recording in
1103
+ Movement Disorder Surgery 8 (2004).
1104
+ 50.
1105
+ Henze, D.A. et al. Intracellular features predicted by extracellular recordings in the hippocampus
1106
+ in vivo. J Neurophysiol 84, 390-400 (2000).
1107
+ 51.
1108
+ Ching, S., Purdon, P.L., Vijayan, S., Kopell, N.J. & Brown, E.N. A neurophysiological-metabolic
1109
+ model for burst suppression. Proc Natl Acad Sci U S A 109, 3095-3100 (2012).
1110
+ 52.
1111
+ Hudetz, A.G., Vizuete, J.A. & Pillay, S. Differential effects of isoflurane on high-frequency and
1112
+ low-frequency gamma oscillations in the cerebral cortex and hippocampus in freely moving rats.
1113
+ Anesthesiology 114, 588-595 (2011).
1114
+ 53.
1115
+ Kozai, T.D., Jaquins-Gerstl, A.S., Vazquez, A.L., Michael, A.C. & Cui, X.T. Brain tissue responses to
1116
+ neural implants impact signal sensitivity and intervention strategies. ACS Chem Neurosci 6, 48-
1117
+ 67 (2015).
1118
+ 54.
1119
+ Aravanis, A.M. et al. An optical neural interface: in vivo control of rodent motor cortex with
1120
+ integrated fiberoptic and optogenetic technology. J Neural Eng 4, S143-156 (2007).
1121
+ 55.
1122
+ Ung, K. & Arenkiel, B.R. Fiber-optic implantation for chronic optogenetic stimulation of brain
1123
+ tissue. J Vis Exp, e50004 (2012).
1124
+ 56.
1125
+ Poduval, R.K. et al. Optical fiber ultrasound transmitter with electrospun carbon nanotube-
1126
+ polymer composite. Appl Phys Lett 110, 223701 (2017).
1127
+ 57.
1128
+ Finlay, M.C. et al. Through-needle all-optical ultrasound imaging in vivo: a preclinical swine
1129
+ study. Light Sci Appl 6, e17103 (2017).
1130
+ 58.
1131
+ Yoo, M., Koo, H., Kim, M., Kim, H.I. & Kim, S. Near-infrared stimulation on globus pallidus and
1132
+ subthalamus. J Biomed Opt 18, 128005 (2013).
1133
+ 59.
1134
+ Abaya, T.V. et al. Characterization of a 3D optrode array for infrared neural stimulation. Biomed
1135
+ Opt Express 3, 2200-2219 (2012).
1136
+ 60.
1137
+ Yoo, S., Mittelstein, D.R., Hurt, R.C., Lacroix, J. & Shapiro, M.G. Focused ultrasound excites
1138
+ cortical neurons via mechanosensitive calcium accumulation and ion channel amplification. Nat
1139
+ Commun 13, 493 (2022).
1140
+
1141
+ 28
1142
+
1143
+ 61.
1144
+ Shi, L., Jiang, Y., Zheng, N., Cheng, J.X. & Yang, C. High-precision neural stimulation through
1145
+ optoacoustic emitters. Neurophotonics 9, 032207 (2022).
1146
+ 62.
1147
+ Pisanello, F. et al. Multipoint-emitting optical fibers for spatially addressable in vivo
1148
+ optogenetics. Neuron 82, 1245-1254 (2014).
1149
+
1150
+ Supplementary Information
1151
+
1152
+ Supplementary Figure 1. Microscope images of deposited carbon black and PDMS composite.
1153
+ The coverage area was controlled through tuning the injection pressure and time. Injection time was
1154
+ varied between 1 second and 2 seconds, and the pressure was varied from 2 psi, 3 psi and 4 psi. Scale bar:
1155
+ 50 µm.
1156
+
1157
+
1158
+ 2 psi
1159
+ 3 psi
1160
+ 4 psi
1161
+ 2 s1030 nm
1162
+ Optical fiber
1163
+ pulsedlaser
1164
+ Function
1165
+ generator
1166
+ mFOE
1167
+ Neurons cultured
1168
+ Micromanipulator
1169
+ loadedwithOGD-1
1170
+ Objective
1171
+ Lens
1172
+ Lens
1173
+ 470nm
1174
+ DM
1175
+ LED
1176
+ Lens
1177
+ CMOS
1178
+ Mirror
1179
+ camera29
1180
+
1181
+ Supplementary Figure 2. Schematic of in vitro mFOE stimulation and calcium imaging set up.
1182
+ Stimulation: 1030 nm pulsed laser is triggered by a function generator and delivered to the mFOE through
1183
+ an optical fiber. Calcium imaging: Oregon green is excited by 470 nm LED and the fluorescence signal is
1184
+ detected through a CMOS camera.
1185
+
1186
+ Supplementary Figure 3. Illustration of the laser pulse train for 5 bursts with 100 ms duration at
1187
+ 1Hz.
1188
+
1189
+
1190
+ Repetition rate: 1.7 kHz
1191
+ Pulse numbers: 170
1192
+ 3 ns pulse width
1193
+ 100 msPre
1194
+ Post30
1195
+
1196
+ Supplementary Fig. 4 Calcium imaging of neurons before and after mFOE stimulation. Scale bar:
1197
+ 100 µm.
1198
+
1199
+ Supplementary Fig. S5 Average calcium traces of laser only control groups. The laser duration was
1200
+ same with three conditions tested in mFOE stimulation (200 ms, 100 ms and 50 ms). Laser light with
1201
+ pulse energy of 41.8 µJ was triggered at the time point labelled by the red bar. Shaded areas: standard
1202
+ deviation. (N=3)
1203
+
1204
+
1205
+
1206
+ 200 ms
1207
+ 100 ms
1208
+ 50 ms
1209
+ 0.1
1210
+ 0.1
1211
+ 0.1
1212
+ 0.05
1213
+ F 0.05
1214
+ F 0.05
1215
+ △F/
1216
+ △F/
1217
+ △F/
1218
+ 0
1219
+ 1
1220
+ 2
1221
+ 3
1222
+ 0
1223
+ 1
1224
+ 2
1225
+ 3
1226
+ 0
1227
+ 1
1228
+ 2
1229
+ 3
1230
+ Time (s)
1231
+ Time (s)
1232
+ Time (s)1.5
1233
+ 1.5
1234
+ 1.5
1235
+ 50 ms
1236
+ 100 ms
1237
+ 200 ms
1238
+ 1
1239
+ 1
1240
+ 1
1241
+ (0。) .
1242
+ (0。)
1243
+ (0。)
1244
+ 0.5
1245
+ 0.5
1246
+ 0.5
1247
+ △T
1248
+ △T
1249
+ △T
1250
+ 0
1251
+ 0
1252
+ 0
1253
+ -0.5
1254
+ -0.5
1255
+ -0.5
1256
+ 0
1257
+ 2
1258
+ 4
1259
+ 0
1260
+ 2
1261
+ 4
1262
+ 0
1263
+ 2
1264
+ 4
1265
+ Time (s)
1266
+ Time (s)
1267
+ Time (s)31
1268
+
1269
+ Supplementary Fig. S6 Temperature change of the optoacoustic emitter integrated on mFOE. The
1270
+ pulse energy was maintained at 41.8 µJ and the burst duration was varied from 50 ms (blue), 100 ms
1271
+ (yellow) to 200 ms (orange). Laser was trigger at 2.5 second as labelled by the red bar.
1272
+
1273
+
1274
+ Supplementary Fig. S7 LFP recording of sham control stimulation experiments.
1275
+ a. Electrophysiological recording under light only stimulations delivered through a bare multifunctional
1276
+ fiber without optoacoustic emitter. b. Simultaneous optoacoustic stimulation and electrophysiological
1277
+ recording of an euthanized mouse. Same laser condition was used: pulse energy of 41.8 µJ, 50 ms burst of
1278
+ pulses, 1 Hz, blue dots indicate the laser onset.
1279
+
1280
+
1281
+ b
1282
+ a
1283
+ 0.5 -
1284
+ 0.50
1285
+ 0.25
1286
+ Voltage (mV)
1287
+ 0.0
1288
+ Voltage (mV)
1289
+ 0.00
1290
+ -0.5.
1291
+ -0.25
1292
+ -1.0.
1293
+ -0.50
1294
+ 0
1295
+ 2
1296
+ 4
1297
+ 6
1298
+ 8
1299
+ 10
1300
+ 0
1301
+ 2
1302
+ 4
1303
+ 6
1304
+ 8
1305
+ 10
1306
+ Time (s)
1307
+ Time (s)
e9E2T4oBgHgl3EQfGgbf/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
edFST4oBgHgl3EQfFzg4/content/tmp_files/2301.13719v1.pdf.txt ADDED
@@ -0,0 +1,815 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arXiv:2301.13719v1 [math.GT] 31 Jan 2023
2
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
3
+ CRISTINA COSTOYA, VICENTE MU˜NOZ, AND ANTONIO VIRUEL
4
+ Abstract. In this paper we solve in the positive the question of whether any finite set
5
+ of integers A, containing the zero, is the mapping degree set between two oriented closed
6
+ connected manifolds of the same dimension. We extend this question to the rational
7
+ setting, where an affirmative answer is also given.
8
+ 1. Introduction
9
+ In this paper, we settle in the positive various questions which have been raised about
10
+ D(M, N), the set of mapping degrees between two oriented closed connected manifolds
11
+ M and N of the same dimension:
12
+ D(M, N) = {d ∈ Z | ∃f : M → N, deg(f) = d}.
13
+ In [14, Problem 1.1], the authors discuss the problem of finding, for any set A ⊂
14
+ Z containing the zero, two oriented closed connected manifolds M and N of the same
15
+ dimension such that A = D(M, N). Note that 0 ∈ A is a necessary condition as the
16
+ constant map M → N is of degree zero.
17
+ A quick argument shows that when A is an infinite set, this problem is solved in the
18
+ negative [14, Theorem 1.3]: there are uncountably many infinite subsets A ⊂ Z containing
19
+ the zero, compared to the countably many mapping degree sets D(M, N) that exist for
20
+ pairs of oriented closed connected manifolds of the same dimension. Hence not every
21
+ infinite set, containing the zero, is realizable as the mapping degree set of manifolds.
22
+ Thus, one might ask:
23
+ Question 1.1 ([14, Problem 1.4]). Let A be a finite set of integers containing the zero.
24
+ Is A = D(M, N) for some oriented closed manifolds M, N of the same dimension?
25
+ Remark 1.2. It is important to notice that if {0} ⊊ A = D(M, N), A finite, for some
26
+ manifolds M and N, then D(M, M) and D(N, N) must both be contained in {0, 1, −1}.
27
+ Otherwise, if there exists g : M → M with | deg(g)| > 1, then for any non-zero degree
28
+ f : M → N (which exists by assumption), the subset {deg(f ◦ gm) | m ∈ N} of D(M, N)
29
+ is unbounded. This leads to a contradiction as A = D(M, N) is finite. The same follows
30
+ for D(N, N).
31
+ 2020 Mathematics Subject Classification. 55M25, 57N65, 55P62, 55R10.
32
+ Key words and phrases. Mapping degree sets, inflexible manifold, fiber bundle, unstable Adams
33
+ operation.
34
+ The first author was partially supported by MINECO (Spain) grants PID2020-115155GB-I00 and
35
+ TED2021-131201B-I00. The second author was partially supported MINECO (Spain) grant PID2020-
36
+ 118452GB-I00. The third author was partially supported by MINECO (Spain) grant PID2020-118753GB-
37
+ I00, and by PAIDI 2020 (Andalusia) grant PROYEXCEL-00827.
38
+ 1
39
+
40
+ 2
41
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
42
+ An oriented closed manifold M satisfying D(M, M) ⊆ {0, 1, −1} is called an inflexible
43
+ manifold [7, Definition 1.4].
44
+ This condition is equivalent to asking that D(M, M) is
45
+ bounded: since it is a multiplicative semi-group, if there exists any ℓ ∈ D(M, M) with
46
+ |ℓ| > 1, then D(M, M) is unbounded. Simply connected inflexible manifolds are rare
47
+ objects that have appeared quite recently in literature using rational homotopy theory
48
+ and surgery theory (see [5] for an account on the simply connected inflexible manifolds
49
+ that are known at present). Not surprisingly, and in lights of Remark 1.2, part of our key
50
+ constructions will use rational homotopy methods.
51
+ The main result in this work answers Question 1.1 positively:
52
+ Theorem A. Let A be a finite set of integers containing the zero. Then, A = D(M, N)
53
+ for some oriented closed connected 3-manifolds M, N.
54
+ The proof of this theorem will be carried out at the end of Section 3. Appealing to [14,
55
+ Example 1.5], we point out that the 3-dimension of the manifolds is the lowest possible.
56
+ A second problem related to Question 1.1 is also treated in this paper. More precisely,
57
+ let the rational mapping degree set between oriented closed connected n-manifolds M, N
58
+ be the following set
59
+ DQ(M, N) = {d ∈ Q | ∃f : (M(0), [M]Q) → (N(0), [N]Q), deg(f) = d},
60
+ where [M] ∈ Hn(M; Z) denotes the cohomological fundamental class of M, [M]Q ∈
61
+ Hn(M; Q) denotes the rational cohomological fundamental class of M, and M(0) the
62
+ rationalization of M. Then, we raise the following question, which can be thought of
63
+ as a rational version of [14, Problem 1.4]:
64
+ Question 1.3. Let A be a finite set of rational numbers containing the zero. Is A =
65
+ DQ(M, N) for some oriented closed connected manifolds M, N of the same dimension?
66
+ In Section 4 we solve this problem in the positive by proving:
67
+ Theorem B. Let A be a finite set of rational numbers containing the zero. Then A =
68
+ DQ(M, N) for some oriented closed manifolds M, N. Moreover, given any integer k ≥ 1,
69
+ the manifolds M, N above can be chosen (30k + 17)-connected.
70
+ The proofs of Theorem A and Theorem B consist of mainly two steps:
71
+ • Arithmetical decomposition of finite sets:
72
+ In Section 2 we show how to decompose
73
+ the candidate A to be realized as the mapping degree set of manifolds, as an in-
74
+ tersection of sums over specifically designed sequences of integers SBi, i = 0, . . . , n
75
+ (see Definition 2.1). Each of those sums gradually approaches A (Proposition 2.2,
76
+ Corollary 2.3).
77
+ • Spherical fibrations: In Sections 3 and 4, we use certain inflexible manifolds (resp.
78
+ inflexible Sullivan algebras) as the basis of spherical fibrations where the total
79
+ spaces are also inflexible manifolds (resp. inflexible Sullivan algebras). Relations
80
+ between connected sums and mapping degree sets (see Propositions 3.3 and 4.3)
81
+ allow one to consider iterated connected sums of the total spaces, in a first stage
82
+ to realize the sums SBi above mentioned, and in a second stage to realize the
83
+ candidate A.
84
+
85
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
86
+ 3
87
+ Looking at the connectivity, while manifolds from Theorem B are simply connected
88
+ (indeed, they are as highly connected as desired) the ones from Theorem A have non-
89
+ trivial fundamental group. In Section 5 we will use unstable Adams operations to prove
90
+ the following results that guarantee that manifolds realizing finite sets of integers can be
91
+ chosen simply connected:
92
+ Theorem C. Suppose that there exists an oriented closed k-connected 2m-manifold Σ,
93
+ m > 1, verifying that Σ(0) is inflexible and πj(Σ(0)) = 0 for j ≥ 2m − 1. Then any finite
94
+ set of integers A containing the zero can be realized as A = D(M, N) for some oriented
95
+ closed k-connected (4m − 1)-manifolds M, N.
96
+ Examples of simply connected manifolds fulfilling the hypotheses of Theorem C can be
97
+ found in [1, Example 3.8] and [7, Theorem 6.8]. Hence, the following holds:
98
+ Corollary D. Any finite set of integers A containing the zero can be realized as A =
99
+ D(M, N) for some oriented closed simply connected manifolds M, N.
100
+ 2. Some arithmetic combinatorics
101
+ In this section we show that every finite set A ⊂ Z (resp. ⊂ Q) containing the zero can
102
+ be expressed as the intersection of sums over certain sequences of integers, that gradually
103
+ approach A. The sequences have an additional property (see Proposition 2.2) that will
104
+ be crucial to prove Theorem C in Section 5 below.
105
+ Following the notation in [7, Section II.1], [14, Section 3], given A, B ⊂ Z (resp. ⊂ Q)
106
+ we write A + B := {a + b : a ∈ A, b ∈ B} ⊂ Z (resp. ⊂ Q).
107
+ Definition 2.1. Let B be a finite sequence of not necessarily pairwise distinct non-zero
108
+ integers (resp. rational numbers). We write
109
+ SB :=
110
+
111
+ b∈B
112
+ {0, b} ⊂ Z (resp. ⊂ Q),
113
+ and we refer to it as the sum over the sequence B.
114
+ Proposition 2.2. Let d1, . . . , dn be pairwise distinct non-zero integers. For any positive
115
+ integer m ≥ 1, there exist finite sequences B(i), i = 0, . . . , n, of not necessarily pairwise
116
+ distinct non-zero integers, such that
117
+ {0, d1, . . . , dn} =
118
+ n�
119
+ i=0
120
+ SB(i).
121
+ Moreover, every element in B(i) can be written as a power ±km for some positive integer
122
+ k coprime to m!.
123
+ Proof. Fix m ≥ 1. Since the construction of B(i), i = 0, . . . , n, depends on the sign of
124
+ the pairwise distinct di ∈ Z, i = 1, . . . , n, we write them as an ordered sequence
125
+ {−ar < . . . < −a1 < 0 < e1 < . . . < es}
126
+ where n = r + s. We assume a0 = 0 = e0.
127
+
128
+ 4
129
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
130
+ In the first place, let B(0) be the sequence consisting of ar copies of −1 = −1m and es
131
+ copies of 1 = 1m. Thus
132
+ {−ar < . . . < es} ⊂ SB(0) = [−ar, es] ∩ Z.
133
+ In the second place, for j = 1, . . . , s, choose a positive kj ∈ Z coprime with m! such that
134
+ km
135
+ j > max{es, ej + ar}. Then, let B(j) be the sequence consisting of km
136
+ j − ej copies of
137
+ −1 = −1m, ej−1 copies of 1 = 1m, and one copy of km
138
+ j . Hence,
139
+ {−ar < . . . < es} ⊂ SB(j) =
140
+
141
+ [−(km
142
+ j − ej), ej−1] ∪ [ej, km
143
+ j + ej−1]
144
+
145
+ ∩ Z.
146
+ Finally, for j = s + 1, . . . , n, choose a positive kj ∈ Z coprime with m! such that km
147
+ j >
148
+ max{ar, aj−s + es}. Then, let B(j) be the sequence consisting of km
149
+ j − aj−s copies of
150
+ 1 = 1m, aj−s−1 copies of −1 = −1m, and one copy of −km
151
+ j . Hence,
152
+ {−ar < . . . < es} ⊂ SB(j) =
153
+
154
+ [−km
155
+ j − aj−s−1, −aj−s] ∪ [−aj−s−1, km
156
+ j − aj−s−1]
157
+
158
+ ∩ Z.
159
+ All of the above implies that
160
+ {0, d1, . . . , dn} = {−ar < . . . < es} =
161
+ n�
162
+ i=0
163
+ SB(i).
164
+
165
+ For A ⊂ Q and λ ∈ Q we write
166
+ λA := {λa : a ∈ A}.
167
+ Notice that if B(i) is a finite sequence of not necessarily pairwise distinct non-zero rational
168
+ numbers, i = 0, . . . , n, for any λ ∈ Q, we have that
169
+ λ
170
+ � n�
171
+ i=0
172
+ SB(i)
173
+
174
+ =
175
+ n�
176
+ i=0
177
+ SλB(i).
178
+ Therefore, the following is a direct consequence of Proposition 2.2:
179
+ Corollary 2.3. Let d1, d2, . . . , dn be pairwise distinct non-zero rational numbers. Then,
180
+ there exist finite sequences B(i), i = 0, . . . , n, of not necessarily pairwise distinct non-zero
181
+ rational numbers, such that
182
+ {0, d1, . . . , dn} =
183
+ n�
184
+ i=0
185
+ SB(i).
