Datasets:

pratyushmaini

/

llm_dataset_inference

like 1

*]{}, [**586**]{}, 356, 2003. Zimmermann, H.-U., and B. Aschenbach, [*A&A*]{}, in press (astro-ph/0304322), 2003. --- abstract: 'Modern robotic systems have become a substitute for humans when it’s necessary to perform risky or exhausting tasks. In such application scenarios, communications between robots and the control center are one of the major problems. The commonly used solution assumes that newer messages are more valuable. We find that it does not hold in many scenarios. In this paper, we propose a novel, resilient buffer management policy called OptSample. We make a new assumption that uniformly sampled messages are the most valuable and define an evaluation function to estimate the profit of the received message sequence. Our OptSample policy can uniformly sample messages and dynamically adjust the sample rate based on the run-time network situation. Our analysis and simulation shows that the OptSample policy can effectively prevent losing long segments of continuous messages and can improve the profit of the received messages. We implement the OptSample policy in ROS, without changing the interface or API for the applications. Our experiments show that the OptSample policy can improve the results of several application scenarios including surveillance, 3D reconstruction, and SLAM.' author: - | Yu-Ping Wang$^{1}$, Zi-Xin Zou$^{1}$, Xu-Qiang Hu$^{1}$, Dinesh Manocha$^{2}$, Lei Qiao$^{3}$, Shi-Min Hu$^{1}$\ \ The video of applications can be found at\ <https://www.youtube.com/watch?v=ji-yvbT1l-w>[^1][^2][^3][^4] bibliography: -'main.bib' title: '**OptSample: A Resilient Buffer Management Policy for Robotic Systems Based on Optimal Message Sampling** ' --- Introduction {#sec:introduction} ============ Modern robotic systems have become a substitute for humans when it’s necessary to perform risky or exhausting tasks such as military operations, exploration, rescue operations, surveillance, or large-scale cleaning operations. In such application scenarios, a control center through which humans can monitor and operate the whole system is usually needed. However, communication issues, such as unstable network connection and limited bandwidth, cause problems. Network connections, especially wireless ones (e.g. over WiFi), are not always stable 

* ] { }, [ * * 586 * * ] { }, 356, 2003. Zimmermann, H.-U., and B. Aschenbach, [ * A&A * ] { }, in press (astro - ph/0304322), 2003. --- abstract:' Modern robotic systems have become a stand-in for world when it ’s necessary to perform risky or draining tasks. In such lotion scenarios, communications between automaton and the control center are one of the major problems. The normally use solution assumes that raw messages are more valuable. We find that it does not hold in many scenarios. In this paper, we propose a fresh, resilient buffer management policy called OptSample. We cause a new assumption that uniformly sample messages are the about valuable and define an evaluation function to estimate the profit of the received message sequence. Our OptSample policy can uniformly sample message and dynamically adjust the sample rate based on the run - time network situation. Our analysis and simulation shows that the OptSample policy can effectively prevent losing long segments of continuous messages and can improve the net income of the meet messages. We follow through the OptSample policy in ROS, without change the interface or API for the applications. Our experiments show that the OptSample policy can better the results of several application scenarios include surveillance, 3D reconstruction, and SLAM.' author: - | Yu - Ping Wang$^{1}$, Zi - Xin Zou$^{1}$, Xu - Qiang Hu$^{1}$, Dinesh Manocha$^{2}$, Lei Qiao$^{3}$, Shi - Min Hu$^{1}$\ \ The video of applications can be found at\ < https://www.youtube.com/watch?v=ji-yvbT1l-w>[^1][^2][^3][^4 ] bibliography: -'main.bib' title:' * * OptSample: A Resilient Buffer Management Policy for Robotic Systems Based on Optimal Message Sampling * *' --- insertion { # sec: introduction } = = = = = = = = = = = = Modern robotic systems have become a stand-in for humans when it ’s necessary to do risky or exhausting tasks such as military operations, exploration, rescue operations, surveillance, or large - plate cleaning operations. In such application scenarios, a control center through which humans can monitor and operate the whole system is normally needed. However, communication issues, such as unstable network connection and circumscribed bandwidth, cause problems. Network connections, especially wireless one (e.g. over WiFi), are not constantly stable

*]{}, [**586**]{}, 356, 2003. Ximmermann, H.-U., and B. Aschtnbach, [*A&A*]{}, in press (astro-'h/0304322), 2003. --- absfract: 'Moaern robotic systems have berome a suvstitute for humans whdn it’s nebessary ti pecform risky or eegaustinn tasid. In wuch applicatipn scenarims, communicatimnr yetween robots and the control centew are ome of the major krobltms. The bonmonly used solution assumes fhat nexer messages arr more valuable. We find thwt ih does not hold in many scenatjos. Un this papef, we propose a novel, rgsilient buffer management polich calked OptSamkls. Ag make a new assuiption that mmiformny sampked messages ave thx mowt valuable and definx an evaluation funcjion to esdijate the profit od rhe rgceivad mdwsaee aeausnce. Okr KptSample lolicy can yniformly sample mexswtes and dynamjcally aqjust the sample rate based on the run-tpme hetwork situation. Our abalysis and simulatioj shows trat the OptSample policy can effectively prevent nosinj uony segmdbtd of continuous messages and can improve the [dogin of the received messages. Wr lmljement the OpjSample pomicy in ROS, withouh changyng tye interfwce pr API for the applications. Our experimvnts show that the OptDample poliey can imprpve the results of sevexal apllication sfenarios jvcluding surveiluanbe, 3D reconstruction, and SLAM.' auehor: - | Bu-Piny Wang$^{1}$, Zk-Xin Zou$^{1}$, Xt-Qiang Hu$^{1}$, Finesm Manocha$^{2}$, Lei Qiao$^{3}$, Dhi-Miu Hu$^{1}$\ \ The video of applications can be found at\ <httks://wfw.ymutube.coi/watcm?v=ji-yvbT1l-w>[^1][^2][^3][^4] biblijgraphy: -'main.bib' jitle: '**OptFampld: A Resilivnt Buffec Management Policy for Rmhotic Systemv Based jn Optimql Messxee Sampling** ' --- Inyroductiou {#sec:untroduction} ============ Modern rocktic systems hart btcime a substituye wor hlmais whqt it’s necessdry go orrforo risky or cxhxustong tasks such as minitady operations, explprwtion, rewcue opewations, surveollance, or large-scwle cneaiing olerwtions. In such application sceharios, a fonbrol center trrounh wrich humanx can monitor and operate the whole sysvem is usually needed. Hiwever, communicatiou lssues, such es unseable netfork connection and oimited bandwidth, cause problems. Networi connactiojs, especially wireless ones (e.g. over WiFi), are not always stable 

*]{}, [**586**]{}, 356, 2003. Zimmermann, H.-U., and [*A&A*]{}, press (astro-ph/0304322), --- abstract: 'Modern substitute humans when it’s to perform risky exhausting tasks. In such application scenarios, between robots and the control center are one of the major problems. The used solution assumes that newer messages are more valuable. We find that it not in scenarios. this paper, we propose a novel, resilient buffer management policy called OptSample. We make a new that uniformly sampled messages are the most valuable define an evaluation function estimate the profit of the message Our OptSample can sample and dynamically adjust sample rate based on the run-time network situation. Our analysis and simulation shows that the OptSample policy effectively prevent segments of messages can the profit of messages. We implement the OptSample policy changing the interface or API for the applications. experiments show the OptSample policy can improve the of several application scenarios including surveillance, 3D reconstruction, SLAM.' author: - | Yu-Ping Wang$^{1}$, Zi-Xin Zou$^{1}$, Xu-Qiang Hu$^{1}$, Dinesh Manocha$^{2}$, Lei Qiao$^{3}$, Shi-Min The video of applications be found at\ bibliography: title: A Buffer Management for Robotic Systems Based on Optimal Message Sampling** ' --- Introduction ============ Modern robotic systems have become a substitute for humans necessary perform risky or tasks such as military exploration, operations, surveillance, or large-scale In application center which can monitor and operate whole system is usually needed. communication issues, such as bandwidth, cause problems. Network connections, especially wireless ones over WiFi), are not always stable

*]{}, [**586**]{}, 356, 2003. Zimmermann, H.-U., and B. AschenbacH, [*A&A*]{}, in press (Astro-Ph/0304322), 2003. --- aBstRaCt: 'MoDern Robotic systems HAve bEcome a substitute for humAns whEn IT’s neCEsSary tO perforM RiSKY or ExHaUstInG TaSks. In SucH applicAtion scenaRioS, cOmmunicationS BeTween robotS anD the control cEntEr are oNe Of tHE majoR prOblemS. The coMMonly uSed solutiOn ASsumes THat neweR MEsSageS are more valuable. WE FiND that it does not Hold in MaNY sCENarIos. in this papeR, wE propOSe a noveL, ReSILIenT Buffer managemEnt policy caLLed optSamPlE. We MAke a neW assuMpTIon That uniformLy saMpled messAges arE The most VAluable And defIne An eValuATiOn FunCtIOn tO EsTimATe tHe profit Of ThE receIved MESSAge sEquEnce. our OpTSample policy Can UnifORmlY sampLe mesSageS aNd dynAmicalLy adjUsT the sample rate bAsed On the run-tIme NeTwoRk SituaTIon. Our AnaLysIs and siMulatioN ShoWs THAT tHe OptSample policy cAn EFFeCtively pRevent LOsInG Long segmEnTs oF conTINuous MessAGeS and can iMprove THe PrOfit of tHe ReceivEd MesSagEs. We iMPlemEnt the optSamplE poliCY in ROS, without cHAnging the inteRFaCE Or apI foR thE applicatioNs. OuR ExpeRimeNTs ShoW That tHe OptsaMPlE Policy can improve the ReSults oF seveRal applicatioN scenarios INCLuding suRveiLLaNCe, 3D reconstructIon, anD SLAM.' authoR: - | yu-Ping WaNg$^{1}$, Zi-XIn Zou$^{1}$, Xu-QIang Hu$^{1}$, DinESH Manocha$^{2}$, lei qiaO$^{3}$, ShI-MiN hU$^{1}$\ \ THe video of applICAtioNs Can be foUnd At\ <https://Www.YouTubE.coM/wAtch?v=ji-yvBT1l-w>[^1][^2][^3][^4] biblIoGrApHy: -'MaiN.bib' tITle: '**OptSaMpLe: A reSilIent BUFfer MaNagemEnt POlIcY For robotic sYsTEMs BaSeD oN OptImaL MEssagE SamPLinG** ' --- IntrodUction {#sec:IntROducTiOn} ============ modern rObotic systems HaVe become a sUbStiTute foR HUmans wheN it’s necessary to perform rISky or exHauSting TaskS such as miLitAry opeRatIOns, expLoratiOn, resCuE opERAtionS, SUrVeiLlAnce, or largE-SCalE cleaNiNg opErationS. In such application SCenArios, a control CenTer tHROuGh wHIcH HumAnS Can MONitor and operate The whole syStEM iS usually neEDed. hoWever, coMmunicaTion iSSues, sucH as unstabLe network CoNnecTIOn aNd limited bAndwidth, Cause probLEms. NeTWoRk conNecTions, eSpEciAlly wIrelesS OneS (e.g. ovEr WiFi), ArE not alWays sTaBle 

*]{}, [**586**]{}, 356, 20 03. Zimme rmann , H .-U ., and B.Aschenbach, [* A &A*] {}, in press (astro-ph /0304 32 2 ), 2 0 03 . -- - abstr a ct : 'Mo de rn ro bo t ic syst ems have b ecome a su bst it ute for huma n swhen it’snec essary to pe rfo rm ris ky or exhau sti ng ta sks. I n suchapplicati on scenar i os, com m u ni cati ons between robot s a n d the controlcenter a r eo n e o f t he major p ro blems . The co m mo n l y us e d solution as sumes thatn ewe r mess ag esa re mor e val ua b le. We find th at i t does no t hold in many scenari os. In th ispape r ,we pr op o sea n ove l , r esilient b uf fer m anag e m e n t po lic y ca lledOptSample. We ma ke a new assu mptio n th at unif ormlysampl ed messages are t he m ost valua ble a ndde finea n eval uat ion functi on to e s tim at e t he profit of the rec ei v e dmessagesequen c e. O u r OptSam pl e p olic y can u nifo r ml y sample messa g es a nd dyna mi callyad jus t t he sa m plerate b ased onthe r u n-time network situation. Ou r a n a ly s is a ndsimulationshow s tha t th e O ptS a mplepolic yc an effectively prevent l osinglongsegments of c ontinuousm e s sages an d ca n i m prove the prof it of the recei v ed messa ges.We imple ment theO p tSamplepol icy in RO S , w ithout changi n g the i nterfac e o r API f orthe ap pli ca tions. Ou r experi me nt ssh owthatt he OptSa mp lepo lic y can improv e the res ul ts ofseveral ap p l icat io nscen ari os incl udin g su rveilla nce, 3D r eco n stru ct io n, andSLAM.' author :- | Yu -P ing Wang$ ^ { 1}$, Zi- Xin Zou$^{1}$, Xu-Qiang Hu$^{1} $,Dines h Ma nocha$^{2 }$, Lei Q iao $ ^{3}$, Shi-M in Hu $^ {1} $ \ \ T he video ofa p pli catio ns can be fou nd at\ <https: / /ww w.youtube.com /wa tch? v = ji -yv b T1 l -w> [^ 1 ][^ 2 ] [^3][^4] biblio graphy: -' ma i n. bib' title : '* *O ptSampl e: A Re silie n t Buffe r Managem ent Polic yforR o bot ic Systems Based o n Optimal Messa g eSampl ing ** ' - -- I ntrod uction {#s ec:in troduc ti on} == ===== == === Mod ern robotic systems hav e beco me asub stitute f orh uma ns when i t’snecessarytoper formris k y orexha u st ing tasks suc h as milit a ry op e r at ions, explo r a t ion , res cue operat ions , surveillance, o r large-scale c lean i n g o per a tion s. In such appli cat io n scenario s, a controlcenter t hr o ugh w hich h umanscan mon i t or and op erat e t he wholesys te m is usu al ly needed . Ho we ver, c ommuni c atio n issues, such asunsta b l e net w ork conn ec tion an d lim ited bandw idth, cause probl ems. Net work co nn ection s,es pecially w i reless on es (e .g. ove rWiFi ),are no t al w a ys st able  

*]{}, [**586**]{},_356, 2003. Zimmermann,_H.-U., and B. Aschenbach,_[*A&A*]{}, in_press_(astro-ph/0304322), 2003. _--- abstract:_'Modern robotic systems_have become a_substitute for humans when_it’s necessary to_perform_risky or exhausting tasks. In such application scenarios, communications between robots and the control_center_are one_of_the_major problems. The commonly used_solution assumes that newer messages_are more_valuable. We find that it does not hold_in_many scenarios. In_this paper, we propose a novel, resilient buffer management_policy called OptSample. We make a_new assumption that_uniformly_sampled_messages are the most_valuable and define an evaluation function_to estimate the profit of the_received message sequence. Our OptSample policy can_uniformly sample messages and dynamically adjust_the sample rate based on_the run-time_network situation. Our analysis and_simulation shows that_the OptSample_policy can effectively_prevent losing long segments of continuous_messages and can_improve the profit of the received_messages._We implement the_OptSample_policy_in ROS,_without changing the_interface_or API_for_the applications. Our experiments show that_the_OptSample policy can improve the results of_several application scenarios including_surveillance,_3D reconstruction, and SLAM.' author: -_| Yu-Ping_Wang$^{1}$, Zi-Xin Zou$^{1}$, Xu-Qiang Hu$^{1}$, Dinesh_Manocha$^{2}$, Lei_Qiao$^{3}$, Shi-Min_Hu$^{1}$\ \ The video of_applications can be found at\ _ <https://www.youtube.com/watch?v=ji-yvbT1l-w>[^1][^2][^3][^4] bibliography: -'main.bib' title: '**OptSample:_A Resilient_Buffer_Management Policy for_Robotic_Systems Based_on Optimal Message Sampling** ' --- Introduction {#sec:introduction} ============ Modern robotic_systems have_become a substitute for humans when_it’s necessary to perform_risky_or exhausting tasks such as military_operations, exploration, rescue operations, surveillance, or_large-scale cleaning operations. In such_application_scenarios,_a control center through which_humans can monitor and operate the_whole system is_usually needed. However, communication issues, such as_unstable_network connection and limited bandwidth, cause_problems. Network_connections, especially wireless ones (e.g. over_WiFi),_are_not always stable 

neuron functionality is shown in Fig. \[fig:neuron\]. This is a current-mode circuit composed of multiple “compartments” or blocks. The *LEAK* block models the neuron’s passive leak conductance, producing exponential sub-threshold dynamics in response to constant input currents. The *AHP* block models the generation of the after hyper-polarizing current in real neurons, responsible for their spike-frequency adaptation behavior. The Na+ and K+ blocks model the effect of Sodium and Potassium channels, responsible for generating action-potentials (spikes) in real neurons. The *HS* block implements handshaking with following encoder block for encoding spike events following protocol. We used an optimized Traff’s current comparator (see *CC* box in the Na+ block) to make an accurate comparison between the neuron $\mathsf{I_{mem}}$ current and a programmable $\mathsf{I_{ref}}$ threshold current, which sets the neuron’s spiking threshold. Current limitation transistors ($\mathsf{M_{C7},M_{C8}}$) are included to reduce static power consumption. We used the same split-transistor sub-threshold technique used in the synapse circuits for enhanced current-mirror operation and for precise control of small currents. We added also several reset transistors, such as $\mathsf{M_{L3}}$ and $\mathsf{M_{NA1}}$ to further reduce power consumption during the spike reset phase. The active area of the neuron is 20$\mu m^{2}$; the neuron capacitance is also implemented using a MIMCAP structure and measures approximately 1.5pF. Fig. \[fig:isynimem\] shows the expected response of the neuron circuit to a constant input current. By tuning the biases that control the neuron’s integration time constant, firing threshold, refractory period and spike-frequency adaptation dynamics, the proposed circuit can reproduce a wide range of spiking behaviors [@Qiao_etal15]. ![Membrane current trace over time.[]{data-label="fig:trace"}](Imem){width="35.00000%"} Simulation results {#sec:results} ================== ![Synapse and neuron response to a 100Hz spike train.[]{data-label="fig:isynimem"}](Isyn_Imem){width="40.00000%"} In Fig. \[fig:isynimem\] we show circuit simulation results of both synaptic and neuron currents, while they are being

neuron functionality is shown in Fig.   \[fig: neuron\ ]. This is a current - mode circuit write of multiple “ compartment ” or blocks. The * LEAK * block models the neuron ’s passive escape conductance, producing exponential sub - threshold dynamics in response to changeless input currents. The * AHP * pulley models the generation of the after hyper - polarize current in real nerve cell, responsible for their spike - frequency adaptation behavior. The Na+ and K+ block model the effect of Sodium and Potassium channels, creditworthy for generating action - potentials (spike) in real neurons. The * HS * blocking implements handshaking with come encoder block for encoding spike events following protocol. We used an optimize Traff ’s current comparator (see * CC * box in the Na+ block) to make an accurate comparison between the neuron $ \mathsf{I_{mem}}$ current and a programmable $ \mathsf{I_{ref}}$ threshold current, which sets the neuron ’s spiking threshold. Current limitation transistors ($ \mathsf{M_{C7},M_{C8}}$) are included to reduce static power consumption. We use the same split - transistor sub - threshold proficiency use in the synapse circuit for enhanced current - mirror operation and for precise control of small currents. We added also respective reset transistors, such as $ \mathsf{M_{L3}}$ and $ \mathsf{M_{NA1}}$ to further reduce power consumption during the spike reset phase. The active area of the neuron is 20$\mu m^{2}$; the neuron capacitance is also implemented using a MIMCAP social organization and measures approximately 1.5pF. Fig.   \[fig: isynimem\ ] show the expect response of the neuron circuit to a constant input current. By tuning the bias that control the neuron ’s integration time constant, displace threshold, refractory period and spike - frequency adaptation dynamics, the proposed tour can reproduce a across-the-board range of spike behaviors   [ @Qiao_etal15 ]. ! [ Membrane current trace over time.[]{data - label="fig: trace"}](Imem){width="35.00000% " } Simulation resultant role { # sec: results } = = = = = = = = = = = = = = = = = = ! [ Synapse and nerve cell response to a 100Hz spike train.[]{data - label="fig: isynimem"}](Isyn_Imem){width="40.00000% " } In Fig.   \[fig: isynimem\ ] we show circuit pretense results of both synaptic and neuron currents, while they are being

nekron functionality is shuwn in Fig. \[fig:neorin\]. Thiv is a current-oode circuit composed of mulviplw “comkcrtments” or blocks. Tfe *LEAK* bpock modwls uhe neuron’s passits leak gjndudbance, 'roducing expongntial sub-thseshold dynamiws iu response to constant input currentf. The *ANP* block models jhe gtnewatikn of the after hyper-polarizing cudrent ii real neurons, tesponsible for their spikf-freeuency adaptation hehavior. Thg Na+ qnd K+ blocks model the effect of Skdium and Potassium channels, rerponsnble for gebeeatltg action-poventiajs (spikes) in real negrons. Tne *HS* block imilemeits yandshaking with follmwing encoder bloch for encmdnng spike events folliwung ptotocml. Wd usdd zn oltimizfd Vraff’s currsnt compararor (see *CC* box in tne Ba+ block) to mzke an ascurate comparison between the neuron $\mdthaf{I_{mem}}$ current and a prigrammable $\mathsf{I_{ref}}$ thresholq current, which sets the neuron’s spiking thresholg. Curcevt oiikrahion transistors ($\mathsf{M_{C7},M_{C8}}$) are included to rqsuve static power gonsumption. We usec hhr same split-trxnsistor aub-threshold technlque usgd in rhe synapfe corcuits for enhanced currenr-mirror operctiin and for precise control of smakl cutrents. We added also sereral deset translstors, sudf as $\mathsf{M_{L3}}$ ana $\mstvsf{M_{NA1}}$ to further reduce pjwer consnmptipn durivg tne spihe reset pjase. Bve active area of hhe nguron hs 20$\mu m^{2}$; thf neuron capacitance is also im'kemented usimg a KIMCAP scructuve and measures approximately 1.5pF. Fig. \[yig:isyvimem\] showa the eepected respjnse of the naoron circuit vo a conseant inpyt currdvt. By tuning tne biases that conteol the neuron’s inbegrajikn time constanc, diring thresholc, rdfrwcnorb perymd and spike-xreqjenzu adaotation dyncmkcs, yhe proposed circuit can reproduce a wide takge of spuking beraviors [@Qiao_etsl15]. ![Membrane current tracx over time.[]{catw-label="fig:trace"}](Imem){width="35.00000%"} Simulatjon resulhs {#fec:results} ================== ![Sygapsc anq neuron rzsponse to a 100Hz spike train.[]{data-label="fig:iwynimem"}](Isyn_Imem){widrh="40.00000%"} In Fig. \[fig:isynimem\] ee show cirruit symulation results of both synqptic and neuron gurrents, while they ars beinc

neuron functionality is shown in Fig. \[fig:neuron\]. a circuit composed multiple “compartments” or the passive leak conductance, exponential sub-threshold dynamics response to constant input currents. The block models the generation of the after hyper-polarizing current in real neurons, responsible their spike-frequency adaptation behavior. The Na+ and K+ blocks model the effect of and channels, for action-potentials (spikes) in real neurons. The *HS* block implements handshaking with following encoder block for encoding events following protocol. We used an optimized Traff’s comparator (see *CC* box the Na+ block) to make accurate between the $\mathsf{I_{mem}}$ and programmable $\mathsf{I_{ref}}$ threshold which sets the neuron’s spiking threshold. Current limitation transistors ($\mathsf{M_{C7},M_{C8}}$) are included to reduce static power consumption. used the sub-threshold technique in synapse for enhanced current-mirror for precise control of small currents. several reset transistors, such as $\mathsf{M_{L3}}$ and $\mathsf{M_{NA1}}$ further reduce consumption during the spike reset phase. active area of the neuron is 20$\mu m^{2}$; neuron capacitance is also implemented using a MIMCAP structure and measures approximately 1.5pF. Fig. \[fig:isynimem\] expected response of the circuit to a input By the that control neuron’s integration time constant, firing threshold, refractory period and spike-frequency adaptation the proposed circuit can reproduce a wide range of spiking ![Membrane trace over time.[]{data-label="fig:trace"}](Imem){width="35.00000%"} results {#sec:results} ================== ![Synapse neuron to a 100Hz spike Fig. we results both and neuron currents, while are being