186
+ 3. Circle bundles over inflexible 2-manifolds: mapping degree set
187
+ This section is devoted to prove Theorem A. As explained in the introduction (see
188
+ Remark 1.2) if we want to realize a finite set of integer strictly containing the zero as a
189
+ mapping degree set D(M, N), then both M and N need to be inflexible manifolds. We
190
+ are going to consider circle bundles over certain inflexible 2-manifolds, with prescribed
191
+ Euler class, whose total space is again an inflexible 3-manifold. These 3-manifolds will be
192
+ used as building blocks to construct, be means of iterated connected sums, the manifolds
193
+ M and N.
194
+ We first collect from literature a couple of results that are needed:
195
+
196
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
197
+ 5
198
+ Lemma 3.1 ([5, Lemma 7.8], [14, Lemma 3.5]). Let M1, M2 and N be oriented closed
199
+ connected n-manifolds. Then
200
+ D(M1, N) + D(M2, N) ⊆ D(M1#M2, N)
201
+ Moreover, if πn−1(N) = 0, then
202
+ D(M1, N) + D(M2, N) = D(M1#M2, N).
203
+ We reformulate [14, Lemma 4.3] as follows:
204
+ Lemma 3.2. Let M and N1, N2 be oriented closed n-manifolds. Then
205
+ D(M, N1#N2) ⊆ D(M, N1) ∩ D(M, N2).
206
+ Using the previous two lemmas, we prove the following result:
207
+ Proposition 3.3. Let M1, M2 and N1, N2 be oriented closed n-manifolds verifying that
208
+ πn−1(Nj) = 0, j = 1, 2, and D(Mi, Nj) = {0}, for i ̸= j. Then
209
+ D(M1#M2, N1#N2) = D(M1, N1) ∩ D(M2, N2).
210
+ Proof. By combining Lemma 3.2 and Lemma 3.1, it follows directly that:
211
+ D(M1#M2, N1#N2) ⊆ D(M1#M2, N1) ∩ D(M1#M2, N2) = D(M1, N1) ∩ D(M2, N2).
212
+ Conversely, let fi : Mi → Ni, i = 1, 2, be maps both of the same degree d. Without
213
+ loss of generality we may assume that fi is cellular, i = 1, 2.
214
+ Therefore it induces a
215
+ commutative diagram of cofibration sequences
216
+ M[n−1]
217
+ i
218
+ Mi
219
+ Sn
220
+ N[n−1]
221
+ i
222
+ Ni
223
+ Sn
224
+ fi
225
+ ˜fi
226
+ where X[n−1] stands for the (n−1)-skeleton of X, and ˜fi is a pointed map of degree d (the
227
+ base points in Sn are the class represented by M[n−1]
228
+ i
229
+ and N[n−1]
230
+ i
231
+ ). Hence, there exists a
232
+ pointed homotopy deforming ˜fi to a pointed map ˜gi such that ˜gi stabilizes the equator
233
+ Sn−1 ⊂ Sn and ˜gi|Sn−1 = g for some fixed g : Sn−1 → Sn−1 of degree d. Then, the pointed
234
+ homotopy deforming ˜fi can be lifted to Mi and defines gi : Mi → Ni that induces a maps
235
+ between disks gi : DMi → DNi such that gi|∂DMi = g, i = 1, 2.
236
+ Finally, gluing M1#M2 along ∂DMi, i = 1, 2, and N1#N2 along ∂DNi, i = 1, 2, give
237
+ rise to a well defined a map
238
+ g1#g2: M1#M2 → N1#N2
239
+ whose degree is precisely d by construction. Therefore
240
+ D(M1, N1) ∩ D(M2, N2) ⊆ D(M1#M2, N1#N2)
241
+ and we conclude the proof.
242
+
243
+ A rational version of Proposition 3.3 will be required in order to prove Theorem B. This
244
+ will be done in Section 4. Although we will not give the details, previous results (Lemma
245
+ 3.1 and Lemma 3.2) can be easily generalized to finite iterated connected sums. Hence,
246
+ following along the lines of the proof in Proposition 3.3 we obtain:
247
+
248
+ 6
249
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
250
+ Corollary 3.4. Let Mi, Ni, i = 1, . . . , r, be oriented closed connected n-manifolds such
251
+ that πn−1(Ni) = 0, i = 1, . . . , r, and D(Mi, Nj) = {0}, for i ̸= j. Then
252
+ D(M1# · · · #Mr, N1# · · · #Nr) =
253
+ r�
254
+ i=1
255
+ D(Mi, Ni).
256
+ We now have all the ingredients to prove our main theorem.
257
+ Proof of Theorem A. Let A = {0, d1, . . . , dn} be a finite set of pairwise distinct integers.
258
+ We need to show that A is realized by two oriented closed 3-manifolds M, N in the sense
259
+ that A = D(M, N).
260
+ For this purpose, we consider an oriented closed hyperbolic surface of genus g > 1, Σg.
261
+ Then, for any i ∈ Z, let Ki be the total space in the circle bundle
262
+ S1 → Ki → Σg
263
+ with Euler number e(Ki) = i. Observe that Ki, i ∈ Z, is an aspherical 3-manifold. The
264
+ mapping degree set between these 3-manifolds is fully described in [14, Lemma 3.4]:
265
+ D(Ki, Kj) =
266
+
267
+ {0, j/i},
268
+ if i|j,
269
+ {0},
270
+ if i̸ |j.
271
+ (1)
272
+ According to Proposition 2.2, for any positive integer m > 0 that we fix, there exist
273
+ finite sequences, B(i), i = 0, . . . , n, of not necessarily pairwise distinct non-zero integers,
274
+ satisfying that
275
+ A =
276
+ n�
277
+ i=0
278
+ SB(i).
279
+ Now, we choose particular pairwise distinct primes q0, q1, . . . , qn fulfilling the condition
280
+ qi > max{|b| : b ∈ B(i)}, i = 0, . . . , n,
281
+ and we denote
282
+ αi = qi
283
+
284
+ b∈B(i)
285
+ b, i = 0, . . . , n.
286
+ Then, we construct the following “intermediate” manifolds (that will serve us to realize
287
+ each of the sums SB(i)), for i = 0, . . . , n:
288
+ Mi =
289
+ #
290
+ b∈B(i)
291
+ Kαi/b
292
+ Ni = Kαi.
293
+ Because Kαi are aspherical 3-manifolds, for i = 0, . . . , n, we have that π2(Kαi) = 0,
294
+ and conditions to apply Lemma 3.1 hold. Therefore:
295
+ D(Mi, Nj) = D( #
296
+ b∈B(i)
297
+ Kαi/b, Kαj) =
298
+
299
+ b∈B(i)
300
+ D(Kαi/b, Kαj).
301
+ Using (1), we then get that, for i = 0, . . . , n,
302
+ D(Mi, Ni) = SB(i) , and
303
+ D(Mi, Nj) = {0}, for i ̸= j.
304
+
305
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
306
+ 7
307
+ Finally, we consider the following iterated connected sums:
308
+ M = M0#M1# . . . #Mn,
309
+ N = N0#N1# . . . #Nn,
310
+ for which all the conditions to apply Corollary 3.4 plainly hold. Hence,
311
+ D(M, N) =
312
+ n�
313
+ i=0
314
+ SB(i) = A,
315
+ and the proof of Theorem A is complete.
316
+
317
+ Remark 3.5. We end this section by pointing out that all the 3-manifolds involved in the
318
+ previous theorem are inflexible (see also Remark 1.2). It is clear, by (1), that Ki, i ∈ Z,
319
+ are inflexible. Now, proceeding along the lines of Theorem A, we apply repeatedly Lemma
320
+ 3.2 and Lemma 3.1 to get the inflexibility property. On the one hand, we obtain that
321
+ D(Mi, Mj) = {0} for i ̸= j, and on the other hand
322
+ D(M, M) ⊆
323
+ n�
324
+ i=0
325
+ D(Mi, Mi).
326
+ Also, by Lemma 3.2,
327
+ D(Mi, Mi) = D(Mi,
328
+ #
329
+ b∈B(i)
330
+ Kαi/b) ⊂
331
+
332
+ b∈B(i)
333
+ D(Mi, Kαi/b)
334
+ and using Lemma 3.1,
335
+ D(Mi, Kαi/b) = D( #
336
+ b′∈B(i)
337
+ Kαi/b′, Kαi/b) =
338
+
339
+ b′∈B(i)
340
+ D(Kαi/b′, Kαi/b).
341
+ Now, by Equation (1), D(Kαi/b′, Kαi/b) is either {0} or {0, b′/b} whenever b|b′. Hence,
342
+ D(Mi, Kαi/b) is bounded, and so D(Mi, Mi) and D(M, M) are bounded. Hence Mi and
343
+ M are inflexible manifolds, i = 1, . . . , n. The same arguments work for N so we conclude.
344
+ 4. Spherical fibrations over inflexible Sullivan models:
345
+ rational mapping degree set
346
+ In this section we prove Theorem B, which can be thought of as the rational version
347
+ of Theorem A. Rational homotoy theory provides an equivalence of categories between
348
+ the category of simply connected rational spaces and the category of certain differential
349
+ graded algebras, the so-called Sullivan minimal models. We refer to [8] for basics facts in
350
+ Rational Homotopy Theory.
351
+ More concretely, if V is a graded rational vector space, we write ΛV for the free com-
352
+ mutative graded algebra on V . A Sullivan model (ΛV, ∂) is a commutative differential
353
+ graded algebra (cdga for short) which is free as commutative graded algebra on a simply
354
+ connected graded vector space V of finite dimension in each degree. It is minimal if in
355
+ addition ∂(W) ⊂ Λ≥2W.
356
+ Now, if M is an oriented closed simply connected manifold, then the cohomology of
357
+ the associated minimal model AM coincides with the rational cohomology of M.
358
+ In
359
+
360
+ 8
361
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
362
+ particular AM has a cohomological fundamental class [AM] ∈ H∗(AM) ∼= H∗(M; Q)
363
+ which is isomorphic to the rational cohomological fundamental class [M]Q of M.
364
+ Ellipticity for a Sullivan minimal model (ΛV, ∂) means that both V and H∗(ΛV ) are
365
+ finite-dimensional.
366
+ Hence, the cohomology is a Poincar´e duality algebra [9] and one
367
+ can easily compute the degree of its fundamental cohomological class [8, Theorem 32.6].
368
+ In particular one can introduce the notion of mapping degree between elliptic Sullivan
369
+ minimal models and also translate the notion of inflexibility:
370
+ Let (ΛV, ∂) be an elliptic Sullivan minimal model. Let µ ∈ (ΛV )n be a representative of
371
+ its cohomological fundamental class. Then (ΛV, ∂) is inflexible if for every cdga-morphism
372
+ ϕ: (ΛV, ∂) → (ΛV, ∂)
373
+ we have deg(ϕ) = 0, ±1, where H([µ]) = deg(ϕ)[µ].
374
+ 4.1. Rational mapping degree set and connected sums. The following results es-
375
+ tablish, under certain restrictions, the relationship between rational mapping degree sets
376
+ and connected sums of manifolds:
377
+ Lemma 4.1 ([7, Lemma II.2]). Let M1, M2 and N be oriented closed n-manifolds with
378
+ πn−1(N(0)) = 0. Then
379
+ DQ(M1#M2, N) = DQ(M1, N) + DQ(M2, N).
380
+ Proof. Under the same assumptions, in [7, Lemma II.2] is asserted that the following
381
+ holds:
382
+ D(M1#M2, N) ⊆ DQ(M1, N) + DQ(M2, N).
383
+ However, a stronger result is demonstrated in the proof. Namely,
384
+ DQ(M1#M2, N) ⊆ DQ(M1, N) + DQ(M2, N).
385
+ Hence, it suffices to prove the other inclusion.
386
+ To that end, one can apply the same
387
+ arguments as in [5, Lemma 7.8]: let q(0) : (M1)(0)#(M2)(0) → (M1)(0) ∨ (M2)(0) denote the
388
+ rationalization of the pinching map. Then for any given maps fi : (Mi)(0) → N(0), the
389
+ composition
390
+ (f1 ∨ f2) ◦ q: (M1)(0)#(M2)(0) → N(0)
391
+ has degree deg(f1) + deg(f2) and the result follows.
392
+
393
+ A precise definition of connected sum in the world of cdga’s:
394
+ Definition 4.2. Let Ai, i = 1, 2, be connected cdgas and let ai ∈ Ai, i = 1, 2, be elements
395
+ of the same degree. The connected sum of the pairs (Ai, [ai]), i = 1, 2, is the dga
396
+ (A1, [a1])#(A2, [a2])
397
+ def
398
+ := (A1 ⊕Q A2)/I ,
399
+ where A1 ⊕Q A2
400
+ def
401
+ := (A1 ⊕ A2)/Q{(1, −1)}, and I ⊂ A1 ⊕Q A2 is the differential ideal
402
+ generated by a1 − a2.
403
+
404
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
405
+ 9
406
+ Connected sums of cdgas provide rational models for connected sums of oriented mani-
407
+ folds. Indeed, for Mi, i = 1, 2 oriented closed simply connected n-manifold, with Sullivan
408
+ minimal model AMi, let mi be a representative of the cohomological fundamental class of
409
+ AMi, for i = 1, 2. By [5, Theorem 7.12]
410
+ (AM1, [m1])#(AM 2, [m2])
411
+ (2)
412
+ is a rational model of M1#M2.
413
+ We use (2) above to prove the rational version of Proposition 3.3:
414
+ Proposition 4.3. Let M1, M2 and N1, N2 be oriented closed simply connected n-manifolds
415
+ such that πn−1(Nj) ⊗ Q = 0, j = 1, 2, and DQ(Mi, Nj) = {0}, i ̸= j. Then
416
+ DQ(M1#M2, N1#N2) = DQ(M1, N1) ∩ DQ(M2, N2).
417
+ Proof. According to Lemma 4.1 and the rational version of Lemma 3.2 (which can be
418
+ proved following the same arguments as in [14, Lemma 4.3]), we get that
419
+ DQ(M1#M2, N1#N2) ⊂ DQ(M1, N1) ∩ DQ(M2, N2).
420
+ Conversely, let (AMi, [mi]) and (ANi, [ni]) be Sullivan minimal models of (Mi, [Mi]) and
421
+ (Ni, [Ni]) respectively, i = 1, 2. For
422
+ d ∈ DQ(M1, N1) ∩ DQ(M2, N2)
423
+ there exists fi : ANi → AMi with fi(ni) = d · mi + αi and where αi is a coboundary,
424
+ i = 1, 2.
425
+ Because (AM1, [m1 + α1])#(AM2, [m2 + α2]) and (AN1, [n1])#(AN2, [n2]) are
426
+ Sullivan minimal models for M1#M2 and N1#N2 respectively, then f1 and f2 give rise to
427
+ a well defined cdga-morphism
428
+ f1#f2 : (AN1, [n1])#(AN2, [n2]) → (AM1, [m1 + α1])#(AM2, [m2 + α2])
429
+ defined by
430
+ (f1#f2)(x) =
431
+
432
+ f1(x),
433
+ if x ∈ AN1,
434
+ f2(x),
435
+ if x ∈ AN2
436
+ and whose degree is deg(f1#f2) = d.
437
+
438
+ Remark 4.4. The previous result can be generalized to an arbitrary finite iterated con-
439
+ nected sum, as in Corollary 3.4. Namely, if Mi, Ni, i = 1, . . . , r, are oriented closed simply
440
+ connected n-manifolds such that πn−1(Nj) = 0, j = 1, . . . , r, and DQ(Mi, Nj) = {0}, for
441
+ i ̸= j, then
442
+ DQ(M1# · · · #Mr, N1# · · · #Nr) =
443
+ r�
444
+ i=1
445
+ DQ(Mi, Ni).
446
+ 4.2. Inflexible Sullivan minimal models of inflexible manifolds. Following the
447
+ same strategy as in Section 3, we consider spherical fibrations over certain elliptic and
448
+ inflexible Sullivan minimal models (Definition 4.5), whose total spaces are the Sullivan
449
+ minimal models of inflexible manifolds (see Lemma 4.7). These manifolds will be the
450
+ building blocks to construct, by means of iterated connected sums, manifolds that realize
451
+ finite sets of rational numbers.
452
+
453
+ 10
454
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
455
+ Definition 4.5. Let (A, ∂) be an elliptic, inflexible Sullivan minimal model of formal
456
+ dimension 2m, m ≥ 1, such that πj(A) = 0 for j ≥ 2m−1. Fix µ ∈ A a representative of
457
+ its cohomological fundamental class. Then for any non-zero q ∈ Q, define the following
458
+ Sullivan minimal model
459
+ (Kq(A), ∂) := (A ⊗ Λ(y2m−1), ∂)
460
+ that extends the differential of A by ∂(y2m−1) = qµ.
461
+ Remark 4.6. Notice that (Kq(A), ∂) is the total space in the rational S2m−1-fiber sequence:
462
+ (Λ(y2m−1), 0) ←− (Kq(A), ∂) ←− (A, ∂),
463
+ whose Euler class is q[µ].
464
+ Lemma 4.7. Let (A, ∂) be an elliptic, inflexible Sullivan minimal model of formal di-
465
+ mension 2m, m ≥ 1, such that πj(A) = 0 for j ≥ 2m − 1. Fix µ ∈ A a representative
466
+ of the fundamental class of A, and let x ∈ A such that ∂(x) = µ2. Then for any non-
467
+ zero q ∈ Q,
468
+
469
+ Kq(A), [y2m−1µ − qx]
470
+
471
+ is the Sullivan minimal model of an oriented closed
472
+ inflexible (4m − 1)-manifold MKq, with the same connectivity as (A, ∂).
473
+ Proof. According to [6, Proposition 3.1], (Kq(A), ∂) is an elliptic Sullivan model of formal
474
+ dimension 4m−1 where y2m−1µ−qx is a representative of its cohomological fundamental
475
+ class. By [6, Lemma 3.2], (Kq(A), ∂) is an inflexible algebra because (A, ∂) is so. Now,
476
+ since its formal dimension is 4m − 1 ≡ 3 mod 4, the obstruction theory of Sullivan [15,
477
+ Theorem (13.2)] and Barge [2, Th´eor`eme 1] guarantees that (Kq(A), [y2m−1µ − qx]) is the
478
+ Sullivan minimal model of an oriented closed simply-connected manifold MKq. Finally,
479
+ by [4, Proposition A.1], MKq and (A, ∂) have the same connectivity.
480
+
481
+ We compute the rational mapping degree set between the manifolds appearing in the
482
+ previous lemma:
483
+ Lemma 4.8. For any non-zero q ∈ Q, let MKq be the oriented closed manifold from
484
+ Lemma 4.7 whose Sullivan minimal model is (Kq(A), ∂) from Definition 4.5. Then
485
+ DQ(MKp, MKq) = {0, q/p}.
486
+ Proof. We follow the ideas in [6, Lemma 3.2].
487
+ Let f : (Kq(A, ∂) → (Kp(A), ∂) be a
488
+ morphism of non-trivial degree d ∈ Q, that is,
489
+ f(y2m−1µ − qx) = d(y2m−1µ − px) + α
490
+ (3)
491
+ where α is a coboundary. By a degree argument, f induces a non-trivial degree morphism
492
+ f|A : (A, ∂) → (A, ∂). On the one hand f(µ) = �dµ + β1 and f(x) = �d 2x + β2 where β1, β2
493
+ are coboundaries, and �d ∈ {−1, 1}. On the other hand, f(y2m−1) = ay2m−1 + γ where
494
+ a ∈ Q and γ is a coboundary.
495
+ Because f(∂y2m−1) = ∂f(y2m−1), we get that ap = q �d and β1 = 0. Hence a = �d (q/p)
496
+ and
497
+ f(y2m−1µ − qx) =
498
+
499
+ (�d q/p y2m−1 + γ
500
+
501
+ (�dµ) − q(�d 2x + β2)
502
+ = (�d 2q/p)(y2m−1µ − px) − qβ2
503
+ = (q/p)(y2m−1µ − px) − qβ2
504
+ (recall �d ∈ {−1, 1}).
505
+ By comparing this equation to (3), we obtain that d = q/p and the proof is complete.
506
+
507
+
508
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
509
+ 11
510
+ We illustrate the existence of elliptic, inflexible Sullivan minimal models satisfying the
511
+ conditions from Definition 4.5 and Lemma 4.7:
512
+ Definition 4.9. Let Γ be a connected finite simple graph with more that one vertex, i.e.,
513
+ |V (Γ)| > 1. Given an integer k ≥ 1, let (Ak(Γ), ∂) be the (30k +17)-connected elliptic and
514
+ inflexible Sullivan algebra constructed in [4, Definition 2.1], whose formal dimension is
515
+ 2m = 540k2+984k +396+|V (Γ)|(360k2+436k +132) and πj
516
+
517
+ Ak(Γ)
518
+
519
+ = 0 for j ≥ 2m−1.
520
+ Fix µ ∈ Ak(Γ) a representative of the cohomological fundamental class. Then for any
521
+ nonz-zero q ∈ Q, define the following Sullivan minimal model
522
+ (Kq(Γ, k), ∂) := (Ak(Γ) ⊗ Λ(y2m−1), ∂)
523
+ that extends the differential of Ak(Γ) by ∂(y2m−1) = qµ.
524
+ Remark 4.10. Because conditions from Lemma 4.7 hold, (Kq(Γ, k), ∂) is a Sullivan model
525
+ of an oriented closed (30k + 17)-connected inflexible (4m − 1)-manifold MKq(Γ,k), where
526
+ 2m = 540k2 + 984k + 396 + |V (Γ)|(360k2 + 436k + 132).
527
+ Lemma 4.11. Let Γ1 and Γ2 be connected finite simple graphs with |V (Γ1)| = |V (Γ2)| > 1.
528
+ Given a positive integer k ≥ 1, and a non-zero pi ∈ Q, i = 1, 2, consider the manifold
529
+ MKpi(Γi,k), i = 1, 2 as in Remark 4.10. Then
530
+ DQ(MKp1(Γ1,k), MKp2(Γ2,k)) =
531
+
532
+ {0, p2/p1},
533
+ if
534
+ Γ1 ∼= Γ2,
535
+ {0},
536
+ otherwise.
537
+ Proof. Let (Kpi(Γi, k), ∂) = (Ak(Γi) ⊗ Λ(yi), ∂), introduced in Definition 4.9, be the Sul-
538
+ livan model of the manifold MKpi(Γi,k), where ∂(yi) = piµi for µi a representative of the
539
+ cohomological fundamental class of Ak(Γi), i = 1, 2. Recall from Lemma 4.7 that for
540
+ xi ∈ Ak(Γi) satisfying ∂(xi) = µ2
541
+ i , the element yiµi − pixi is a representative of the
542
+ cohomological fundamental class of (Kpi(Γi, k), ∂), i = 1, 2.
543
+ With these constructions in mind, we follow the ideas from [6, Lemma 3.2]. Consider
544
+ a morphism of non-trivial degree d ∈ Q:
545
+ f : (Kp2(Γ2, k), ∂) → (Kp1(Γ1, k), ∂).
546
+ Then f(y2µ2 − p2x2) = d(y1µ1 − p1x1) + α with α a coboundary. By a degree argument,
547
+ the morphism f induces a non-trivial degree morphism
548
+ f|Ak(Γ2): (Ak(Γ2), ∂) → (Ak(Γ1), ∂).
549
+ Focusing specifically on this former morphism, the arguments in [4, Lemma 2.12] (see
550
+ also [6, Remark 2.8]), show that it is induced by a graph full monomorphism σ: Γ1 → Γ2.
551
+ Now, since |V (Γ1)| = |V (Γ2)|, σ is indeed an isomorphism of graphs, and f(µ2) = µ1 + β1
552
+ and f(x2) = x1 + β2 with β1, β2 coboundaries, by [4, Lemma 2.12].
553
+ Finally, by another degree reasoning argument, one obtains that f(y2) = ay1 + γ where
554
+ a is a non-zero rational number, and γ is a coboundary. We conclude as in the proof of
555
+ Lemma 4.8.
556
+
557
+
558
+ 12
559
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
560
+ 4.3. Proof of Theorem B. Let A = {0, d1, . . . , dn} where d1, d2, . . . , dn are pairwise
561
+ different non-zero rational numbers. Fix an integer k ≥ 1. According to Corollary 2.3,
562
+ there exist finite sequences of not necessarily pairwise distinct non-zero rational numbers
563
+ B(i), i = 0, . . . , n, such that
564
+ A =
565
+ n�
566
+ i=0
567
+ SB(i).
568
+ Choose Γ0, Γ1, . . . , Γn, pairwise non-isomorphic connected finite simple graphs, such
569
+ that |V (Γi)| = |V (Γj)| > 1 for every i, j = 0, . . . , n. According to Remark 4.10, we define
570
+ the (30k + 17)-connected manifolds
571
+ Mi =
572
+ #
573
+ b∈B(i)
574
+ MKb−1(Γi,k)
575
+ Ni = MK1(Γi,k),
576
+ for i = 0, . . . , n. By Lemmas 4.11 and the rational version of Lemma 3.2 (which can be
577
+ proved following the same arguments as in [14, Lemma 4.3]), we have that
578
+ DQ(Mi, Ni) = SB(i) , and
579
+ DQ(Mi, Nj) = {0}, for i ̸= j.
580
+ Finally, define
581
+ M = M0#M1# . . . #Mn,
582
+ N = N0#N1# . . . #Nn.
583
+ and use Proposition 4.3 (see also Remark 4.4) to get
584
+ DQ(M, N) =
585
+ N�
586
+ i=0
587
+ SB(i) = A.
588
+ 5. From unstable Adams operations to mapping degree sets
589
+ We recall the basics on unstable Adams operations following Jackowski-McCLure-
590
+ Oliver’s work [10, 11]. Given a compact connected Lie group G, a self-map f : BG → BG
591
+ is called an unstable Adams operation of degree r ≥ 0, if H2i(f; Q) is the multiplication
592
+ by ri for each i > 0 [10, p. 183]. For a given simple Lie group G with Weyl group WG,
593
+ an unstable Adams operation of degree r > 0 exists if and only if (r, |WG|) = 1, and
594
+ moreover, this operation is unique [10, Theorem 2]. In particular, when G = SO(2m − 1)
595
+ or G = SO(2m), m > 1, unstable Adams operations of degree r > 0 exist if r and m!
596
+ are coprime numbers. In what follows, we denote by ϕr the unstable Adams operation
597
+ of degree r > 0 on BSO(2m − 1) and BSO(2m). Notice that since they are unique, then
598
+ ϕs ◦ ϕr = ϕrs.
599
+ Henceforward, (Σ, [Σ]) is a fixed oriented closed connected 2m-manifold whose ratio-
600
+ nalization (Σ(0), [Σ]Q) is inflexible and πj(Σ(0)) = 0 for j ≥ 2m − 1. Let (AΣ, ∂) be a
601
+ Sullivan minimal model of Σ. Denote by π: Σ → S2m the map obtained by collapsing the
602
+ (2m − 1)-skeleton of Σ.
603
+
604
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
605
+ 13
606
+ Lemma 5.1. Let X2m ∈ H2m(BSO(2m); Z) be the Euler class of the spherical fiber se-
607
+ quence
608
+ S2m−1 → BSO(2m − 1) → BSO(2m),
609
+ thus X2m is a torsion free integral cohomology class [3, Theorem 1.5, Equation (2.1)], and
610
+ S2m is thought of as an oriented closed manifold. There exists ι: S2m → BSO(2m), a tor-
611
+ sion free element in π2m(BSO(2m)), and a non-zero integer κ ∈ Z such that H∗(ι; Z)(X2m) =
612
+ κ[S2m].
613
+ Proof. Recall that π2m(BSO(2m)) ∼= π2m−1(SO(2m)). By [13, Corollary IV.6.14] (see also
614
+ [12, p. 161]), π2m−1(SO(2m)) contains a copy of Z inducing the p-local (thus rational)
615
+ splitting SO(2m) ≃(p) SO(2m − 1) × S2m−1 [13, Corollary IV.6.21]. Let ι be a generator
616
+ of such a copy of Z in π2m(BSO(2m)).
617
+ By construction, H∗(ι; Q) is non-trivial on the Euler class of the rational fiber sequence
618
+ S2m−1
619
+ (0)
620
+ → BSO(2m − 1)(0) → BSO(2m)(0),
621
+ which is just X2m ⊗Q 1. Therefore, H∗(ι; Z)(X2m) = κ[S2m] for some non-zero κ ∈ Z.
622
+
623
+ Definition 5.2. Given any integers r > 0, m > 1, with r coprime to m!, we define:
624
+ (1) The oriented (4m − 1)-manifold Erm as the total space in the principal spherical
625
+ SO(2m)-fiber bundle
626
+ S2m−1
627
+ S2m−1
628
+ Erm
629
+ BSO(2m − 1)
630
+ Σ
631
+ BSO(2m),
632
+
633
+ φr
634
+ (4)
635
+ where φr = ϕr ◦ ι ◦ π.
636
+ (2) The oriented (4m−1)–manifold E−rm obtained by reversing the original orientation
637
+ on the manifold Erm above introduced.
638
+ Remark 5.3. The Euler class of the spherical fiber bundle over Σ given in diagram (4) is
639
+ κrm[Σ] by construction.
640
+ Recall from the beginning of this section that (Σ, [Σ]) is a fixed oriented closed connected
641
+ 2m-manifold where (AΣ, ∂) is its Sullivan minimal model.
642
+ Lemma 5.4. Let Erm be the manifold introduced in Definition 5.2. A Sullivan mini-
643
+ mal model of Erm is Kκrm(AΣ) as given in Definition 4.5. Therefore Erm is rationally
644
+ equivalent to MKκrm, the manifold given in Lemma 4.7.
645
+ Proof. As it was pointed out in Remark 4.6, Kκrm(AΣ) is a Sullivan minimal model for
646
+ the total space in a rational S2m−1-fiber sequence whose Euler class is κrm[Σ]Q. It coin-
647
+ cides with the Euler class of the rationalization of the spherical SO(2m)-fiber bundle in
648
+ diagram (4). Therefore Kκrm(AΣ) is a Sullivan minimal model for Erm.
649
+
650
+
651
+ 14
652
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
653
+ Lemma 5.5. Let i, j, m be positive integers, m > 1, such that (i, m!) = (j, m!) = 1, and
654
+ let Erm, r = i, j, be the (4m − 1)-manifold introduced in Definition 5.2. Then
655
+ D(Eim, Ejm) =
656
+
657
+ {0, (j/i)m},
658
+ if i|j,
659
+ {0},
660
+ if i̸ |j.
661
+ Proof. By Lemma 5.4, the manifolds Erm and MKκrm are rationally equivalent, for every
662
+ 0 < r ∈ Z. Therefore:
663
+ D(Eim, Ejm) ⊂ DQ(Eim, Ejm) ∩ Z = DQ(MKκim, MKκjm) ∩ Z
664
+ = {0, (j/i)m} ∩ Z (by Lemma 4.8)
665
+ =
666
+
667
+ {0, (j/i)m},
668
+ if i|j,
669
+ {0},
670
+ if i̸ |j.
671
+ The proof will be completed if we construct a map f : Eim → Ejm of degree (j/i)m
672
+ when i|j. To this end, let us suppose that j = di, d ∈ Z, and recall that unstable Adams
673
+ operations satisfy that ϕj = ϕd ◦ ϕi.
674
+ Therefore, by construction (see Definition 5.2)
675
+ φj = ϕd ◦ φi.
676
+ Let f : Eim → Ejm be the map obtained by the universal property of
677
+ pullbacks in the following commutative diagram:
678
+ Eim
679
+ BSO(2m − 1)
680
+ Ejm
681
+ BSO(2m − 1)
682
+ Σ
683
+ BSO(2m)
684
+ BSO(2m)
685
+ f
686
+ ϕd
687
+
688
+ φi
689
+ ϕd
690
+ (5)
691
+ Diagram (5) gives rise to a commutative diagram of spherical fiber sequences
692
+ S2m−1
693
+ S2m−1
694
+ Eim
695
+ Ejm
696
+ Σ
697
+ Σ,
698
+ �f
699
+ f
700
+ (6)
701
+ whose associated Serre spectral sequences (Sss) can be compared via the edge morphisms
702
+ given by naturality: the Sss associated to the left (resp. right) side of diagram (6) is fully
703
+ determined by the differential
704
+ d2m([S2m−1]) = κim[Σ] (resp. d2m([S2m−1]) = κjm[Σ]),
705
+ and since by naturality
706
+ d2m
707
+
708
+ H∗( �f)([S2m−1])
709
+
710
+ = H∗(IdΣ)
711
+
712
+ d2m([S2m−1])
713
+
714
+ we obtain that deg( �f) = (j/i)m.
715
+
716
+ FINITE SETS CONTAINING ZERO ARE MAPPING DEGREE SETS
717
+ 15
718
+ Now, the cohomological fundamental class [Eim] (resp. [Ejm]) is represented by the
719
+ class [S2m−1]⊗[Σ] in the E2m−1,2m
720
+
721
+ -term of the Sss associated to the left (resp. right) fiber
722
+ sequence in diagram (6). Hence by naturality
723
+ H∗(f)([Ejm]) = H∗( �f)([S2m−1]) ⊗ H∗(IdΣ)([Σ])]
724
+ =
725
+
726
+ (j/i)m[S2m−1]
727
+
728
+ ⊗ [Σ]
729
+ = (j/i)m[Eim]
730
+ and therefore deg(f) = (j/i)m.
731
+
732
+ Remark 5.6. Notice that manifolds Erm and E−rm differ in just the orientation. Hence,
733
+ for any other oriented closed connected (4m − 1)-manifold N, the mapping set degree is
734
+ D(E−rm, N) = −D(Erm, N) and D(N, E−rm) = −D(N, Erm).
735
+ Proof of Theorem C. Let Σ be an oriented closed k-connected 2m-manifold verifying
736
+ that Σ(0) is inflexible and πj(Σ(0)) = 0 for j ≥ 2m − 1. Let A = {0, d1, . . . , dn} where
737
+ d1, d2, . . . , dn are pairwise different non-zero integers.
738
+ According to Proposition 2.2, there exist finite sequences B(i), i = 0, . . . , n, of not
739
+ necessarily pairwise distinct non-zero integers, such that every element in B(i) can be
740
+ written as ±rm for 0 < r ∈ Z with (r, m!) = 1, and
741
+ A =
742
+ n�
743
+ i=0
744
+ SB(i).
745
+ Choose pairwise distinct prime numbers q0, q1, . . . , qn, in such a way that
746
+ qj > max{|b| : b ∈ B(i), i = 0, . . . , n}
747
+ and (qj, m!) = 1, for j = 0, . . . , n. Let αi = qm
748
+ i
749
+
750
+ b∈B(i)
751
+ b, for every i = 0, . . . , n. Notice
752
+ that αi and αi/b, b ∈ B(i), are integers that can be written up to a sign as rm for some
753
+ positive integer r such that (r, m!) = 1
754
+ Following the notation in Definition 5.2, we define the following (4m − 1)-manifolds
755
+ Mi =
756
+ #
757
+ b∈B(i)
758
+ Eαi/b
759
+ Ni = Eαi
760
+ for i = 0, . . . , n. According to Lemma 5.5 and Lemma 3.1, we deduce that
761
+ D(Mi, Ni) = SB(i) , and
762
+ D(Mi, Nj) = {0}, for i ̸= j.
763
+ Finally, we construct
764
+ M = M0#M1# . . . #Mn,
765
+ N = N0#N1# . . . #Nn,
766
+ and, according to Corollary 3.4, we obtain that
767
+ D(M, N) =
768
+ N
769
+
770
+ i=0
771
+ SB(i) = A.
772
+
773
+ 16
774
+ C. COSTOYA, V. MU˜NOZ, AND A. VIRUEL
775
+ References
776
+ [1] M. Amann. Degrees of self-maps of simply connected manifolds. Int. Math. Res. Not., 2015(18):8545–
777
+ 8589, 2015.
778
+ [2] J. Barge. Structures diff´erentiables sur les types d’homotopie rationnelle simplement connexes. Ann.
779
+ Sci. ´Ec. Norm. Sup´er. (4), 9:469–501, 1976.
780
+ [3] E. H. j. Brown. The cohomology of BSOnand BOnwith integer coefficients. Proc. Am. Math. Soc.,
781
+ 85:283–288, 1982.
782
+ [4] C. Costoya, D. M´endez, and A. Viruel. Homotopically rigid Sullivan algebras and their applications.
783
+ In An alpine bouquet of algebraic topology, volume 708 of Contemp. Math., pages 103–121. Amer.
784
+ Math. Soc., Providence, RI, 2018.
785
+ [5] C. Costoya, V. Mu˜noz, and A. Viruel. On Strongly Inflexible Manifolds. International Mathematics
786
+ Research Notices, 03 2022. rnac064.
787
+ [6] C. Costoya and A. Viruel. Every finite group is the group of self-homotopy equivalences of an elliptic
788
+ space. Acta Math., 213(1):49–62, 2014.
789
+ [7] D. Crowley and C. L¨oh. Functorial seminorms on singular homology and (in)flexible manifolds.
790
+ Algebr. Geom. Topol., 15(3):1453–1499, 2015.
791
+ [8] Y. F´elix, S. Halperin, and J.-C. Thomas. Rational homotopy theory, volume 205 of Springer-Verlag.
792
+ Springer-Verlag, New York, 2001.
793
+ [9] S. Halperin. Finiteness in the minimal models of sullivan. Trans. Amer. Math. Soc., 230:173–199,
794
+ 1977.
795
+ [10] S. Jackowski, J. E. McClure, and B. Oliver. Homotopy classification of self-maps of BG via G-actions.
796
+ I. Ann. Math. (2), 135(1):183–226, 1992.
797
+ [11] S. Jackowski, J. E. McClure, and B. Oliver. Homotopy classification of self-maps of BG via G-actions.
798
+ II. Ann. Math. (2), 135(2):227–270, 1992.
799
+ [12] M. A. Kervaire. Some non-stable homotopy groups of Lie groups. Ill. J. Math., 4:161–169, 1960.
800
+ [13] M. Mimura and H. Toda. Topology of Lie groups, I and II. Transl. from the Jap. by Mamoru Mimura
801
+ and Hirosi Toda, volume 91 of Transl. Math. Monogr. Providence, RI: American Mathematical
802
+ Society, 1991.
803
+ [14] C. Neofytidis, S. Wang, and Z. Wang. Realising sets of integers as mapping degree sets. Bull. Lond.
804
+ Math. Soc. (to appear), 2022.
805
+ [15] D. Sullivan. Infinitesimal computations in topology. Publ. Math., Inst. Hautes ´Etud. Sci., 47:269–331,
806
+ 1977.
807
+ CITIC, Departamento de Computaci´on, Universidade da Coru˜na, 15071-A Coru˜na, Spain.
808
+ Email address: [email protected]
809
+ Departamento de ´Algebra, Geometr´ıa y Topolog´ıa, Universidad Complutense de Madrid,
810
+ Plaza de las Ciencias, 3, 28040-Madrid, Spain
811
+ Email address: [email protected]
812
+ Departamento de ´Algebra, Geometr´ıa y Topolog´ıa, Universidad de M´alaga, Campus
813
+ de Teatinos, s/n, 29071-M´alaga, Spain
814
+ Email address: [email protected]
815
+
edFST4oBgHgl3EQfFzg4/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
f9AzT4oBgHgl3EQfavwP/content/tmp_files/2301.01372v1.pdf.txt ADDED
@@ -0,0 +1,2306 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Spatially Varying Anisotropy for Gaussian Random Fields
2
+ in Three-Dimensional Space
3
+ Martin Outzen Berild∗and Geir-Arne Fuglstad
4
+ Department of Mathematical Sciences,
5
+ Norwegian University of Science and Technology, Norway
6
+ Abstract
7
+ Isotropic covariance structures can be unreasonable for phenomena in
8
+ three-dimensional spaces. We construct a class of non-stationary anisotropic
9
+ Gaussian random fields (GRFs) in three dimensions through stochastic par-
10
+ tial differential equations allowing for Gaussian Markov random field approx-
11
+ imations. The class is proven in a simulation study where we explore the
12
+ amount of data required to estimate these models. Then, we apply it to an
13
+ ocean mass outside Trondheim, Norway, based on simulations from a numer-
14
+ ical ocean model. And our model outperforms a stationary anisotropic GRF
15
+ on predictions using in-situ measurements collected with an autonomous
16
+ underwater vehicle.