neuron functionality is showN in Fig. \[fig:nEuron\]. thiS is A cUrreNt-moDe circuit compoSEd of Multiple “compartments” or BlockS. THE *LEAk* BlOck moDels the NEuRON’s pAsSiVe lEaK CoNductAncE, producIng exponenTiaL sUb-threshold dYNaMics in respOnsE to constant iNpuT curreNtS. ThE *aHP* blOck ModelS the geNEratioN of the aftEr HYper-poLArizing CURrEnt iN real neurons, respoNSiBLe for their spikE-frequEnCY aDAPtaTioN behavior. THe na+ and k+ Blocks mODeL THE efFEct of Sodium anD Potassium cHAnnEls, resPoNsiBLe for gEneraTiNG acTion-potentiAls (sPikes) in reAl neurONs. The *HS* BLock impLementS haNdsHakiNG wItH foLlOWinG EnCodER blOck for enCoDiNg spiKe evENTS FollOwiNg prOtocoL. We used an optiMizEd TrAFf’s CurreNt comParaToR (see *Cc* box in The Na+ BlOck) to make an accuRate ComparisoN beTwEen ThE neurON $\mathsF{I_{mEm}}$ cUrrent aNd a progRAmmAbLE $\MAtHsf{I_{ref}}$ threshold cuRrENT, wHich sets The neuROn’S sPIking thrEsHolD. CurRENt limItatIOn TransistOrs ($\matHSf{m_{C7},m_{C8}}$) are inClUded to ReDucE stAtic pOWer cOnsumpTion. We usEd the SAme split-transiSTor sub-threshoLD tECHnIQue uSed In the synapsE cirCUits For eNHaNceD CurreNt-mirRoR OpERation and for precise CoNtrol oF smalL currents. We adDed also sevERAL reset trAnsiSToRS, such as $\mathsf{M_{l3}}$ and $\mAthsf{M_{NA1}}$ to FUrther reDuce pOwer consUmption duRINg the spiKe rEseT phAse. tHE aCtive area of thE NEuroN iS 20$\mu m^{2}$; the NeuRon capaCitAncE is AlsO iMplementeD using a MiMcAp sTrUctUre anD Measures ApProXiMatEly 1.5pF. fIg. \[fig:iSynimEm\] shOwS tHE exPected rESpONSe of ThE nEuroN ciRcUit to A conSTanT input cUrrent. By tUniNG the BiAsEs that cOntrol the neurOn’S integratiOn TimE constANT, firing tHreshold, refractory perioD And spikE-frEquenCy adAptation dYnaMics, thE prOPosed cIrcuit Can rePrOduCE A wide RANgE of SpIking behavIORs [@QIao_etAl15]. ![membRane curRent trace over time.[]{dATa-lAbel="fig:trace"}](IMem){WidtH="35.00000%"} sImUlaTIoN ResUlTS {#seC:REsults} ================== ![Synapse anD neuron resPoNSe To a 100Hz spike TRaiN.[]{dAta-labeL="fig:isyNimem"}](iSyn_Imem){Width="40.00000%"} In FiG. \[fig:isyniMeM\] we sHOW ciRcuit simulAtion resUlts of botH SynapTIc And neUroN curreNtS, whIle thEy are bEIng

neuron functionality is s hown in Fi g. \[ fig :ne ur on\] . Th is is a curren t -mod e circuit composed ofmulti pl e “co m pa rtmen ts” orb lo c k s.Th e*LE AK * b lockmod els the neuron’spas si ve leak cond u ct ance, prod uci ng exponenti alsub-th re sho l d dyn ami cs in respo n se toconstantin p ut cur r ents. T h e * AHP* block models the ge n eration of the after h y pe r - pol ari zing curre nt in r e al neur o ns , r esp o nsible for th eir spike-f r equ ency a da pta t ion be havio r. The Na+ and K+ blo cks model the e f fect of Sodiumand Po tas siu m ch a nn el s,re s pon s ib lef orgenerati ng a ction -pot e n t i als(sp ikes ) inreal neurons. Th e *H S * b lockimple ment shands haking with f ollowing encode r bl ock for e nco di ngsp ike e v ents f oll owi ng prot ocol. W e us ed a n o ptimized Traff’s c ur r e nt compara tor (s e e*C C * box in t heNa+b l ock)to m a ke an accu rate c o mp ar ison be tw een th eneu ron $\ma t hsf{ I_{mem }}$ curr ent a n d a programmab l e $\mathsf{I_ { re f } }$ thre sho ld current, whi c h se ts t h eneu r on’sspiki ng th r eshold. Current lim it ationtrans istors ($\mat hsf{M_{C7} , M _ {C8}}$)arei nc l uded to reduce stat ic power c o nsumptio n. We used th e same sp l i t-transi sto r s ub- thr e s ho ld techniqueu s ed i nthe syn aps e circu its fo r e nha nc ed curren t-mirror o pe ra ti onand f o r precis econ tr olof sm a ll cur rents . We a dd e d a lso sev e ra l rese ttr ansi sto rs , suc h as $\m athsf{M _{L3}}$ a nd$ \mat hs f{ M_{NA1} }$ to further r educe powe rcon sumpti o n duringthe spike reset phase.T he acti veareaof t he neuron is 20$\m u m ^ {2}$;the ne uronca pac i t ancei s a lso i mplemented u sin g a M IM CAPstructu re and measures ap p rox imately 1.5pF . F ig.\ [ fi g:i s yn i mem \] sho w s the expected r esponse of t h eneuron cir c uit t o a con stant i nputc urrent. By tunin g the bia se s th a t co ntrol theneuron’s integrat i on ti m econst ant , firi ng th resho ld, re f rac toryperiod a nd spi ke-fr eq uency ad aptation dynamics, thepropos ed ci rcu it can re pro d uce a wide r ange of spikin g b eha viors  [@ Q iao_e tal1 5 ]. ! [ Membr anec urrent tr a ce ov e r t ime.[]{data - l a bel ="fig :tr a ce"}]( Imem ){width="35.00000 % "} Simulation res u l ts{#s e c:re su lts} ========= === == = = == ![Sy na pse and neu ron resp on s e toa 100H z spik e train . [ ]{ d ata-la bel= "fi g:isynime m"} ]( I syn_Ime m) {w i dth="4 0.00 00 0%"} In Fig .  \[f i g :isynimem\] we s how c i r cuits imu latio nresults of b oth synapt ic and neur on cur rent s, wh ile the yare be ing

neuron_functionality is_shown in Fig. \[fig:neuron\]. This_is a_current-mode_circuit composed_of_multiple “compartments” or_blocks. The *LEAK*_block models the neuron’s_passive leak conductance,_producing_exponential sub-threshold dynamics in response to constant input currents. The *AHP* block models the_generation_of the_after_hyper-polarizing_current in real neurons, responsible_for their spike-frequency adaptation behavior._The Na+_and K+ blocks model the effect of Sodium_and_Potassium channels, responsible_for generating action-potentials (spikes) in real neurons. The *HS*_block implements handshaking with following encoder_block for encoding_spike_events_following protocol. We used_an optimized Traff’s current comparator (see_*CC* box in the Na+ block)_to make an accurate comparison between the_neuron $\mathsf{I_{mem}}$ current and a programmable_$\mathsf{I_{ref}}$ threshold current, which sets_the neuron’s_spiking threshold. Current limitation transistors_($\mathsf{M_{C7},M_{C8}}$) are included_to reduce_static power consumption._We used the same split-transistor sub-threshold_technique used in_the synapse circuits for enhanced current-mirror_operation_and for precise_control_of_small currents._We added also_several_reset transistors,_such_as $\mathsf{M_{L3}}$ and $\mathsf{M_{NA1}}$ to further_reduce_power consumption during the spike reset phase._The active area of_the_neuron is 20$\mu m^{2}$;_the neuron capacitance is also_implemented using a MIMCAP structure and_measures approximately_1.5pF. Fig. \[fig:isynimem\]_shows the expected response of the neuron circuit to a constant_input current. By tuning the biases_that control the neuron’s_integration time_constant,_firing threshold, refractory_period_and spike-frequency_adaptation dynamics, the proposed circuit can reproduce_a wide_range of spiking behaviors [@Qiao_etal15]. ![Membrane current trace_over time.[]{data-label="fig:trace"}](Imem){width="35.00000%"} Simulation results {#sec:results} ================== ![Synapse_and_neuron response to a 100Hz spike_train.[]{data-label="fig:isynimem"}](Isyn_Imem){width="40.00000%"} In Fig. \[fig:isynimem\] we show circuit simulation_results of both synaptic and_neuron_currents,_while they are being

Thus, $\kappa^{s}(H_{n}; K_{1,3})\geq \lceil\frac{n}{2}\rceil$. $\kappa(H_{n}; K_{1,3})$ and $\kappa^{s}(H_{n}; K_{1,3})$ --------------------------------------------------------- .2cm \[3-upper\] $\kappa(H_{n}; K_{1,3})\leq \lceil\frac{n}{2}\rceil$ for $n\geq 4$. For any $u=u_1\cdots u_n\in V(H_n)$, $N_{H_n}(u)=\{u^1,\ldots,u^n\}$. Let $T_i$ be the subgraph induced by $\{u^i,u^{i,1}, u^{i+1}, u^{i,1,1}\}$ for $1\leq i \leq n-2$. If $n$ is odd, we let $T_n$ be the subgraph induced by $\{u^n, u^{n-1,1},u^{n,1^{*}}, u^{n,2^{*}}\}$; and if $n$ is even, let $T_{n-1}$ the subgraph induced by $\{u^{n-1}, u^{n-1,1}, u^{n}, u^{n-1,1,1^{*}}\}$. Then $T_i$ is a $K_{1,3}$ by Proposition \[star\](a-c) for $1\leq i\leq n$, and $V(T_i)\cap V(T_j)=\emptyset$ by Proposition \[vertex\] for $1\leq i\neq j\leq n$. Set $\mathcal{S} = \{T_1, T_3,\ldots, T_{n-2}, T_n\}$ if $n$ is odd; and $\mathcal{S} = \{T_1, T_3,\ldots, T_{n-3}, T_{n-1}\}$ if $n$ is even. Then, in either case, $H_n-\mathcal{S}$ is disconnected, one component is $\{u\}$ and $|\mathcal{S}| =\lceil\frac{n}{2} \rceil$. Thus, $\kappa(H_{n}; K_{1,3})\leq \lceil\frac{n}{2

Thus, $ \kappa^{s}(H_{n }; K_{1,3})\geq \lceil\frac{n}{2}\rceil$. $ \kappa(H_{n }; K_{1,3})$ and $ \kappa^{s}(H_{n }; K_{1,3})$ --------------------------------------------------------- .2 cm \[3 - upper\ ] $ \kappa(H_{n }; K_{1,3})\leq \lceil\frac{n}{2}\rceil$ for $ n\geq 4$. For any $ u = u_1\cdots u_n\in V(H_n)$, $ N_{H_n}(u)=\{u^1,\ldots, u^n\}$. Let $ T_i$ be the subgraph induced by $ \{u^i, u^{i,1 }, u^{i+1 }, u^{i,1,1}\}$ for $ 1\leq i \leq n-2$. If $ n$ is curious, we lease $ T_n$ be the subgraph induced by $ \{u^n, u^{n-1,1},u^{n,1^ { * } }, u^{n,2^{*}}\}$; and if $ n$ is even, let $ T_{n-1}$ the subgraph induced by $ \{u^{n-1 }, u^{n-1,1 }, u^{n }, u^{n-1,1,1^{*}}\}$. Then $ T_i$ is a $ K_{1,3}$ by Proposition   \[star\](a - speed of light) for $ 1\leq i\leq n$, and $ V(T_i)\cap V(T_j)=\emptyset$ by Proposition   \[vertex\ ] for $ 1\leq i\neq j\leq n$. Set $ \mathcal{S } = \{T_1, T_3,\ldots, T_{n-2 }, T_n\}$ if $ n$ is odd; and $ \mathcal{S } = \{T_1, T_3,\ldots, T_{n-3 }, T_{n-1}\}$ if $ n$ is even. Then, in either case, $ H_n-\mathcal{S}$ is disconnect, one part is $ \{u\}$ and $ |\mathcal{S}| = \lceil\frac{n}{2 } \rceil$. Thus, $ \kappa(H_{n }; K_{1,3})\leq \lceil\frac{n}{2

Thud, $\kappa^{s}(H_{n}; K_{1,3})\geq \lceil\frag{n}{2}\rceil$. $\kappa(H_{n}; K_{1,3})$ and $\ka'pa^{s}(H_{n}; I_{1,3})$ --------------------------------------------------------- .2cm \[3-upper\] $\kappa(H_{n}; K_{1,3})\leq \lceil\frac{n}{2}\rceip$ dor $n\teq 4$. For any $u=u_1\cdots u_n\kn V(H_n)$, $N_{H_j}(u)=\{u^1,\ldots,y^n\}$. Ltt $T_i$ be the subgczph indmeed bg $\{u^i,u^{n,1}, n^{i+1}, u^{i,1,1}\}$ for $1\leq i \leq n-2$. If $t$ is odd, we led $G_n$ be the subgraph induced by $\{u^n, u^{n-1,1},u^{n,1^{*}}, u^{n,2^{*}}\}$; and iv $n$ is even, lej $T_{n-1}$ uhe subfgakh induced by $\{u^{n-1}, u^{n-1,1}, u^{n}, u^{n-1,1,1^{*}}\}$. Then $T_i$ is a $K_{1,3}$ by Propositipn \[star\](a-c) for $1\leq i\leq n$, anf $V(T_l)\cap V(T_j)=\emptyset$ bj Propositiin \[vewrex\] for $1\leq k\neq j\leq n$. Set $\mathcal{A} = \{T_1, T_3,\ldots, T_{n-2}, T_n\}$ if $n$ is odd; avd $\machcal{S} = \{T_1, T_3,\odits, J_{n-3}, T_{n-1}\}$ if $n$ iw eveg. Then, in eibner cave, $H_n-\mayhcal{S}$ is discpnnxctee, one component is $\{u\}$ end $|\mathcal{S}| =\lceil\frwc{n}{2} \rceil$. Tkus, $\kappa(H_{n}; K_{1,3})\leq \lceio\feac{n}{2

Thus, $\kappa^{s}(H_{n}; K_{1,3})\geq \lceil\frac{n}{2}\rceil$. $\kappa(H_{n}; K_{1,3})$ and --------------------------------------------------------- \[3-upper\] $\kappa(H_{n}; \lceil\frac{n}{2}\rceil$ for $n\geq V(H_n)$, Let $T_i$ be subgraph induced by u^{i+1}, u^{i,1,1}\}$ for $1\leq i \leq If $n$ is odd, we let $T_n$ be the subgraph induced by $\{u^n, u^{n,2^{*}}\}$; and if $n$ is even, let $T_{n-1}$ the subgraph induced by $\{u^{n-1}, u^{n}, Then is $K_{1,3}$ by Proposition \[star\](a-c) for $1\leq i\leq n$, and $V(T_i)\cap V(T_j)=\emptyset$ by Proposition \[vertex\] for $1\leq j\leq n$. Set $\mathcal{S} = \{T_1, T_3,\ldots, T_{n-2}, if $n$ is odd; $\mathcal{S} = \{T_1, T_3,\ldots, T_{n-3}, if is even. in case, is disconnected, one is $\{u\}$ and $|\mathcal{S}| =\lceil\frac{n}{2} \rceil$. Thus, $\kappa(H_{n}; K_{1,3})\leq \lceil\frac{n}{2

Thus, $\kappa^{s}(H_{n}; K_{1,3})\geq \lceil\frac{N}{2}\rceil$. $\kappA(H_{n}; K_{1,3})$ aNd $\kAppA^{s}(h_{n}; K_{1,3})$ --------------------------------------------------------- .2cM \[3-uppEr\] $\kappa(H_{n}; K_{1,3})\leq \lCEil\fRac{n}{2}\rceil$ for $n\geq 4$. For any $U=u_1\cdoTs U_N\in V(h_N)$, $N_{h_n}(u)=\{u^1,\lDots,u^n\}$. LET $T_I$ BE thE sUbGraPh INdUced bY $\{u^i,U^{i,1}, u^{i+1}, u^{i,1,1}\}$ fOr $1\leq i \leq n-2$. if $n$ Is Odd, we let $T_n$ be THe Subgraph inDucEd by $\{u^n, u^{n-1,1},u^{n,1^{*}}, u^{n,2^{*}}\}$; And If $n$ is eVeN, leT $t_{n-1}$ the SubGraph InduceD By $\{u^{n-1}, u^{n-1,1}, U^{n}, u^{n-1,1,1^{*}}\}$. Then $T_I$ iS A $K_{1,3}$ by PrOPositioN \[STaR\](a-c) fOr $1\leq i\leq n$, and $V(T_i)\cAP V(t_J)=\emptyset$ by ProPositiOn \[VErTEX\] foR $1\leQ i\neq j\leq n$. seT $\mathCAl{S} = \{T_1, T_3,\ldOTs, t_{N-2}, t_N\}$ if $N$ Is odd; and $\mathcAl{S} = \{T_1, T_3,\ldots, T_{N-3}, t_{n-1}\}$ iF $n$ is evEn. theN, In eithEr casE, $H_N-\MatHcal{S}$ is discOnneCted, one coMponenT Is $\{u\}$ and $|\mAThcal{S}| =\lCeil\frAc{n}{2} \RceIl$. ThUS, $\kApPa(H_{N}; K_{1,3})\LEq \lCEiL\frAC{n}{2

Thus, $\kappa^{s}(H_{n}; K_{1,3})\ geq \ lce il\ fr ac{n }{2} \rceil$. $\ka p pa(H _{n}; K_{1,3})$ and $\ kappa ^{ s }(H_ { n} ; K_{ 1,3})$- -- - - --- -- -- --- -- - -- ----- --- ------- ---------- --- -- ------- .2c m \[3-upper\ ] $ \kappa(H_{n} ; K _{1,3} )\ leq \lcei l\f rac{n }{2}\r c eil$ f or $n\geq 4 $ . For any $u= u _ 1\ cdot s u_n\in V(H_n)$, $N _ {H_n}(u)=\{u^1 ,\ldot s, u ^n \ } $.Let $T_i$ beth e sub g raph in d uc e d by$ \{u^i,u^{i,1} , u^{i+1},u ^{i ,1,1}\ }$ fo r $1\le q i \ le q n- 2$. If $n$is o dd, we le t $T_n $ be the subgrap h indu ced by $\{ u ^n ,u^{ n- 1 ,1} , u^ {n, 1 ^{* }}, u^{n ,2 ^{ *}}\} $; a n d i f $n $ i s ev en, l et $T_{n-1}$the sub g rap h ind ucedby $ \{ u^{n- 1}, u^ {n-1, 1} , u^{n}, u^{n-1 ,1,1 ^{*}}\}$. Th en $T _i $ isa $K_{1 ,3} $ b y Propo sition\ [st ar \ ] ( a- c) for $1\leq i\le qn $ ,and $V(T _i)\ca p V (T _ j)=\empt ys et$ byP r oposi tion \[ vertex\] for $ 1 \l eq i\neqj\ leq n$ . Se t $ \math c al{S } = \{ T_1, T_3 ,\ldo t s, T_{n-2}, T_ n \}$ if $n$ is od d ; a n d $\ mat hcal{S} = \ {T_1 , T_3 ,\ld o ts , T _ {n-3} , T_{ n- 1 }\ } $ if $n$ is even. T he n, ineithe r case, $H_n- \mathcal{S } $ is disco nnec t ed , one component is $ \{u\}$ and $|\mathc al{S} | =\lcei l\frac{n} { 2 } \rceil $.Thu s,$\k a p pa (H_{n}; K_{1, 3 } )\le q\lceil\ fra c{n}{2

Thus, $\kappa^{s}(H_{n};_K_{1,3})\geq \lceil\frac{n}{2}\rceil$. $\kappa(H_{n};_K_{1,3})$ and $\kappa^{s}(H_{n}; K_{1,3})$ --------------------------------------------------------- .2cm \[3-upper\]_$\kappa(H_{n}; K_{1,3})\leq_\lceil\frac{n}{2}\rceil$_for $n\geq_4$. For_any $u=u_1\cdots u_n\in_V(H_n)$, $N_{H_n}(u)=\{u^1,\ldots,u^n\}$. Let_$T_i$ be the subgraph_induced by $\{u^i,u^{i,1},_u^{i+1},_u^{i,1,1}\}$ for $1\leq i \leq n-2$. If $n$ is odd, we let $T_n$ be_the_subgraph induced_by_$\{u^n,_u^{n-1,1},u^{n,1^{*}}, u^{n,2^{*}}\}$; and if $n$_is even, let $T_{n-1}$ the_subgraph induced_by $\{u^{n-1}, u^{n-1,1}, u^{n}, u^{n-1,1,1^{*}}\}$. Then $T_i$ is_a_$K_{1,3}$ by Proposition \[star\](a-c)_for $1\leq i\leq n$, and $V(T_i)\cap V(T_j)=\emptyset$ by Proposition \[vertex\]_for $1\leq i\neq j\leq n$. Set $\mathcal{S}_= \{T_1, T_3,\ldots,_T_{n-2},_T_n\}$_if $n$ is odd;_and $\mathcal{S} = \{T_1, T_3,\ldots, T_{n-3},_T_{n-1}\}$ if $n$ is even. Then,_in either case, $H_n-\mathcal{S}$ is disconnected, one_component is $\{u\}$ and $|\mathcal{S}| =\lceil\frac{n}{2}_\rceil$. Thus, $\kappa(H_{n}; K_{1,3})\leq \lceil\frac{n}{2

relatively light sparticles, if the present deviation from the SM of $2.6\sigma$ persists. At large ${\mbox{$ \tan\beta~ $}}$ a global fit including both $b\to X_s\gamma$ and $a_\mu$ as well as the present Higgs limit of 113.5 GeV leaves a quite large region in the CMSSM parameter space. Here we left the trilinear coupling to be a free parameter, which affects both the Higgs limit constraint and the [$b \to X_s \gamma~ $]{}constraint, but in opposite ways, so that the preferred region is similar for the no-scale models with $A_0=0$ and models which leave $A_0$ free. The 95% lower limit on $m_{1/2}$ is $ 300$ GeV (see Figs. \[f6\]+\[f7\]), which implies that the lightest chargino (neutralino) is above 240(120) GeV. The 95% upper limit on $m_{1/2}$ is determined by the lower limit on $a_\mu^{SUSY}$ and therefor depends on (see Fig. \[f2\]). For =35(50) one finds $m_{1/2}\leq 610(720)$ GeV, which implies that the lightest chargino is below 500(590) GeV and the lightest neutralino is below 260(310) GeV. Acknowledgements {#acknowledgements.unnumbered} ================ D.K. would like to thank the Heisenberg-Landau Programme, RFBR grant \# 99-02-16650 and DFG grant \# 436/RUS/113/626 for financial support and the Karlsruhe University for hospitality during completion of this work. [99]{} H. N. Brown et al., hep-ex/0102017, Phys. Rev. Lett. [**86**]{} (2001) 2227. F. J. Yndurain, hep-ph/0102231. A. Czarnecki and W. Marchiano, hep-ph/0102122. U. Chattopadhyay and P. Nath, hep-ph/0102157;\ T. Ibrahim, U. Chattopadhyay, P. Nath hep-ph/0102324. L. Everett, G. L. Kane, S. Rigolin and L.Wang, hep

relatively light sparticles, if the present deviation from the SM of $ 2.6\sigma$ persists. At large $ { \mbox{$ \tan\beta~ $ } } $ a ball-shaped paroxysm including both $ b\to X_s\gamma$ and $ a_\mu$ as well as the present Higgs terminus ad quem of 113.5 GeV leaves a quite big region in the CMSSM parameter outer space. Here we left the trilinear coupling to be a complimentary parameter, which affects both the Higgs terminus ad quem constraint and the [ $ b \to X_s \gamma~ $ ] { } constraint, but in opposite ways, so that the preferable region is similar for the no - scale models with $ A_0=0 $ and models which exit $ A_0 $ free. The 95% lower limit on $ m_{1/2}$ is $ 300 $ GeV (learn Figs. \[f6\]+\[f7\ ]), which implies that the lightest chargino (neutralino) is above 240(120) GeV. The 95% upper limit on $ m_{1/2}$ is determined by the low limit on $ a_\mu^{SUSY}$ and therefor depends on (see Fig. \[f2\ ]). For = 35(50) one finds $ m_{1/2}\leq 610(720)$ GeV, which implies that the light chargino is below 500(590) GeV and the lightest neutralino is below 260(310) GeV. Acknowledgements { # acknowledgements.unnumbered } = = = = = = = = = = = = = = = = D.K. would like to thank the Heisenberg - Landau Programme, RFBR grant \ # 99 - 02 - 16650 and DFG grant \ # 436 / RUS/113/626 for fiscal documentation and the Karlsruhe University for hospitality during completion of this work. [ 99 ] { } H. N. Brown et al. , hep - ex/0102017, Phys. Rev. Lett. [ * * 86 * * ] { } (2001) 2227. F. J. Yndurain, hep - ph/0102231. A. Czarnecki and W. Marchiano, hep - ph/0102122. U. Chattopadhyay and P. Nath, hep - ph/0102157;\ T. Ibrahim, U. Chattopadhyay, P. Nath hep - ph/0102324. L. Everett, G. L. Kane, S. Rigolin and L.Wang, hep