17
+ Keywords: Spatial non-stationarity; spatially-varying anisotropy; stochastic par-
18
+ tial differential equations; Gaussian Markov random fields.
19
+ ∗Corresponding author, [email protected]
20
+ 1
21
+ arXiv:2301.01372v1 [stat.ME] 3 Jan 2023
22
+
23
+ 1
24
+ Introduction
25
+ Gaussian random fields (GRFs) are a powerful tool for spatial and spatio-temporal
26
+ geostatistical modeling (Diggle et al., 1998; Cressie and Wikle, 2015). When the
27
+ key goal is predictions at unobserved locations, i.e., kriging, isotropic covariance
28
+ functions often perform well, and more flexible covariance structures should be
29
+ used with care (Fuglstad et al., 2015b). However, the screening effect in kriging
30
+ (Stein, 2002) is not relevant in other settings where the primary goal is the esti-
31
+ mated covariance structure. E.g., to describe internal variability in a climate model
32
+ ensemble (Castruccio et al., 2019), or to produce a spatial prior based on numerical
33
+ simulations that will later be used to guide autonomous sampling (Fossum et al.,
34
+ 2021; Foss et al., 2021). For the former, Fuglstad and Castruccio (2020); Hu et al.
35
+ (2021) demonstrated that flexible covariance structures can perform better than
36
+ stationary covariance structures.
37
+ There are many approaches to constructing flexible covariance structures (Samp-
38
+ son, 2010; Salvaña and Genton, 2021; Schmidt et al., 2011). Some early approaches
39
+ are the deformation method (Sampson and Guttorp, 1992) and kernel convolutions
40
+ (Paciorek and Schervish, 2006), but they both involve the covariances between any
41
+ pair of locations. This means standard implementations are infeasible for large
42
+ datasets. There are many ways to overcome such computational issues in spatial
43
+ statistics and some are applicable for flexible covariance structures (Heaton et al.,
44
+ 2019). The stochastic partial differential equation (SPDE) approach (Lindgren
45
+ et al., 2011) is interesting because it directly gives rise to computationally efficient
46
+ models and easily extends to non-stationary covariance models.
47
+ However, increasing the degree of flexibility in the covariance structure requires
48
+ increasing the number of parameters. The common isotropic Matérn covariance
49
+ functions (Stein, 2012) are parametrized through 3 parameters: marginal vari-
50
+ ance, range, and smoothness. Flexible models can have 100s or more parameters
51
+ 2
52
+
53
+ (Fuglstad et al., 2015b). An appealing way to reduce dimensionality is to describe
54
+ the covariance structure through covariates (Schmidt et al., 2011; Neto et al., 2014;
55
+ Ingebrigtsen et al., 2014, 2015; Risser and Calder, 2015).
56
+ The aforementioned works are all considering flexible covariance structures in
57
+ two-dimensional space, and while the methods can be extended to three-dimensional
58
+ space, the literature is sparse. For example, the SPDE approach has been used
59
+ for simple anisotropic covariance structures in the context of fMRI data from the
60
+ brain (Sidén et al., 2021), and more complex covariance structures in the context of
61
+ astronomy (Lee and Gammie, 2021), though this was two-dimensional space and
62
+ time treated as three-dimensional space. However, spatially varying anisotropy
63
+ in the SPDE approach (Fuglstad et al., 2015a) has not been extended to three-
64
+ dimensional space.
65
+ The aim of this paper is to develop a new method for spatially varying anisotropy
66
+ in three-dimensional space through the SPDE approach.
67
+ A key advantage is
68
+ that the formulation as an SPDE guarantees a valid covariance structure, and
69
+ the main challenge is how to describe and parametrize non-stationary covariance
70
+ structures. Fuglstad et al. (2015a) used one vector field to describe spatially vary-
71
+ ing anisotropy, but in three dimensions, two spatially varying orthogonal vector
72
+ fields are necessary for full generality.
73
+ In a simulation study, we investigate how much data is necessary to recover
74
+ parameters for three different model complexities: stationary isotropic, station-
75
+ ary anisotropic, and non-stationary anisotropic. We then estimate GRF priors to
76
+ encode knowledge about the ocean from a numerical forecast generated by the nu-
77
+ merical model SINMOD by SINTEF. A stationary GRF prior and a non-stationary
78
+ GRF prior are updated based on in-situ measurements by an autonomous under-
79
+ water vehicle (AUV), and we evaluate the predictive ability during a mission in
80
+ Trondheimsfjorden, Norway, on May 27, 2021. Improved predictions are key, for
81
+ 3
82
+
83
+ example, in autonomous sampling of the oceans (Fossum et al., 2019, 2021), but
84
+ current approaches in autonomous ocean sampling are limited to stationary GRFs.
85
+ In Section 2, we describe how to model anisotropy and non-stationarity in three
86
+ dimensions using SPDEs. Then in Section 3, we describe how to perform inference
87
+ for the new model in a computationally efficient way. In Section 4, we describe
88
+ the simulation study and discuss the results, and continue with the application to
89
+ sampling in the ocean in Section 5. We end with a discussion in Section 6.
90
+ 2
91
+ Constructing SPDEs with spatially varying anisotropy
92
+ 2.1
93
+ Existing models
94
+ The Matérn covariance function on R3 is given by
95
+ r(s1, s2) =
96
+ σ2
97
+ 2ν−1Γ(ν)(κ||s1 − s2||)νKν(κ||s1 − s2||),
98
+ s1, s2 ∈ R3,
99
+ (1)
100
+ where ||·|| is the Euclidean distance in R3, σ > 0 is the marginal standard deviation,
101
+ Kν is the modified Bessel function of the second kind and order ν > 0, and κ > 0
102
+ is an inverse spatial scale parameter. As discussed in Lindgren et al. (2011), GRFs
103
+ with this covariance function is the stationary solutions of the SPDE
104
+ (κ2 − ∇ · ∇)α/2(τu(s)) = W(s),
105
+ s ∈ R3,
106
+ (2)
107
+ where α = ν + 3/2, τ =
108
+
109
+ 8πκ/σ, ∇ · ∇ is the Laplacian, and W is a standard
110
+ Gaussian white noise process.
111
+ Lindgren et al. (2011) proposed to introduce non-stationarity by allowing κ
112
+ and τ to vary in space (Ingebrigtsen et al., 2014, 2015) or by deformations of space
113
+ (Hildeman et al., 2021). Fuglstad et al. (2015a,b) consider a version of the SPDE,
114
+ where the Laplacian is replaced by an anisotropic Laplacian where the direction
115
+ and degree of anisotropy vary spatially. This was further extended to spherical
116
+ 4
117
+
118
+ geometry in Fuglstad and Castruccio (2020); Hu et al. (2021). However, all of
119
+ these works were in two-dimensional base spaces, and only simpler models have
120
+ been applied for three-dimensional base spaces (Sidén et al., 2021).
121
+ The key idea in Fuglstad et al. (2015a) was to replace ∇ · ∇ by ∇ · H(s)∇,
122
+ where H(s) is everywhere a symmetric positive definite 2 × 2 matrix that controls
123
+ the strength and direction of anisotropy. The matrix-valued function was specified
124
+ as H(s) = γ(s)I2 + v(s)v(s)T, s ∈ R2, where γ(·) is a positive function and v(·)
125
+ is a vector field. This allows γ(·) to control the baseline strength of dependence in
126
+ all directions, and v(·) to control the strength and direction of additional spatial
127
+ dependence. However, the same parametrization in R3 is not sufficiently general
128
+ to control anisotropy fully.
129
+ 2.2
130
+ Stationary anisotropy in R3
131
+ We follow the idea in Fuglstad et al. (2015a) for R2, and change the SPDE in
132
+ Equation (2) to
133
+ (κ2 − ∇ · H∇)u(s) = W(s), s ∈ R3,
134
+ (3)
135
+ where ∇·H∇ is an anisotropic Laplacian and the symmetric positive definite 3×3
136
+ matrix H controls the anisotropy. The parameter τ has been dropped since κ and
137
+ H together control both marginal variance and correlation.
138
+ As shown in Appendix A.1, the resulting marginal variance is
139
+ σ2
140
+ m =
141
+ 1
142
+ 8πκ
143
+
144
+ det(H)
145
+ (4)
146
+ and the covariance function is explicitly known as
147
+ r(s1, s2) =
148
+ 1
149
+ 8πκ
150
+
151
+ det(H)
152
+ exp
153
+
154
+ −κ||H−1/2(s1 − s2)||)
155
+
156
+ (5)
157
+ for s1, s2 ∈ R3. The latter is derived in Appendix A.2. This corresponds to geo-
158
+ metric anisotropy in the Matérn covariance function with smoothness ν = 1/2. To
159
+ 5
160
+
161
+ understand the behavior of the covariance function, it is useful to think about H
162
+ in terms of its eigenvalue decomposition. Let ˜v1, ˜v2, and ˜v3 be orthonormal eigen-
163
+ vectors corresponding to eigenvalues λ1, λ2 and λ3, respectively. Then Figure 1
164
+ shows an example of the 0.37 level iso-correlation surface that will arise from the
165
+ covariance function in Equation (5). The semi-axes of the ellipsoid in the figure
166
+ are v1 = (√λ1/κ)˜v1, v2 = (√λ2/κ)˜v2, and v3 = (√λ3/κ)˜v3, which by evaluating
167
+ the covariance function with either of these semi-axes will yield the relationship
168
+ and the iso-correlation level r(v)/σ2
169
+ m = e−1 ≈ 0.37.
170
+ v
171
+ v
172
+ v
173
+ 1
174
+ 2
175
+ 3
176
+ Figure 1: Iso-correlation surface at the ∼0.37 level of Equation (5), where v1, v2,
177
+ and v3 are the eigenvectors of H with lengths √λ1/κ, √λ2/κ and √λ3/κ.
178
+ We generalize the parametrization described in Section 2.2 and H is decom-
179
+ posed as
180
+ H = γI3 + vvT + ωωT.
181
+ (6)
182
+ where v = (vx, vy, vz)T ∈ R3 and w = (ωx, ωy, ωz)T ∈ R3, v ⊥ ω, and γ > 0.
183
+ The eigenvalue decomposition of H has eigenvalues λ1 = γ, λ2 = γ + ||v||2 and
184
+ 6
185
+
186
+ λ3 = γ + ||w||2 with the corresponding eigenvectors v1 = v × ω, v2 = v and
187
+ v3 = ω, respectively. We construct ω by a linear combination of two orthogonal
188
+ vectors in the plane with v as normal vector. First, let ω1 = (−vy, vx, 0)T, which
189
+ satisfies v ⊥ ω1. Second, let ω2 = v × ω1 = (−vzvx, −vzvy, v2
190
+ x + v2
191
+ y)T, which also
192
+ satisfies v ⊥ ω2. We parametrize ω through
193
+ ω = ρ1
194
+ ω1
195
+ ||ω1|| + ρ2
196
+ ω2
197
+ ||ω2||,
198
+ (7)
199
+ where ρ1, ρ2 ∈ R which works whenever vx = vy ̸= 0. An alternative solution is to
200
+ use Euler-Rodrigues parametrization (Euler, 1771; Rodrigues, 1840) to obtain both
201
+ v and ω; however, in this case, the parameters are less interpretable and the issue
202
+ is simply nullified by numerical optimization with appropriate initial parameter
203
+ values.
204
+ The above parametrization for H uses six parameters, γ, vx, vy, vz, ρ1, and
205
+ ρ2, to describe all forms of geometric anisotropy. The parameterization is inter-
206
+ pretable: 1) γ controls the isotropic effect, 2) vx, vy, and vz controls one anisotropy
207
+ in one direction, and 3) ρ1 and ρ2 controls anisotropy in a second direction orthog-
208
+ onal to the first. Lastly, κ simultaneously controls scaling of spatial dependence
209
+ equally in all directions, and the variance of the GRF together with the six other
210
+ parameters as seen in Equation (4).
211
+ 2.3
212
+ Spatially varying anisotropy on bounded domain D ⊂ R3
213
+ Non-stationarity and spatially varying anisotropy is achieved by making the coef-
214
+ ficients in Equation (3) spatially varying,
215
+ (κ(s)2 − ∇ · H(s)∇)u(s) = W(s),
216
+ s ∈ R3,
217
+ (8)
218
+ where κ(·) is a positive function, and H is a spatially varying symmetric positive
219
+ definite 3 × 3 matrix.
220
+ Heuristically, one can imagine that the SPDE is gluing
221
+ 7
222
+
223
+ together different local behavior described by ellipsoids, as discussed in Section
224
+ 2.2, to a valid non-stationary covariance structure.
225
+ In practice, we need to limit Equation (8) to a bounded domain to parametrize
226
+ the non-stationarity. The SPDE we propose is
227
+ (κ(s)2 − ∇ · H(s)∇)u(s) = W(s),
228
+ s ∈ D ⊂ R3,
229
+ (9)
230
+ where D is bounded, and we enforce the boundary condition
231
+ (H(s)∇u(s))Tn(s),
232
+ s ∈ ∂D,
233
+ where n(s) is the outward normal vector of D. This corresponds to no flux through
234
+ the boundary. The effect of the boundary conditions is increased marginal variance
235
+ on the boundary and increased spatial dependency due to the “reflective” boundary
236
+ condition. As discussed in Lindgren et al. (2011); Fuglstad et al. (2015b), one can
237
+ extend the domain D outside the area with observations to reduce boundary effects,
238
+ or one can consider the boundary effects a feature that the non-stationary model
239
+ can adjust for if necessary.
240
+ 3
241
+ Estimating SPDEs with spatially varying anisotropy
242
+ 3.1
243
+ Parameterizing the non-stationarity
244
+ Before using the SPDE in Equation (8) in inference, we parametrize the non-
245
+ stationarity through a finite number of parameters.
246
+ This involves expanding
247
+ log(κ(·)), log(γ(·)), vx(·), vy(·), vz(·), ρ1(·), and ρ2(·) in basis functions.
248
+ The
249
+ log-transform is used for κ(·) and γ(·) since they must be positive functions.
250
+ Let g : R3 → R denote a generic function that we want to expand in a basis,
251
+ and let p > 0 the number of basis functions.
252
+ We use basis splines similar to
253
+ Fuglstad et al. (2015b), and set
254
+ g(s) = f(s)Tαg,
255
+ (10)
256
+ 8
257
+
258
+ where αg ∈ Rp, and f(s) = (f1(s), . . . , fp(s))T is a p-dimensional vector with the
259
+ basis functions evaluated at location s.
260
+ In this paper, we will use rectangular domains D = [A1, B1]×[A2, B2]×[A3, B3],
261
+ and a basis constructed as a tensor product of three one-dimensional B-splines.
262
+ This means that p = m3, where m > 0 is the number of basis functions used in each
263
+ dimension. We use clamped splines where the derivative is 0 at each boundary,
264
+ and the construction of the clamped one-dimensional B-splines is discussed in
265
+ Appendix A.3.
266
+ Figure 2 shows an example of the resulting basis functions in
267
+ 1-dimension.
268
+ 0.0
269
+ 0.2
270
+ 0.4
271
+ 0.6
272
+ 0.8
273
+ 1.0
274
+ 0.0
275
+ 0.2
276
+ 0.4
277
+ 0.6
278
+ 0.8
279
+ 1.0
280
+ Figure 2: Clamped B-spline basis with three basis functions in 1D.
281
+ Let Bx,i denote the i-th basis function of the second-order basis in the x-
282
+ dimension, and similarly By,j and Bz,k for the y- and z-dimension. The resulting
283
+ tree-dimensional basis is then
284
+ fijk (s) = Bx,i(s1) · By,j(s2) · Bz,k(s3),
285
+ s = (s1, s2, s3)T ∈ D,
286
+ (11)
287
+ for all combinations i, j, k ∈ {1, . . . , m}. This means that αg ∈ Rm3, and m3
288
+ parameters must be estimated for each of the seven functions described at the
289
+ start of the section.
290
+ 9
291
+
292
+ Figure 3: Parameterized function representation with B-splines in 3D.
293
+ In Sections 4 and 5, we use m = p3 = 33 = 27. For a total of 189 parameters
294
+ in the seven functions. When data is sparse, such a model can easily result in
295
+ overfitting (Fuglstad et al., 2015b), and it is necessary to introduce penalties on
296
+ the seven functions. In Fuglstad et al. (2015b), this was achieved by a hierarchical
297
+ model where
298
+ τg∆g(s) = Wg(s),
299
+ s ∈ D,
300
+ together with Neumann boundary conditions of zero derivatives on the boundary
301
+ of the domain.
302
+ However, this requires selecting a reasonable value for τg > 0
303
+ for each of the seven functions and is computationally expensive if it is done
304
+ using cross-validation. However, in the context of this paper, we are constructing
305
+ a stochastic model that mimics the behavior of a densely “observed�� numerical
306
+ simulation model and does not include penalties beyond the restriction of using
307
+ 27 basis functions. We demonstrate the ability of this model to be estimated in
308
+ our context in the simulation study in Section 4, and also investigate the amount
309
+ of data needed to estimate the model.
310
+ 10
311
+
312
+ 40
313
+ 35
314
+ 30
315
+ 25
316
+ 20
317
+ L
318
+ 15
319
+ 10
320
+ 10
321
+ 15
322
+ 0
323
+ 25
324
+ 25
325
+ 35
326
+ AO3.2
327
+ Hierarchical model and discretization
328
+ Consider a bounded domain D ⊂ R3, and observations y = (y1, y2, . . . , yn) made
329
+ at locations s1, s2, . . . , sn ∈ D. We assume a Gaussian observation model
330
+ yi|η(si), σ2
331
+ N ∼ N(η(si), σ2
332
+ N),
333
+ i = 1, . . . , n,
334
+ where σ2
335
+ N > 0 is the nugget variance and
336
+ η(s) = x(s)Tβ + u(s),
337
+ s ∈ D,
338
+ describes true spatial variation as a combination of covariates and a GRF. Here
339
+ x(·) is a spatially varying vector of k covariates, β ∈ Rk are the coefficients of
340
+ the covariates, and u(·) is a GRF with spatially varying anisotropy as presented
341
+ in Section 2.
342
+ As described in Appendix B, the GRF u(·) is discretized using a regular grid
343
+ with l cells, and we get a Gaussian Markov random field w = (w1, . . . , wl)T. Let
344
+ θ be the vector of all parameters controlling u(·), then
345
+ w|θ ∼ Nl(0, Q−1),
346
+ where dependence on θ is suppressed for Q, and Q is a l × l precision matrix
347
+ with a three-dimensional spatial sparsity structure.
348
+ The vector w is linked to
349
+ u(·) through a linear transformation u(s) = a(s)Tw, where a has only one non-
350
+ zero entry corresponding to which grid cell location s belongs. This gives u =
351
+ (u(s1), . . . , u(sn))T = Aw, where the n × l matrix A only has one non-zero entry
352
+ on each row.
353
+ The coefficients of the fixed effect, β, is assigned the weak penalty β ∼
354
+ NK(0, V IK) for a fixed V > 0. Thus we can write y as
355
+ y = Xβ + Aw + ϵ,
356
+ (12)
357
+ 11
358
+
359
+ where X is the design matrix of covariates, and ϵ ∼ Nn(0, Inσ2
360
+ N) is an n-dimensional
361
+ vector of random noise. This gives rise to the hierarchical formulation
362
+ y|β, w, σ2
363
+ N ∼ Nn(Xβ + Aw, σ2
364
+ NIn),
365
+ β ∼ Nk(0, V Ik),
366
+ w|θ ∼ Nl(0, Q−1).
367
+ Let s∗ ∈ D be an unobserved location. After parameters ˆθ and ˆ
368
+ σ2
369
+ N are esti-
370
+ mated, one can predict the underlying value η(s∗) = x(s∗)Tβ + a(s∗)Tw or a new
371
+ observation y∗ = x(s∗)Tβ + a(s∗)Tw + ϵ∗, where ϵ∗ ∼ N(0, ˆ
372
+ σ2
373
+ N) is a new nugget.
374
+ The predictions are made using the conditional distributions η(s∗)|y, θ = ˆθ, σ2
375
+ N =
376
+ ˆ
377
+ σ2
378
+ N and y∗|y, θ = ˆθ, σ2
379
+ N = ˆ
380
+ σ2
381
+ N. The estimation of parameters is detailed in the next
382
+ section.
383
+ 3.3
384
+ Parameter inference
385
+ Simplify notation by letting z = (uT, βT)T. Then
386
+ z|θ ∼ N(0, Q−1
387
+ z ), where Qz =
388
+
389
+ �Q
390
+ 0
391
+ 0
392
+ V Ik
393
+
394
+ � .
395
+ Let S =
396
+
397
+ A
398
+ X
399
+
400
+ , then the observation model can be rewritten as
401
+ y|z, σ2
402
+ N ∼ Nn(Sz, Inσ2
403
+ N).
404
+ (13)
405
+ Using this notation the log-likelihood can be expressed as
406
+ log π(θ, σ2
407
+ N|y) = Const + log π(θ, σ2
408
+ N) + 1
409
+ 2 log det (Qz) − n
410
+ 2 log(σ2
411
+ N)
412
+ − 1
413
+ 2 log det (QC) − 1
414
+ 2µT
415
+ CQCµC −
416
+ 1
417
+ 2σ2
418
+ N
419
+ (y − SµC)T(y − SµC).
420
+ (14)
421
+ Here dependence on θ is suppressed for µC, Qz and QC, and π(θ, σ2
422
+ N) can be used
423
+ to assign a penalty on θ, e.g., like the random-walk penalty used in Fuglstad et al.
424
+ 12
425
+
426
+ (2015b). The conditional precision matrix QC is
427
+ QC = Qz + STS/σ2
428
+ N
429
+ (15)
430
+ and µC is the conditional mean,
431
+ µC = Q−1
432
+ C STy/σ2
433
+ N.
434
+ (16)
435
+ Parameter inference is done by maximizing Equation (14) with respect to θ
436
+ and σ2
437
+ N. The parameter vector θ includes all coefficients for the basis functions,
438
+ and when using 27 basis functions for each function,
439
+ θ =
440
+
441
+ αlog(κ2), αlog γ, αvx, αvy, αvz, αρ1, αρ2
442
+
443
+ ,
444
+ has 189 parameters. The parameter space is challenging to search and we use an
445
+ analytical expression for the gradient in the optimization algorithm. The deriva-
446
+ tion of the analytical gradient involves many nested chain rules and a technique
447
+ to calculate a partial inverse of sparse matrices (Rue and Held, 2010), see Ap-
448
+ pendix A.5 for a complete description.
449
+ 4
450
+ Simulation study
451
+ In this section, we perform a simulation study to investigate the amount of data
452
+ required to acquire reasonable parameter estimates of models with varying com-
453
+ plexity that are specified through the SPDE. A comparison of these estimates is
454
+ made from simulated data generated from three different parametrizations of the
455
+ covariance structures.
456
+ The observation model for the different parametrizations is
457
+ ymod = Awmod + ϵ,
458
+ (17)
459
+ 13
460
+
461
+ where wmod is the GMRF controlled by the parameters θmod in the respective
462
+ models, and ϵ is the independent noise term with mean zero and standard deviation
463
+ σN = 0.1 which is identical for all the parametrizations. Furthermore, the models
464
+ are discretized on the same domain with a grid of size (M, N, P) = (30, 30, 30)
465
+ resulting in a total of 27000 grid nodes where the center of which is our spatial
466
+ locations s ∈ D = [A1, B1] × [A2, B2] × [A3, B3] = [0, 40] × [0, 40] × [0, 40].
467
+ The first and simplest model is a Stationary Isotropic (SI) model which has a
468
+ covariance structure controlled by the three parameters θSI = (log κ2, log γ, log σ2
469
+ N),
470
+ that is assigned to the values κ2 = 0.2, γ = 2.5 and σN = 0.1. The resulting spatial
471
+ range is 10.59 with a marginal variance of 0.023.
472
+ The second is a Stationary Anisotropic (SA) model composed of the 8 parame-
473
+ ters θSA = (log κ2, log γ, vx, vy, vz, ρ1, ρ2, log σ2
474
+ N) set to κ2 = 0.35, γ = 0.5, vx = 1.9,
475
+ vy = 1.4, vz = 0.4, ρ1 = 1.4, ρ2 = 0.6 and σN = 0.1. This results in spatial ranges
476
+ of 10.08 along the x-dimension, 6.75 along y, and 3.88 along z with a marginal
477
+ variance of 0.023.
478
+ The parameters of these first two models are simply assigned some reason-
479
+ able value; however, the third and most complex model with a non-stationary
480
+ anisotropic covariance and a total of 190 parameters, they are much more trou-
481
+ blesome to select. Therefore, functions are chosen to assign the parameter val-
482
+ ues in θNA throughout the domain D such that the dependency directions im-
483
+ itate a vortex.
484
+ Using these functions and evaluating them at the spatial lo-
485
+ cations in the discretization the parameters of the B-splines, described in Sec-
486
+ tion 3.1, are found by optimization. These aforementioned parameters are θNA =
487
+
488
+ αlog(κ2), αlog γ, αvx, αvy, αvz, αρ1, αρ2, log σN
489
+
490
+ with σN = 0.1, and the resulting
491
+ covariance structure can be viewed in Figure 4.
492
+ We will now examine the extent of data required to fit back the parameters
493
+ of the three models described above. First, we simulate multiple datasets from
494
+ 14
495
+
496
+ (a) Correlation
497
+ (b) Marginal Variance
498
+ Figure 4: Spatial correlation at location [26,26,20] (a) and variance of the spatial
499
+ effect (b) in the non-stationary anisotropic model.
500
+ the observation model, Equation (17), with a different number of observed spatial
501
+ locations and realizations (replicated observations of these spatial locations). The
502
+ number of spatial locations varies between 100, 10000, and 27000 (all), and the
503
+ number of realization range between 1, 10, and 100, so nine different combinations
504
+ of dataset sizes. Furthermore, we want to perform 100 different trials for each of
505
+ these combinations, and thereby have 900 total datasets per model. Also, note
506
+ that the observed spatial locations are randomly chosen in each trial. From this,
507
+ some statistics can be recovered about the model estimates that can give insight
508
+ into the applicability of the different parameterizations.
509
+ Table 1 shows the root mean square error (RMSE) between the set parameter
510
+ values in each model and their values inferred by the different datasets. This was
511
+ obtained using the inference method described in Section 3.3 with the observation
512
+ model in Equation (17) for each respective parametrization and trial. The columns
513
+ describe the different number of observation locations (No. loc.) and the number
514
+ 15
515
+
516
+ 40
517
+ 0.8
518
+ 30
519
+ 0.6
520
+ - 40
521
+ 0.4
522
+ - 30
523
+ 20
524
+ X
525
+ 0.2
526
+ 10
527
+ a
528
+ 0
529
+ y0.09
530
+ 40
531
+ 0.08
532
+ 30
533
+ 0.07
534
+ 0.06
535
+ 20-
536
+ 0.05
537
+ 10
538
+ 0.04
539
+ &o -
540
+ 40
541
+ 0.03
542
+ 30-
543
+ 30
544
+ 20
545
+ y
546
+ 20
547
+ x
548
+ 0.02
549
+ 10
550
+ 10
551
+ 0
552
+ 0.01Table 1: The Root Mean Square Error (RMSE) of parameter estimates in the
553
+ stationary isotropic, stationary anisotropic, and non-stationary anisotropic model
554
+ from 100 independent trials for each combination of dataset sizes; the number of
555
+ observed locations (No. loc.) and the number of replicated observations of these
556
+ locations (No. real.).
557
+ No. loc.
558
+ 100
559
+ 10000
560
+ 27000
561
+ No. real.
562
+ 1
563
+ 10
564
+ 100
565
+ 1
566
+ 10
567
+ 100
568
+ 1
569
+ 10
570
+ 100
571
+ Stat. Iso.
572
+ log κ
573
+ 0.763 0.168 0.047 0.123
574
+ log γ
575
+ 0.626 0.164 0.062 0.032
576
+ log τ
577
+ 2.670 0.674 0.182 0.049
578
+ Stationary Anisotropic
579
+ log κ
580
+ 0.876
581
+ 0.195
582
+ 0.081 0.094 0.038
583
+ log γ
584
+ 8.289
585
+ 5.601
586
+ 0.463 0.228 0.079
587
+ |vx|
588
+ 1.208
589
+ 0.785
590
+ 0.440 0.200 0.070
591
+ |vy|
592
+ 1.040
593
+ 0.679
594
+ 0.354 0.152 0.035
595
+ |vz|
596
+ 1.091
597
+ 0.498
598
+ 0.214 0.075 0.027
599
+ |ρ1|
600
+ 0.977
601
+ 0.801
602
+ 0.249 0.129 0.038
603
+ |ρ2|
604
+ 1.337
605
+ 0.489
606
+ 0.275 0.078 0.027
607
+ log τ
608
+ 1.977
609
+ 1.352
610
+ 0.182 0.189 0.028
611
+ Non-Stationary Anisotropic
612
+ log κ
613
+ 2.572 0.811 0.356 0.269
614
+ log γ
615
+ 2.615 1.173 0.694 0.585
616
+ |vx|
617
+ 1.929 0.742 0.531 0.509
618
+ |vy|
619
+ 2.699 0.668 0.453 0.432
620
+ |vz|
621
+ 1.591 0.610 0.343 0.296
622
+ |ρ1|
623
+ 0.144 0.714 0.287 0.210
624
+ |ρ2|
625
+ 0.420 0.604 0.376 0.344
626
+ log τ
627
+ 1.152 0.017 0.005 0.005
628
+ 16
629
+
630
+ of realizations (No. real.), and the different blocks represent the different models.
631
+ The columns highlighted in bold for each respective model are the ones we have
632
+ deemed as reasonable parameter estimates. Also, note that some parts of the table
633
+ are omitted to simplify the presentation of the results for the reader as the full
634
+ table does not affect the conclusion of this study. From Table 1 we observe that the
635
+ (simple) stationary models, SI and SA, require very little data. In fact, observing
636
+ under 1% of the grid for 10 realizations or more is good enough for the SI and the
637
+ SA only requires some more realizations to attain similar parameter accuracy.
638
+ On the other hand, the most flexible parameterization, the NA model, requires
639
+ much more data and only reaches reasonable parameter accuracy when the whole
640
+ grid is observed with 10 or more realizations. Now there is a large discrepancy
641
+ between 10000 observed points ( 37%) and 27000 (100%), so it could be interesting
642
+ to investigate where in this range reasonable estimates are obtained. However, we
643
+ have not chosen to explore this here. We also want to note that these estimates
644
+ will change with the complexity of the covariance structure and with the initial
645
+ values in the optimization.
646
+ 5
647
+ GRF prior for statistical sampling of the ocean
648
+ 5.1
649
+ Aim
650
+ Forecasts produced by numerical ocean models describe realistic behavior for the
651
+ ocean, but local behavior such as plumes created by freshwater discharge from a
652
+ river into the ocean are hard to accurately forecast. However, we can construct
653
+ a prior based on the numerical ocean model that informs prior beliefs about the
654
+ ocean, which can aid AUVs to more effectively sample the ocean. In this paper,
655
+ the goal is to determine the three-dimensional extent of a freshwater plume in the
656
+ ocean, and we assume operation time is short enough to justify a purely spatial
657
+ 17
658
+
659
+ prior that does not assume dynamical changes in time.
660
+ There are two steps in our approach. Step 1 is to estimate a stationary GRF
661
+ prior and a non-stationary GRF prior based on a simulation from the numerical
662
+ ocean model as described in Section 5.2. Step 2 is to combine each of the estimated
663
+ priors with an observation model, and evaluate the predictive ability on in-situ
664
+ observations from AUV as described in Section 5.3. The GRFs that we estimate
665
+ based on the numerical ocean model can be viewed as statistical emulators of the
666
+ ocean.
667
+ 5.2
668
+ The numerical ocean model and the GRF prior
669
+ The model training data used in this application is from a forecast produced by
670
+ the ocean model SINMOD. Data is provided by SINTEF Ocean which developed
671
+ and ran the simulation. SINMOD is a three-dimensional numerical ocean model
672
+ based on primitive equations that are solved using finite difference methods on a
673
+ regular grid with horizontal cell sizes of 20km×20km and is nested in several steps
674
+ down to 32m × 32m. Moreover, it uses z* vertical layers which allow for varying
675
+ grid resolutions depending on the depth and help capture the higher variability of
676
+ the surface. SINMOD is driven by atmospheric forces, freshwater outflows, and
677
+ tides, and it provides numerical simulations of multiple variables such as salinity,
678
+ temperature, and currents.
679
+ The reader is referred to Slagstad and McClimans
680
+ (2005) for a more detailed description of the method.
681
+ The area of operation is located in Trondheimsfjorden at Ladehammaren just
682
+ outside of Trondheim, Norway, and the operation date, the time measurements are
683
+ collected with the AUV, is May 27, 2021, between 10:30 and 14:30. The outlined
684
+ area in Figure 5 indicates the operational area which covers 1408m × 1408m in
685
+ the horizontal plane. At the southeast side of this field, the Nidelva river flows
686
+ into the fjord.
687
+ This causes a very dynamic salinity field that is unfeasible to
688
+ 18
689
+
690
+ Figure 5: The area of operation in Trondheimsfjorden at Ladehammaren just
691
+ outside of Trondheim, Norway. The compass shows the cardinal directions relative
692
+ to the map.
693
+ describe with a stationary covariance model. Therefore, we will use the numerical
694
+ simulations from SINMOD to estimate a non-stationary GRF. As demonstrated
695
+ in the simulation study, complex covariance structures can reliably be estimated
696
+ based on such dense data.
697
+ In this application, we will focus on univariate modeling of the salinity and
698
+ we choose the fine-scale horizontal grid sizes hx = 32 m hy = 32 m, which in total
699
+ gives N = 45 and M = 45 grid nodes for both the numerical and the statistical
700
+ model. Moreover, in the vertical plane, we use 1-meter increments between the
701
+ depth layers, i.e., hz = 1 m.
702
+ To avoid any major effects of the boundaries in
703
+ this direction P = 11 depth layers are used resulting in a depth range of 0.5m
704
+ 19
705
+
706
+ Ostmarkneset
707
+ Munkholmen
708
+ Korsvika
709
+ Ladehammaren
710
+ 6668
711
+ Trondheim
712
+ Traante
713
+ 6668
714
+ Reina
715
+ Brattora
716
+ Trondheim sentralstasjon
717
+ 706
718
+ moen
719
+ 6692
720
+ 706
721
+ Sjobadet
722
+ 706
723
+ 6690
724
+ Skansen
725
+ Kuhauoer
726
+ 6650
727
+ 6666
728
+ 6650
729
+ 6650
730
+ Rosenheto 10.5m. SINMOD outputs zt, t = 0, 1, 2, . . . , 143, which are vectors of salinity
731
+ values in all cells in the three-dimensional grid at different time points throughout
732
+ the whole May 27, 2021. The timesteps are 10 minutes, and Figure 6 shows five
733
+ timesteps from SINMOD for the top six depth layers during the operation. Note
734
+ Figure 6: Five timesteps of the dataset simulated with the numerical ocean model
735
+ SINMOD for May 27, 2021. The timestamps are displayed over their respective
736
+ timesteps. The N-arrow is the cardinal north.
737
+ that the varying vertical layers in the numerical model are either with 0.5m or 1m
738
+ increments, so the SINMOD simulations don’t require any additional modification
739
+ to fit within our statistical model.
740
+ We first estimate the model
741
+ zt = Φzt−1 + ϵt,
742
+ t = 1, . . . , 143,
743
+ where Φ is a diagonal matrix of AR(1) coefficients. The diagonal entries of Φ are
744
+ estimated with maximum likelihood separately for each spatial location such that
745
+ ˆΦii = �143
746
+ t=1 zt,izt−1,i/ �143
747
+ t=1 z2
748
+ t−1,i for i = 1, . . . , NMP, where zt,i is the value in cell
749
+ i at time t. We then compute empirical innovations ˆϵt = zt− ˆΦzt−1, t = 1, . . . , 143.
750
+ 20
751
+
752
+ 10:30
753
+ Depth:
754
+ 0.5
755
+ 30
756
+ N
757
+ 1.5
758
+ 25
759
+ 20
760
+ 2.5
761
+ 15
762
+ 3.5
763
+ 10
764
+ 4.5
765
+ 5
766
+ 5.5
767
+ 014:30
768
+ Depth:
769
+ 0.5
770
+ 30
771
+ N
772
+ 1.5
773
+ 25
774
+ 20
775
+ 2.5
776
+ 15
777
+ 3.5
778
+ 10
779
+ 4.5
780
+ 5
781
+ 5.5
782
+ 013:30
783
+ Depth:
784
+ 0.5
785
+ 30
786
+ N
787
+ 1.5
788
+ 25
789
+ 20
790
+ 2.5
791
+ 15
792
+ 3.5
793
+ 10
794
+ 4.5
795
+ 5
796
+ 5.5
797
+ 012:30
798
+ Depth:
799
+ 0.5
800
+ 30
801
+ N
802
+ 1.5
803
+ 25
804
+ 20
805
+ 2.5
806
+ 15
807
+ 3.5
808
+ 10
809
+ 4.5
810
+ 5
811
+ 5.5
812
+ 011:30
813
+ Depth:
814
+ 0.5
815
+ 30
816
+ N
817
+ 1.5
818
+ 25
819
+ 20
820
+ 2.5
821
+ 15
822
+ 3.5
823
+ 10
824
+ 4.5
825
+ 5
826
+ 5.5
827
+ 0These empirical innovations describe the spatial covariance structure for short-term
828
+ changes in salinity.
829
+ We fit the flexible non-stationary anisotropic model with 190 parameters, ˆθNA =
830
+ (αlog κ, αlog γ, αvx, αvy, αvz, αρ1, αρ2, log σ2
831
+ N), and the stationary anisotropic model
832
+ with 8 parameters, ˆθSA = (log κ2, log γ, vx, vy, vz, ρ1, ρ2, log σ2
833
+ N), to the assumed in-
834
+ dependent realization from a GRF ˆϵ1, . . . .ˆϵ143. Note that there are NMP = 22275
835
+ spatial locations and the 144 empirical innovations cover the whole day of May 27,
836
+ 2021. Figures 7b show the resulting variance of the spatial effect and Figure 7c
837
+ the spatial correlation with location (x, y, z) = (22, 10, 0) of the non-stationary
838
+ anisotropic model. The same figures of the stationary anisotropic model can be
839
+ found in Appendix C, Figure S3.
840
+ (a) SINMOD prior
841
+ (b) Marginal Variance
842
+ (c) Correlation
843
+ Figure 7:
844
+ Prior field (a) found from SINMOD simulations, the variance of the
845
+ spatial effect (b) and spatial correlation of point [22,10,0] (marked) (c) in the
846
+ non-stationary anisotropic model. The N-arrow is the cardinal north.