repatively light sparticler, if the presenj eeviatmon froj the SM of $2.6\sigma$ persists. At large ${\muox{$ \ran\beuc~ $}}$ a global fit incljding botj $b\to X_s\tamme$ and $a_\mu$ as well as the presehb Higys limit of 113.5 GeV leaves a xuite large reciun in the CMSSM parameter space. Here re left tje trilinear cjuplpnd to be a free parameter, which affects both tie Higgs limit vonstraint and the [$b \to D_s \gwmma~ $]{}constraint, buh in opposijs wwts, so that tfe preferrtd region is similar for the no-scale models with $A_0=0$ and modgla avich leave $E_0$ free. Nhe 95% lower limit on $m_{1/2}$ is $ 300$ GrV (see Figs. \[f6\]+\[f7\]), whmch umplies that the lighvest chargino (neutrajino) is atore 240(120) GeV. The 95% upper linir on $k_{1/2}$ is detdemived bb tge lowfr mimit on $a_\ju^{SUSY}$ and rherefor depends on (sqv Fig. \[f2\]). For =35(50) kne figdf $m_{1/2}\leq 610(720)$ GeV, which implies that the lighuest dhargino is below 500(590) GeV qnd the lightest neuttalino is felow 260(310) GeV. Acknowledgements {#acknowledgements.unnumbared} ================ D.I. wobod liyw ho thank the Heisenberg-Landau Programme, RFBR ddamt \# 99-02-16650 and DFG grakt \# 436/RUS/113/626 for financoap xopport and the Karlsxhhs University for hlspitalyty dyring comkletipn of this work. [99]{} H. N. Brown er al., hep-ex/0102017, Pkys. Rev. Lett. [**86**]{} (2001) 2227. F. J. Yudurain, hep-pk/0102231. A. Czsrnecli and W. Marchiano, hep-pk/0102122. U. Chzttopadhyay and P. Naff, hep-ph/0102157;\ T. Ibrahim, U. Bhatdopadhyay, P. Nath hep-ph/0102324. L. Ederett, G. O. Kaue, S. Rigulin and L.Rang, hep

relatively light sparticles, if the present deviation SM $2.6\sigma$ persists. large ${\mbox{$ \tan\beta~ both X_s\gamma$ and $a_\mu$ well as the Higgs limit of 113.5 GeV leaves quite large region in the CMSSM parameter space. Here we left the trilinear to be a free parameter, which affects both the Higgs limit constraint and [$b X_s $]{}constraint, in opposite ways, so that the preferred region is similar for the no-scale models with $A_0=0$ models which leave $A_0$ free. The 95% lower on $m_{1/2}$ is $ GeV (see Figs. \[f6\]+\[f7\]), which that lightest chargino is 240(120) The 95% upper on $m_{1/2}$ is determined by the lower limit on $a_\mu^{SUSY}$ and therefor depends on (see Fig. \[f2\]). =35(50) one 610(720)$ GeV, implies the chargino is below and the lightest neutralino is below {#acknowledgements.unnumbered} ================ D.K. would like to thank the Programme, RFBR \# 99-02-16650 and DFG grant \# for financial support and the Karlsruhe University for during completion of this work. [99]{} H. N. Brown et al., hep-ex/0102017, Phys. Rev. Lett. 2227. F. J. Yndurain, A. Czarnecki and Marchiano, U. and Nath, hep-ph/0102157;\ Ibrahim, U. Chattopadhyay, P. Nath hep-ph/0102324. L. Everett, G. L. Kane, Rigolin and L.Wang, hep

relatively light sparticles, If the preseNt devIatIon FrOm thE SM oF $2.6\sigma$ persists. aT larGe ${\mbox{$ \tan\beta~ $}}$ a global fiT inclUdINg boTH $b\To X_s\gAmma$ and $A_\Mu$ AS WelL aS tHe pReSEnT HiggS liMit of 113.5 Gev leaves a quIte LaRge region in tHE CmSSM parameTer Space. Here we lEft The triLiNeaR CouplIng To be a Free paRAmeter, Which affeCtS Both thE higgs liMIT cOnstRaint and the [$b \to X_s \gAMmA~ $]{}Constraint, but iN opposItE WaYS, So tHat The preferrEd RegioN Is similAR fOR THe nO-Scale models wiTh $A_0=0$ and modelS WhiCh leavE $A_0$ FreE. the 95% lowEr limIt ON $m_{1/2}$ iS $ 300$ GeV (see Figs. \[F6\]+\[f7\]), whIch implieS that tHE lighteST chargiNo (neutRalIno) Is abOVe 240(120) gev. ThE 95% uPPer LImIt oN $M_{1/2}$ is DeterminEd By The loWer lIMIT On $a_\mU^{SUsY}$ anD therEfor depends on (See fig. \[f2\]). fOr =35(50) oNe finDs $m_{1/2}\leQ 610(720)$ GeV, WhIch imPlies tHat thE lIghtest chargino Is beLow 500(590) GeV and The LiGhtEsT neutRAlino iS beLow 260(310) geV. AcknOwledgeMEntS {#aCKNOwLedgements.unnumberEd} ================ d.k. WoUld like tO thank THe heISenberg-LAnDau progRAMme, RFbR grANt \# 99-02-16650 And DFG grAnt \# 436/RUS/113/626 FOr FiNancial SuPport aNd The karLsruhE univErsity For hospiTalitY During completiON of this work. [99]{} H. N. bRoWN Et AL., hep-Ex/0102017, PHys. Rev. Lett. [**86**]{} (2001) 2227. F. j. YndURain, Hep-pH/0102231. a. CZarNEcki aNd W. MaRcHIaNO, hep-ph/0102122. U. ChattopadhyaY aNd P. NatH, hep-pH/0102157;\ T. Ibrahim, U. ChaTtopadhyay, p. nATh hep-ph/0102324. L. everETt, g. l. Kane, S. Rigolin aNd L.WaNg, hep

relatively light sparticl es, if the pres ent de vi atio n fr om the SM of $ 2 .6\s igma$ persists. At la rge $ {\ m box{ $ \ tan\b eta~ $} } $a glo ba lfit i n cl uding bo th $b\t o X_s\gamm a$an d $a_\mu$ as we ll as thepre sent Higgs l imi t of 1 13 .5G eV le ave s a q uite l a rge re gion in t he CMSSMp aramete r sp ace. Here we left the tr i linear couplin g to b ea f r e e p ara meter, whi ch affe c ts both th e H igg s limit constr aint and th e [$ b \to X_s \gamma ~ $]{ }c o nst raint, butin o pposite w ays, s o that t h e prefe rred r egi onis s i mi la r f or the no -sc a lemodels w it h$A_0= 0$ a n d m odel s w hich leav e $A_0$ free. T he 9 5 % l owerlimit on$m _{1/2 }$ is$ 300 $GeV (see Figs.\[f6 \]+\[f7\] ),wh ich i mplie s thatthe li ghtestchargin o (n eu t r a li no) is above 240(1 20 ) Ge V. The 9 5% upp e rli m it on $m _{ 1/2 }$ i s deter mine d b y the lo wer li m it o n $a_\m u^ {SUSY} $and th erefo r dep ends o n (see F ig. \ [ f2\]). For =35 ( 50) one finds $m _ { 1/ 2 }\le q 6 10(720)$ Ge V, w h ichimpl i es th a t the ligh te s tc hargino is below 50 0( 590) G eV an d the lightes t neutrali n o is below 260 ( 31 0 ) GeV. Acknow ledge ments {#ac k nowledge ments .unnumbe red} ==== = = ======== == D. K.wou l d l ike to thankt h e He is enberg- Lan dau Pro gra mme , R FBR g rant \# 9 9-02-166 50 a nd D FGgrant \# 436/R US /11 3/ 626 forf inanci al su ppor tan d th e Karls r uh e Univ er si ty f orho spita lity dur ing com pletion o f t h is w or k. [99]{ } H. N. Brown e t al., hep -e x/0 102017 , Phys. Re v. Lett. [**86**]{} (20 0 1) 2227 . F . J.Yndu rain, hep -ph /01022 31. A. Cza rnecki andW. Ma r c hiano , he p-p h/ 0102122. U . Cha ttopa dh yayand P.Nath, hep-ph/01021 5 7;\ T. Ibrahim,U.Chat t o pa dhy a y, P.Na t h h e p -ph/0102324. L. Everett,G. L. Kane, S.R igo li n and L .Wang,hep

relatively_light sparticles,_if the present deviation_from the_SM_of $2.6\sigma$_persists. At_large ${\mbox{$ \tan\beta~_$}}$ a global_fit including both $b\to_X_s\gamma$ and $a_\mu$_as_well as the present Higgs limit of 113.5 GeV leaves a quite large region_in_the CMSSM_parameter_space._Here we left the trilinear_coupling to be a free_parameter, which_affects both the Higgs limit constraint and the_[$b_ \to _X_s \gamma~ $]{}constraint, but in opposite ways, so that_the preferred region is similar for_the no-scale models_with_$A_0=0$_and models which leave_$A_0$ free. The 95% lower limit on_$m_{1/2}$ is $ 300$ GeV (see_Figs. \[f6\]+\[f7\]), which implies that the lightest_chargino (neutralino) is above 240(120) GeV._The 95% upper limit on_$m_{1/2}$ is_determined by the lower limit_on $a_\mu^{SUSY}$ and_therefor depends_on (see Fig._\[f2\]). For =35(50) one finds $m_{1/2}\leq_610(720)$ GeV, which_implies that the lightest chargino is_below_500(590) GeV and_the_lightest_neutralino is_below 260(310) GeV. Acknowledgements_{#acknowledgements.unnumbered} ================ D.K._would like_to_thank the Heisenberg-Landau Programme, RFBR grant_\#_99-02-16650 and DFG grant \# 436/RUS/113/626 for_financial support and the_Karlsruhe_University for hospitality during_completion of this work. [99]{} H._N. Brown et al., hep-ex/0102017, Phys._Rev. Lett._[**86**]{} (2001)_2227. F. J. Yndurain, hep-ph/0102231. A. Czarnecki and W. Marchiano, hep-ph/0102122._U. Chattopadhyay and P. Nath, hep-ph/0102157;\ T._Ibrahim, U. Chattopadhyay, P._Nath hep-ph/0102324._L._Everett, G. L._Kane,_S. Rigolin_and L.Wang, hep

= +0.6 for all \[Fe/H\] $\leq$ –1.0. Since, at the abundance of And I derived here, $M_{V}$(RR) on our adopted scale is very close to 0.6, no adjustment of the MK90 result is required. To compare our distance to And I with that for M31 however, requires an M31 modulus that is on the same distance scale. There are two appropriate determinations. First, Pritchet & van den Bergh (1988) have determined the mean magnitudes of RR Lyrae stars in an M31 halo field. Adopting a mean abundance of \[Fe/H\] $\approx$ –1.0 for these stars (Pritchet & van den Bergh 1988), then yields an M31 distance modulus $(m-M)_{0}$ = 24.35 $\pm$ 0.15 on our adopted scale[^6]. Second, Mould & Kristian (1986) give I = 20.55 $\pm$ 0.15 mag for the apparent I magnitude of the red giant branch tip in their M31 halo field. Assuming a reddening of E(B–V) = 0.08 $\pm$ 0.02 for this field and noting that $M_{I}(RGB_{tip}) \approx$ –4.05 $\pm$ 0.10 for old metal-poor red giants (Da Costa & Armandroff 1990), then yields $(m-M)_{0}$ = 24.45 $\pm$ 0.20 for M31 on our adopted distance scale. Together these determinations suggest $(m-M)_{0}$ = 24.40 $\pm$ 0.13 for M31 or a distance of 760 $\pm$ 45 kpc. Thus, on the basis of these data, And I is apparently some 50 $\pm$ 50 kpc beyond M31 along the line-of-sight. The recent HST results of Ajhar [*et al. *]{}(1996) however, give a somewhat contradictory picture. Ajhar [*et al. *]{}(1996) give $V_{HB}$ values for three globular clusters and two field regions in M31 based on WFPC2 images in the F555W and F814W filters. Considering first the two metal-poor globular clusters (K105 and K219) where no adjustment is necessary to convert $V_{HB}$ to $V_{RR}$, and noting that we adopt a reddening E(B

= +0.6 for all \[Fe / H\ ] $ \leq$ – 1.0. Since, at the abundance of And I derived here, $ M_{V}$(RR) on our adopted scale is very close to 0.6, no allowance of the MK90 resultant role is required. To compare our distance to And I with that for M31 however, requires an M31 modulus that is on the same distance plate. There are two appropriate determinations. First, Pritchet & van den Bergh (1988) have decide the mean magnitudes of RR Lyrae star in an M31 halo field. assume a mean abundance of \[Fe / H\ ] $ \approx$ – 1.0 for these stars (Pritchet & van den Bergh 1988), then render an M31 distance modulus $ (m - M)_{0}$ = 24.35 $ \pm$ 0.15 on our adopted scale[^6 ]. Second, Mould & Kristian (1986) give I = 20.55 $ \pm$ 0.15 mag for the apparent I magnitude of the red giant outgrowth tip in their M31 halo field. bear a reddening of E(B – V) = 0.08 $ \pm$ 0.02 for this field and noting that $ M_{I}(RGB_{tip }) \approx$ – 4.05 $ \pm$ 0.10 for old metallic element - poor red giants (Da Costa & Armandroff 1990), then yields $ (m - M)_{0}$ = 24.45 $ \pm$ 0.20 for M31 on our espouse distance scale. Together these determinations suggest $ (m - M)_{0}$ = 24.40 $ \pm$ 0.13 for M31 or a distance of 760 $ \pm$ 45 kpc. Thus, on the basis of these data, And I is apparently some 50 $ \pm$ 50 kpc beyond M31 along the line - of - sight. The recent HST results of Ajhar [ * et al.   * ] { } (1996) however, give a somewhat contradictory picture. Ajhar [ * et al.   * ] { } (1996) move over $ V_{HB}$ values for three globular bunch and two sphere region in M31 based on WFPC2 images in the F555W and F814W filters. Considering first the two metal - poor ball-shaped clusters (K105 and K219) where no adjustment is necessary to convert $ V_{HB}$ to $ V_{RR}$, and noting that we adopt a reddening E(B

= +0.6 vor all \[Fe/H\] $\leq$ –1.0. Since, au the abundance oy And I derivsd here, $O_{V}$(RR) on our adopted scale is vwry coose to 0.6, no adjustment of the MN90 result is cequired. To compacs our dlftandc to Cnv I with that fpr M31 howevar, requires an M31 mldulus that is on the same distance scale. Yhfre are two apkroprpaee dsnevminations. First, Pritchet & van dsn Bergi (1988) have determimed the mean magnitudes of RR Pyrae stars in an L31 halo fiele. Adjpting a mean abundance of \[Fe/H\] $\apprkx$ –1.0 for these stars (Pritchet & vxn deu Bergh 1988), thgu yifnds an M31 diwtancv modulus $(m-M)_{0}$ = 24.35 $\pm$ 0.15 on our adppted scale[^6]. Segond, Kouod & Kristian (1986) give I = 20.55 $\pm$ 0.15 mag for the wpparent H jagnitude of the eee giatt bsancf tio ih vhejr M31 hwlo field. Asshming a redeening of E(B–V) = 0.08 $\pm$ 0.02 fjg this field znd noeigg that $M_{I}(RGB_{tip}) \approx$ –4.05 $\pm$ 0.10 for old meual-pokr red giants (Da Costa & Armandroff 1990), then yiepds $(m-M)_{0}$ = 24.45 $\[m$ 0.20 for M31 on our adopted distance scale. Together dhese aettrninatkind suggest $(m-M)_{0}$ = 24.40 $\pm$ 0.13 for M31 or a distance of 760 $\pm$ 45 kkc. Nhus, on the basis of these dstw, Sgd I is appargntly some 50 $\pm$ 50 kpc beyond M31 along jhe libe-of-sight. Uhe rrcent HST results of Ajhar [*wt al. *]{}(1996) howeveg, gice a somewhat contxadictory pieture. Sjhar [*et al. *]{}(1996) give $V_{HB}$ values yor thdee globulag clustera and two field rdgipnv in M31 based on WFPC2 imagef in the H555W anb F814W filgers. Consiqering firdt thc two metal-poor glohular cnusters (K105 wnd K219) where no adjustment is nerxssary to confest $N_{HB}$ to $V_{RX}$, and koting that we wdopt a reddennng E(B

= +0.6 for all \[Fe/H\] $\leq$ –1.0. the of And derived here, $M_{V}$(RR) very to 0.6, no of the MK90 is required. To compare our distance And I with that for M31 however, requires an M31 modulus that is the same distance scale. There are two appropriate determinations. First, Pritchet & van Bergh have the magnitudes of RR Lyrae stars in an M31 halo field. Adopting a mean abundance of \[Fe/H\] –1.0 for these stars (Pritchet & van den 1988), then yields an distance modulus $(m-M)_{0}$ = 24.35 0.15 our adopted Second, & (1986) give I 20.55 $\pm$ 0.15 mag for the apparent I magnitude of the red giant branch tip in their halo field. reddening of = $\pm$ for this field that $M_{I}(RGB_{tip}) \approx$ –4.05 $\pm$ 0.10 red giants (Da Costa & Armandroff 1990), then $(m-M)_{0}$ = $\pm$ 0.20 for M31 on our distance scale. Together these determinations suggest $(m-M)_{0}$ = $\pm$ 0.13 for M31 or a distance of 760 $\pm$ 45 kpc. Thus, on the these data, And I apparently some 50 50 beyond along line-of-sight. The HST results of Ajhar [*et al. *]{}(1996) however, give a somewhat picture. Ajhar [*et al. *]{}(1996) give $V_{HB}$ values for three and field regions in based on WFPC2 images the and F814W filters. Considering two globular K219) no is necessary to convert to $V_{RR}$, and noting that adopt a reddening E(B

= +0.6 for all \[Fe/H\] $\leq$ –1.0. Since, at the abuNdance of AnD I derIveD heRe, $m_{V}$(RR) On ouR adopted scale iS Very Close to 0.6, no adjustment of tHe MK90 rEsULt is REqUired. to compaRE oUR DisTaNcE to anD i wIth thAt fOr M31 howeVer, requireS an m31 mOdulus that is ON tHe same distAncE scale. There aRe tWo apprOpRiaTE deteRmiNatioNs. FirsT, pritchEt & van den BErGH (1988) have dETermineD THe Mean Magnitudes of RR LyrAE sTArs in an M31 halo fiEld. AdoPtINg A MEan AbuNdance of \[Fe/h\] $\aPprox$ –1.0 FOr these STaRS (pRitCHet & van den BergH 1988), then yields AN M31 dIstancE mOduLUs $(m-M)_{0}$ = 24.35 $\pm$ 0.15 On our AdOPteD scale[^6]. SeconD, MouLd & KristiaN (1986) give I = 20.55 $\PM$ 0.15 mag for THe apparEnt I maGniTudE of tHE rEd GiaNt BRanCH tIp iN TheIr M31 halo fIeLd. assumIng a REDDEninG of e(B–V) = 0.08 $\pM$ 0.02 for tHis field and noTinG thaT $m_{I}(RgB_{tip}) \ApproX$ –4.05 $\pm$ 0.10 fOr Old meTal-pooR red gIaNts (Da Costa & ArmanDrofF 1990), then yielDs $(m-m)_{0}$ = 24.45 $\pM$ 0.20 foR M31 On our ADopted DisTanCe scale. togetheR TheSe DETErMinations suggest $(m-M)_{0}$ = 24.40 $\Pm$ 0.13 FOR M31 Or a distaNce of 760 $\pM$ 45 KpC. THUs, on the bAsIs oF theSE Data, ANd I iS ApParently Some 50 $\pm$ 50 KPc BeYond M31 alOnG the liNe-Of-sIghT. The rECent hST resUlts of AjHar [*et AL. *]{}(1996) however, give a sOMewhat contradICtORY pICturE. AjHar [*et al. *]{}(1996) give $v_{HB}$ vALues For tHReE glOBular ClustErS AnD Two field regions in M31 bAsEd on WFpC2 imaGes in the F555W and f814W filters. CONSIdering fIrst THe TWo metal-poor gloBular Clusters (K105 aND K219) where nO adjuStment is Necessary TO Convert $V_{hB}$ tO $V_{Rr}$, anD noTINg That we adopt a rEDDeniNg e(B

= +0.6 for all \[Fe/H\] $ \leq$ –1.0 . Sin ce, at t he a bund ance of And Id eriv ed here, $M_{V}$(RR) o n our a d opte d s caleis very cl o s e t o0. 6,no ad justm ent of the MK90 resu ltis required. T ocompare ou r d istance to A ndI with t hat for M 31howev er, re q uiresan M31 mo du l us tha t is ont h esame distance scale.T he r e are two appr opriat ed et e r min ati ons. First ,Pritc h et & va n d e n Ber g h (1988) have determined the meanma gni t udes o f RRLy r aestars in an M31 halo fie ld. Ad o pting a mean ab undanc e o f \ [Fe/ H \] $ \ap pr o x$– 1. 0 f o r t hese sta rs ( Pritc het& v a n de n B ergh 1988 ), then yield s a n M3 1 di stanc e mod ulus $ (m-M) _{0}$= 24. 35 $\pm$ 0.15 onouradopted s cal e[ ^6] .Secon d , Moul d & Kr istian(1986)g ive I = 20 .55 $\pm$ 0.15 mag f o r t he appar ent Im ag ni t ude of t he re d gi a n t bra ncht ip in thei r M31h al ofield.As suming a re dde ningo f E( B–V) = 0.08 $\ pm$ 0 . 02 for this fi e ld and noting th a t $ M _{I} (RG B_{tip}) \a ppro x $ –4 .05$ \p m$0 .10 f or ol dm et a l-poor red giants ( Da Costa & Ar mandroff 1990 ), then yi e l d s $(m-M) _{0} $ = 24.45 $\pm$ 0. 20 fo r M31 on o u r adopte d dis tance sc ale. Toge t h er these de ter min ati o n ssuggest $(m-M ) _ {0}$ = 24.40$\p m$ 0.13 fo r M 31oradistanceof 760 $ \p m$ 4 5kpc . Thu s , on the b asi softhese data,And I isap pa r ent ly some 50 $ \pm$ 5 0kpcbey on d M31 alo n g t he line -of-sight . T he r ec en t HST r esults of Ajh ar [*et al.*] {}( 1996)h o wever, g ive a somewhat contradi c tory pi ctu re. A jhar [*et al.  *] {}(199 6)g ive $V _{HB}$ valu es fo r three g lo bul ar clustersa n d t wo fi el d re gions i n M31 based on WFP C 2 i mages in theF55 5W a n d F 814 W f i lte rs . Co n s idering first t he two met al - po or globula r cl us ters (K 105 and K219 ) whereno adjust ment is n ec essa r y to convert $ V_{HB}$to $V_{RR } $, an d n oting th at wead opt a re ddenin g E( B