847
+ 21
848
+
849
+ Depth:
850
+ 0.5
851
+ N
852
+ 0.8
853
+ 1.5
854
+ 2.5
855
+ 0.6
856
+ 3.5
857
+ 0.4
858
+ 4.5
859
+ 0.2
860
+ 5.5
861
+ 0Depth:
862
+ 0.5
863
+ N
864
+ 30
865
+ 1.5
866
+ 25
867
+ 2.5
868
+ 20
869
+ 3.5
870
+ 15
871
+ 4.5
872
+ 10
873
+ 5.5
874
+ 5Depth:
875
+ 0.5
876
+ N
877
+ 0.7
878
+ 1.5
879
+ 0.6
880
+ 0.5
881
+ 2.5
882
+ 0.4
883
+ 3.5
884
+ 0.3
885
+ 4.5
886
+ 0.2
887
+ 5.5
888
+ 0.1In the next step, we construct the expected value of the GRF using the time
889
+ average of the whole day, µ = �143
890
+ t=0 zt/144. The mean is shown in Figure 7a
891
+ and shows the overall tendency for freshwater near the river outlet and saltwater
892
+ further out in the ocean. We choose the prior
893
+ η = µ + e,
894
+ (18)
895
+ where we combine the fixed mean vector, µ, with a new realization, e, of the
896
+ estimated stationary anisotropic model or the non-stationary anisotropic model.
897
+ This is a spatial prior on a 32 m × 32 m × 1 m resolution.
898
+ 5.3
899
+ In-situ data collection and emulator evaluation
900
+ In-situ measurements were made with the AUV on May 27, 2021, between 10:30
901
+ and 14:30. The AUV followed 9 pre-planned paths within the area of operation:
902
+ two intersects at 0.5m depth one northbound and one north-westbound starting
903
+ from the river, two zig-zags in each depth layer (0.5m,2m,5m), and one up and
904
+ down pattern in depth ranging from 0.5m to 10.5m moving north-westbound start-
905
+ ing from the river. Figure 8 displays the locations of the measurements in the top
906
+ 5 layers of the field.
907
+ The AUV is moving at 1.5 m/s and continuously samples the salinity. This
908
+ means that multiple measurements are made within each 32 m × 32 m × 1 m grid
909
+ cell. Measurements are represented as yi, i = 1, . . . , nobs, whereby yi is the average
910
+ value measured in grid cell i. We combine these measurements with the prior in
911
+ Equation (18) using
912
+ yi|η, σ2
913
+ N
914
+ ind
915
+ ∼ N(aT
916
+ i η, σ2
917
+ meas),
918
+ i = 1, . . . , nobs,
919
+ η ∼ N(µ, Q−1
920
+ Prior),
921
+ where ai selects the correct grid cell, Q−1
922
+ Prior is the estimated precision matrix
923
+ for the GMRF, and the Gaussian likelihood with nugget variance σ2
924
+ meas describes
925
+ 22
926
+
927
+ Figure 8: Measurement locations of the AUV in the top 6 depth layers of the spatial
928
+ field on May 27th, 2021, in Trondheimsfjorden at Ladehammaren just outside of
929
+ Trondheim, Norway. The N-arrow is the cardinal north.
930
+ measurement noise and sub-grid variation. In general, we would estimate σ2
931
+ meas
932
+ using a trial run, but in this case, we estimated σ2
933
+ meas using the average empirical
934
+ variance over all observed grid cells in the total dataset. Note that we have not
935
+ accounted for the uncertainty in the AUVs positions in these models. As the AUV
936
+ dive, it loses its GPS signal and only relies on estimated location.
937
+ When the
938
+ GPS signal is returned a linear interpolation is made to account for drift but no
939
+ uncertainty is included.
940
+ We evaluated the two priors, or emulators, by randomly ordering the 9 seg-
941
+ ments and then sequentially including more and more observations for predicting
942
+ the remaining hold-out data. The random permutation of the segments was done
943
+ repeatedly to determine the variation in scores over different paths. This scheme
944
+ 23
945
+
946
+ 0.5m
947
+ Nevaluates the AUVs’ ability to predict future observations while maintaining the
948
+ sequential structure of measurements.
949
+ Figure 9 shows that the non-stationary
950
+ model provides a better prior for the salinity in the ocean than the stationary
951
+ model. The differences are largest when little data is available, which is consistent
952
+ with the idea that the prior is most important in this case. The non-stationary
953
+ model can leverage knowledge about which areas are most uncertain using the
954
+ spatially varying marginal variance and update the prior based on expected simi-
955
+ larities from the spatially varying anisotropy. The improvements are seen both in
956
+ point predictions through RMSE and in predictive distributions as measured by
957
+ CPRS (Gneiting and Raftery, 2007).
958
+ 6
959
+ Discussion
960
+ We extend the class of SPDE-based GRFs introduced in Fuglstad et al. (2015a)
961
+ to three-dimensional space by overcoming two key issues: parametrization and
962
+ computation. For the former, we developed a specification of spatially varying
963
+ anisotropy through a spatially varying baseline isotropic dependence, and two
964
+ orthogonal spatially varying vector fields that describe extra dependence. This
965
+ allows for an interpretable description of the 3×3 positive definite matrix describing
966
+ anisotropy. For the latter, we use a finite volume method to construct a GMRF
967
+ that approximates the solution of the SPDE.
968
+ The specification of spatially varying marginal variance and spatially varying
969
+ anisotropy requires specifying 7 spatially varying real functions. In this paper,
970
+ we expand each function with a clamped B-spline basis. If each function uses
971
+ P 3 basis functions, this gives in total 7P 3 coefficients. As demonstrated in the
972
+ simulation study, an unpenalized estimation of these parameters requires a densely
973
+ observed area and multiple realizations. Application of the new models in data-
974
+ 24
975
+
976
+ Figure 9: The root mean square error (RMSE, top) and the continuous ranked
977
+ probability score (CRPS, bottom) of predictions from the stationary anisotropic
978
+ (orange) and non-stationary anisotropic models (blue) given different proportions
979
+ of observed data (5%, 95%). The error bars are the standard deviations of the
980
+ different measures under random permutations of the 9 segments.
981
+ sparse situations will require penalties that restrict the regularity of the 7 spatially
982
+ varying functions. However, more research is needed to come up with a practical
983
+ 25
984
+
985
+ 2.5
986
+ 2.0
987
+ 1.0
988
+ 0.5
989
+ 1.2
990
+ 1.0
991
+ CRPS
992
+ B'0
993
+ 0.6
994
+ 0.4
995
+ 0.2
996
+ 5%10%15%20%25%30%35%40%45%50%55%60%65%70%75%80%85%90%95%
997
+ StationaryAnisotropic
998
+ Non-stationaryAnisotropicway to determine the appropriate strength of penalization for each of the functions.
999
+ While we did not experience any practical issues with the chosen way to de-
1000
+ scribe the two orthogonal vector fields, the construction has a “gimbal lock” type
1001
+ issue. If one vector field points exactly along the z-axis, there is no unique choice
1002
+ for the second vector field. A potential way to avoid this issue is by describing
1003
+ the orientation of the two orthogonal vector fields through quaternions or Euler-
1004
+ Rodrigues parameters.
1005
+ Moving from two-dimensional space to three-dimensional space introduces an
1006
+ asymptotically higher computation cost as a function of grid size. For a regular
1007
+ three-dimensional grid with N nodes, the computational cost is O(N 2) compared
1008
+ to O(N 3/2) in two-dimensional space. This increased computational cost arises
1009
+ from increased fill-in in the Cholesky factor. However, the application demon-
1010
+ strates that the use of a grid size of N = 22275 is unproblematic even for real-time
1011
+ updates on an AUV.
1012
+ For the predictions of salinity in the Trondheim’s fjord, we see the highest
1013
+ improvement of the complex GRF prior compared to an isotropic GRF, for sparse
1014
+ in-situ measurements. As more data is collected, the difference between the models
1015
+ decreases. This suggests that the key advantage of training the more complex
1016
+ GRF is to encode prior physical knowledge so that we can more effectively update
1017
+ knowledge about unobserved locations. Salinity was used as an example, but in
1018
+ general, the same approach could be used to map other biologically interesting
1019
+ quantities such as phytoplankton (Fossum et al., 2019). The GRFs developed in
1020
+ this paper are a step forward in quantifying beliefs about unobserved regions in
1021
+ the ocean, which is essential for optimal decisions and more effective autonomous
1022
+ sampling (Fossum et al., 2021).
1023
+ In future work, it would be interesting to add a dynamic component to the
1024
+ model to capture physical processes such as diffusion and advection. However,
1025
+ 26
1026
+
1027
+ this substantially increases computational cost, and it is not clear to which degree
1028
+ an advection field from a numerical model should be trusted and which boundary
1029
+ conditions are best in an advection-dominated problem. The new class of GRFs
1030
+ shows great promise for encoding prior knowledge about a phenomenon in a com-
1031
+ putationally efficient way. However, overfitting is an important issue, and we must
1032
+ consider ways to penalize the complexity. In particular, we need to consider ways
1033
+ to allow flexibility in an area where it is needed such as a river outlet, and restrict
1034
+ flexibility in areas where we expect stationarity.
1035
+ Acknowledgments
1036
+ Berild and Fuglstad are supported by the Research Council of Norway, project
1037
+ number 305445. The authors are grateful to Ingrid Ellingsen and SINTEF for
1038
+ providing the simulations from the numerical ocean model SINMOD.
1039
+ 27
1040
+
1041
+ A.
1042
+ General properties
1043
+ A.1
1044
+ Marginal Variance
1045
+ Here, we will derive the expression for the marginal variance in a general sense and
1046
+ then specify it for three-dimensional spaces with exponential covariance functions.
1047
+ The SPDE considered in this work is
1048
+ (κ2 − ∇ · H∇)α/2u(s) = W(s),
1049
+ (S1)
1050
+ where s ∈ D ⊆ Rd a spatial location in the domain of dimension d and α = ν +d/2
1051
+ where ν > 0 is the smoothness. Any solution of this SPDE is a Matérn field and
1052
+ let σm > 0 be its marginal standard deviation; then, its covariance function is
1053
+ r(s1, s2) =
1054
+ σ2
1055
+ m
1056
+ 2ν−1Γ(ν)(κ||H−1/2(s1 − s2)||)νKν(κ||H−1/2(s1 − s2)||).
1057
+ (S2)
1058
+ The transfer function of the SPDE is
1059
+ g(w) = (κ2 + wTHw)−α/2.
1060
+ Using this and by including the spectral density of standard Gaussian white noise
1061
+ in Rd is (2π)−d, the spectral density of the solution of the SPDE is
1062
+ fS(w) = (2π)−d(κ2 + wTHw)−α.
1063
+ Lastly, to find the marginal variance of the field the integral of the spectral density
1064
+ is made over Rd as
1065
+ σ2
1066
+ m =
1067
+
1068
+ Rd fS(w)dw.
1069
+ Including the change of variables w = κH−1/2z the expression becomes
1070
+ σ2
1071
+ m = (2π)−d
1072
+
1073
+ Rd(κ2 + κ2zTz)−α det(κH−1/2)dz
1074
+ = (2π)−d
1075
+
1076
+ Rd κd−2α(1 + zTz)−α det(H)−1/2dz
1077
+ α=ν+d/2
1078
+ =
1079
+ (2π)−dκ−2ν det(H)−1/2
1080
+
1081
+ Rd(1 + zTz)−αdz,
1082
+ (S3)
1083
+ 28
1084
+
1085
+ which by specifying a exponential covariance in R3 with α = 2, ν = 1/2 and d = 3
1086
+ is
1087
+ σ2
1088
+ m =
1089
+ 1
1090
+ 8πκ
1091
+
1092
+ det(H)
1093
+ .
1094
+ Note that the integral in Equation (S3) is solved by converting to polar coordinates
1095
+ as
1096
+
1097
+ R3
1098
+ 1
1099
+ (1 + zTz)2dz =
1100
+ � π
1101
+ 0
1102
+ sin(φ)dφ
1103
+ � 2π
1104
+ 0
1105
+
1106
+ � ∞
1107
+ 0
1108
+ ρ2
1109
+ (1 + ρ2)2dρ = π2.
1110
+ A.2
1111
+ Covariance function
1112
+ Evaluating Equation (S2) at ν = 1/2 and including the expression for the marginal
1113
+ variance the covariance function can be formalized as
1114
+ r(s1, s2) =
1115
+
1116
+ 2
1117
+ π
1118
+ 1
1119
+ 8πκ
1120
+
1121
+ det(H)
1122
+
1123
+ κ||H−1/2(s1 − s2)||K 1
1124
+ 2(κ||H−1/2(s1 − s2)||).
1125
+ Then, consider the modified Bessel function of the second kind
1126
+ Kn(z) =
1127
+ � π
1128
+ 2z
1129
+ e−z
1130
+ (n − 1
1131
+ 2)!
1132
+ � ∞
1133
+ 0
1134
+ e−ttn−1/2
1135
+
1136
+ 1 − t
1137
+ 2z
1138
+ �n−1/2
1139
+ dt,
1140
+ and evaluate this at order 1/2 gives
1141
+ K 1
1142
+ 2(z) =
1143
+ � π
1144
+ 2ze−z.
1145
+ The covariance function can then be formalized as
1146
+ r (s1, s2) =
1147
+
1148
+ 2
1149
+ πσ2
1150
+ m
1151
+
1152
+ κ||H−1/2(s1 − s2)||
1153
+ ×
1154
+
1155
+ π
1156
+ 2 · κ||H−1/2(s1 − s2)|| exp
1157
+
1158
+ −κ||H−1/2(s1 − s2)||
1159
+
1160
+ =σ2
1161
+ m exp
1162
+
1163
+ −κ||H−1/2(s1 − s2)||
1164
+
1165
+ .
1166
+ (S4)
1167
+ A.3
1168
+ One-dimensional clamped B-splines
1169
+ We illustrate the construction of 1-dimensional splines B-splines using the interval
1170
+ [A, B] ∈ R. Let A = t0 < t1 < · · · < tm = B be the knot points. Then the
1171
+ 29
1172
+
1173
+ zero-order B-splines are constructed recursively as
1174
+ Bi,0(t) =
1175
+
1176
+
1177
+
1178
+
1179
+
1180
+ 1,
1181
+ ti ≤ t ≤ ti+1,
1182
+ 0,
1183
+ otherwise,
1184
+ ,
1185
+ t ∈ [A, B],
1186
+ for i = 0, . . . , p − 1.
1187
+ Let r denote the order of the B-splines.
1188
+ The first- and
1189
+ second-order basis splines are constructed as
1190
+ Bi,r(t) =
1191
+ t − ti
1192
+ ti+r − ti
1193
+ Bi,r−1(t) +
1194
+ ti+r+1 − t
1195
+ ti+r+1 − ti+1
1196
+ Bi+1,r−1(t),
1197
+ t ∈ [A, B],
1198
+ for i = 0, . . . , p − r − 1.
1199
+ Using the r-order B-spline basis, we construct a function g : [A, B] → R by
1200
+ g(t) =
1201
+ p−r−1
1202
+
1203
+ i=0
1204
+ αiBi,r(t).
1205
+ where α0, . . . , αp−r−1 ∈ R are coefficients. We use a clamped spline where g′(A) =
1206
+ g′(B) = 0 and need the additional requirement that α0 = α1 and αp−r−2 = αp−r−1.
1207
+ A.4
1208
+ Integrated likelihood
1209
+ The distribution of z = (u, β) is given by
1210
+ z|θ ∼ N(0, Q−1
1211
+ z ),
1212
+ and the observation model is
1213
+ y|z, θ, σ2
1214
+ N ∼ Nn(Sz, Inσ2
1215
+ N).
1216
+ 30
1217
+
1218
+ From this the distribution of z given some observations y is
1219
+ π(z|θ, σ2
1220
+ N, y) ∝ π(z, θ, σ2
1221
+ N, y)
1222
+ = π(θ, σ2
1223
+ N)π(z|θ)π(y|θ, σ2
1224
+ N, z)
1225
+ ∝ exp
1226
+
1227
+ −1
1228
+ 2zTQzz − 1
1229
+ 2(y − Sz)TInσ−2
1230
+ N (y − Sz)
1231
+
1232
+ ∝ exp
1233
+
1234
+ −1
1235
+ 2
1236
+
1237
+ zT �
1238
+ Qz + σ−2
1239
+ N STS
1240
+
1241
+ z − 2zTSTy · σ−2
1242
+ N
1243
+ ��
1244
+ ∝ exp
1245
+
1246
+ −1
1247
+ 2(z − µC)TQC(z − µC)
1248
+
1249
+
1250
+ z|θ, σ2
1251
+ N, y ∼ Nn
1252
+
1253
+ µC, Q−1
1254
+ C
1255
+
1256
+ Here, QC = Qz +STS·σ−2
1257
+ N is the conditional precision matrix and µC = Q−1
1258
+ C STy·
1259
+ σ−2
1260
+ N is the conditional mean.
1261
+ Then, integrating out z from the joint distribution gives
1262
+ π(θ, σ2
1263
+ N, y) = π(θ, z, σ2
1264
+ N, y)
1265
+ π(z|θ, σ2
1266
+ N, y)
1267
+ = π(θ, σ2
1268
+ N)π(z|θ)π(y|θ, σ2
1269
+ N, z)
1270
+ π(z|θ, σ2
1271
+ N, y)
1272
+ ,
1273
+ where the left-hand side does not depend on z such that it may be evaluated for
1274
+ any given value. Let us evaluate it for z = µC such that
1275
+ π(θ, σ2
1276
+ N, y) ∝π(θ, σ2
1277
+ N)π(z = µC|θ)π(y|θ, σ2
1278
+ N, z = µC)
1279
+ π(z = µC|θ, σ2
1280
+ N, y)
1281
+ ∝π(θ)|Qz|1/2|In · σ−2
1282
+ N |1/2
1283
+ |QC|1/2
1284
+ exp
1285
+
1286
+ −1
1287
+ 2µT
1288
+ CQzµC
1289
+
1290
+ × exp
1291
+
1292
+ −1
1293
+ 2(y − SµC)TIn · σ−2
1294
+ N (y − SµC)
1295
+
1296
+ .
1297
+ The last term π(z|θ, σ2
1298
+ N, y) is removed since it is equal to 1. Thereby, conditioning
1299
+ 31
1300
+
1301
+ on y and taking the log we have the log-likelihood
1302
+ log(π(θ, σ2
1303
+ N|y)) =Constant + log(π(θ, σ2
1304
+ N)) + 1
1305
+ 2 log(det(Qz)) + n
1306
+ 2 log(σ−2
1307
+ N )
1308
+ − 1
1309
+ 2 log(det(QC)) − 1
1310
+ 2µT
1311
+ CQzµC −
1312
+ 1
1313
+ 2 · σ2
1314
+ N
1315
+ (y − SµC)T(y − SµC).
1316
+ (S5)
1317
+ A.5
1318
+ Gradient of the log-likelihood
1319
+ This section is similar to the derivation of the gradient presented in the supple-
1320
+ mentary material of Fuglstad et al. (2015b).
1321
+ log(π(θ, τN|y)) =Constant + log(π(θ, τN)) + 1
1322
+ 2 log(det(Qz)) + n
1323
+ 2 log(σ−2
1324
+ N )
1325
+ − 1
1326
+ 2 log(det(QC)) + 1
1327
+ 2µT
1328
+ CQCµC − τN
1329
+ 2 yTy.
1330
+ Note that the last two terms are rewritten for simplicity in the gradient calculation
1331
+ and that the variance of the Gaussian noise term, σ2
1332
+ N is re-parametrized with its
1333
+ inverse τN = 1/σ2
1334
+ N (precision). Derivatives of the log-likelihood are taken with
1335
+ respect to θi, the elements of θ, and the precision on log scale as log(τN).
1336
+ The first term is a constant and therefore its derivative is zero with respect to
1337
+ any of the parameters. The next term, the penalty or the prior of the parameters,
1338
+ is not used in this paper and otherwise depends on the choice of penalty so gradient
1339
+ calculation is not specified for this term.