=_+0.6 for_all \[Fe/H\] $\leq$ –1.0._Since, at_the_abundance of_And_I derived here,_$M_{V}$(RR) on our_adopted scale is very_close to 0.6,_no_adjustment of the MK90 result is required. To compare our distance to And I with_that_for M31_however,_requires_an M31 modulus that is_on the same distance scale._There are_two appropriate determinations. First, Pritchet & van den_Bergh_(1988) have determined_the mean magnitudes of RR Lyrae stars in an_M31 halo field. Adopting a mean_abundance of \[Fe/H\]_$\approx$_–1.0_for these stars (Pritchet_& van den Bergh 1988), then_yields an M31 distance modulus $(m-M)_{0}$_= 24.35 $\pm$ 0.15 on our adopted_scale[^6]. Second, Mould & Kristian (1986)_give I = 20.55 $\pm$_0.15 mag_for the apparent I magnitude_of the red_giant branch_tip in their_M31 halo field. Assuming a reddening_of E(B–V) =_0.08 $\pm$ 0.02 for this field_and_noting that $M_{I}(RGB_{tip})_\approx$_–4.05_$\pm$ 0.10_for old metal-poor_red_giants (Da_Costa_& Armandroff 1990), then yields $(m-M)_{0}$_=_24.45 $\pm$ 0.20 for M31 on our_adopted distance scale. Together_these_determinations suggest $(m-M)_{0}$ =_24.40 $\pm$ 0.13 for M31_or a distance of 760 $\pm$_45 kpc._Thus, on_the basis of these data, And I is apparently some 50_$\pm$ 50 kpc beyond M31 along_the line-of-sight. The recent HST_results of_Ajhar_[*et al. *]{}(1996) however,_give_a somewhat_contradictory picture. Ajhar [*et al. *]{}(1996) give $V_{HB}$_values for_three globular clusters and two field_regions in M31 based_on_WFPC2 images in the F555W and_F814W filters. Considering first the two_metal-poor globular clusters (K105 and_K219)_where_no adjustment is necessary to_convert $V_{HB}$ to $V_{RR}$, and noting_that we adopt_a reddening E(B

as implications for future RG/EB studies. Observations {#data} ============ *Kepler* light curves {#kepler} --------------------- Our light curves are from the *Kepler* Space Telescope in long-cadence mode (one data point every 29.4 minutes), and span 17 quarters—roughly four years—with only occasional gaps. These light curves are well-suited for red giant asteroseismology, as main sequence stars with convective envelopes oscillate too rapidly to be measured with *Kepler* long-cadence data. When studying long-period eclipsing binaries, it is important to remove instrumental effects in the light curve while preserving the astrophysically interesting signal. In this work, we prioritize preserving eclipses. Our detrending algorithm uses the simple aperture photometry (SAP) long-cadence *Kepler* data for quarters 0–17. First, any observations with NaNs are removed, and observations from different quarters are put onto the same median level so that the eclipses line up. The out-of-eclipse portions of the light curve are flattened, which removes any out-of-eclipse variability. For eclipse modeling, we use only the portions of the light curve that lie within one eclipse duration of the start and end of each eclipse. This differs from the light curve processing needed for asteroseismology, [which typically]{} “fills” the eclipses to minimize their effect on the power spectrum [@gau14]. The processed light curve is presented in Figure \[fig:keplerfig\]. The top panel shows the entire detrended light curve, while the middle and bottom panels indicate the regions near each eclipse used in this work. We adopt the convention that the “primary” eclipse is the deeper of the two, when Star 1 is eclipsing Star 2. The geometry of the system creates partial eclipses with different depths due to similarly-sized stars in an eccentric orbit viewed with an inclination less than 90 degrees. [For comparison, we present the detrended light curve with eclipses removed in Figure \[fig:lcfig2\]. The system shows out-of-eclipse photometric modulations on the order of $2 \%$.]{} {width="1.8\columnwidth"} {width="1.7\columnwidth"} Ground-based

as implications for future RG / EB studies. Observations { # data } = = = = = = = = = = = = * Kepler * light curve { # kepler } --------------------- Our faint curves are from the * Kepler * Space Telescope in long - cadence modality (one data point every 29.4 minute), and cross 17 quarters — roughly four years — with only occasional col. These light curves are well - suit for crimson giant asteroseismology, as main sequence stars with convective envelopes hover too rapidly to be measured with * Kepler * farseeing - cadence data. When study long - time period eclipsing binaries, it is important to remove instrumental impression in the light curve while preserving the astrophysically interesting signal. In this work, we prioritize preserving eclipse. Our detrending algorithm use the simple aperture photometry (SAP) long - cadence * Kepler * data for quarters 0–17. First, any observations with NaNs are removed, and observations from unlike quarter are put onto the like medial level so that the eclipses line up. The out - of - eclipse portions of the clean curvature are flattened, which removes any knocked out - of - eclipse variability. For eclipse modeling, we use only the portions of the light bend that lie within one eclipse duration of the start and end of each eclipse. This differs from the clean curvature processing needed for asteroseismology, [ which typically ] { } “ fills ” the eclipses to minimize their effect on the world power spectrum [ @gau14 ]. The processed light curve is presented in Figure \[fig: keplerfig\ ]. The top panel shows the entire detrended light bend, while the middle and bottom gore indicate the regions near each eclipse use in this work. We assume the convention that the “ chief ” eclipse is the deeper of the two, when Star 1 is eclipse Star 2. The geometry of the system creates partial eclipses with different depth due to similarly - sized stars in an bizarre orbit viewed with an inclination less than 90 degrees. [ For comparison, we confront the detrended unaccented curve with eclipses remove in Figure \[fig: lcfig2\ ]. The system shows out - of - eclipse photometric modulation on the order of $ 2 \%$. ] { } ! [ image](fig1.png){width="1.8\columnwidth " } ! [ image](fig2 - eps - converted - to.pdf){width="1.7\columnwidth " } Ground - based

as implications for future RG/EB studies. Observatimns {#dafa} ============ *Kepler* light curves {#kepler} --------------------- Our lighv cueves qre from the *Kepler* Spxce Telesbope in ling-cedence mode (one vzta poikc evedn 29.4 miuuves), and span 17 qoarters—roughny four years—whtf lnly occasional gaps. These light cuwves arr aell-suited for red dianf asteroseismology, as main sequencs stars with convectove envelopes oscillate tol raoidly to be measurfd with *Kepoer* jing-cadence dxta. When stlbying long-pgriod eclipsing binaries, it is ioportcnt to remoce indjrumental efhects pn the light glrve whhle prexerving the asbrophbsicqlly interesting signel. In this work, we ptioritize [rzserving eclipses. Our dwtrenging algueitfm hsxs fhe silplx aperture lhotometry (WAP) long-cadence *Kepkew* data for quadters 0–17. Fyrst, any observations with NaNs are remmves, and observations from different quarters ate put ontj the same median level so that the eclipses line up. Tie ouu-on-ccliowe portions of the light curve are flattened, whydh rvmoves any out-of-eglipse variability. Flr gclipse modelivg, we bae only the portions of the lighr curve trat kie within one eclipse durarion of the frart and end of eaeh eclipse. Tkis digfers from the light curve pxocessjng needed vor asterkreismology, [which tyiicanly]{} “fills” the eclipses to iinimize vheir effect on jhe powqr spectrul [@gau14]. Bve processed light curvg is psesented ij Figure \[fig:keplerfig\]. The top peiel shows the ettige detrenbed linht curve, while the middle anb bottom paneus indicatv the regmons near eash eclipse usaf in this wock. We ado[t tye cinventiuv that the “prikary” ecliist is the dweper of the two, wmen Sjad 1 is eclipsing Stqr 2. The geometru ow tre sbstem wreates parthal dclklses dith differcnt depyhs due to similarly-vizes stars in an eccemtvic orbit viewed rith an inclimation less than 90 fegrexs. [For compsrifon, we present the detrended ljght curvf wlth eclipses wemoyed yn Figure \[yig:lcfig2\]. The system shows out-of-eclipse piotometric modulations in the order of $2 \%$.]{} {width="1.8\cplumnridth"} {width="1.7\columnwidth"} Nround-based

as implications for future RG/EB studies. Observations *Kepler* curves {#kepler} Our light curves Telescope long-cadence mode (one point every 29.4 and span 17 quarters—roughly four years—with occasional gaps. These light curves are well-suited for red giant asteroseismology, as main stars with convective envelopes oscillate too rapidly to be measured with *Kepler* long-cadence When long-period binaries, is important to remove instrumental effects in the light curve while preserving the astrophysically interesting signal. this work, we prioritize preserving eclipses. Our detrending uses the simple aperture (SAP) long-cadence *Kepler* data for 0–17. any observations NaNs removed, observations from different are put onto the same median level so that the eclipses line up. The out-of-eclipse portions of light curve which removes out-of-eclipse For modeling, we use portions of the light curve that eclipse duration of the start and end of eclipse. This from the light curve processing needed asteroseismology, [which typically]{} “fills” the eclipses to minimize effect on the power spectrum [@gau14]. The processed light curve is presented in Figure \[fig:keplerfig\]. panel shows the entire light curve, while middle bottom indicate regions near eclipse used in this work. We adopt the convention that the eclipse is the deeper of the two, when Star 1 Star The geometry of system creates partial eclipses different due to similarly-sized stars eccentric viewed less 90 [For comparison, we present detrended light curve with eclipses in Figure \[fig:lcfig2\]. The on the order of $2 \%$.]{} {width="1.8\columnwidth"} {width="1.7\columnwidth"}

as implications for future RG/eB studies. OBservAtiOns {#DaTa} ============ *KePler* Light curves {#kepLEr} --------------------- OuR light curves are from the *kepleR* SPAce TELeScope In long-cADeNCE moDe (OnE daTa POiNt eveRy 29.4 mInutes), aNd span 17 quarTerS—rOughly four yeARs—With only ocCasIonal gaps. TheSe lIght cuRvEs aRE well-SuiTed foR red giANt asteRoseismolOgY, As main SEquence STArS witH convective enveloPEs OScillate too rapIdly to Be MEaSURed WitH *Kepler* lonG-cAdencE Data. WheN StUDYIng LOng-period ecliPsing binariES, it Is impoRtAnt TO removE instRuMEntAl effects in The lIght curve While pREservinG The astrOphysiCalLy iNterEStInG siGnAL. In THiS woRK, we PrioritiZe PrEservIng eCLIPSes. OUr dEtreNding Algorithm uses The SimpLE apErturE photOmetRy (sAP) loNg-cadeNce *KePlEr* data for quarteRs 0–17. FiRst, any obsErvAtIonS wIth NanS are reMovEd, aNd obserVations FRom DiFFEReNt quarters are put onTo THE sAme mediaN level SO tHaT The eclipSeS liNe up. tHE out-oF-eclIPsE portionS of the LIgHt Curve arE fLattenEd, WhiCh rEmoveS Any oUt-of-ecLipse varIabilITy. For eclipse moDEling, we use onlY ThE POrTIons Of tHe light curvE thaT Lie wIthiN OnE ecLIpse dUratiOn OF tHE start and end of each eClIpse. ThIs difFers from the liGht curve prOCESsing neeDed fOR aSTeroseismology, [Which Typically]{} “fILls” the ecLipseS to minimIze their eFFEct on the PowEr sPecTruM [@GAu14]. the processed lIGHt cuRvE is presEntEd in FigUre \[Fig:KepLerFiG\]. The top paNel shows ThE eNtIrE deTrendED light cuRvE, whIlE thE middLE and boTtom pAnelS iNdICatE the regIOnS NEar eAcH eClipSe uSeD in thIs woRK. We Adopt thE conventiOn tHAt thE “pRiMary” eclIpse is the deepEr Of the two, whEn staR 1 is eclIPSing Star 2. the geometry of the system cREates paRtiAl eclIpseS with diffEreNt deptHs dUE to simIlarly-Sized StArs IN An eccENTrIc oRbIt viewed wiTH An iNclinAtIon lEss than 90 Degrees. [For comparisON, we Present the detRenDed lIGHt CurVE wITh eClIPseS REmoved in Figure \[fIg:lcfig2\]. The SySTeM shows out-oF-EclIpSe photoMetric mOdulaTIons on tHe order of $2 \%$.]{} {WidtH="1.8\COluMnwidth"} {wIDtH="1.7\coluMnwIdth"} GrOuNd-bAsed

as implications for futur e RG/EB st udies . Obs er vati ons{#data} ====== = ==== = *Kepler* light curv es {# ke p ler} -- ----- ------- - -- - - -- O ur li gh t c urves ar e fromthe *Keple r*Sp ace Telescop e i n long-cad enc e mode (onedat a poin teve r y 29. 4 m inute s), an d span17 quarte rs — roughl y four y e a rs —wit h only occasional ga p s. These light curve sa re w ell -su ited for r ed gian t astero s ei s m o log y , as main seq uence stars wit h conv ec tiv e envel opesos c ill ate too rap idly to be me asured with *K e pler* l ong-ca den cedata . Wh enst u dyi n glon g -pe riod ecl ip si ng bi nari e s , it i s i mpor tantto remove ins tru ment a l e ffect s intheli ght c urve w hilepr eserving the as trop hysically in te res ti ng si g nal. I n t his work,we prio r iti ze p r es erving eclipses. O ur d et rendingalgori t hm u s es the s im ple ape r t ure p hoto m et ry (SAP) long- c ad en ce *Kep le r* dat afor qu arter s 0–1 7. Fir st, anyobser v ations with Na N s are removed , a n d o b serv ati ons from di ffer e nt q uart e rs ar e putontoth e s a me median level soth at the ecli pses line up. The out-o f - e clipse p orti o ns of the light c urveare flatte n ed, whic h rem oves any out-of-e c l ipse var iab ili ty. Fo r ec lipse modelin g , weus e onlythe portio nsofthe li gh t curve t hat liewi th in o neeclip s e durati on of t hestart and en d ofeach e cl i pse . Thisd if f e rs f ro mthelig ht curv e pr o ces sing ne eded forast e rose is mo logy, [ which typical ly ]{} “fills ”the eclip s e s to min imize their effect on t h e power sp ectru m [@ gau14]. The proce sse d light curve is p re sen t e d inF i gu re\[ fig:kepler f i g\] . The t op p anel sh ows the entire det r end ed light curv e,whil e th e m i dd l e a nd bot t o m panels indica te the reg io n snear eache cli ps e usedin this work . We ado pt the co nventionth at t h e “p rimary” ec lipse is the deep e r oft he two, wh en Sta r1 i s ecl ipsing Sta r 2.The ge om etry o f the s ystem cr eates partial eclipseswith d iffer ent depths d uet o s imilarly- size d stars in an ec centr ico rbitview e dwit h an i ncli n ation les s t han 9 0degrees. [F o r com paris on, we pre sent the detrended li g ht curve withecli p s esrem o vedin Figure \[fig: lcf ig 2 \ ]. The s ys tem shows o ut-of-ec li p se ph otomet ric mo dulatio n s o n the o rder of $2 \%$.] {} ! [ image]( fi g1 . png){w idth =" 1.8\co lumnwi d th"} {w id th="1.7 \ colu mnwidth"} Ground-bas ed

as_implications for_future RG/EB studies. Observations {#data} ============ *Kepler*_light curves_{#kepler} --------------------- Our_light curves_are_from the *Kepler*_Space Telescope in_long-cadence mode (one data_point every 29.4_minutes),_and span 17 quarters—roughly four years—with only occasional gaps. These light curves are well-suited_for_red giant_asteroseismology,_as_main sequence stars with convective_envelopes oscillate too rapidly to_be measured_with *Kepler* long-cadence data. When studying long-period eclipsing binaries,_it_is important to_remove instrumental effects in the light curve while preserving_the astrophysically interesting signal. In this_work, we prioritize_preserving_eclipses._Our detrending algorithm uses_the simple aperture photometry (SAP) long-cadence_*Kepler* data for quarters 0–17. First,_any observations with NaNs are removed, and_observations from different quarters are put_onto the same median level_so that_the eclipses line up. The_out-of-eclipse portions of_the light_curve are flattened,_which removes any out-of-eclipse variability. For_eclipse modeling, we_use only the portions of the_light_curve that lie_within_one_eclipse duration_of the start_and_end of_each_eclipse. This differs from the light_curve_processing needed for asteroseismology, [which typically]{} “fills”_the eclipses to minimize_their_effect on the power_spectrum [@gau14]. The processed light curve_is presented in Figure \[fig:keplerfig\]. The_top panel_shows the_entire detrended light curve, while the middle and bottom panels indicate_the regions near each eclipse used_in this work. We_adopt the_convention_that the “primary”_eclipse_is the_deeper of the two, when Star 1_is eclipsing_Star 2. The geometry of the_system creates partial eclipses_with_different depths due to similarly-sized stars_in an eccentric orbit viewed with_an inclination less than 90_degrees._[For_comparison, we present the detrended_light curve with eclipses removed in_Figure \[fig:lcfig2\]. The_system shows out-of-eclipse photometric modulations on the_order_of $2 \%$.]{} {width="1.8\columnwidth"} {width="1.7\columnwidth"} Ground-based

alpha}_a)} \cdot \frac{p(X_i | A = 0, X_i; \widehat{\alpha}_m)}{p(M_i | A = 1, X_i; \widehat{\alpha}_m)} \ Y_i - \frac{ \mathbb I(A_i = 0)}{ p(A_i = 0 | X_i; \widehat{\alpha}_a)} \ Y_i \right). \label{eqn-IPW} \end{aligned}$$ ]{}Since solving the constrained MLE problem using this estimator entails only restricting parameters of $A$ and $M$ models, predicting a new instance is done using $\mathbb{E}[Y | X] = \sum_{A, M} \mathbb{E}[Y | A, M, X] \ p(M | A, X) \ p(A | X).$ ***Mixed approach***: The third way of computing the NDE is using $A$ and $Y$ models. In this estimator, we fit the models $p(A | X; \alpha_a)$ and $\mathbb{E}[Y | A, M, X; \alpha_y]$ by MLE, as usual, and combine the edge G-formula and IPW in the following way: [ $$\begin{aligned} \mathbb{P}_n \left(\frac{\mathbb I(A_i = 0)}{p(A_i = 0 | X_i ; \widehat{\alpha}_a)} \ \mathbb{E}[Y_i | A = 1, M_i, X_i; \widehat{\alpha}_y] - \mathbb{E}[Y_i | A = 0, M_i; \widehat{\alpha}_y] \right), \label{eqn-mixed} \end{aligned}$$ ]{} Since solving the constrained MLE problem using this estimator entails only restricting parameters of $A$ and $Y$ models, predicting a new instance is done using $\mathbb{E}[Y | M, X] = \sum_{A} \mathbb{E}[Y | A, M, X] \ \frac{p(M | A, X) \cdot p(A | X)}{\sum_A p(M | A, X) \cdot p(A | X)}.$ ***Aug

alpha}_a) } \cdot \frac{p(X_i | A = 0, X_i; \widehat{\alpha}_m)}{p(M_i | A = 1, X_i; \widehat{\alpha}_m) } \ Y_i - \frac { \mathbb I(A_i = 0) } { p(A_i = 0 | X_i; \widehat{\alpha}_a) } \ Y_i \right). \label{eqn - IPW } \end{aligned}$$ ] { } Since solving the constrained MLE problem using this calculator fee-tail only restricting argument of $ A$ and $ M$ model, predicting a new example is dress using $ \mathbb{E}[Y | ten ] = \sum_{A, M } \mathbb{E}[Y | A, M, X ] \ p(M | A, X) \ p(A | X).$ * * * Mixed approach * * *: The third room of computing the NDE is using $ A$ and $ Y$ model. In this estimator, we fit the models $ p(A | X; \alpha_a)$ and $ \mathbb{E}[Y | A, M, X; \alpha_y]$ by MLE, as usual, and combine the edge gram - formula and IPW in the following way: [ $ $ \begin{aligned } \mathbb{P}_n \left(\frac{\mathbb I(A_i = 0)}{p(A_i = 0 | X_i; \widehat{\alpha}_a) } \ \mathbb{E}[Y_i | A = 1, M_i, X_i; \widehat{\alpha}_y ] - \mathbb{E}[Y_i | A = 0, M_i; \widehat{\alpha}_y ] \right), \label{eqn - mixed } \end{aligned}$$ ] { } Since clear the constrained MLE problem using this estimator entail only restricting parameters of $ A$ and $ Y$ models, predicting a newfangled instance is done using $ \mathbb{E}[Y | M, X ] = \sum_{A } \mathbb{E}[Y | A, M, X ] \ \frac{p(M | A, X) \cdot p(A | X)}{\sum_A p(M | A, X) \cdot p(A | X)}.$ * * * Aug

alpja}_a)} \cdot \frac{p(X_i | A = 0, X_i; \widehat{\alpha}_m)}{p(M_i | A = 1, X_i; \wisehat{\alpfa}_m)} \ Y_i - \frac{ \mathbb I(A_i = 0)}{ p(A_i = 0 | X_i; \widehat{\alpha}_a)} \ Y_k \rigjt). \lavel{ewb-IPW} \env{zligned}$$ ]{}Since dolvnnj the constraingd MLE problam using this asgilator entails only restricting paraieters pf $A$ and $M$ modelf, prtdistinf a new instance is done using $\matgbb{E}[Y | Q] = \sum_{A, M} \mathbb{E}[U | A, M, X] \ p(M | A, X) \ p(A | X).$ ***Mided wpproach***: The third way of comkhtigt the NDE is using $A$ and $Y$ models. Jn this estimator, we fit the moaels $'(A | X; \alpha_q)$ qnd $\kathbb{E}[Y | A, M, X; \wlpha_y]$ by MLC, as usgal, and combine the ecge G-firmula and IPW in the following way: [ $$\bedin{aligneg} \mathbb{P}_n \ledt(\drac{\mdthbt I(A_k = 0)}{p(X_i = 0 | X_j ; \widfhav{\alpha}_a)} \ \mafhbb{E}[Y_i | A = 1, M_i, X_i; \widehat{\alphs}_y] - \mathbb{E}[Y_i | Z = 0, M_i; \wydehat{\alpha}_y] \right), \label{eqn-mixed} \ens{aligned}$$ ]{} Since solving the constrained MLE kroblem usyng this estimator entails only restricting paramaters uf $C$ and $H$ mldels, predicting a new instance is done using $\jauhbn{E}[Y | M, X] = \sum_{A} \mcthbb{E}[Y | A, M, X] \ \grwc{l(I | A, X) \cdot p(X | X)}{\sum_A l(M | A, X) \cdot p(A | X)}.$ ***Wug

alpha}_a)} \cdot \frac{p(X_i | A = 0, | = 1, \widehat{\alpha}_m)} \ Y_i 0)}{ = 0 | \widehat{\alpha}_a)} \ Y_i \label{eqn-IPW} \end{aligned}$$ ]{}Since solving the constrained problem using this estimator entails only restricting parameters of $A$ and $M$ models, a new instance is done using $\mathbb{E}[Y | X] = \sum_{A, M} \mathbb{E}[Y A, X] p(M A, X) \ p(A | X).$ ***Mixed approach***: The third way of computing the NDE is $A$ and $Y$ models. In this estimator, we the models $p(A | \alpha_a)$ and $\mathbb{E}[Y | A, X; by MLE, usual, combine edge G-formula and in the following way: [ $$\begin{aligned} \mathbb{P}_n \left(\frac{\mathbb I(A_i = 0)}{p(A_i = 0 | X_i ; \widehat{\alpha}_a)} \mathbb{E}[Y_i | 1, M_i, \widehat{\alpha}_y] \mathbb{E}[Y_i A = 0, \right), \label{eqn-mixed} \end{aligned}$$ ]{} Since solving problem using this estimator entails only restricting parameters $A$ and models, predicting a new instance is using $\mathbb{E}[Y | M, X] = \sum_{A} \mathbb{E}[Y A, M, X] \ \frac{p(M | A, X) \cdot p(A | X)}{\sum_A p(M | A, p(A | X)}.$ ***Aug

alpha}_a)} \cdot \frac{p(X_i | A = 0, X_i; \widehAt{\alpha}_m)}{p(M_I | A = 1, X_i; \wIdeHat{\AlPha}_m)} \ y_i - \frAc{ \mathbb I(A_i = 0)}{ p(A_i = 0 | x_I; \widEhat{\alpha}_a)} \ Y_i \right). \label{Eqn-IPw} \eND{aliGNeD}$$ ]{}SincE solvinG ThE COnsTrAiNed mLe PrOblem UsiNg this eStimator enTaiLs Only restrictINg Parameters Of $A$ And $M$ models, prEdiCting a NeW inSTance Is dOne usIng $\matHBb{E}[Y | X] = \sUm_{A, M} \mathbB{E}[y | a, M, X] \ p(M | A, x) \ P(A | X).$ ***MixeD APpRoacH***: The third way of comPUtINg the NDE is usinG $A$ and $Y$ MoDElS. iN thIs eStimator, we FiT the mODels $p(A | X; \ALpHA_A)$ And $\MAthbb{E}[Y | A, M, X; \alpHa_y]$ by MLE, as uSUal, And comBiNe tHE edge G-FormuLa ANd IpW in the follOwinG way: [ $$\begin{AligneD} \Mathbb{P}_N \Left(\fraC{\mathbB I(A_I = 0)}{p(A_I = 0 | X_i ; \wIDeHaT{\alPhA}_A)} \ \maTHbB{E}[Y_I | a = 1, M_i, x_i; \widehaT{\aLpHa}_y] - \maThbb{e}[y_I | a = 0, m_i; \wiDehAt{\alPha}_y] \rIght), \label{eqn-mIxeD} \end{ALigNed}$$ ]{} SiNce soLvinG tHe conStrainEd MLE PrOblem using this eStimAtor entaiLs oNlY reStRictiNG paramEteRs oF $A$ and $Y$ mOdels, prEDicTiNG A NeW instance is done usiNg $\MAThBb{E}[Y | M, X] = \suM_{A} \mathBB{E}[y | A, m, x] \ \frac{p(M | A, x) \cDot P(A | X)}{\sUM_a p(M | A, X) \Cdot P(a | X)}.$ ***aug

alpha}_a)} \cdot \frac{p(X _i | A = 0 , X_i ; \ wid eh at{\ alph a}_m)}{p(M_i | A =1, X_i; \widehat{\alph a}_m) }\ Y_i - \ frac{ \ m at h b b I (A _i =0) } {p(A_i =0 | X_i ; \widehat {\a lp ha}_a)} \ Y _ i \righ t). \label{ eqn -IPW} \ e nd{al ign ed}$$ ]{}Si n ce sol ving theco n strain e d MLE p r o bl em u sing this estimat o re ntails only re strict in g p a r ame ter s of $A$ a nd $M$m odels,p re d i c tin g a new instan ce is doneu sin g $\ma th bb{ E }[Y |X] =\s u m_{ A, M} \math bb{E }[Y | A,M, X]\ p(M |A , X) \p(A |X). $ ***M i xe dapp ro a ch* * *: Th e th ird wayof c omput ingt h e NDEisusin g $A$ and $Y$ mode ls. Int his esti mator , we f it th e mode ls $p (A | X; \alpha_a) $ an d $\mathb b{E }[ Y | A , M,X ; \alp ha_ y]$ by MLE , as us u al, a n d co mbine the edge G-f or m u la and IPW in th e f ol l owing wa y: [$$\b e g in{al igne d } \mat hbb{P} _ n \left (\ frac{\ ma thb b I (A_i= 0)} {p(A_i = 0 | X _i ;\ widehat{\alpha } _a)} \ \mathb b {E } [ Y_ i | A =1, M_i, X_i ; \w i deha t{\a l ph a}_ y ] -\math bb { E} [ Y_i | A = 0, M_i; \ wi dehat{ \alph a}_y] \right) , \la b e l {eqn-mix ed} \end{aligned}$ $ ]{} Since sol v ing theconst rained M LE proble m using th isest ima tor e nt ails only res t r icti ng parame ter s of $A $ a nd$Y$ mo de ls, predi cting ane win st anc e isd one usin g$\m at hbb {E}[Y | M, X ] = \ sum_ {A }\ mat hbb{E}[ Y | A , M, X ]\ \f rac {p (M |A, X ) \c dot p(A | X)}{\s um_ A p(M | A , X) \c dot p(A | X)} .$ ***Aug

alpha}_a)} \cdot_\frac{p(X_i |_A = 0, X_i;_\widehat{\alpha}_m)}{p(M_i |_A_= 1,_X_i;_\widehat{\alpha}_m)} \ Y_i _ _- \frac{ \mathbb I(A_i_= 0)}{ p(A_i_=_0 | X_i; \widehat{\alpha}_a)} \ Y_i \right). __\label{eqn-IPW} ___\end{aligned}$$ ]{}Since solving the constrained_MLE problem using this estimator_entails only_restricting parameters of $A$ and $M$ models, predicting_a_new instance is_done using $\mathbb{E}[Y | X] = \sum_{A, M} \mathbb{E}[Y_| A, M, X] \ p(M_| A, X)_\_p(A_| X).$ ***Mixed approach***: The_third way of computing the NDE_is using $A$ and $Y$ models._In this estimator, we fit the models_$p(A | X; \alpha_a)$ and $\mathbb{E}[Y_| A, M, X; \alpha_y]$_by MLE,_as usual, and combine the_edge G-formula and_IPW in_the following way:_[ $$\begin{aligned} \mathbb{P}_n _ _\left(\frac{\mathbb I(A_i = 0)}{p(A_i = 0_|_X_i ; \widehat{\alpha}_a)}_\_\mathbb{E}[Y_i_| A_= 1, M_i,_X_i;_\widehat{\alpha}_y] _-_\mathbb{E}[Y_i | A = 0, M_i;_\widehat{\alpha}_y]_\right), \label{eqn-mixed} _ \end{aligned}$$ ]{}_Since_solving the constrained MLE_problem using this estimator entails_only restricting parameters of $A$ and_$Y$ models,_predicting a_new instance is done using $\mathbb{E}[Y | M, X] = \sum_{A}_\mathbb{E}[Y | A, M, X] \_\frac{p(M | A, X)_\cdot p(A_|_X)}{\sum_A p(M |_A,_X) \cdot_p(A | X)}.$ ***Aug