1340
+ To continue note the derivatives of the precision matrix
1341
+ ∂QC
1342
+ ∂θi
1343
+ = ∂Qz
1344
+ ∂θi
1345
+ and
1346
+ ∂QC
1347
+ ∂ log(τN) = STSτN,
1348
+ which is used in the following derivations. First, the derivatives with respect to θi
1349
+ are considered. The derivative of the log determinant terms are
1350
+
1351
+ ∂θi
1352
+ (log(det(Q)) − log(det(QC))) =Tr
1353
+
1354
+ Q−1∂Q
1355
+ ∂θi
1356
+
1357
+ − Tr
1358
+
1359
+ Q−1
1360
+ C
1361
+ ∂Q
1362
+ ∂θi
1363
+
1364
+ =Tr
1365
+
1366
+ (Q−1 − Q−1
1367
+ C )∂Q
1368
+ ∂θi
1369
+
1370
+ ,
1371
+ 32
1372
+
1373
+ and the derivative of the quadratic terms are
1374
+
1375
+ ∂θi
1376
+ �1
1377
+ 2yTyτN + 1
1378
+ 2µT
1379
+ CQCµC
1380
+
1381
+ = ∂
1382
+ ∂θi
1383
+ �1
1384
+ 2µT
1385
+ CQCµC
1386
+
1387
+ = − 1
1388
+ 2yTτNSQ−1
1389
+ C
1390
+ �∂QC
1391
+ ∂θi
1392
+
1393
+ Q−1
1394
+ C STτNy
1395
+ = − 1
1396
+ 2µT
1397
+ C
1398
+ �∂Q
1399
+ ∂θi
1400
+
1401
+ µC.
1402
+ Then, combining these the derivative of the log-likelihood with respect to θi is
1403
+
1404
+ ∂θi
1405
+ log(π(θ, τN|y)) = ∂
1406
+ ∂θi
1407
+ log(π(θ, τN))+Tr
1408
+
1409
+ (Q−1 − Q−1
1410
+ C )∂Q
1411
+ ∂θi
1412
+
1413
+ −1
1414
+ 2µT
1415
+ C
1416
+ �∂Q
1417
+ ∂θi
1418
+
1419
+ µC
1420
+ Next, the derivative with respect to the log precision, log τN, is considered.
1421
+ The derivative of the log determinant terms are
1422
+
1423
+ ∂ log(τN)
1424
+ �n
1425
+ 2 log(τN) − 1
1426
+ 2 log(det(QC))
1427
+
1428
+ =n
1429
+ 2 − 1
1430
+ 2Tr
1431
+
1432
+ Q−1
1433
+ C
1434
+
1435
+ ∂ log(τN)QC
1436
+
1437
+ =n
1438
+ 2 − 1
1439
+ 2Tr
1440
+
1441
+ Q−1
1442
+ C STS · τN
1443
+
1444
+ Further, the derivative of 1/2yTy · τN with respect to log(τN) is just the same
1445
+ expression so the remaining quadratic term becomes
1446
+ ∂ 1
1447
+ 2µT
1448
+ CQCµC
1449
+ ∂ log(τN)
1450
+ =∂ 1
1451
+ 2yTτNSQ−1
1452
+ C STτNy
1453
+ ∂ log(τN)
1454
+ =yTτNSQ−1
1455
+ C ST
1456
+ ∂τN
1457
+ ∂ log(τN)y − 1
1458
+ 2yTτNSQ−1
1459
+ C
1460
+ ∂QC
1461
+ ∂ log(τN)Q−1
1462
+ C STτNy
1463
+ =µT
1464
+ CSTτNy − 1
1465
+ 2µT
1466
+ CSTSµCτN,
1467
+ and then, by adding the last quadratic term, the expression simplifies to
1468
+ −1/2yTy · τN + µT
1469
+ CSTy · τN − 1
1470
+ 2µT
1471
+ CSTSµC · τN = −1
1472
+ 2(y − SµC)T(y − SµC) · τN.
1473
+ Finally, combining all these terms we have the derivative of the log-likelihood with
1474
+ respect to log(τN):
1475
+ ∂ log(π(θ, τN|y))
1476
+ ∂ log(τN))
1477
+ =∂ log(π(θ, τN)
1478
+ ∂ log(τN)
1479
+ + n
1480
+ 2 − 1
1481
+ 2Tr
1482
+
1483
+ Q−1
1484
+ C STS · τN
1485
+
1486
+ − 1
1487
+ 2(y − SµC)T(y − SµC) · τN
1488
+ 33
1489
+
1490
+ Note that the derivative of QC can be calculated quickly and it is derived from
1491
+ a series of chain rules; first on QC, then on A and AH, and finally within H. The
1492
+ most computationally heavy calculation in the gradient of the log-likelihood is to
1493
+ calculate the inverses in the difference Q−1 − Q−1
1494
+ C . However, since this term is
1495
+ multiplied with the derivative of Q with respect to θi, which carries the non-zero
1496
+ structure of Q, only elements of Q−1 and Q−1
1497
+ C which correspond to the non-zero
1498
+ structure of Q need to be calculated. This is done by calculating a partial inverse
1499
+ of two matrices as described in Rue and Held (2010).
1500
+ B.
1501
+ Derivation
1502
+ B.1
1503
+ Discretization
1504
+ To find the local solution of the SPDE the domain D = [A1, B1]×[A2, B2]×[A3, B3]
1505
+ is divided into equally sized rectangular cubes or cells. We use M cells to divide
1506
+ [A1, B1] in the x-direction, N cells on [A2, B1] in y-direction and P cells on [A3, B3]
1507
+ in z-direction. The cells have sides parallel to each axis of size hx = (B1 − A1)/M,
1508
+ hy = (B2 − A2)/N, and hz = (B3 − A3)/P. The cells are assigned an index with
1509
+ regards to their cell number along each axes starting from number 0; i ∈ [0, M]
1510
+ along x, j ∈ [0, N] along y, and k ∈ [0, P] along z. For a specific cell, its domain
1511
+ can be denoted as
1512
+ Ei,j,k = [ihx, (i + 1)hx] × [jhy, (j + 1)hy] × [khz, (k + 1)hz],
1513
+ and Figure S1 shows this cell and its closest neighbors. Furthermore, as a regular
1514
+ grid is employed the volume of a cell is V = hxhyhz.
1515
+ To further define the local solution of the SPDE we denote the faces of a grid
1516
+ cell as σF
1517
+ i,j,k (front), σB
1518
+ i,j,k (back), σL
1519
+ i,j,k (left), σR
1520
+ i,j,k (right), σU
1521
+ i,j,k (up) and σD
1522
+ i,j,k
1523
+ (down) with their respective face centers si,j−1/2,k, si,j+1/2,k, si−1/2,j,k, si+1/2,j,k,
1524
+ 34
1525
+
1526
+ Figure S1: One cell Ei,j,k in the discretization with its closest neighbours; Ei+1,j,k,
1527
+ Ei−1,j,k, Ei,j+1,k, Ei,j−1,k, Ei,j,k+1, and Ei,j,k−1.
1528
+ si,j,k+1/2 and si,j,k−1/2. Figure S2 describes the different faces of a cell.
1529
+ B.2
1530
+ Local solution of the SPDE
1531
+ Note that this description is an extension to three dimensions of the derivation
1532
+ described in Fuglstad et al. (2015a), and the reader is referred to there for fur-
1533
+ ther details. To locally solve the SPDE a finite volume scheme is derived. First,
1534
+ 35
1535
+
1536
+ I
1537
+ 1Figure S2: One cell Ei,j,k of the discretization with all its faces; σF
1538
+ i,j,k (front),
1539
+ σB
1540
+ i,j,k (back), σL
1541
+ i,j,k (left), σR
1542
+ i,j,k (right), σU
1543
+ i,j,k (up), and σD
1544
+ i,j,k (down) each with its
1545
+ respective face centres.
1546
+ Equation (S1) is integrated over a cell Ei,j,k as
1547
+
1548
+ Eijk
1549
+ κ2(s)ds −
1550
+
1551
+ Eijk
1552
+ ∇ · H(s)∇u(s)ds =
1553
+
1554
+ Eijk
1555
+ W(s)ds,
1556
+ (S6)
1557
+ where ds is a volume element. The integral of the Gaussian white noise on the
1558
+ right-hand side is a Gaussian variable with mean zero and variance equal to the
1559
+ volume of a cell which is independent of neighboring cells. Let zijk be an standard
1560
+ Gaussian variable; then, Equation (S6) becomes
1561
+
1562
+ Eijk
1563
+ κ2(s)ds −
1564
+
1565
+ Eijk
1566
+ ∇ · H(s)∇u(s)ds =
1567
+
1568
+ V zijk.
1569
+ 36
1570
+
1571
+ Then, applying the divergence theorem to the second integral with the divergence
1572
+ operator gives
1573
+
1574
+ Eijk
1575
+ κ2(s)ds −
1576
+
1577
+ ∂Eijk
1578
+ (H(s)∇u(s))Tn(s)dσ =
1579
+
1580
+ V zijk.
1581
+ The first integral is approximated by letting k2
1582
+ ijk be the average value of the con-
1583
+ tinuous function κ2(s) within a cell, i.e. κ2
1584
+ ijk = 1/V
1585
+
1586
+ Eijk κ2(s)ds, resulting in
1587
+ V κ2
1588
+ ijkuijk −
1589
+
1590
+ ∂Eijk
1591
+ (H(s)∇u(s))Tn(s)dσ =
1592
+
1593
+ V zijk.
1594
+ (S7)
1595
+ To describe the solution of the second integral it is divided into integrals over
1596
+ each surface as
1597
+
1598
+ ∂Eijk
1599
+ (H(s)∇u(s))Tn(s)dσ = W L
1600
+ ijk + W R
1601
+ ijk + W B
1602
+ ijk + W F
1603
+ ijk + W U
1604
+ ijk + W D
1605
+ ijk,
1606
+ (S8)
1607
+ or W dir
1608
+ ijk =
1609
+
1610
+ σdir
1611
+ ijk(H(s)∇u(s))Tn(s)dσ, where dir denotes the surface; R (posi-
1612
+ tive x-direction), L (negative x-direction), B (positive y-direction), F (negative
1613
+ y-direction), U (positive z-direction), and D (negative z-direction). Now, an ap-
1614
+ proximation of this surface integral over each face is required. It is assumed that
1615
+ the gradient of u(s) is constant over each face and equal to the value at the center
1616
+ of each face. The resulting scheme for the gradient on each face is described in
1617
+ Table S1.
1618
+ Furthermore, let H be approximated by its value at the center of the
1619
+ face, and then, we have the approximation
1620
+ W dir
1621
+ ijk =
1622
+
1623
+ σdir
1624
+ ijk
1625
+ ∇u(s)TH(s)n(s)dσ
1626
+ ≈∇u(cdir
1627
+ ijk)TH(cdir
1628
+ ijk)n(cdir
1629
+ ijk)
1630
+
1631
+ σdir
1632
+ ijk
1633
+
1634
+ =∇u(cdir
1635
+ ijk)TH(cdir
1636
+ ijk)n(cdir
1637
+ ijk)A(σdir
1638
+ ijk),
1639
+ (S9)
1640
+ where cdir
1641
+ ijk is the center of face dir in the cell Eijk, and A(σdir
1642
+ ijk) is the area of the
1643
+ face. Combining Equation (S9) with the scheme of ∇u(cdir
1644
+ ijk) from Table S1, and
1645
+ 37
1646
+
1647
+ Face
1648
+ Scheme
1649
+ σR
1650
+ i,j,k
1651
+
1652
+ ∂xu(si+1/2,j,k) ≃
1653
+ 1
1654
+ hx (u(si+1,j,k) − u(si,j,k))
1655
+
1656
+ ∂yu(si+1/2,j,k) ≃
1657
+ 1
1658
+ 4hy (u(si+1,j+1,k) + u(si,j+1,k) − u(si+1,j−1,k) − u(si,j−1,k))
1659
+
1660
+ ∂zu(si+1/2,j,k) ≃
1661
+ 1
1662
+ 4hz (u(si+1,j,k+1) + u(si,j,k+1) − u(si+1,j,k−1) − u(si,j,k−1))
1663
+ σL
1664
+ i,j,k
1665
+
1666
+ ∂xu(si−1/2,j,k) ≃
1667
+ 1
1668
+ hx (u(si,j,k) − u(si−1,j,k))
1669
+
1670
+ ∂yu(si−1/2,j,k) ≃
1671
+ 1
1672
+ 4hy (u(si,j+1,k) + u(si−1,j+1,k) − u(si,j−1,k) − u(si−1,j−1,k))
1673
+
1674
+ ∂zu(si−1/2,j,k) ≃
1675
+ 1
1676
+ 4hz (u(si,j,k+1) + u(si−1,j,k+1) − u(si,j,k−1) − u(si−1,j,k−1))
1677
+ σB
1678
+ i,j,k
1679
+
1680
+ ∂xu(si,j+1/2,k) ≃
1681
+ 1
1682
+ 4hx (u(si+1,j+1,k) + u(si+1,j,k) − u(si−1,j+1,k) − u(si−1,j,k))
1683
+
1684
+ ∂yu(si,j+1/2,k) ≃
1685
+ 1
1686
+ hy (u(si,j+1,k) − u(si,j,k))
1687
+
1688
+ ∂zu(si,j+1/2,k) ≃
1689
+ 1
1690
+ 4hz (u(si,j+1,k+1) + u(si,j,k+1) − u(si,j+1,k−1) − u(si,j,k−1))
1691
+ σF
1692
+ i,j,k
1693
+
1694
+ ∂xu(si,j−1/2,k) ≃
1695
+ 1
1696
+ 4hx (u(si+1,j,k) + u(si+1,j−1,k) − u(si−1,j,k) − u(si−1,j−1,k))
1697
+
1698
+ ∂yu(si,j−1/2,k) ≃
1699
+ 1
1700
+ hy (u(si,j,k) − u(si,j−1,k))
1701
+
1702
+ ∂zu(si,j−1/2,k) ≃
1703
+ 1
1704
+ 4hz (u(si,j,k+1) + u(si,j−1,k+1) − u(si,j,k−1) − u(si,j−1,k−1))
1705
+ σU
1706
+ i,j,k
1707
+
1708
+ ∂xu(si,j,k+1/2) ≃
1709
+ 1
1710
+ 4hz (u(si+1,j,k+1) + u(si+1,j,k) − u(si−1,j,k+1) − u(si−1,j,k))
1711
+
1712
+ ∂yu(si,j,k+1/2) ≃
1713
+ 1
1714
+ 4hy (u(si,j+1,k+1) + u(si,j+1,k) − u(si,j−1,k+1) − u(si,j−1,k))
1715
+
1716
+ ∂zu(si,j,k+1/2) ≃
1717
+ 1
1718
+ hx (u(si,j,k+1) − u(si,j,k))
1719
+ σD
1720
+ i,j,k
1721
+
1722
+ ∂xu(si,j,k−1/2) ≃
1723
+ 1
1724
+ 4hz (u(si+1,j,k) + u(si+1,j,k−1) − u(si−1,j,k) − u(si−1,j,k−1))
1725
+
1726
+ ∂yu(si,j,k−1/2) ≃
1727
+ 1
1728
+ 4hy (u(si,j+1,k) + u(si,j+1,k−1) − u(si,j−1,k) − u(si,j−1,k−1))
1729
+
1730
+ ∂zu(si,j,k−1/2) ≃
1731
+ 1
1732
+ hx (u(si,j,k) − u(si,j,k−1))
1733
+ Table S1: Numerical scheme of the partial derivative with respect to x, y and z of
1734
+ uijk on the different faces of cell Eijk.
1735
+ denoting the components of H as
1736
+ H(s) =
1737
+
1738
+ ����
1739
+ H11(s)
1740
+ H12(s)
1741
+ H13(s)
1742
+ H21(s)
1743
+ H22(s)
1744
+ H23(s)
1745
+ H31(s)
1746
+ H32(s)
1747
+ H33(s)
1748
+
1749
+ ����
1750
+ 38
1751
+
1752
+ the approximations for each face become
1753
+ ˆW R
1754
+ i,j,k =
1755
+ hyhz
1756
+
1757
+ H11(si+1/2,j,k)u(si+1,j,k) − u(si,j,k)
1758
+ hx
1759
+
1760
+ +
1761
+ hyhz
1762
+
1763
+ H21(si+1/2,j,k)u(si+1,j+1,k) + u(si,j+1,k) − u(si+1,j−1,k) − u(si,j−1,k)
1764
+ 4hy
1765
+
1766
+ +
1767
+ hyhz
1768
+
1769
+ H31(si+1/2,j,k)u(si+1,j,k+1) + u(si,j,k+1) − u(si+1,j,k−1) − u(si,j,k−1)
1770
+ 4hz
1771
+
1772
+ ,
1773
+ ˆW L
1774
+ i,j,k =
1775
+ hyhz
1776
+
1777
+ H11(si−1/2,j,k)u(si−1,j,k) − u(si,j,k)
1778
+ hx
1779
+
1780
+ +
1781
+ hyhz
1782
+
1783
+ H21(si−1/2,j,k)u(si,j−1,k) + u(si−1,j−1,k) − u(si,j+1,k) − u(si−1,j+1,k)
1784
+ 4hy
1785
+
1786
+ +
1787
+ hyhz
1788
+ ���
1789
+ H31(si−1/2,j,k)u(si,j,k−1) + u(si−1,j,k−1) − u(si,j,k+1) − u(si−1,j,k+1)
1790
+ 4hz
1791
+
1792
+ ,
1793
+ ˆW B
1794
+ i,j,k =
1795
+ hxhz
1796
+
1797
+ H12(si,j+1/2,k)u(si+1,j+1,k) + u(si+1,j,k) − u(si−1,j+1,k) − u(si−1,j,k)
1798
+ 4hx
1799
+
1800
+ +
1801
+ hxhz
1802
+
1803
+ H22(si,j+1/2,k)u(si,j+1,k) − u(si,j,k)
1804
+ hy
1805
+
1806
+ +
1807
+ hxhz
1808
+
1809
+ H32(si,j+1/2,k)u(si,j+1,k+1) + u(si,j,k+1) − u(si,j+1,k−1) − u(si,j,k−1)
1810
+ 4hz
1811
+
1812
+ ,
1813
+ ˆW F
1814
+ i,j,k =
1815
+ hxhz
1816
+
1817
+ H12(si,j−1/2,k)u(si−1,j,k) + u(si−1,j−1,k) − u(si+1,j,k) − u(si+1,j−1,k)
1818
+ 4hx
1819
+
1820
+ +
1821
+ hxhz
1822
+
1823
+ H22(si,j−1/2,k)u(si,j−1,k) − u(si,j,k)
1824
+ hy
1825
+
1826
+ +
1827
+ hxhz
1828
+
1829
+ H32(si,j−1/2,k)u(si,j,k−1) + u(si,j−1,k−1) − u(si,j,k+1) − u(si,j−1,k+1)
1830
+ 4hz
1831
+
1832
+ ,
1833
+ 39
1834
+
1835
+ ˆW U
1836
+ i,j,k =
1837
+ hxhy
1838
+
1839
+ H13(si,j,k+1/2)u(si+1,j,k+1) + u(si+1,j,k) − u(si−1,j,k+1) − u(si−1,j,k)
1840
+ 4hx
1841
+
1842
+ +
1843
+ hxhy
1844
+
1845
+ H23(si,j,k+1/2)u(si,j+1,k+1) + u(si,j+1,k) − u(si,j−1,k+1) − u(si,j−1,k)
1846
+ 4hy
1847
+
1848
+ +
1849
+ hxhy
1850
+
1851
+ H33(si,j,k+1/2)u(si,j,k+1) − u(si,j,k)
1852
+ hz
1853
+
1854
+ ,
1855
+ ˆW D
1856
+ i,j,k =
1857
+ hxhy
1858
+
1859
+ H13(si,j,k−1/2)u(si−1,j,k) + u(si−1,j,k−1) − u(si+1,j,k) − u(si+1,j,k−1)
1860
+ 4hx
1861
+
1862
+ +
1863
+ hxhy
1864
+
1865
+ H23(si,j,k−1/2)u(si,j−1,k) + u(si,j−1,k−1) − u(si,j+1,k) − u(si,j+1,k−1)
1866
+ 4hy
1867
+
1868
+ +
1869
+ hxhy
1870
+
1871
+ H33(si,j,k−1/2)u(si,j,k−1) − u(si,j,k)
1872
+ hz
1873
+
1874
+ ,
1875
+ ˆW T
1876
+ i,j,k =
1877
+ hxhy
1878
+
1879
+ H13(si,j,k+1/2)u(si+1,j,k+1) + u(si+1,j,k) − u(si−1,j,k+1) − u(si−1,j,k)
1880
+ 4hx
1881
+
1882
+ +
1883
+ hxhy
1884
+
1885
+ H23(si,j,k+1/2)u(si,j+1,k+1) + u(si,j+1,k) − u(si,j−1,k+1) − u(si,j−1,k)
1886
+ 4hy
1887
+
1888
+ +
1889
+ hxhy
1890
+
1891
+ H33(si,j,k+1/2)u(si,j,k+1) − u(si,j,k)
1892
+ hz
1893
+
1894
+ ,
1895
+ ˆW B
1896
+ i,j,k =
1897
+ hxhy
1898
+
1899
+ H13(si,j,k−1/2)u(si−1,j,k) + u(si−1,j,k−1) − u(si+1,j,k) − u(si+1,j,k−1)
1900
+ 4hx
1901
+
1902
+ +
1903
+ hxhy
1904
+
1905
+ H23(si,j,k−1/2)u(si,j−1,k) + u(si,j−1,k−1) − u(si,j+1,k) − u(si,j+1,k−1)
1906
+ 4hy
1907
+
1908
+ +
1909
+ hxhy
1910
+
1911
+ H33(si,j,k−1/2)u(si,j,k−1) − u(si,j,k)
1912
+ hz
1913
+
1914
+ .