$f_{n-1}^i(t):=\mu( Q_{n-1}^i(t))$, where $Q_{n-1}^i(t)$ denotes the $t$-homotetic cube of $Q_{n-1}^{i}$ with respect to the center $x(Q_{n-1}^{i})$. Let $t_{n-1}^{i}= \sup\{t\in[0,1]\,;\, \mu( Q_{n-1}^i(t))< m-\alpha-(\mu(Q_{n-1}^1)+\cdots+\mu(Q_{n-1}^{i-1}))\}$. If the function $f_{n-1}^i$ is continuous at $t_{n-1}^{i}$, then we consider as “canonical” set $H_Q=(Q_n(t_n^0))^\circ\cup \left(\cup_{j=1}^{i-1} Q_{n-1}^j\right)\cup (Q_{n-1}^i(t_{n-1}^{i}))^\circ$ and we are done. If not, we can continue iterating. If this iteration stops at some stage, then we are done. We next show that the iteration has to stop at some stage. Assume that this is not the case. Then, finally, we can find a line segment $Q_1$ and a function $f_1(t):=\mu(Q_1(t))$ such that the function $f_1$ is discontinuous at some $t$, which is equivalent to saying that there is a point $x\in Q_1$ such that $\mu(\{x\})>0$. And this contradicts the assumption that $\mu$ has no point masses. Therefore, the iteration has to stop. Observe that $H_Q$ is always a subset of $Q$, and it is a cube with part of its boundary. We will call such set a “canonical” extended cube. Finally, observe that the “canonical” method chosen to construct the sets $H_Q$ give that if $0<m_1<m_2<\mu(Q)$, then the corresponding “canonical” extended cubes $H_Q(m_1)$ and $H_Q(m_2)$ satisfy $H_Q(m_1)\subset H_Q(m_2)$. \[

$ f_{n-1}^i(t):=\mu (Q_{n-1}^i(t))$, where $ Q_{n-1}^i(t)$ denotes the $ t$-homotetic cube of $ Q_{n-1}^{i}$ with respect to the center $ x(Q_{n-1}^{i})$. lease $ t_{n-1}^{i}= \sup\{t\in[0,1]\,;\, \mu (Q_{n-1}^i(t) ) < m-\alpha-(\mu(Q_{n-1}^1)+\cdots+\mu(Q_{n-1}^{i-1}))\}$. If the affair $ f_{n-1}^i$ is continuous at $ t_{n-1}^{i}$, then we consider as “ canonical ” set $ H_Q=(Q_n(t_n^0))^\circ\cup \left(\cup_{j=1}^{i-1 } Q_{n-1}^j\right)\cup (Q_{n-1}^i(t_{n-1}^{i}))^\circ$ and we are perform. If not, we can continue iterating. If this iteration stop at some stage, then we are done. We next prove that the iteration has to stop at some stagecoach. Assume that this is not the case. Then, ultimately, we can find a line segment $ Q_1 $ and a function $ f_1(t):=\mu(Q_1(t))$ such that the function $ f_1 $ is discontinuous at some $ t$, which is equivalent to say that there is a point $ x\in Q_1 $ such that $ \mu(\{x\})>0$. And this contradicts the presumption that $ \mu$ has no point masses. Therefore, the iteration has to discontinue. Observe that $ H_Q$ is always a subset of $ Q$, and it is a cube with part of its boundary. We will shout such set a “ canonical ” extended cube. Finally, respect that the “ canonic ” method chosen to construct the sets $ H_Q$ give that if $ 0 < m_1 < m_2<\mu(Q)$, then the corresponding “ canonical ” extended cube $ H_Q(m_1)$ and $ H_Q(m_2)$ satisfy $ H_Q(m_1)\subset H_Q(m_2)$. \ [

$f_{n-1}^l(t):=\mu( Q_{n-1}^i(t))$, where $Q_{n-1}^i(t)$ denutes the $t$-homotgtuc cubx of $Q_{n-1}^{j}$ with rdspect to the center $x(Q_{n-1}^{i})$. Lev $t_{n-1}^{u}= \sup\{u\pn[0,1]\,;\, \mu( Q_{n-1}^i(t))< m-\alpha-(\mu(Q_{n-1}^1)+\zdots+\mu(Q_{n-1}^{p-1}))\}$. If the duncuion $f_{n-1}^i$ is contiihous at $t_{n-1}^{i}$, tgcn we ronsider as “canpnical” set $H_Q=(Q_n(t_n^0))^\circ\cup \ldfc(\cup_{j=1}^{i-1} Q_{n-1}^j\right)\cup (Q_{n-1}^i(t_{n-1}^{i}))^\circ$ and we are dome. If not, we can conuinte ifvrcting. If this iteration stops af some vtage, then we are done. We next show that the iteration has to dtop at somg stwte. Assume thxt this is not the casg. Then, finally, we can find a lind segkent $Q_1$ and a fujwtion $f_1(t):=\mu(Q_1(v))$ such that the fukbtion $f_1$ is disvontinuous at xomx $t$, qhich is equivalent tm saying that therg is a poitt $x\in Q_1$ such that $\nu(\{z\})>0$. And thiv covrraaicus vhe assumotikn that $\mu$ has no poibt masses. Therefore, trv iteration hzs to ftjp. Observe that $H_Q$ is always a subset ox $Q$, and it is a cube with part of its boundary. Ae will cwll such set a “canonical” extended cube. Finally, obsarve vhxt uhc “zqnlnical” method chosen to construct the sets $H_Q$ fife that if $0<m_1<m_2<\mu(Q)$, then the cprgexkonding “canonizal” excsnsed cubes $H_Q(m_1)$ and $J_Q(m_2)$ satysfy $Y_Q(m_1)\subset H_Q(m_2)$. \[

$f_{n-1}^i(t):=\mu( Q_{n-1}^i(t))$, where $Q_{n-1}^i(t)$ denotes the $t$-homotetic $Q_{n-1}^{i}$ respect to center $x(Q_{n-1}^{i})$. Let If function $f_{n-1}^i$ is at $t_{n-1}^{i}$, then consider as “canonical” set $H_Q=(Q_n(t_n^0))^\circ\cup \left(\cup_{j=1}^{i-1} (Q_{n-1}^i(t_{n-1}^{i}))^\circ$ and we are done. If not, we can continue iterating. If this stops at some stage, then we are done. We next show that the has stop some Assume that this is not the case. Then, finally, we can find a line segment $Q_1$ a function $f_1(t):=\mu(Q_1(t))$ such that the function $f_1$ discontinuous at some $t$, is equivalent to saying that is point $x\in such $\mu(\{x\})>0$. this contradicts the that $\mu$ has no point masses. Therefore, the iteration has to stop. Observe that $H_Q$ is always subset of it is cube part its boundary. We such set a “canonical” extended cube. the “canonical” method chosen to construct the sets give that $0<m_1<m_2<\mu(Q)$, then the corresponding “canonical” extended $H_Q(m_1)$ and $H_Q(m_2)$ satisfy $H_Q(m_1)\subset H_Q(m_2)$. \[

$f_{n-1}^i(t):=\mu( Q_{n-1}^i(t))$, where $Q_{n-1}^i(t)$ denoteS the $t$-homotEtic cUbe Of $Q_{N-1}^{i}$ With RespEct to the center $X(q_{n-1}^{i})$. LEt $t_{n-1}^{i}= \sup\{t\in[0,1]\,;\, \mu( Q_{n-1}^i(t))< m-\alphA-(\mu(Q_{n-1}^1)+\CdOTs+\mu(q_{N-1}^{i-1}))\}$. if the FunctioN $F_{n-1}^I$ IS coNtInUouS aT $T_{n-1}^{I}$, then We cOnsider As “canonicaL” seT $H_q=(Q_n(t_n^0))^\circ\cup \LEfT(\cup_{j=1}^{i-1} Q_{n-1}^j\rIghT)\cup (Q_{n-1}^i(t_{n-1}^{i}))^\ciRc$ aNd we arE dOne. iF not, wE caN contInue itERating. if this iteRaTIon stoPS at some STAgE, theN we are done. We next sHOw THat the iteratioN has to StOP aT SOme StaGe. Assume thAt This iS Not the cASe. tHEN, fiNAlly, we can find A line segmenT $q_1$ anD a funcTiOn $f_1(T):=\Mu(Q_1(t))$ suCh thaT tHE fuNction $f_1$ is diSconTinuous at Some $t$, wHIch is eqUIvalent To sayiNg tHat TherE Is A pOinT $x\IN Q_1$ sUCh ThaT $\Mu(\{x\})>0$. and this cOnTrAdictS the ASSUMptiOn tHat $\mU$ has nO point masses. THerEforE, The IteraTion hAs to StOp. ObsErve thAt $H_Q$ iS aLways a subset of $Q$, And iT is a cube wIth PaRt oF iTs bouNDary. We WilL caLl such sEt a “canoNIcaL” eXTENdEd cube. Finally, obserVe THAt The “canonIcal” meTHoD cHOsen to coNsTruCt thE SEts $H_Q$ Give THaT if $0<m_1<m_2<\mu(Q)$, Then thE CoRrEspondiNg “CanoniCaL” exTenDed cuBEs $H_Q(M_1)$ and $H_Q(M_2)$ satisfy $h_Q(m_1)\suBSet H_Q(m_2)$. \[

$f_{n-1}^i(t):=\mu( Q_{n- 1}^i(t))$, wher e $ Q_{ n- 1}^i (t)$ denotes the $ t $-ho motetic cube of $Q_{n- 1}^{i }$ with re spect to the ce n t er$x (Q _{n -1 } ^{ i})$. Le t $t_{n -1}^{i}= \ sup \{ t\in[0,1]\,; \ ,\mu( Q_{n- 1}^ i(t))< m-\al pha -(\mu( Q_ {n- 1 }^1)+ \cd ots+\ mu(Q_{ n -1}^{i -1}))\}$. I f the f u nction$ f _{ n-1} ^i$ is continuous at $t_{n-1}^{i}$, thenwe co n s ide r a s “canonic al ” set $H_Q=(Q _ n( t _ n ^0) ) ^\circ\cup \l eft(\cup_{j = 1}^ {i-1}Q_ {n- 1 }^j\ri ght)\ cu p (Q _{n-1}^i(t_ {n-1 }^{i}))^\ circ$a nd we a r e done. If no t,wecanc on ti nue i t era t in g.I f t his iter at io n sto ps a t s o me s tag e, t hen w e are done. Wenext sho w tha t the ite ra tionhas to stop a t some stage. A ssum e that th isis no tthe c a se. Th en, fi nally,we canf ind a l i ne segment $Q_1$ and a f un ction $f _1(t): = \m u( Q _1(t))$su chthat t he fu ncti o n$f_1$ is disco n ti nu ous atso me $t$ ,whi chis eq u ival ent to sayingthatt here is a poin t $x\in Q_1$ s u ch t ha t $\m u(\ {x\})>0$. A nd t h is c ontr a di cts the a ssump ti o nt hat $\mu$ has no po in t mass es. T herefore, the iteration h a s to sto p. O b se r ve that $H_Q$is al ways a sub s et of $Q $, an d it isa cube wi t h part of it s b oun dar y . W e will call s u c h se ta “cano nic al” ext end edcub e. F inally, o bserve t ha tth e“ca nonic a l” metho dcho se n t o con s tructthe s ets$H _Q $ gi ve that if $ 0<m_ 1< m_ 2<\m u(Q )$ , the n th e co rrespon ding “can oni c al”ex te nded cu bes $H_Q(m_1) $and $H_Q(m _2 )$satisf y $H_Q(m_1 )\subset H_Q(m_2)$. \[

$f_{n-1}^i(t):=\mu(_Q_{n-1}^i(t))$, where_$Q_{n-1}^i(t)$ denotes the $t$-homotetic_cube of_$Q_{n-1}^{i}$_with respect_to_the center $x(Q_{n-1}^{i})$._Let $t_{n-1}^{i}= \sup\{t\in[0,1]\,;\,_\mu( Q_{n-1}^i(t))< m-\alpha-(\mu(Q_{n-1}^1)+\cdots+\mu(Q_{n-1}^{i-1}))\}$. If_the function $f_{n-1}^i$_is_continuous at $t_{n-1}^{i}$, then we consider as “canonical” set $H_Q=(Q_n(t_n^0))^\circ\cup \left(\cup_{j=1}^{i-1} Q_{n-1}^j\right)\cup (Q_{n-1}^i(t_{n-1}^{i}))^\circ$ and_we_are done._If_not,_we can continue iterating. If_this iteration stops at some_stage, then_we are done. We next show that the iteration_has_to stop at_some stage. Assume that this is not the case._Then, finally, we can find a_line segment $Q_1$_and_a_function $f_1(t):=\mu(Q_1(t))$ such that_the function $f_1$ is discontinuous at_some $t$, which is equivalent to_saying that there is a point $x\in_Q_1$ such that $\mu(\{x\})>0$. And this_contradicts the assumption that $\mu$_has no_point masses. Therefore, the iteration_has to stop._Observe that_$H_Q$ is always_a subset of $Q$, and it_is a cube_with part of its boundary. We_will_call such set_a_“canonical”_extended cube. Finally,_observe that the_“canonical”_method chosen_to_construct the sets $H_Q$ give that_if_$0<m_1<m_2<\mu(Q)$, then the corresponding “canonical” extended cubes_$H_Q(m_1)$ and $H_Q(m_2)$ satisfy_$H_Q(m_1)\subset_H_Q(m_2)$. \[

$$\label{eqn.G6} G_{6} (e^{\pi i/3})=G_{6} (e^{2\pi i/3})=\sum_{m,n}^{*} \frac{1}{m^{6}+n^{6}} >0.$$ Consequently, if $g_{3} (\mathcal{L})=-C<0$ then $\mathcal{L}$ must be of the form for some $\omega$ such that $\omega^{6}<0$, that is, $\omega$ is a positive multiple of a primitive $12$th root of unity. Thus, in the case $g_{3} (\mathcal{L})=-C<0$ the lattice $\mathcal{L}$ must have the form with $\omega = |\omega |e^{\pi i/6}$. Assume now that $\mathcal{L}$ is of the form for some $\omega$ satisfying. Then by the *addition law* for the elliptic curve $y^{2}=4x^{3}-g_{3}$ (cf. [@K84], section I.7; see especially Problem 8), $$\mathcal{P}_{\mathcal{L}} (\sqrt{3}|\omega |/3)=\mathcal{P}_{\mathcal{L}} (2\sqrt{3}|\omega |/3)=0.$$ Since $\mathcal{P}_{\mathcal{L}}$ has degree $2$, it has only two zeros in a fundamental parallelogram, and by equation the derivatives $\mathcal{P}_{\mathcal{L}}' (\sqrt{3}|\omega |/3)=-\mathcal{P}_{\mathcal{L}}' (2\sqrt{3}|\omega |/3)$ must be the two square roots of $C$. It is easily seen that the unique solution of with initial conditions $u (y_{0})=0$ and ${u'(y_{0})}>0$ must be strictly increasing, with increasing derivative, on any interval $(y_{0},y_{1})$ on which the solution $u$ is well-defined and finite. This implies that $\mathcal{P}_{\mathcal{L}}' (\sqrt{3}|\omega |/3)$ is *negative*, and hence $\mathcal{P}_{\mathcal{L}}' (2\sqrt{3}|\omega |/3)$ is *positive*. It then follows that $u$ is strictly increasing in $(2\sqrt{3}|\omega |/3,\sqrt{3}|\omega |)$.

$ $ \label{eqn. G6 } G_{6 } (e^{\pi i/3})=G_{6 } (e^{2\pi i/3})=\sum_{m, n}^ { * } \frac{1}{m^{6}+n^{6 } } > 0.$$ Consequently, if $ g_{3 } (\mathcal{L})=-C<0 $ then $ \mathcal{L}$ must be of the form for some $ \omega$ such that $ \omega^{6}<0 $, that is, $ \omega$ is a positive multiple of a primitive $ 12$th root of oneness. therefore, in the case $ g_{3 } (\mathcal{L})=-C<0 $ the lattice $ \mathcal{L}$ must get the shape with $ \omega = |\omega |e^{\pi i/6}$. Assume now that $ \mathcal{L}$ is of the shape for some $ \omega$ satisfying. Then by the * summation law * for the egg-shaped curvature $ y^{2}=4x^{3}-g_{3}$ (cf. [ @K84 ], incision I.7; see especially Problem 8), $ $ \mathcal{P}_{\mathcal{L } } (\sqrt{3}|\omega |/3)=\mathcal{P}_{\mathcal{L } } (2\sqrt{3}|\omega |/3)=0.$$ Since $ \mathcal{P}_{\mathcal{L}}$ has academic degree $ 2 $, it has only two zero in a fundamental parallelogram, and by equation the derived function $ \mathcal{P}_{\mathcal{L } }' (\sqrt{3}|\omega |/3)=-\mathcal{P}_{\mathcal{L } }' (2\sqrt{3}|\omega |/3)$ must be the two square roots of $ C$. It is easily examine that the unique solution of with initial conditions $ u (y_{0})=0 $ and $ { u'(y_{0})}>0 $ must be strictly increase, with increasing derivative, on any interval $ (y_{0},y_{1})$ on which the solution $ u$ is well - defined and finite. This implies that $ \mathcal{P}_{\mathcal{L } }' (\sqrt{3}|\omega |/3)$ is * damaging *, and hence $ \mathcal{P}_{\mathcal{L } }' (2\sqrt{3}|\omega |/3)$ is * positive *. It then follows that $ u$ is strictly increasing in $ (2\sqrt{3}|\omega |/3,\sqrt{3}|\omega |)$.

$$\lahel{eqn.G6} G_{6} (e^{\pi i/3})=G_{6} (e^{2\pi l/3})=\sum_{m,n}^{*} \frac{1}{m^{6}+n^{6}} >0.$$ Consexuentlg, if $g_{3} (\mxthcal{L})=-C<0$ then $\mathcal{L}$ must ue od the form for some $\omega$ sjch that $\lmega^{6}<0$, thqt iw, $\omega$ is e positiyz mulflple mh a primitive $12$tm root of utity. Thus, in tve ccse $g_{3} (\mathcal{L})=-C<0$ the lattice $\mathcal{L}$ iust hafe the form with $\omebw = |\ojvgc |e^{\pi i/6}$. Assume now that $\mathcal{L}$ is of uhe form for some $\pmega$ satisfying. Then by tje *afdition law* for thf elliptic xurvq $y^{2}=4x^{3}-g_{3}$ (cf. [@K84], seztion I.7; set zspecially Kroblem 8), $$\mathcal{P}_{\mathcal{L}} (\sqrt{3}|\omdga |/3)=\mathcal{P}_{\narhcwn{L}} (2\sqrt{3}|\omege |/3)=0.$$ Sinbe $\mathcal{P}_{\mabncal{L}}$ vas degtee $2$, it has onky vwo zeros in a fundamentan parallelogram, anq by equadikn the derivativew $\nathcdl{P}_{\mdthcxo{L}}' (\rqru{3}|\omxga |/3)=-\mathcwl{P}_{\jathcal{L}}' (2\sqdt{3}|\omega |/3)$ muwt be the two squart rjits of $C$. It ia easijy seen that the unique solution of with pnitjal conditions $u (y_{0})=0$ and ${y'(y_{0})}>0$ must be strictly ijcreasing, with increasing derivative, on any interval $(y_{0},y_{1})$ on whici ghe sjuytlon $u$ is well-defined and finite. This implies egau $\msthcal{P}_{\mathcal{J}}' (\sqrt{3}|\omega |/3)$ ls *gegative*, and fence $\matgcal{P}_{\mathcal{L}}' (2\sqrt{3}|\lmega |/3)$ ys *powitive*. It them follows that $u$ is strictlt increasing un $(2\sqrt{3}|\omega |/3,\sqrt{3}|\olega |)$.

$$\label{eqn.G6} G_{6} (e^{\pi i/3})=G_{6} (e^{2\pi i/3})=\sum_{m,n}^{*} \frac{1}{m^{6}+n^{6}} if (\mathcal{L})=-C<0$ then must be of such $\omega^{6}<0$, that is, is a positive of a primitive $12$th root of Thus, in the case $g_{3} (\mathcal{L})=-C<0$ the lattice $\mathcal{L}$ must have the form $\omega = |\omega |e^{\pi i/6}$. Assume now that $\mathcal{L}$ is of the form some satisfying. by *addition law* for the elliptic curve $y^{2}=4x^{3}-g_{3}$ (cf. [@K84], section I.7; see especially Problem 8), $$\mathcal{P}_{\mathcal{L}} |/3)=\mathcal{P}_{\mathcal{L}} (2\sqrt{3}|\omega |/3)=0.$$ Since $\mathcal{P}_{\mathcal{L}}$ has degree $2$, has only two zeros a fundamental parallelogram, and by the $\mathcal{P}_{\mathcal{L}}' (\sqrt{3}|\omega (2\sqrt{3}|\omega must the two square of $C$. It is easily seen that the unique solution of with initial conditions $u (y_{0})=0$ and must be with increasing on interval on which the is well-defined and finite. This implies |/3)$ is *negative*, and hence $\mathcal{P}_{\mathcal{L}}' (2\sqrt{3}|\omega |/3)$ *positive*. It follows that $u$ is strictly increasing $(2\sqrt{3}|\omega |/3,\sqrt{3}|\omega |)$.

$$\label{eqn.G6} G_{6} (e^{\pi i/3})=G_{6} (e^{2\pi i/3})=\sum_{m,n}^{*} \fRac{1}{m^{6}+n^{6}} >0.$$ ConseQuentLy, iF $g_{3} (\mAtHcal{l})=-C<0$ thEn $\mathcal{L}$ must BE of tHe form for some $\omega$ such That $\oMeGA^{6}<0$, thaT Is, $\Omega$ Is a posiTIvE MUltIpLe Of a PrIMiTive $12$tH roOt of uniTy. Thus, in thE caSe $G_{3} (\mathcal{L})=-C<0$ thE LaTtice $\mathcAl{L}$ Must have the fOrm With $\omEgA = |\omEGa |e^{\pi I/6}$. AsSume nOw that $\MAthcal{l}$ is of the fOrM For somE $\Omega$ saTISfYing. then by the *addition LAw* FOr the elliptic cUrve $y^{2}=4x^{3}-G_{3}$ (cF. [@k84], sECTioN I.7; sEe especialLy problEM 8), $$\mathcaL{p}_{\mATHCal{l}} (\Sqrt{3}|\omega |/3)=\mathCal{P}_{\mathcal{l}} (2\SqrT{3}|\omega |/3)=0.$$ siNce $\MAthcal{p}_{\mathCaL{l}}$ haS degree $2$, it haS onlY two zeros In a funDAmental PArallelOgram, aNd bY eqUatiON tHe DerIvATivES $\mAthCAl{P}_{\Mathcal{L}}' (\SqRt{3}|\Omega |/3)=-\MathCAL{p}_{\MathCal{l}}' (2\sqrT{3}|\omegA |/3)$ must be the two SquAre rOOts Of $C$. It Is easIly sEeN that The uniQue soLuTion of with initiAl coNditions $u (Y_{0})=0$ anD ${u'(Y_{0})}>0$ muSt Be strICtly inCreAsiNg, with iNcreasiNG deRiVATIvE, on any interval $(y_{0},y_{1})$ on WhICH tHe solutiOn $u$ is wELl-DeFIned and fInIte. this IMPlies That $\MAtHcal{P}_{\matHcal{L}}' (\sQRt{3}|\OmEga |/3)$ is *neGaTive*, anD hEncE $\maThcal{p}_{\MathCal{L}}' (2\sqRt{3}|\omega |/3)$ iS *posiTIve*. It then folloWS that $u$ is stricTLy INCrEAsinG in $(2\Sqrt{3}|\omega |/3,\sqRt{3}|\omEGa |)$.

$$\label{eqn.G6} G_{6 } (e^{\pii/3}) =G_ {6} ( e^{2 \pii/3})=\sum_{m, n }^{* } \frac{1}{m^{6}+n ^{6}} > 0 .$$C on seque ntly, i f $ g _ {3} ( \m ath ca l {L })=-C <0$ then $ \mathcal{L }$mu st be of the fo rm for som e $ \omega$ such th at $\o me ga^ { 6}<0$ , t hat i s, $\o m ega$ i s a posit iv e multi p le of a p ri miti ve $12$th root of un i ty. Thus, in t he cas e$ g_ { 3 } ( \ma thcal{L})= -C <0$ t h e latti c e$ \ m ath c al{L}$ must h ave the for m wi th $\o me ga= |\ome ga |e ^{ \ pii/6}$. Ass umenow that$\math c al{L}$i s of th e form fo r s ome$ \o me ga$ s a tis f yi ng. The n by the * ad ditio n la w * f or t heelli pticcurve $y^{2}= 4x^ {3}- g _{3 }$ (c f. [@ K84] ,secti on I.7 ; see e specially Probl em 8 ), $$\mat hca l{ P}_ {\ mathc a l{L}}(\s qrt {3}|\om ega |/3 )= \ m a th cal{P}_{\mathcal{L }} ( 2\ sqrt{3}| \omega |/ 3) = 0.$$ Sin ce $\ math c a l{P}_ {\ma t hc al{L}}$has de g re e$2$, it h as onl ytwo ze ros i n a f undame ntal par allel o gram, and by e q uation the de r iv a t iv e s $\ mat hcal{P}_{\m athc a l{L} }' ( \ sq rt{ 3 }|\om ega | /3 ) =- \ mathcal{P}_{\mathca l{ L}}' ( 2\sqr t{3}|\omega | /3)$ mustb e the twosqua r er oots of $C$. I t iseasily see n that th e uni que solu tion of w i t h initia l c ond iti ons $ u(y_{0})=0$ an d ${u' (y _{0})}> 0$must be st ric tly in cr easing, w ith incr ea si ng d eri vativ e , on any i nte rv al$(y_{ 0 },y_{1 })$ o n wh ic ht hesolutio n $ u $ iswe ll -def ine dand f init e . T his imp lies that $\ m athc al {P }_{\mat hcal{L}}' (\s qr t{3}|\omeg a|/3 )$ is* n egative* , and hence $\mathcal{P } _{\math cal {L}}' (2\ sqrt{3}|\ ome ga |/3 )$i s *pos itive* . Itth enf o llows t ha t $ u$ is strict l y in creas in g in $(2\sq rt{3}|\omega |/3,\ s qrt {3}|\omega |) $.

$$\label{eqn.G6} _ _ G_{6} (e^{\pi i/3})=G_{6}_(e^{2\pi i/3})=\sum_{m,n}^{*} __ _\frac{1}{m^{6}+n^{6}}_>0.$$ Consequently, if_$g_{3} (\mathcal{L})=-C<0$ then_$\mathcal{L}$ must be of_the form for_some_$\omega$ such that $\omega^{6}<0$, that is, $\omega$ is a positive multiple of a primitive_$12$th_root of_unity._Thus,_in the case $g_{3} (\mathcal{L})=-C<0$ the_lattice $\mathcal{L}$ must have the_form with_$\omega = |\omega |e^{\pi i/6}$. Assume now that $\mathcal{L}$_is_of the form_for some $\omega$ satisfying. Then by the *addition law*_for the elliptic curve $y^{2}=4x^{3}-g_{3}$ (cf._[@K84], section I.7;_see_especially_Problem 8), $$\mathcal{P}_{\mathcal{L}} (\sqrt{3}|\omega _ |/3)=\mathcal{P}_{\mathcal{L}} (2\sqrt{3}|\omega |/3)=0.$$_Since $\mathcal{P}_{\mathcal{L}}$ has degree $2$, it_has only two zeros in a fundamental_parallelogram, and by equation the derivatives_$\mathcal{P}_{\mathcal{L}}' (\sqrt{3}|\omega |/3)=-\mathcal{P}_{\mathcal{L}}' (2\sqrt{3}|\omega |/3)$ must_be the_two square roots of $C$._It is easily_seen that_the unique solution_of with initial conditions $u (y_{0})=0$_and ${u'(y_{0})}>0$ must_be strictly increasing, with increasing derivative,_on_any interval $(y_{0},y_{1})$_on_which_the solution_$u$ is well-defined_and_finite. This_implies_that $\mathcal{P}_{\mathcal{L}}' (\sqrt{3}|\omega |/3)$ is *negative*,_and_hence $\mathcal{P}_{\mathcal{L}}' (2\sqrt{3}|\omega |/3)$ is *positive*. It_then follows that $u$_is_strictly increasing in $(2\sqrt{3}|\omega_|/3,\sqrt{3}|\omega |)$.

is injective possibly with the exception of ${{\boldsymbol{\varphi}}}^*(0)={{\boldsymbol{\varphi}}}^*(1)$ in case ${{\boldsymbol{\varphi}}}$ was closed. In other words, $\Gamma^*$ is obtained by extending $\Gamma$ beyond the beginning point ${{\boldsymbol{\varphi}}}(0)$ and the end point ${{\boldsymbol{\varphi}}}(1)$ by straight line segments of strictly positive lengths belonging to the lines tangent to $\Gamma$ at these two points, respectively. We need to introduce a notation. Suppose $\Gamma\subset{{\mathbb{R}}}^2$ and ${{\boldsymbol{y}}}\in \Gamma$ are such that $\Gamma$ is a $C^2$ curve in a neighborhood of ${{\boldsymbol{y}}}$. Let ${{\boldsymbol{u}}}$ be a unit normal vector to $\Gamma$ at ${{\boldsymbol{y}}}$, and in the case $\Gamma=\partial D$ for a domain $D$, we choose ${{\boldsymbol{u}}}$ to be pointing outwards from $D$. For any $r>0$, we denote by $$\label{eqn:Bpm def} B_+(r,{{\boldsymbol{y}}},\Gamma):=rB+{{\boldsymbol{y}}}+ r{{\boldsymbol{u}}}{{\quad\text{and}\quad}} B_-(r,{{\boldsymbol{y}}},\Gamma):=rB+{{\boldsymbol{y}}}- r{{\boldsymbol{u}}}$$ the two discs of radius $r$ tangent to $\Gamma$ at ${{\boldsymbol{y}}}$. We need that closed $C^2$ curves without corner points possess a [*rolling disc property*]{}. Namely, if $\Gamma$ is such a curve, then there exists $r=r(\Gamma)>0$ such that for any $0<r'\le r$ and any point ${{\boldsymbol{y}}}\in\Gamma$ we have $B_\pm(r',{{\boldsymbol{y}}},\Gamma)\cap \Gamma=\{{{\boldsymbol{y}}}\}$. While in a neighborhood of ${{\boldsymbol{y}}}$ such a statement follows from standard differential geometry (curvature is separated from zero), the stated above global version follows from a generalization of Blaschke’s rolling theorem [@Wa]\*[Th. 1 (iii) and (v)]{}. In the proof of the main result we will extend the rolling disc property to certain non-closed $C^2$ curves. We will need the following analog of Bernstein and Markov inequalities due to Vidensk

is injective possibly with the exception of $ { { \boldsymbol{\varphi}}}^*(0)={{\boldsymbol{\varphi}}}^*(1)$ in case $ { { \boldsymbol{\varphi}}}$ was closed. In other words, $ \Gamma^*$ is prevail by unfold $ \Gamma$ beyond the beginning point $ { { \boldsymbol{\varphi}}}(0)$ and the end point $ { { \boldsymbol{\varphi}}}(1)$ by true line segments of rigorously positive lengths belonging to the credit line tangent to $ \Gamma$ at these two points, respectively. We need to precede a notation. Suppose $ \Gamma\subset{{\mathbb{R}}}^2 $ and $ { { \boldsymbol{y}}}\in \Gamma$ are such that $ \Gamma$ is a $ C^2 $ bend in a vicinity of $ { { \boldsymbol{y}}}$. Let $ { { \boldsymbol{u}}}$ be a unit normal vector to $ \Gamma$ at $ { { \boldsymbol{y}}}$, and in the case $ \Gamma=\partial D$ for a domain $ D$, we choose $ { { \boldsymbol{u}}}$ to be pointing outwards from $ D$. For any $ r>0 $, we announce by $ $ \label{eqn: Bpm def } B_+(r,{{\boldsymbol{y}}},\Gamma):=rB+{{\boldsymbol{y}}}+ r{{\boldsymbol{u}}}{{\quad\text{and}\quad } } B_-(r,{{\boldsymbol{y}}},\Gamma):=rB+{{\boldsymbol{y}}}- r{{\boldsymbol{u}}}$$ the two discs of radius $ r$ tangent to $ \Gamma$ at $ { { \boldsymbol{y}}}$. We need that close $ C^2 $ curves without corner point own a [ * rolling disc property * ] { }. Namely, if $ \Gamma$ is such a curve, then there exists $ r = r(\Gamma)>0 $ such that for any $ 0 < r'\le r$ and any distributor point $ { { \boldsymbol{y}}}\in\Gamma$ we have $ B_\pm(r',{{\boldsymbol{y}}},\Gamma)\cap \Gamma=\{{{\boldsymbol{y}}}\}$. While in a neighborhood of $ { { \boldsymbol{y}}}$ such a statement follows from standard differential geometry (curvature is separated from zero), the stated above global version follows from a generalization of Blaschke ’s rolling theorem   [ @Wa]\*[Th.   1   (iii)   and   (v) ] { }. In the proof of the main result we will extend the rolling phonograph record property to certain non - shut $ C^2 $ curve. We will want the following analog of Bernstein and Markov inequality due to Vidensk

is injective possibly with the exception of ${{\boldvymbol{\barphi}}}^*(0)={{\boudsymbol{\varphi}}}^*(1)$ in case ${{\boldsbmboo{\varpyi}}}$ was closed. In other words, $\Galma^*$ is ovtaiied by extending $\Gamma$ bcvond fme beyiining point ${{\bolcsymbol{\var[hi}}}(0)$ and the eng oonnt ${{\boldsymbol{\varphi}}}(1)$ by straight linq segmemtd of strictly kositpvq lehgths belonging to the lines tangeht to $\Gemma$ at these teo points, respectively. We nfed ho introduce a notwtion. Suppowe $\Gwnma\subset{{\matfbb{R}}}^2$ and ${{\boldsymbol{y}}}\in \Gamma$ are such that $\Gamma$ is a $C^2$ cuxve in a neugybogvood of ${{\bolvsymboj{y}}}$. Let ${{\boldsnkbol{u}}}$ te a unot normal vectpr vo $\Gqmma$ at ${{\boldsymbol{y}}}$, aid in the case $\Gamma=\kartial D$ xox a domain $D$, we choosw ${{\voldsimbol{g}}}$ to ve ooihtmng outwagds from $D$. Fod any $r>0$, we eenote by $$\label{eqn:Bkm qvg} B_+(r,{{\boldsymbom{y}}},\Gammw):=rF+{{\boldsymbol{y}}}+ r{{\boldsymbol{u}}}{{\quad\text{and}\quag}} B_-(d,{{\boldsymbol{y}}},\Gamma):=rB+{{\boldwymbol{y}}}- r{{\boldsymbol{u}}}$$ jhe two difcs of radius $r$ tangent to $\Gamma$ at ${{\boldsymbol{y}}}$. We need ghau gjuwef $C^2$ curves without corner points possess a [*rojmimg disc property*]{}. Namely, if $\Balms$ is such a cutve, theu tgere exists $r=r(\Gammw)>0$ such jhat fir any $0<r'\lt r$ amd any point ${{\boldsymbol{y}}}\in\Gqmma$ we have $V_\pm(r',{{\boldsymbol{y}}},\Gamla)\cap \Gamma=\{{{\yoldsykbol{y}}}\}$. While in a neighborhoob of ${{\bkldsymbol{y}}}$ duch a stzgement follows ffom sdandard differential geomeery (curvavure ns separxted from sero), the shated above global versiln foplmws from a generalization of Blaschke’s rolling theorem [@Ws]\*[Tv. 1 (iip) and (v)]{}. In che prpof of the mayn result we wnll exteud the rolling dpsc propecty to certayn non-closed $W^2$ curves. We winl need ehe dolliwing avxlog of Bernstrin and Mcxkov ineqyalities due to Vivensy

is injective possibly with the exception of case was closed. other words, $\Gamma^*$ beyond beginning point ${{\boldsymbol{\varphi}}}(0)$ the end point by straight line segments of strictly lengths belonging to the lines tangent to $\Gamma$ at these two points, respectively. need to introduce a notation. Suppose $\Gamma\subset{{\mathbb{R}}}^2$ and ${{\boldsymbol{y}}}\in \Gamma$ are such that is $C^2$ in neighborhood of ${{\boldsymbol{y}}}$. Let ${{\boldsymbol{u}}}$ be a unit normal vector to $\Gamma$ at ${{\boldsymbol{y}}}$, and in case $\Gamma=\partial D$ for a domain $D$, we ${{\boldsymbol{u}}}$ to be pointing from $D$. For any $r>0$, denote $$\label{eqn:Bpm def} r{{\boldsymbol{u}}}{{\quad\text{and}\quad}} r{{\boldsymbol{u}}}$$ two discs of $r$ tangent to $\Gamma$ at ${{\boldsymbol{y}}}$. We need that closed $C^2$ curves without corner points possess a disc property*]{}. $\Gamma$ is a then exists $r=r(\Gamma)>0$ such any $0<r'\le r$ and any point $B_\pm(r',{{\boldsymbol{y}}},\Gamma)\cap \Gamma=\{{{\boldsymbol{y}}}\}$. While in a neighborhood of ${{\boldsymbol{y}}}$ a statement from standard differential geometry (curvature is from zero), the stated above global version follows a generalization of Blaschke’s rolling theorem [@Wa]\*[Th. 1 (iii) and (v)]{}. In the proof of result we will extend rolling disc property certain $C^2$ We need the analog of Bernstein and Markov inequalities due to Vidensk

is injective possibly with thE exception Of ${{\bolDsyMboL{\vArphI}}}^*(0)={{\bolDsymbol{\varphi}}}^*(1)$ iN Case ${{\Boldsymbol{\varphi}}}$ was cloSed. In OtHEr woRDs, $\gamma^*$ Is obtaiNEd BY ExtEnDiNg $\GAmMA$ bEyond The BeginniNg point ${{\bolDsyMbOl{\varphi}}}(0)$ and tHE eNd point ${{\bolDsyMbol{\varphi}}}(1)$ by StrAight lInE seGMents Of sTrictLy posiTIve lenGths belonGiNG to the LInes tanGENt To $\GaMma$ at these two poinTS, rESpectively. We neEd to inTrODuCE A noTatIon. Suppose $\gaMma\suBSet{{\mathBB{R}}}^2$ AND ${{\BolDSymbol{y}}}\in \GammA$ are such thaT $\gamMa$ is a $C^2$ CuRve IN a neigHborhOoD Of ${{\bOldsymbol{y}}}$. LEt ${{\boLdsymbol{u}}}$ Be a uniT Normal vECtor to $\GAmma$ at ${{\BolDsyMbol{Y}}}$, AnD iN thE cASe $\GAMmA=\paRTiaL D$ for a doMaIn $d$, we chOose ${{\BOLDSymbOl{u}}}$ To be PointIng outwards frOm $D$. for aNY $r>0$, wE denoTe by $$\lAbel{EqN:Bpm dEf} B_+(r,{{\boLdsymBoL{y}}},\Gamma):=rB+{{\boldsyMbol{Y}}}+ r{{\boldsymBol{U}}}{{\qUad\TeXt{and}\QUad}} B_-(r,{{\bOldSymBol{y}}},\GamMa):=rB+{{\bolDSymBoL{Y}}}- R{{\BoLdsymbol{u}}}$$ the two disCs OF RaDius $r$ tanGent to $\gAmMa$ AT ${{\boldsymBoL{y}}}$. WE neeD THat clOsed $c^2$ CuRves withOut corNEr PoInts posSeSs a [*rolLiNg dIsc PropeRTy*]{}. NaMely, if $\gamma$ is sUch a cURve, then there exISts $r=r(\Gamma)>0$ sucH ThAT FoR Any $0<r'\Le r$ And any point ${{\BoldSYmboL{y}}}\in\gAmMa$ wE Have $B_\Pm(r',{{\boLdSYmBOl{y}}},\Gamma)\cap \Gamma=\{{{\bolDsYmbol{y}}}\}$. while In a neighborhoOd of ${{\boldsyMBOL{y}}}$ such a sTateMEnT Follows from staNdard DifferentiAL geometrY (curvAture is sEparated fROM zero), the StaTed AboVe gLOBaL version folloWS From A gEneraliZatIon of BlAscHke’S roLliNg Theorem [@Wa]\*[th. 1 (iii) and (V)]{}. IN tHe PrOof Of the MAin resulT wE wiLl ExtEnd thE RollinG disc PropErTy TO ceRtain noN-ClOSEd $C^2$ cUrVeS. We wIll NeEd the FollOWinG analog Of BernsteIn aND MarKoV iNequaliTies due to VideNsK

is injective possibly wit h the exce ption of ${ {\ bold symb ol{\varphi}}}^ * (0)= {{\boldsymbol{\varphi} }}^*( 1) $ inc as e ${{ \boldsy m bo l { \va rp hi }}} $w as clos ed. In oth er words,$\G am ma^*$ is obt a in ed by exte ndi ng $\Gamma$bey ond th ebeg i nning po int $ {{\bol d symbol {\varphi} }} ( 0)$ an d the en d po int${{\boldsymbol{\v a rp h i}}}(1)$ by st raight l i ne s egm ent s of stric tl y pos i tive le n gt h s bel o nging to thelines tange n t t o $\Ga mm a$a t thes e two p o int s, respecti vely . We nee d to i n troduce a notat ion. S upp ose $\G a mm a\ sub se t {{\ m at hbb { R}} }^2$ and $ {{ \bold symb o l { y }}}\ in\Gam ma$ a re such that$\G amma $ is a $C ^2$ c urve i n a n eighbo rhood o f ${{\boldsymbo l{y} }}$. Let${{ \b old sy mbol{ u }}}$ b e a un it norm al vect o r t o$ \ G am ma$ at ${{\boldsym bo l { y} }}$, and in th e c as e $\Gamma =\ par tial D $ for a d o ma in $D$,we cho o se $ {{\bold sy mbol{u }} }$tobe po i ntin g outw ards fro m $D$ . For any $r>0$ , we denote by $$ \ l ab e l{eq n:B pm def} B_+ (r,{ { \bol dsym b ol {y} } },\Ga mma): =r B +{ { \boldsymbol{y}}}+ r {{ \bolds ymbol {u}}}{{\quad\ text{and}\ q u a d}} B_- (r,{ { \b o ldsymbol{y}}}, \Gamm a):=rB+{{\ b oldsymbo l{y}} }- r{{\b oldsymbol { u }}}$$ th e t wodis cso f r adius $r$ tan g e nt t o$\Gamma $ a t ${{\b old sym bol {y} }} $. We ne ed thatcl os ed $ C^2 $ cur v es witho ut co rn erpoint s posse ss a[*ro ll in g di sc prop e rt y * ]{}. N am ely, if $ \Gamm a$ i s su ch a cu rve, then th e re e xi st s $r=r( \Gamma)>0$ su ch that foran y $ 0<r'\l e r$ and a ny point ${{\boldsymbol { y}}}\in \Ga mma$we h ave $B_\p m(r ',{{\b old s ymbol{ y}}},\ Gamma )\ cap \ Gamma = \ {{ {\b ol dsymbol{y} } } \}$ . Whi le ina neigh borhood of ${{\bol d sym bol{y}}}$ suc h a sta t e me ntf ol l ows f r oms t andard differen tial geome tr y ( curvaturei s s ep aratedfrom ze ro),t he stat ed aboveglobal ve rs ionf o llo ws from agenerali zation of Blasc h ke ’s ro lli ng the or em[@Wa] \*[Th. 1 ( iii)and (v )] {}. In thepr oof of t he main result we willextend therol ling disc pr o per ty to cer tain non-close d $ C^2 $ cur ves . Wewill ne edt he fo llow i ng analog of Be r n st ein and Mar k o v in equal iti e s dueto V idensk

is_injective possibly_with the exception of_${{\boldsymbol{\varphi}}}^*(0)={{\boldsymbol{\varphi}}}^*(1)$ in_case_${{\boldsymbol{\varphi}}}$ was_closed._In other words,_$\Gamma^*$ is obtained_by extending $\Gamma$ beyond_the beginning point_${{\boldsymbol{\varphi}}}(0)$_and the end point ${{\boldsymbol{\varphi}}}(1)$ by straight line segments of strictly positive lengths belonging_to_the lines_tangent_to_$\Gamma$ at these two points,_respectively. We need to introduce a_notation. Suppose_$\Gamma\subset{{\mathbb{R}}}^2$ and ${{\boldsymbol{y}}}\in \Gamma$ are such that $\Gamma$_is_a $C^2$ curve_in a neighborhood of ${{\boldsymbol{y}}}$. Let ${{\boldsymbol{u}}}$ be a_unit normal vector to $\Gamma$ at_${{\boldsymbol{y}}}$, and in_the_case_$\Gamma=\partial D$ for a_domain $D$, we choose ${{\boldsymbol{u}}}$ to_be pointing outwards from $D$. For_any $r>0$, we denote by $$\label{eqn:Bpm def} B_+(r,{{\boldsymbol{y}}},\Gamma):=rB+{{\boldsymbol{y}}}+_r{{\boldsymbol{u}}}{{\quad\text{and}\quad}} B_-(r,{{\boldsymbol{y}}},\Gamma):=rB+{{\boldsymbol{y}}}- r{{\boldsymbol{u}}}$$ the two discs_of radius $r$ tangent to_$\Gamma$ at_${{\boldsymbol{y}}}$. We need that closed $C^2$_curves without corner_points possess_a [*rolling disc_property*]{}. Namely, if $\Gamma$ is such_a curve, then_there exists $r=r(\Gamma)>0$ such that for_any_$0<r'\le r$ and_any_point_${{\boldsymbol{y}}}\in\Gamma$ we_have $B_\pm(r',{{\boldsymbol{y}}},\Gamma)\cap \Gamma=\{{{\boldsymbol{y}}}\}$._While_in a_neighborhood_of ${{\boldsymbol{y}}}$ such a statement follows_from_standard differential geometry (curvature is separated from_zero), the stated above_global_version follows from a_generalization of Blaschke’s rolling theorem [@Wa]\*[Th. 1 (iii) and (v)]{}._In the proof of the main_result we_will extend_the rolling disc property to certain non-closed $C^2$ curves. We will need_the following analog of Bernstein and_Markov inequalities due to_Vidensk

d_{A,\Phi}:W^{k+2,p} \to W^{k,p}$ is Fredholm with index zero by Theorem \[thm:Gilkey\_1-4-5\_Sobolev\], the following operator is invertible, $$d_{A,\Phi}^*d_{A,\Phi}:(\Ker d_{A,\Phi}^*d_{A,\Phi})^\perp \cap W^{k+2,p} \to (\Ker d_{A,\Phi}^*d_{A,\Phi})^\perp \cap W^{k,p}.$$ Thus, $\xi = 0$ and the operator is injective. For surjectivity, suppose $\chi = d_{A,\Phi}\zeta \perp \Ran d_{A,\Phi}\cap W^{k,p}$ for $\zeta \in W^{k+1,p}$. We may again assume without loss of generality that $\zeta \perp \Ker d_{A,\Phi}^*d_{A,\Phi}$. If $\chi\perp d_{A,\Phi}d_{A,\Phi}^*d_{A,\Phi}\xi$ for all $\xi \in W^{k+3,p}$, then $(d_{A,\Phi}^*d_{A,\Phi})^2\zeta = 0$, and we again find that $\zeta = 0$ and so $\chi=0$. Since $d_{A,\Phi}d_{A,\Phi}^*: W^{k+2,p} \to W^{k,p}$ has closed range (because it is a Fredholm operator), this implies that the operator is surjective and thus invertible by the Open Mapping Theorem. According to Proposition \[prop:Fredholmness\_and\_index\_Laplace\_operator\_on\_W1p\_pairs\_slice\], the operator is Fredholm. Consequently, $\sM'(A,\Phi): \Ker d_{A,\Phi}^*\cap W^{k+2,p} \to \Ker d_{A,\Phi}^*\cap W^{k,p}$ is Fredholm by virtue of the direct sum decomposition and invertibility of the operator. We compute indices, $$\begin{aligned} {}& \Ind\left(\sM'(A,\Phi) + d_{A,\Phi}d_{A,\Phi}^*:W^{k+2,p} \to W^{k,p}\right) \\ &\quad = \Ind\begin{pmatrix}\sM