1915
+ Next, a vectorization of the discretization is made; first moving along the z-
1916
+ direction, then along x-direction, and lastly along the y-direction. Let us denote
1917
+ this with the common index l = j · M · P + i · P + k so sijk = sj·M·P+i·P+k = sl
1918
+ which gives u(sijk) = ul and κ2(sijk) = κ2
1919
+ l , and let the last index be L = (N −
1920
+ 40
1921
+
1922
+ 1)MP + (M − 1)P + P − 1. Further, the vectorization results in the linear system
1923
+ of equations
1924
+ (DV Dκ2 − AH)u = D1/2
1925
+ V z,
1926
+ (S10)
1927
+ where DV = V · IMNP, Dκ2 = [κ2
1928
+ 0, . . . , κ2
1929
+ l , . . . , κ2
1930
+ L] IMNP, and z ∼ N(0, IMNP).
1931
+ For simplicity the indices of the neighbors are denoted kp = k + 1, kn = k − 1,
1932
+ jp = j + 1, jn = j − 1, ip = i + 1, and in = i − 1. The development of AH is done
1933
+ by the sum ˆW L
1934
+ ijk + ˆW R
1935
+ ijk + ˆW B
1936
+ ijk + ˆW F
1937
+ ijk + ˆW U
1938
+ ijk + ˆW D
1939
+ ijk and accounting for the index
1940
+ in uijk to form the linear relationship. In the following, non-zero elements of the
1941
+ (jMN + iP + k)-th row of AH are formalized, and the index in (AH)_ denotes
1942
+ the column being assigned. The resulting coefficient with the point itself is
1943
+ (AH)j·M·P+i·P+k = − hyhz
1944
+ hx
1945
+
1946
+ H11(si+1/2,j,k) + H11(si−1/2,j,k)
1947
+
1948
+ − hxhz
1949
+ hy
1950
+
1951
+ H22(si,j+1/2,k) + H22(si,j−1/2,k)
1952
+
1953
+ − hxhy
1954
+ hz
1955
+
1956
+ H33(si,j,k+1/2) + H22(si,j,k−1/2)
1957
+
1958
+ ,
1959
+ with the six closest neighbors are
1960
+ (AH)j·M·P+i·P+kp =hxhy
1961
+ hz
1962
+ H33(si,j,k+1/2)
1963
+ + hy
1964
+ 4
1965
+
1966
+ H31(si+1/2,j,k) − H31(si−1/2,j,k)
1967
+
1968
+ + hx
1969
+ 4
1970
+
1971
+ H32(si,j+1/2,k) − H32(si,j−1/2,k)
1972
+
1973
+ (AH)j·M·P+i·P+kn =hxhy
1974
+ hz
1975
+ H33(si,j,k−1/2)
1976
+ − hy
1977
+ 4
1978
+
1979
+ H31(si+1/2,j,k) − H31(si−1/2,j,k)
1980
+
1981
+ − hx
1982
+ 4
1983
+
1984
+ H32(si,j+1/2,k) − H32(si,j−1/2,k)
1985
+
1986
+ (AH)j·M·P+ip·P+k =hzhy
1987
+ hx
1988
+ H11(si+1/2,j,k)
1989
+ + hy
1990
+ 4
1991
+
1992
+ H12(si,j,k+1/2) − H12(si,j,k−1/2)
1993
+
1994
+ + hz
1995
+ 4
1996
+
1997
+ H13(si,j+1/2,k) − H13(si,j−1/2,k)
1998
+
1999
+ 41
2000
+
2001
+ (AH)j·M·P+in·P+k =hzhy
2002
+ hx
2003
+ H11(si−1/2,j,k)
2004
+ − hy
2005
+ 4
2006
+
2007
+ H12(si,j,k+1/2) − H12(si,j,k−1/2)
2008
+
2009
+ − hz
2010
+ 4
2011
+
2012
+ H13(si,j+1/2,k) − H13(si,j−1/2,k)
2013
+
2014
+ (AH)jp·M·P+i·P+k =hxhz
2015
+ hy
2016
+ H22(si,j+1/2,k)
2017
+ + hx
2018
+ 4
2019
+
2020
+ H23(si,j,k+1/2) − H23(si,j,k−1/2)
2021
+
2022
+ + hz
2023
+ 4
2024
+
2025
+ H21(si+1/2,j,k) − H21(si−1/2,j,k)
2026
+
2027
+ (AH)jn·M·P+i·P+k =hxhz
2028
+ hy
2029
+ H22(si,j−1/2,k)
2030
+ − hx
2031
+ 4
2032
+
2033
+ H23(si,j,k+1/2) − H23(si,j,k−1/2)
2034
+
2035
+ − hz
2036
+ 4
2037
+
2038
+ H21(si+1/2,j,k) − H21(si−1/2,j,k)
2039
+
2040
+ ,
2041
+ and with the twelve closest diagonals are
2042
+ (AH)j·M·P+ip·P+kp = hy
2043
+ 4
2044
+
2045
+ H31(si+1/2,j,k) + H13(si,j,k+1/2)
2046
+
2047
+ ,
2048
+ (AH)j·M·P+in·P+kn = hy
2049
+ 4
2050
+
2051
+ H31(si−1/2,j,k) + H13(si,j,k−1/2)
2052
+
2053
+ ,
2054
+ (AH)j·M·P+in·P+kp = −hy
2055
+ 4
2056
+
2057
+ H31(si−1/2,j,k) + H13(si,j,k+1/2)
2058
+
2059
+ (AH)j·M·P+ip·P+kn = −hy
2060
+ 4
2061
+
2062
+ H31(si+1/2,j,k) + H13(si,j,k−1/2)
2063
+
2064
+ ,
2065
+ (AH)jp·M·P+i·P+kp = hx
2066
+ 4
2067
+
2068
+ H32(si,j+1/2,k) + H23(si,j,k+1/2)
2069
+
2070
+ ,
2071
+ (AH)jn·M·P+i·P+kn = hx
2072
+ 4
2073
+
2074
+ H32(si,j−1/2,k) + H23(si,j,k−1/2)
2075
+
2076
+ ,
2077
+ (AH)jn·M·P+i·P+kp = −hx
2078
+ 4
2079
+
2080
+ H32(si,j−1/2,k) + H23(si,j,k+1/2)
2081
+
2082
+ ,
2083
+ (AH)jp·M·P+i·P+kn = −hx
2084
+ 4
2085
+
2086
+ H32(si,j+1/2,k) + H23(si,j,k−1/2)
2087
+
2088
+ ,
2089
+ (AH)jp·M·P+ip·P+k = hz
2090
+ 4
2091
+
2092
+ H21(si+1/2,j,k) + H12(si,j+1/2,k)
2093
+
2094
+ ,
2095
+ 42
2096
+
2097
+ (AH)jn·M·P+in·P+k = hz
2098
+ 4
2099
+
2100
+ H21(si−1/2,j,k) + H12(si,j−1/2,k)
2101
+
2102
+ ,
2103
+ (AH)jn·M·P+ip·P+k = −hz
2104
+ 4
2105
+
2106
+ H21(si+1/2,j,k) + H12(si,j−1/2,k)
2107
+
2108
+ ,
2109
+ (AH)jp·M·P+in·P+k = −hz
2110
+ 4
2111
+
2112
+ H21(si−1/2,j,k) + H12(si,j+1/2,k)
2113
+
2114
+ .
2115
+ Note that the corner points are not included in this scheme.
2116
+ Denoting A =
2117
+ DV Dκ2 − AH, Equation (S10) can be written as
2118
+ z = D−1/2
2119
+ V
2120
+ Au,
2121
+ and thus, the joint distribution of u is
2122
+ π(u) ∝ π(z) ∝ exp
2123
+
2124
+ −1
2125
+ 2zTz
2126
+
2127
+ π(u) ∝ exp
2128
+
2129
+ −1
2130
+ 2uTATD−1
2131
+ V Au
2132
+
2133
+ π(u) ∝ exp
2134
+
2135
+ −1
2136
+ 2uTQu
2137
+
2138
+ .
2139
+ Here, Q = ATD−1
2140
+ V A which is a sparse matrix of 93 non-zero elements per row. This
2141
+ corresponds to the point, the 18 closest neighbors, and their 18 closest neighbors.
2142
+ Then removing duplicates results in 93 points.
2143
+ C.
2144
+ Additional figures
2145
+ In the application, Section 5, we estimate the parameters of a non-stationary
2146
+ anisotropic and stationary anisotropic model on a simulated dataset from the nu-
2147
+ merical ocean model SINMOD. The resulting properties of the non-stationary
2148
+ model are presented in Figure 7 in Section 5.2 since this is the main focus of the
2149
+ applications. The properties of the stationary anisotropic model fit on the same
2150
+ dataset are presented in Figure S3. The marginal variance in Figure S3b, which
2151
+ should be constant for this stationary model, shows some variability caused by the
2152
+ 43
2153
+
2154
+ boundary conditions. Notice that this boundary effect is also bigger in the direc-
2155
+ tion of the strongest dependency directions seen in the south and north corners.
2156
+ Notice also the large discrepancies between the correlations in these two models,
2157
+ Figure S3c and Figure 7c, as the stationary anisotropic model kind of captures an
2158
+ average correlation within the field.
2159
+ (a) SINMOD prior
2160
+ (b) Marginal Variance
2161
+ (c) Correlation
2162
+ Figure S3:
2163
+ Prior field (a) found from SINMOD simulations, the variance of the
2164
+ spatial effect (b) and spatial correlation of point [22,10,0] (c) in the stationary
2165
+ anisotropic model. The N-arrow shows the cardinal north.
2166
+ 44
2167
+
2168
+ Depth:
2169
+ 0.5
2170
+ N
2171
+ 0.7
2172
+ 1.5
2173
+ 0.6
2174
+ 2.5
2175
+ 0.5
2176
+ 3.5
2177
+ 0.4
2178
+ 4.5
2179
+ 0.3
2180
+ 5.5
2181
+ 0.2Depth:
2182
+ 0.5
2183
+ N
2184
+ 0.8
2185
+ 1.5
2186
+ 2.5
2187
+ 0.6
2188
+ 3.5
2189
+ 0.4
2190
+ 4.5
2191
+ 0.2
2192
+ 5.5
2193
+ 0Depth:
2194
+ 0.5
2195
+ N
2196
+ 30
2197
+ 1.5
2198
+ 25
2199
+ 2.5
2200
+ 20
2201
+ 3.5
2202
+ 15
2203
+ 4.5
2204
+ 10
2205
+ 5.5
2206
+ 5References
2207
+ Castruccio, S., Hu, Z., Sanderson, B., Karspeck, A., and Hammerling, D. (2019).
2208
+ Reproducing internal variability with few ensemble runs. Journal of Climate,
2209
+ 32(24):8511–8522.
2210
+ Cressie, N. and Wikle, C. K. (2015). Statistics for spatio-temporal data. John
2211
+ Wiley & Sons.
2212
+ Diggle, P. J., Tawn, J. A., and Moyeed, R. A. (1998). Model-based geostatistics.
2213
+ Journal of the Royal Statistical Society: Series C (Applied Statistics), 47(3):299–
2214
+ 350.
2215
+ Euler, L. (1771). Problema algebraicum ob affectiones prorsus singulares memo-
2216
+ rabile. Novi Commentarii academiae scientiarum Petropolitanae, pages 75–106.
2217
+ Foss, K. H., Berget, G. E., and Eidsvik, J. (2021). Using an autonomous underwa-
2218
+ ter vehicle with onboard stochastic advection-diffusion models to map excursion
2219
+ sets of environmental variables. Environmetrics, page e2702. In press.
2220
+ Fossum, T. O., Fragoso, G. M., Davies, E. J., Ullgren, J. E., Mendes, R., Johnsen,
2221
+ G., Ellingsen, I., Eidsvik, J., Ludvigsen, M., and Rajan, K. (2019). Toward adap-
2222
+ tive robotic sampling of phytoplankton in the coastal ocean. Science Robotics,
2223
+ 4(27):eaav3041.
2224
+ Fossum, T. O., Travelletti, C., Eidsvik, J., Ginsbourger, D., and Rajan, K. (2021).
2225
+ Learning excursion sets of vector-valued Gaussian random fields for autonomous
2226
+ ocean sampling. The Annals of Applied Statistics, 15(2):597 – 618.
2227
+ Fuglstad, G.-A. and Castruccio, S. (2020). Compression of climate simulations
2228
+ with a nonstationary global spatiotemporal spde model. The Annals of Applied
2229
+ Statistics, 14(2):542–559.
2230
+ 45
2231
+
2232
+ Fuglstad, G.-A., Lindgren, F., Simpson, D., and Rue, H. (2015a). Exploring a
2233
+ new class of non-stationary spatial gaussian random fields with varying local
2234
+ anisotropy. Statistica Sinica, pages 115–133.
2235
+ Fuglstad, G.-A., Simpson, D., Lindgren, F., and Rue, H. (2015b).
2236
+ Does non-
2237
+ stationary spatial data always require non-stationary random fields?
2238
+ Spatial
2239
+ Statistics, 14:505–531.
2240
+ Gneiting, T. and Raftery, A. E. (2007). Strictly Proper Scoring Rules, Prediction,
2241
+ and Estimation. Journal of the American Statistical Association, 102(477).
2242
+ Heaton, M. J., Datta, A., Finley, A. O., Furrer, R., Guinness, J., Guhaniyogi,
2243
+ R., Gerber, F., Gramacy, R. B., Hammerling, D., Katzfuss, M., Lindgren, F.,
2244
+ Nychka, D. W., Sun, F., and Zammit-Mangion, A. (2019). A case study compe-
2245
+ tition among methods for analyzing large spatial data. Journal of Agricultural,
2246
+ Biological and Environmental Statistics, 24:398–425.
2247
+ Hildeman, A., Bolin, D., and Rychlik, I. (2021). Deformed spde models with an
2248
+ application to spatial modeling of significant wave height. Spatial Statistics,
2249
+ 42:100449.
2250
+ Hu, W., Fuglstad, G.-A., and Castruccio, S. (2021). A stochastic locally diffusive
2251
+ model with neural network-based deformations for global sea surface tempera-
2252
+ ture. Stat. In press.
2253
+ Ingebrigtsen, R., Lindgren, F., and Steinsland, I. (2014).
2254
+ Spatial models with
2255
+ explanatory variables in the dependence structure. Spatial Statistics, 8:20–38.
2256
+ Ingebrigtsen, R., Lindgren, F., Steinsland, I., and Martino, S. (2015). Estimation
2257
+ of a non-stationary model for annual precipitation in southern norway using
2258
+ replicates of the spatial field. Spatial Statistics, 14:338–364.
2259
+ 46
2260
+
2261
+ Lee, D. and Gammie, C. F. (2021). Disks as inhomogeneous, anisotropic gaussian
2262
+ random fields. The Astrophysical Journal, 906(1):39.
2263
+ Lindgren, F., Rue, H., and Lindström, J. (2011). An explicit link between Gaussian
2264
+ fields and Gaussian Markov random fields: the stochastic partial differential
2265
+ equation approach. Journal of the Royal Statistical Society: Series B (Statistical
2266
+ Methodology), 73(4):423–498.
2267
+ Neto, J. H. V., Schmidt, A. M., and Guttorp, P. (2014). Accounting for spatially
2268
+ varying directional effects in spatial covariance structures. Journal of the Royal
2269
+ Statistical Society: Series C: Applied Statistics, pages 103–122.
2270
+ Paciorek, C. J. and Schervish, M. J. (2006). Spatial modelling using a new class
2271
+ of nonstationary covariance functions. Environmetrics: The official journal of
2272
+ the International Environmetrics Society, 17(5):483–506.
2273
+ Risser, M. D. and Calder, C. A. (2015). Regression-based covariance functions for
2274
+ nonstationary spatial modeling. Environmetrics, 26(4):284–297.
2275
+ Rodrigues (1840). Des lois géométriques qui régissent les déplacements d’un sys-
2276
+ tème solide dans l’espace, et de la variation des coordonnées provenant de ces
2277
+ déplacements considérés indépendamment des causes qui peuvent les produire.
2278
+ Journal de Mathématiques Pures et Appliquées, pages 380–440.
2279
+ Rue, H. and Held, L. (2010). Markov Random Fields. Chapman & Hall/CRC
2280
+ Handbooks of Modern Statistical Methods, pages 171–200.
2281
+ Salvaña, M. L. O. and Genton, M. G. (2021). Lagrangian spatio-temporal non-
2282
+ stationary covariance functions. In Advances in Contemporary Statistics and
2283
+ Econometrics, pages 427–447. Springer.
2284
+ 47
2285
+
2286
+ Sampson, P. D. (2010).
2287
+ Constructions for nonstationary spatial processes.
2288
+ In
2289
+ Gelfand, A. E., Diggle, P. J., Fuentes, M., and Guttorp, P., editors, Handbook
2290
+ of Spatial Statistics, chapter 9, pages 119–130. CRC Press, Boca Rotan, FL.
2291
+ Sampson, P. D. and Guttorp, P. (1992). Nonparametric estimation of nonstation-
2292
+ ary spatial covariance structure. Journal of the American Statistical Association,
2293
+ 87(417):108–119.
2294
+ Schmidt, A. M., Guttorp, P., and O’Hagan, A. (2011). Considering covariates in
2295
+ the covariance structure of spatial processes. Environmetrics, 22(4):487–500.
2296
+ Sidén, P., Lindgren, F., Bolin, D., Eklund, A., and Villani, M. (2021). Spatial 3D
2297
+ Matérn priors for fast whole-brain fMRI analysis. Bayesian Analysis. In press.
2298
+ Slagstad, D. and McClimans, T. A. (2005). Modeling the ecosystem dynamics
2299
+ of the Barents sea including the marginal ice zone: I. Physical and chemical
2300
+ oceanography. Journal of Marine Systems, 58(1):1–18.
2301
+ Stein, M. L. (2002). The screening effect in kriging. The Annals of Statistics,
2302
+ 30(1):298–323.
2303
+ Stein, M. L. (2012). Interpolation of spatial data: some theory for kriging. Springer
2304
+ Science & Business Media.
2305
+ 48
2306
+
f9AzT4oBgHgl3EQfavwP/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff
 
gdE2T4oBgHgl3EQfyQgz/content/tmp_files/2301.04118v1.pdf.txt ADDED
The diff for this file is too large to render. See raw diff
 
gdE2T4oBgHgl3EQfyQgz/content/tmp_files/load_file.txt ADDED
The diff for this file is too large to render. See raw diff