d_{A,\Phi}:W^{k+2,p } \to W^{k, p}$ is Fredholm with index zero by Theorem \[thm: Gilkey\_1 - 4 - 5\_Sobolev\ ], the following operator is invertible, $ $ d_{A,\Phi}^*d_{A,\Phi}:(\Ker d_{A,\Phi}^*d_{A,\Phi})^\perp \cap W^{k+2,p } \to (\Ker d_{A,\Phi}^*d_{A,\Phi})^\perp \cap W^{k, p}.$$ Thus, $ \xi = 0 $ and the hustler is injective. For surjectivity, presuppose $ \chi = d_{A,\Phi}\zeta \perp \Ran d_{A,\Phi}\cap W^{k, p}$ for $ \zeta \in W^{k+1,p}$. We may again assume without loss of generalization that $ \zeta \perp \Ker d_{A,\Phi}^*d_{A,\Phi}$. If $ \chi\perp d_{A,\Phi}d_{A,\Phi}^*d_{A,\Phi}\xi$ for all $ \xi \in W^{k+3,p}$, then $ (d_{A,\Phi}^*d_{A,\Phi})^2\zeta = 0 $, and we again find that $ \zeta = 0 $ and indeed $ \chi=0$. Since $ d_{A,\Phi}d_{A,\Phi}^ *: W^{k+2,p } \to W^{k, p}$ has closed range (because it is a Fredholm hustler), this incriminate that the operator is surjective and therefore invertible by the Open Mapping Theorem. harmonize to Proposition \[prop: Fredholmness\_and\_index\_Laplace\_operator\_on\_W1p\_pairs\_slice\ ], the operator is Fredholm. Consequently, $ \sM'(A,\Phi ): \Ker d_{A,\Phi}^*\cap W^{k+2,p } \to \Ker d_{A,\Phi}^*\cap W^{k, p}$ is Fredholm by merit of the direct sum decomposition and invertibility of the operator. We calculate indices, $ $ \begin{aligned } { } & \Ind\left(\sM'(A,\Phi) + d_{A,\Phi}d_{A,\Phi}^*:W^{k+2,p } \to W^{k, p}\right) \\ & \quad = \Ind\begin{pmatrix}\sM

d_{A,\Pji}:W^{k+2,p} \to W^{k,p}$ is Fredholm with index zero by Thxorem \[tgm:Gilkey\_1-4-5\_Robolev\], the following operatlr is ibvertible, $$d_{A,\Phi}^*d_{A,\Phi}:(\Kef d_{A,\Phi}^*d_{A,\Ihi})^\perp \cqp W^{j+2,p} \to (\Ker d_{E,\Lhi}^*d_{A,\Phl})^\'erp \dwp W^{n,'}.$$ Thus, $\xi = 0$ and the operador is injectiee. Flr surjectivity, suppose $\chi = d_{A,\Phi}\zqta \perl \Gan d_{A,\Phi}\cap W^{h,p}$ fpw $\zefa \in W^{k+1,p}$. We may again assume withkut losv of generaliyy that $\zeta \perp \Ker d_{A,\Phl}^*d_{A,\Pji}$. If $\chi\perp d_{A,\Phl}d_{A,\Phi}^*d_{A,\Phi}\zi$ fje all $\xi \in D^{k+3,p}$, then $(d_{A,\Phi}^*d_{A,\Phi})^2\zetz = 0$, and we again find that $\zeta = 0$ anb so $\chi=0$. Sibcw $d_{W,\Khi}d_{A,\Phi}^*: W^{k+2,p} \to W^{h,p}$ has closed range (tecause it is a Fredhplm opwrator), this implies tiat the operator is furjectiva cnd thus invertible bt rhe Oken Mdppivt Tfeodek. Adcordijg vo Propositjon \[prop:Freeholmness\_and\_index\_Laklasv\_pperator\_on\_W1p\_lairs\_sjise\], the operator is Fredholm. Consequentlj, $\sM'(Z,\Phi): \Ker d_{A,\Phi}^*\cap W^{k+2,p} \ti \Ker d_{A,\Phi}^*\cap W^{k,p}$ is Fredholm by virtue of the direct sum decomposition and ineertiuiuitv of tfw lperator. We compute indices, $$\begin{aligned} {}& \Ind\lqrt(\xM'(S,\Phi) + d_{A,\Phi}d_{A,\Pmi}^*:W^{k+2,p} \to W^{k,p}\right) \\ &\qiaf = \Ynd\begin{pmatrkx}\sM

d_{A,\Phi}:W^{k+2,p} \to W^{k,p}$ is Fredholm with index Theorem the following is invertible, $$d_{A,\Phi}^*d_{A,\Phi}:(\Ker d_{A,\Phi}^*d_{A,\Phi})^\perp W^{k,p}.$$ Thus, $\xi 0$ and the is injective. For surjectivity, suppose $\chi d_{A,\Phi}\zeta \perp \Ran d_{A,\Phi}\cap W^{k,p}$ for $\zeta \in W^{k+1,p}$. We may again assume loss of generality that $\zeta \perp \Ker d_{A,\Phi}^*d_{A,\Phi}$. If $\chi\perp d_{A,\Phi}d_{A,\Phi}^*d_{A,\Phi}\xi$ for all \in then = and we again find that $\zeta = 0$ and so $\chi=0$. Since $d_{A,\Phi}d_{A,\Phi}^*: W^{k+2,p} \to W^{k,p}$ closed range (because it is a Fredholm operator), implies that the operator surjective and thus invertible by Open Theorem. According Proposition the is Fredholm. Consequently, \Ker d_{A,\Phi}^*\cap W^{k+2,p} \to \Ker d_{A,\Phi}^*\cap W^{k,p}$ is Fredholm by virtue of the direct sum decomposition and of the compute indices, {}& + \to W^{k,p}\right) \\ \Ind\begin{pmatrix}\sM

d_{A,\Phi}:W^{k+2,p} \to W^{k,p}$ is Fredholm wiTh index zerO by ThEorEm \[tHm:gilkEy\_1-4-5\_SoBolev\], the followINg opErator is invertible, $$d_{A,\PhI}^*d_{A,\PhI}:(\KER d_{A,\PHI}^*d_{a,\Phi})^\pErp \cap W^{K+2,P} \tO (\kEr d_{a,\PHi}^*D_{A,\PHi})^\PErP \cap W^{K,p}.$$ THus, $\xi = 0$ anD the operatOr iS iNjective. For sURjEctivity, suPpoSe $\chi = d_{A,\Phi}\zeTa \pErp \Ran D_{A,\phi}\CAp W^{k,p}$ For $\Zeta \iN W^{k+1,p}$. We MAy agaiN assume wiThOUt loss OF generaLITy That $\Zeta \perp \Ker d_{A,\Phi}^*d_{a,\phI}$. if $\chi\perp d_{A,\Phi}D_{A,\Phi}^*d_{a,\PHI}\xI$ FOr aLl $\xI \in W^{k+3,p}$, then $(D_{A,\phi}^*d_{A,\pHi})^2\zeta = 0$, aND wE AGAin FInd that $\zeta = 0$ anD so $\chi=0$. Since $D_{a,\PhI}d_{A,\Phi}^*: w^{k+2,P} \to w^{K,p}$ has cLosed RaNGe (bEcause it is a fredHolm operaTor), thiS Implies THat the oPeratoR is SurJectIVe AnD thUs INveRTiBle BY thE Open MapPiNg theorEm. AcCORDIng tO PrOposItion \[Prop:FredholmnEss\_And\_iNDex\_laplaCe\_opeRatoR\_oN\_W1p\_paIrs\_sliCe\], the OpErator is FredholM. ConSequently, $\SM'(A,\phI): \KeR d_{a,\Phi}^*\cAP W^{k+2,p} \to \ker D_{A,\PHi}^*\cap W^{k,P}$ is FredHOlm By VIRTuE of the direct sum decOmPOSiTion and iNvertiBIlItY Of the opeRaTor. we coMPUte inDiceS, $$\BeGin{alignEd} {}& \Ind\lEFt(\SM'(a,\Phi) + d_{A,\PHi}D_{A,\Phi}^*:W^{K+2,p} \To W^{K,p}\rIght) \\ &\qUAd = \InD\begin{Pmatrix}\sm

d_{A,\Phi}:W^{k+2,p} \to W ^{k,p}$ is Fred hol m w it h in dexzero by Theore m \[t hm:Gilkey\_1-4-5\_Sobo lev\] ,t he f o ll owing operat o ri s in ve rt ibl e, $$ d_{A, \Ph i}^*d_{ A,\Phi}:(\ Ker d _{A,\Phi}^*d _ {A ,\Phi})^\p erp \cap W^{k+2 ,p} \to ( \K erd _{A,\ Phi }^*d_ {A,\Ph i })^\pe rp \cap W ^{ k ,p}.$$ Thus, $ \ x i= 0$ and the operator is injective. For surje ct i vi t y , s upp ose $\chi=d_{A, \ Phi}\ze t a\ p e rp\ Ran d_{A,\Phi }\cap W^{k, p }$for $\ ze ta\ in W^{ k+1,p }$ . We may againassu me withou t loss of gene r ality t hat $\ zet a \ perp \K er d_ {A , \Ph i }^ *d_ { A,\ Phi}$. I f$\ chi\p erpd _ { A ,\Ph i}d _{A, \Phi} ^*d_{A,\Phi}\ xi$ for all $\xi \inW^{k +3 ,p}$, then$(d_{ A, \Phi}^*d_{A,\Ph i})^ 2\zeta =0$, a ndwe agai n findtha t $ \zeta = 0$ and so$\ c h i =0 $. Since $d_{A,\Ph i} d _ {A ,\Phi}^* : W^{k + 2, p} \to W^{k ,p }$hasc l osedrang e ( becauseit isa F re dholm o pe rator) ,thi s i mplie s tha t theoperator is s u rjective and t h us invertible by t he Open Ma pping Theor em.Acco rdin g t o P r oposi tion\[ p ro p :Fredholmness\_and\ _i ndex\_ Lapla ce\_operator\ _on\_W1p\_ p a i rs\_slic e\], th e operator is F redho lm. Conseq u ently, $ \sM'( A,\Phi): \Ker d_{ A , \Phi}^*\ cap W^ {k+ 2,p } \t o \Ker d_{A,\ P h i}^* \c ap W^{k ,p} $ is Fr edh olm by vi rt ue of the directsu mde co mpo sitio n and inv er tib il ity of t h e oper ator. Weco mp u teindices , $ $ \ begi n{ al igne d}{} & \In d\le f t(\ sM'(A,\ Phi) + d_ {A, \ Phi} d_ {A ,\Phi}^ *:W^{k+2,p} \ to W^{k,p}\r ig ht) \\ &\ q u ad = \In d\begin{pmatrix}\sM

d_{A,\Phi}:W^{k+2,p} \to_W^{k,p}$ is_Fredholm with index zero_by Theorem_\[thm:Gilkey\_1-4-5\_Sobolev\],_the following_operator_is invertible, $$d_{A,\Phi}^*d_{A,\Phi}:(\Ker_d_{A,\Phi}^*d_{A,\Phi})^\perp \cap W^{k+2,p}_\to (\Ker d_{A,\Phi}^*d_{A,\Phi})^\perp \cap_W^{k,p}.$$ Thus, $\xi_=_0$ and the operator is injective. For surjectivity, suppose $\chi = d_{A,\Phi}\zeta \perp \Ran_d_{A,\Phi}\cap_W^{k,p}$ for_$\zeta_\in_W^{k+1,p}$. We may again assume_without loss of generality that_$\zeta \perp_\Ker d_{A,\Phi}^*d_{A,\Phi}$. If $\chi\perp d_{A,\Phi}d_{A,\Phi}^*d_{A,\Phi}\xi$ for all $\xi_\in_W^{k+3,p}$, then $(d_{A,\Phi}^*d_{A,\Phi})^2\zeta_= 0$, and we again find that $\zeta =_0$ and so $\chi=0$. Since $d_{A,\Phi}d_{A,\Phi}^*:_W^{k+2,p} \to W^{k,p}$_has_closed_range (because it is_a Fredholm operator), this implies that_the operator is surjective and thus_invertible by the Open Mapping Theorem. According to_Proposition \[prop:Fredholmness\_and\_index\_Laplace\_operator\_on\_W1p\_pairs\_slice\], the operator is Fredholm._Consequently, $\sM'(A,\Phi): \Ker d_{A,\Phi}^*\cap W^{k+2,p}_\to \Ker_d_{A,\Phi}^*\cap W^{k,p}$ is Fredholm by_virtue of the_direct sum_decomposition and invertibility_of the operator. We compute indices,_$$\begin{aligned} {}& \Ind\left(\sM'(A,\Phi) +_d_{A,\Phi}d_{A,\Phi}^*:W^{k+2,p} \to W^{k,p}\right) \\ &\quad = \Ind\begin{pmatrix}\sM

LeNet [@SzegedyLJSRAEVR15] and ResNet [@HeZRS16]. Besides the success on image categorization, features extracted from the last few layers are applied for generic tasks. [@DonahueJVHZTD14] adopts the deep features from the last two layers in AlexNet and shows the impressive performance on visual recognition with different applications. After that, [@QianJZL15] applies deep features for distance metric learning and achieves the overwhelming performance to the hand-crafted features on fine-grained visual categorization. [@mormont2018comparison] compares deep features from different neural networks and ResNet shows the best results. Besides the model pre-trained on ImageNet, [@ZhouLXTO14] proposes to learn deep features with the large-scale scene data set to improve the performance on the scene recognition task. All of these work directly extract features from the model learned with ERM as the objective. In contrast, we develop an algorithm that is tailored to learn robust deep representations. Note that deep features can be extracted from multiple layers of deep models and we focus on the layer before the final fully-connected layer in this work. Recently, distributionally robust optimization that aims to optimize the worst-case performance has attracted much attention [@ChenLSS17; @NamkoongD16; @sinha2018certifiable]. [@NamkoongD16] proposes to optimize the performance with worst-case distribution over examples that is derived from the empirical distributions. [@ChenLSS17] extends the problem to the non-convex loss function, but they require a near-optimal oracle for the non-convex problem to learn the robust model. [@sinha2018certifiable] introduces the adversarial perturbation on each example for robustness. Most of these algorithms only consider the example-level robustness. In contrast, we propose the hierarchically robust optimization, that considers example-level and concept-level robustness simultaneously, to learn the representations for real applications. Hierarchical Robustness ======================= Problem Formulation {#sec:method} ------------------- Let $\x_i$ denote an image and $y_i\in \R$ be its corresponding label. Given a benchmark data set $\{\x_i, y_i\}$ where $i=1,\ldots, N$, the parameters $\theta$ in a deep neural network can be learned by solving the optimization problem $$\begin{aligned} \

LeNet   [ @SzegedyLJSRAEVR15 ] and ResNet   [ @HeZRS16 ]. Besides the success on image categorization, features educe from the final few layers are applied for generic tasks. [ @DonahueJVHZTD14 ] adopt the deep features from the concluding two layers in AlexNet and shows the impressive operation on ocular recognition with unlike applications. After that, [ @QianJZL15 ] applies deep features for distance measured learning and achieves the overpowering performance to the hand - craft features on fine - grained ocular categorization. [ @mormont2018comparison ] compares bass features from unlike neural network and ResNet shows the best results. Besides the exemplar pre - trained on ImageNet, [ @ZhouLXTO14 ] proposes to learn deep features with the large - scale scene data set to improve the performance on the scene recognition task. All of these work directly educe features from the model memorize with ERM as the aim. In contrast, we develop an algorithm that is tailored to learn robust deep representations. Note that cryptic features can be extracted from multiple layers of deep models and we focus on the level before the final fully - connected layer in this work. Recently, distributionally robust optimization that aims to optimize the bad - case performance has attracted much attention   [ @ChenLSS17; @NamkoongD16; @sinha2018certifiable ]. [ @NamkoongD16 ] proposes to optimize the operation with worst - case distribution over examples that is derived from the empirical distributions. [ @ChenLSS17 ] extends the problem to the non - convex loss routine, but they require a dear - optimal oracle for the non - convex trouble to learn the robust model. [ @sinha2018certifiable ] introduces the adversarial disturbance on each example for robustness. Most of these algorithm only consider the example - horizontal surface robustness. In contrast, we project the hierarchically robust optimization, that consider example - level and concept - level robustness simultaneously, to memorize the representations for real applications. Hierarchical Robustness = = = = = = = = = = = = = = = = = = = = = = = Problem Formulation { # sec: method } ------------------- permit $ \x_i$ denote an image and $ y_i\in \R$ be its corresponding label. Given a benchmark data set $ \{\x_i, y_i\}$ where $ i=1,\ldots, N$, the parameter $ \theta$ in a deep neural network can be learned by solving the optimization problem $ $ \begin{aligned } \

LeNft [@SzegedyLJSRAEVR15] and RerNet [@HeZRS16]. Besidgs the snccess kn image categorization, features extcactwd frim the last few layers are applped for gwnermc tasks. [@DonahueOBHZTD14] adopts fme dezp features from the last dwo layers in DldxUet and shows the impressive performwnce on vlsual recognityon eyth spfnerent applications. After that, [@QjanJZL15] epplies deep festures for distance metric leagning and achieves the overwhgmmigt performancd to the hand-crafted fgatures on fine-grained visual cagegornzation. [@mornobt2018clkparison] conparef deep featuvvs from differrnt neural neteorns qnd ResNet shows the uest results. Besides the moden 're-trained on ImageNer, [@ZhouLFTO14] [ropuwes to lxarh deep feetures with the large-sxale scene data set tj improve the lerforiagce on the scene recognition task. All ox tgese work directly extrqct features from the model lewrned with ERM as the objective. In contrast, we deeelop xn cogorigym that is tailored to learn robust deep represqhtstpons. Note that decp features can be edttwcted from muutiple lagers of deep modeld and wg focuw on the jayet before the final fully-conbected layer un this work. Recentpy, distribucionalky ronust optimization that cims tk optimize hhe worst-dxse performance fas adtracted much attention [@ChegLSS17; @NamkiongB16; @sinha2018cdrtigiable]. [@NamkoongD16] proposes to optimize thf peryormatce with wlrst-case distribution over exam'kes that is cesivvd from tke emplrical distribueions. [@ChenLSS17] gxtends tke proclem to thv non-convxx loss funceion, but they tequire a neac-optimal jracoe fir the vun-convex problrm to leagn the robust model. [@sinhe2018cerjiriable] introduczr the adversariak pdrttrhavion jt each exampne fur fpbustvess. Most on tfese algorithms only convided the example-level rjbustnesw. In conerast, we proppse the hierarchicwlly cobust optikizwtion, that considers example-lebel and clnccpt-level robuftnews simultanepusly, to learn the representations for ceal applications. Hieratchical Robustness ======================= Proyltm Formulatioi {#sec:mqthod} ------------------- Let $\f_i$ denote an image abd $y_i\in \R$ be its gorresponding label. Giben a tenchlark data set $\{\x_i, y_i\}$ where $i=1,\ldots, N$, the parameters $\theta$ in a deep neural betwork can be leadned by smlriny the o[timmzetion problem $$\begpn{aligned} \

LeNet [@SzegedyLJSRAEVR15] and ResNet [@HeZRS16]. Besides the image features extracted the last few tasks. adopts the deep from the last layers in AlexNet and shows the performance on visual recognition with different applications. After that, [@QianJZL15] applies deep features distance metric learning and achieves the overwhelming performance to the hand-crafted features on visual [@mormont2018comparison] deep from different neural networks and ResNet shows the best results. Besides the model pre-trained on ImageNet, proposes to learn deep features with the large-scale data set to improve performance on the scene recognition All these work extract from model learned with as the objective. In contrast, we develop an algorithm that is tailored to learn robust deep representations. that deep be extracted multiple of models and we the layer before the final fully-connected work. Recently, distributionally robust optimization that aims to the worst-case has attracted much attention [@ChenLSS17; @NamkoongD16; [@NamkoongD16] proposes to optimize the performance with worst-case over examples that is derived from the empirical distributions. [@ChenLSS17] extends the problem to the function, but they require near-optimal oracle for non-convex to the model. [@sinha2018certifiable] the adversarial perturbation on each example for robustness. Most of these only consider the example-level robustness. In contrast, we propose the optimization, considers example-level and robustness simultaneously, to learn representations real applications. Hierarchical Robustness Formulation ------------------- an and \R$ be its corresponding Given a benchmark data set y_i\}$ where $i=1,\ldots, N$, deep neural network can be learned by solving optimization problem $$\begin{aligned} \

LeNet [@SzegedyLJSRAEVR15] and ReSNet [@HeZRS16]. BEsideS thE suCcEss oN imaGe categorizatiON, feaTures extracted from the lAst feW lAYers ARe AppliEd for geNErIC TasKs. [@doNahUejvHzTD14] adOptS the deeP features fRom ThE last two layeRS iN AlexNet anD shOws the impresSivE perfoRmAncE On visUal RecogNition WIth difFerent appLiCAtions. aFter thaT, [@qIaNJZL15] Applies deep featurES fOR distance metriC learnInG AnD AChiEveS the overwhElMing pERformanCE tO THE haND-crafted featuRes on fine-grAIneD visuaL cAteGOrizatIon. [@moRmONt2018cOmparison] coMparEs deep feaTures fROm diffeREnt neurAl netwOrkS anD ResnEt ShOws ThE BesT ReSulTS. BeSides the MoDeL pre-tRainED ON imagENeT, [@ZhoULXTO14] Proposes to leaRn dEep fEAtuRes wiTh the LargE-sCale sCene daTa set To Improve the perfoRmanCe on the scEne ReCogNiTion tASk. All oF thEse Work dirEctly exTRacT fEATUrEs from the model learNeD WItH ERM as thE objecTIvE. IN Contrast, We DevElop AN AlgorIthm THaT is tailoRed to lEArN rObust deEp RepresEnTatIonS. Note THat dEep feaTures can Be extRActed from multiPLe layers of deeP MoDELs ANd we FocUs on the layeR befORe thE finAL fUllY-ConneCted lAyER iN This work. Recently, disTrIbutioNally Robust optimizAtion that aIMS To optimiZe thE WoRSt-case performaNce haS attracted MUch attenTion [@CHenLSS17; @NaMkoongD16; @siNHA2018certifiAblE]. [@NaMkoOngd16] PRoPoses to optimiZE The pErFormancE wiTh worst-CasE diStrIbuTiOn over exaMples thaT iS dErIvEd fRom thE EmpiricaL dIstRiButIons. [@CHEnLSS17] eXtendS the PrObLEm tO the non-COnVEX losS fUnCtioN, buT tHey reQuirE A neAr-optimAl oracle fOr tHE non-CoNvEx problEm to learn the rObUst model. [@siNhA2018ceRtifiaBLE] introduCes the adversarial perturBAtion on EacH examPle fOr robustnEss. most of TheSE algorIthms oNly coNsIdeR THe exaMPLe-LevEl Robustness. iN ConTrast, We PropOse the hIerarchically robusT OptImization, that ConSideRS ExAmpLE-lEVel AnD ConCEPt-level robustneSs simultanEoUSlY, to learn thE RepReSentatiOns for rEal apPLicatioNs. HierarcHical RobuStNess ======================= pROblEm FormulatIon {#sec:meThod} ------------------- Let $\x_i$ DEnote AN iMage aNd $y_I\in \R$ be ItS coRrespOnding LAbeL. GiveN a bencHmArk datA set $\{\x_I, y_I\}$ where $i=1,\lDots, N$, the parameters $\theta$ In a deeP neurAl nEtwork can Be lEArnEd by solviNg thE optimizatIon ProBlem $$\bEgiN{AlignEd} \

LeNet [@SzegedyLJSRAEVR15] and ResNe t [@H eZR S16 ]. Bes ides the success o n ima ge categorization, fea tures e x trac t ed from the la s tf e w l ay er s a re ap plied fo r gener ic tasks.[@D on ahueJVHZTD14 ] a dopts thedee p features f rom the l as t t w o lay ers in A lexNet and sh ows the i mp r essive perform a n ce onvisual recognitio n w i th different a pplica ti o ns . Aft erthat, [@Qi an JZL15 ] applie s d e e p fe a tures for dis tance metri c le arning a nda chieve s the o v erw helming per form ance to t he han d -crafte d featur es onfin e-g rain e dvi sua lc ate g or iza t ion . [@morm on t2 018co mpar i s o n ] co mpa resdeepfeatures from di ffer e ntneura l net work sand R esNetshows t he best results . Be sides the mo de l p re -trai n ed onIma geN et, [@Z houLXTO 1 4]pr o p o se s to learn deep fe at u r es with th e larg e -s ca l e sceneda tasett o impr ovet he perform ance o n t he scenere cognit io n t ask . All of t hese w ork dire ctlye xtract feature s from the mod e ll e ar n ed w ith ERM as the obj e ctiv e. I n c ont r ast,we de ve l op an algorithm that i stailor ed to learn robust deep repr e s e ntations . No t et hat deep featu res c an be extr a cted fro m mul tiple la yers of d e e p models an d w e f ocu s on the layer be f o re t he finalful ly-conn ect edlay erin this wor k. Rece nt ly ,di str ibuti o nally ro bu stop tim izati o n that aims toop ti m ize the wo r st - c asepe rf orma nce h as at trac t edmuch at tention [ @Ch e nLSS 17 ;@Namkoo ngD16; @sinha 20 18certifia bl e]. [@Nam k o ongD16]proposes to optimize th e perfor man ce wi th w orst-case di stribu tio n overexampl es th at is d erive d fr omth e empirica l dis tribu ti ons. [@Chen LSS17] extends the pro blem to the n on- conv e x l oss fu n cti on , bu t they require anear-optim al or acle for t h e n on -convex proble m tol earn th e robustmodel. [@ si nha2 0 1 8ce rtifiable] introdu ces the a d versa r ia l per tur bation o n e ach e xample for robu stness .Most o f the se algorit hms only consider the e xample -leve l r obustness . I n co ntrast, w e pr opose thehie rar chica lly robus t op t im iza t ion,that considers ex amp l e -l evel and co n c e pt- level ro b ustnes s si multaneously, tol earn the repre sent a t ion s f o r re al applications. H ie r a rchicalRo bustness == ======== == = ===== ===== Probl em Form u l at i on {#s ec:m eth od} ----- --- -- - ------- - L e t $\x_ i$ d en ote an image and$ y _i\in \R$ be its corr e s pondi n g l abel. G iven ab ench mark dataset $\{\x_i , y_i\ }$ w here$i=1,\l do ts, N$ , t he parameter s $\theta$ in a deep n eu ralnet work c an b e learn ed b ysol ving theo p ti m iz at i onprob lem $ $\ begi n{aligned } \

LeNet [@SzegedyLJSRAEVR15] and_ResNet [@HeZRS16]. Besides_the success on image_categorization, features_extracted_from the_last_few layers are_applied for generic_tasks. [@DonahueJVHZTD14] adopts the_deep features from_the_last two layers in AlexNet and shows the impressive performance on visual recognition with_different_applications. After_that,_[@QianJZL15]_applies deep features for distance_metric learning and achieves the_overwhelming performance_to the hand-crafted features on fine-grained visual categorization._[@mormont2018comparison]_compares deep features_from different neural networks and ResNet shows the best_results. Besides the model pre-trained on_ImageNet, [@ZhouLXTO14] proposes_to_learn_deep features with the_large-scale scene data set to improve_the performance on the scene recognition_task. All of these work directly extract_features from the model learned with_ERM as the objective. In_contrast, we_develop an algorithm that is_tailored to learn_robust deep_representations. Note that_deep features can be extracted from_multiple layers of_deep models and we focus on_the_layer before the_final_fully-connected_layer in_this work. Recently, distributionally_robust_optimization that_aims_to optimize the worst-case performance has_attracted_much attention [@ChenLSS17; @NamkoongD16; @sinha2018certifiable]. [@NamkoongD16] proposes to_optimize the performance with_worst-case_distribution over examples that_is derived from the empirical_distributions. [@ChenLSS17] extends the problem to_the non-convex_loss function,_but they require a near-optimal oracle for the non-convex problem to_learn the robust model. [@sinha2018certifiable] introduces_the adversarial perturbation on_each example_for_robustness. Most of_these_algorithms only_consider the example-level robustness. In contrast, we_propose the_hierarchically robust optimization, that considers example-level_and concept-level robustness simultaneously,_to_learn the representations for real applications. Hierarchical_Robustness ======================= Problem Formulation {#sec:method} ------------------- Let $\x_i$ denote an_image and $y_i\in \R$ be_its_corresponding_label. Given a benchmark data_set $\{\x_i, y_i\}$ where $i=1,\ldots, N$,_the parameters $\theta$_in a deep neural network can be_learned_by solving the optimization problem $$\begin{aligned} \

T_{b}(v) \simeq \tau_{\rm tot} \cdot T_{\rm ex} \label{eqn_Tmb_thin}\end{aligned}$$ indicating that the $T_{\rm ex}$ and $\tau_{\rm tot}$ can be determined arbitrarily. Therefore, we conduct non-LTE analysis in Section \[subsec::non-lte\] to derive the physical conditions. {width="\linewidth"} $N_2H^+$ RADEX Non-LTE Modeling {#subsec::non-lte} ------------------------------- Using RADEX [@radex], a non-LTE radiative transfer code, we construct spectra models in $\rm N_2H^+$ 3–2 and 1–0. The synthetic spectra are constructed with the equation $$\begin{aligned} &&T_{b}(v) = \Psi \left( \frac{\sum J(T_{\rm ex}^i)\,\tau_i(v)}{\sum \tau_i(v)} - J(T_{\rm bg}) \right) \left( 1 - e^{-\sum \tau_i(v)} \right) \nonumber \\ \label{Tb_non-lte}\end{aligned}$$ where $T_{\rm ex}^i$ and $\tau_i(v)$ represent the excitation temperature and optical depth for all hyperfine components in the 3–2 or 1–0 transitions, and $\Psi$ is the beam filling factor. In our models, the beam filling factors are assumed to be unity. We also construct an intensity ratio model by dividing the integrated spectra model in $\rm N_2H^+$ 3–2 with that in $\rm N_2H^+$ 1–0. The constructed models can be represented by a three-dimensional grid, where the three axes are $\rm H_2$ density ($n_{\rm H_2}$) ranging from $10^4$ to $10^9$ $\rm cm^{-3}$, kinetic temperature ($T_{\rm kin}$) ranging from $8$ to $60$ K, and the ratio of $\rm N_2H^+$ column density to linewidth ($N(\rm{N_2H^+})$/$\Delta v$) ranging from $5\times10^{7}$ to $5\times10^{8}$ $\rm s\ cm^{-3}$. The step sizes of the grid are 1

T_{b}(v) \simeq \tau_{\rm tot } \cdot T_{\rm ex } \label{eqn_Tmb_thin}\end{aligned}$$ indicating that the $ T_{\rm ex}$ and $ \tau_{\rm tot}$ can be determined arbitrarily. consequently, we impart non - LTE analysis in Section \[subsec::non - lte\ ] to derive the forcible conditions. ! [ image](non_lte_model_bary.pdf){width="\linewidth " } $ N_2H^+$ RADEX Non - LTE Modeling { # subsec::non - lte } ------------------------------- use RADEX [ @radex ], a non - LTE radiative transfer code, we reconstruct spectra model in $ \rm N_2H^+$ 3–2 and 1–0. The man-made spectra are constructed with the equality $ $ \begin{aligned } & & T_{b}(v) = \Psi \left (\frac{\sum J(T_{\rm ex}^i)\,\tau_i(v)}{\sum \tau_i(v) } - J(T_{\rm bg }) \right) \left (1 - e^{-\sum \tau_i(v) } \right) \nonumber \\ \label{Tb_non - lte}\end{aligned}$$ where $ T_{\rm ex}^i$ and $ \tau_i(v)$ represent the excitation temperature and optical depth for all hyperfine component in the 3–2 or 1–0 transitions, and $ \Psi$ is the beam filling factor. In our models, the radio beam filling factors are assumed to be oneness. We also construct an intensity proportion model by separate the integrated spectra model in $ \rm N_2H^+$ 3–2 with that in $ \rm N_2H^+$ 1–0. The constructed model can be represented by a three - dimensional grid, where the three axes are $ \rm H_2 $ density ($ n_{\rm H_2}$) ranging from $ 10 ^ 4 $ to $ 10 ^ 9 $ $ \rm cm^{-3}$, kinetic temperature ($ T_{\rm kin}$) range from $ 8 $ to $ 60 $ K, and the ratio of $ \rm N_2H^+$ column density to linewidth ($ N(\rm{N_2H^+})$/$\Delta v$) ranging from $ 5\times10^{7}$ to $ 5\times10^{8}$ $ \rm s\ cm^{-3}$. The step sizes of the grid are 1

T_{b}(v) \simeq \tau_{\rm tot} \cdot T_{\ro ex} \label{eqn_Tmb_jhun}\end{anigned}$$ indicatkng that the $T_{\rm ex}$ and $\tau_{\rl rot}$ cqn be determined arbitfarily. Thvrefore, ww coiduct non-LTE analysis in Sectikk \[subvxc::non-lte\] to derlve the phyvical conditiots. {width="\linewidth"} $N_2R^+$ RADEX Nln-LTE Modeling {#subxqc::noh-lte} ------------------------------- Using RADEX [@radex], a non-LTE radjative uransfer code, we cpnstruct spectra models in $\rm J_2H^+$ 3–2 and 1–0. The synthftic spectrq arq constructed with the tqbation $$\begih{aligned} &&T_{b}(v) = \Psi \left( \frac{\sum J(G_{\rm er}^i)\,\tau_i(v)}{\sum \jcy_i(v)} - J(T_{\rm bg}) \rijht) \lest( 1 - e^{-\sum \tam_p(v)} \righd) \nonumbrr \\ \label{Tb_non-lbe}\end{eligbed}$$ where $T_{\rm ex}^i$ and $\tau_i(v)$ represent tre excitadikn temperature ane ipticdl dapth dor alm iypsrfine cojponents ih the 3–2 or 1–0 rransitions, and $\Psi$ if the beam filming fwceor. In our models, the beam filling factmrs are assumed to be unitt. We also construct aj intensiey ratio model by dividing the integrated spectra modem in $\ri B_2H^+$ 3–2 with that in $\rm N_2H^+$ 1–0. The constructed models cwh ne represented bn a three-dimensionsl gtyd, where the jhree arss are $\rm H_2$ density ($j_{\rm H_2}$) rwngint from $10^4$ tj $10^9$ $\rk cm^{-3}$, kinetic temperature ($T_{\rn kin}$) ranginy feom $8$ to $60$ K, and the ratio of $\ro N_2H^+$ colukn density to linewidth ($N(\rm{H_2H^+})$/$\Delta v$) rwnging frko $5\times10^{7}$ to $5\times10^{8}$ $\rm s\ cm^{-3}$. The step sizes of the drid are 1

T_{b}(v) \simeq \tau_{\rm tot} \cdot T_{\rm ex} that $T_{\rm ex}$ $\tau_{\rm tot}$ can conduct analysis in Section to derive the conditions. {width="\linewidth"} $N_2H^+$ RADEX Non-LTE Modeling ------------------------------- Using RADEX [@radex], a non-LTE radiative transfer code, we construct spectra models $\rm N_2H^+$ 3–2 and 1–0. The synthetic spectra are constructed with the equation &&T_{b}(v) \Psi \frac{\sum ex}^i)\,\tau_i(v)}{\sum \tau_i(v)} - J(T_{\rm bg}) \right) \left( 1 - e^{-\sum \tau_i(v)} \right) \nonumber \\ \label{Tb_non-lte}\end{aligned}$$ where ex}^i$ and $\tau_i(v)$ represent the excitation temperature and depth for all hyperfine in the 3–2 or 1–0 and is the filling In models, the beam factors are assumed to be unity. We also construct an intensity ratio model by dividing the integrated model in 3–2 with in N_2H^+$ The constructed models represented by a three-dimensional grid, where are $\rm H_2$ density ($n_{\rm H_2}$) ranging from to $10^9$ cm^{-3}$, kinetic temperature ($T_{\rm kin}$) ranging $8$ to $60$ K, and the ratio of N_2H^+$ column density to linewidth ($N(\rm{N_2H^+})$/$\Delta v$) ranging from $5\times10^{7}$ to $5\times10^{8}$ $\rm s\ cm^{-3}$. sizes of the grid 1

T_{b}(v) \simeq \tau_{\rm tot} \cdot T_{\rm ex} \Label{eqn_TmB_thin}\End{AliGnEd}$$ inDicaTing that the $T_{\rm EX}$ and $\Tau_{\rm tot}$ can be determineD arbiTrARily. tHeReforE, we condUCt NON-LTe aNaLysIs IN SEctioN \[suBsec::non-Lte\] to derivE thE pHysical condiTIoNs. {WidTh="\lineWiDth"} $n_2h^+$ RADEx NoN-LTE MOdelinG {#Subsec::Non-lte} ------------------------------- UsiNg raDEX [@raDEx], a non-Lte RaDiatIve transfer code, we COnSTruct spectra moDels in $\Rm n_2h^+$ 3–2 aND 1–0. the SynThetic specTrA are cONstructED wITH The EQuation $$\begin{aLigned} &&T_{b}(v) = \PsI \LefT( \frac{\sUm j(T_{\rM Ex}^i)\,\tau_I(v)}{\sum \TaU_I(v)} - J(t_{\rm bg}) \right) \lEft( 1 - e^{-\Sum \tau_i(v)} \rIght) \noNUmber \\ \laBEl{Tb_non-Lte}\end{AliGneD}$$ wheRE $T_{\Rm Ex}^i$ AnD $\Tau_I(V)$ rEprESenT the exciTaTiOn temPeraTURE And oPtiCal dEpth fOr all hyperfinE coMponENts In the 3–2 Or 1–0 traNsitIoNs, and $\psi$ is tHe beaM fIlling factor. In oUr moDels, the beAm fIlLinG fActorS Are assUmeD to Be unity. we also cONstRuCT AN iNtensity ratio model By DIViDing the iNtegraTEd SpECtra modeL iN $\rm n_2H^+$ 3–2 wiTH That iN $\rm N_2h^+$ 1–0. thE construCted moDElS cAn be repReSented By A thRee-DimenSIonaL grid, wHere the tHree aXEs are $\rm H_2$ densitY ($N_{\rm H_2}$) ranging frOM $10^4$ tO $10^9$ $\RM cM^{-3}$, KineTic Temperature ($t_{\rm kIN}$) ranGing FRoM $8$ to $60$ k, And thE ratiO oF $\Rm n_2h^+$ column density to linEwIdth ($N(\rM{N_2H^+})$/$\DeLta v$) ranging frOm $5\times10^{7}$ to $5\tIMES10^{8}$ $\rm s\ cm^{-3}$. ThE steP SiZEs of the grid are 1

T_{b}(v) \simeq \tau_{\rmtot} \cdot T_{\ rmex} \ labe l{eq n_Tmb_thin}\en d {ali gned}$$ indicating tha t the $ T _{\r m e x}$ a nd $\ta u _{ \ r m t ot }$ ca nb edeter min ed arbi trarily. T her ef ore, we cond u ct non-LTE a nal ysis in Sect ion \[sub se c:: n on-lt e\] to d erivet he phy sical con di t ions.![image ] ( no n_lt e_model_bary.pdf) { wi d th="\linewidth "} $N _2 H ^+ $ RAD EXNon-LTE Mo de ling{ #subsec : :n o n - lte } ------------ ----------- - --- ---- Us ing RADEX[@rad ex ] , a non-LTE ra diat ive trans fer co d e, we c o nstruct spect ramod elsi n$\ rmN_ 2 H^+ $ 3 –2a nd1–0. The s yn theti c sp e c t r a ar e c onst ructe d with the eq uat ion$ $\b egin{ align ed}&& T_{b} (v) =\Psi\l eft( \frac{\sum J(T _{\rm ex} ^i) \, \ta u_ i(v)} { \sum \ tau _i( v)} - J (T_{\rm bg} )\ r i gh t) \left( 1 - e^{- \s u m \ tau_i(v) } \rig h t) \ n onumber\\ \l abel { T b_non -lte } \e nd{align ed}$$w he re $T_{\r mex}^i$ a nd$\t au_i( v )$ r eprese nt the e xcita t ion temperatur e and opticald ep t h f o r al l h yperfine co mpon e ntsin t h e3–2 or 1– 0 tra ns i ti o ns, and $\Psi$ is t he beamfilli ng factor. In our model s , the beam fil l in g factors are a ssume d to be un i ty. We a lso c onstruct an inten s i ty ratio mo del by di v i di ng the integr a t ed s pe ctra mo del in $\r m N _2H ^+$ 3– 2with that in $\rm N _2 H^ +$ 1– 0. T h e constr uc ted m ode ls ca n be re prese nted b ya th ree-dim e ns i o nalgr id , wh ere t he th reea xes are $\ rm H_2$ d ens i ty ( $n _{ \rm H_2 }$) ranging f ro m $10^4$ t o$10 ^9$ $\ r m cm^{-3} $, kinetic temperature( $T_{\rm ki n}$)rang ing from$8$ to $6 0$K , andthe ra tio o f$\r m N_2H^ + $ c olu mn density t o lin ewidt h($N( \rm{N_2 H^+})$/$\Delta v$) ran ging from $5\ tim es10 ^ { 7} $ t o $ 5 \ti me s 10^ { 8 }$ $\rm s\ cm^{ -3}$. Thest e psizes of t h e g ri d are 1

T_{b}(v) \simeq_\tau_{\rm tot}_\cdot T_{\rm ex} \label{eqn_Tmb_thin}\end{aligned}$$ indicating_that the_$T_{\rm_ex}$ and_$\tau_{\rm_tot}$ can be_determined arbitrarily. Therefore,_we conduct non-LTE analysis_in Section \[subsec::non-lte\]_to_derive the physical conditions. {width="\linewidth"} $N_2H^+$ RADEX Non-LTE Modeling {#subsec::non-lte} ------------------------------- Using RADEX [@radex], a non-LTE radiative transfer_code,_we construct_spectra_models_in $\rm N_2H^+$ 3–2 and_1–0. The synthetic spectra are_constructed with_the equation $$\begin{aligned} &&T_{b}(v) = \Psi \left( \frac{\sum J(T_{\rm_ex}^i)\,\tau_i(v)}{\sum_\tau_i(v)} - J(T_{\rm_bg}) \right) \left( 1 - e^{-\sum \tau_i(v)} \right) \nonumber \\ \label{Tb_non-lte}\end{aligned}$$_where $T_{\rm ex}^i$ and $\tau_i(v)$ represent_the excitation temperature_and_optical_depth for all hyperfine_components in the 3–2 or 1–0_transitions, and $\Psi$ is the beam_filling factor. In our models, the beam_filling factors are assumed to be_unity. We also construct an_intensity ratio_model by dividing the integrated_spectra model in_$\rm N_2H^+$_3–2 with that_in $\rm N_2H^+$ 1–0. The constructed models_can be represented_by a three-dimensional grid, where the_three_axes are $\rm_H_2$_density_($n_{\rm H_2}$)_ranging from $10^4$_to_$10^9$ $\rm_cm^{-3}$,_kinetic temperature ($T_{\rm kin}$) ranging from_$8$_to $60$ K, and the ratio of_$\rm N_2H^+$ column density_to_linewidth ($N(\rm{N_2H^+})$/$\Delta v$) ranging_from $5\times10^{7}$ to $5\times10^{8}$ $\rm_s\ cm^{-3}$. The step sizes of_the grid_are 1

vertex ordering can make a substantial difference in the size of the search space, but that there is no “best” ordering. Sequential Branch and Bound --------------------------- In we give the underlying sequential procedure. The variable $C$ is a growing clique, and $P$ contains undecided vertices which could potentially be added to $C$. The best solution found so far is stored in $C_{max}$. The important part, for this discussion, is that we have a $\KwSty{for}$ loop, and that the $\FuncSty{expand}$ function calls itself recursively inside this loop. Some of these recursive calls are then avoided, by using a greedy graph colouring to prove that the current portion of the search space could not possibly lead to a better solution than the one in $C_{max}$ being found. (Graph $G$, Set $C$, Set $P$, Set $C_{max}$) (Graph $G$) $\rightarrow$ Set We may view the recursive calls made by a branch and bound algorithm as forming a tree, as in. Here, each node in the tree represents a recursive call; triangles represent large subtrees. We mark by $\star$ the location of an optimal solution. Nodes shown in medium blue are those which cannot be eliminated by the bound, regardless of the strength of the incumbent—we say such nodes are *ineliminable*. Nodes in light blue are those which could be eliminated by the bound, if $\star$ has been found—such nodes are *eliminable*. The tree is traversed in a manner similar to a depth-first search, exploring subtrees from left to right. Note that in a sequential run, the leftmost light blue subtree will *not* be eliminated by the bound, since the search will not yet have found $\star$. In other words, not all eliminable nodes are necessarily eliminated. We call eliminable nodes that are not eliminated in the sequential run *avoidable*. In maximum clique terms, $\star$ is the location of a maximum clique, whose size is denoted $\omega$. The medium blue nodes are then the nodes which must be explored to prove that there is no clique of size $\omega + 1$ in the graph. Light blue nodes are those that would be eliminated by the bound, if the algorithm were to be initialised with an incumbent size of $\omega$ rather than $0$. For now we assume there is

vertex ordering can make a substantial difference in the size of the search quad, but that there be no “ best ” ordering. Sequential Branch and Bound --------------------------- In we give the underlie sequential procedure. The variable star $ C$ is a growing clique, and $ P$ contains undecided vertex which could potentially be added to $ C$. The best solution find so far is store in $ C_{max}$. The authoritative part, for this discussion, is that we have a $ \KwSty{for}$ loop, and that the $ \FuncSty{expand}$ function calls itself recursively inside this cringle. Some of these recursive calls are then avoided, by using a avid graph colouring to prove that the current portion of the search space could not possibly leave to a better solution than the one in $ C_{max}$ being found. (Graph $ G$, Set $ C$, Set $ P$, Set $ C_{max}$) (Graph $ G$) $ \rightarrow$ Set We may view the recursive call made by a branch and bound algorithm as forming a tree, as in. Here, each node in the tree constitute a recursive call; triangles represent large subtrees. We mark by $ \star$ the location of an optimal solution. Nodes shown in medium blue are those which cannot be extinguish by the bound, regardless of the strength of the incumbent — we suppose such node are * ineliminable *. Nodes in light amobarbital sodium are those which could be eliminated by the bound, if $ \star$ has been found — such nodes are * eliminable *. The tree is traverse in a manner similar to a depth - first search, exploring subtrees from left to right. notice that in a sequential run, the leftmost light blue subtree will * not * be eliminated by the bound, since the search will not yet have find $ \star$. In other words, not all eliminable nodes are necessarily eliminated. We bid eliminable nodes that are not eliminated in the sequential run * avoidable *. In maximum clique terms, $ \star$ is the location of a maximal clique, whose size is denoted $ \omega$. The medium blue nodes are then the nodes which must be research to prove that there embody no clique of size $ \omega + 1 $ in the graph. Light gloomy nodes are those that would be eliminate by the bound, if the algorithm were to be initialised with an incumbent size of $ \omega$ quite than $ 0$. For now we bear there is

vegtex ordering can make a substantial diyderencx in ths size ow the search space, but that vherw is bo “best” ordering. Sequengial Branbh and Boynd --------------------------- Ii we give the unvsrlying sequehbial 'ricedure. The vatiable $C$ is d growing cliqge, aud $P$ contains undecided vertices whish coulc ootentially be addtd eo $C$. Nht best solution found so far is sfored ii $C_{max}$. The impottant part, for this discusdion, is that we have a $\KwSty{for}$ liop, wbd that the $\WuncSty{expand}$ function calls itself recursively insidd thix loop. Somg kf jhese recursmve cajls are then avoided, by usimg a greedy grsph cooouring to prove that the current portijn of the szarch space could not pissibny laad gi a beutec sklutioj tian the one in $C_{max}$ beung found. (Graph $G$, Seu $C$, Wet $P$, Set $C_{mas}$) (Grapr $D$) $\rightarrow$ Set We may view the recursine czlls made by a branch abd bound algorithm as forming w tree, as in. Here, each node in the tree representv a rxcjrsnyc cauo; hriangles represent large subtrees. We mark by $\atsr$ the location jf an optimsl spjution. Nodes rhown nh jedium blue are thlse whish cabnot be ejimimated by the bound, regardlews of the stgengrh of the incumbenc—we say such nodgs are *ineliminable*. Nodes in uighf blue are hhose whidf could be elimivatvd bf the bound, if $\star$ has beqn found—snch npdes ard *elominabje*. The tref is bsaversed in a mannfr silinar to a dfpth-first search, exploring subtcxes from left tm rpght. Note that in a sequentyal run, the leytmost lnght buue subtrev will *nov* be eliminaeed by the bogjd, since the search rill not yet haxd found $\star$. Im other words, not aol eliminable nodex ate necessarily elnounated. We call rlioinwbpe njges that are not elkkinatdd in the scqudntisl run *avoidable*. In mdximhm clique terms, $\stsr$ is the oocation of a maximum clique, whose size is dxnoted $\omegs$. Tre medium blue nodes are then fhe nodes whlch must be evploved jo prove thct there is no clique of size $\omega + 1$ in the graph. Light boue nodes are those uhat would be elimynated by the bound, if the altorithm were to bt initialised with an ihcumbett sixe of $\omega$ rather than $0$. For now we assume there is

vertex ordering can make a substantial difference size the search but that there Branch Bound --------------------------- In give the underlying procedure. The variable $C$ is a clique, and $P$ contains undecided vertices which could potentially be added to $C$. best solution found so far is stored in $C_{max}$. The important part, for discussion, that have $\KwSty{for}$ loop, and that the $\FuncSty{expand}$ function calls itself recursively inside this loop. Some of these calls are then avoided, by using a greedy colouring to prove that current portion of the search could possibly lead a solution the one in being found. (Graph $G$, Set $C$, Set $P$, Set $C_{max}$) (Graph $G$) $\rightarrow$ Set We may view recursive calls a branch bound as a tree, as each node in the tree represents triangles represent large subtrees. We mark by $\star$ location of optimal solution. Nodes shown in medium are those which cannot be eliminated by the regardless of the strength of the incumbent—we say such nodes are *ineliminable*. Nodes in light those which could be by the bound, $\star$ been nodes *eliminable*. The is traversed in a manner similar to a depth-first search, exploring from left to right. Note that in a sequential run, light subtree will *not* eliminated by the bound, the will not yet have In words, nodes necessarily We call eliminable nodes are not eliminated in the run *avoidable*. In maximum location of a maximum clique, whose size is $\omega$. The medium blue nodes are then nodes which must be explored to prove that there is no clique size $\omega in the graph. Light blue nodes are those would be eliminated by bound, if the algorithm were to be initialised with incumbent of $\omega$ than $0$. For we assume there

vertex ordering can make a subStantial diFfereNce In tHe Size Of thE search space, buT That There is no “best” ordering. SEquenTiAL BraNCh And BoUnd --------------------------- In we GIvE THe uNdErLyiNg SEqUentiAl pRocedurE. The variabLe $C$ Is A growing cliqUE, aNd $P$ containS unDecided vertiCes Which cOuLd pOTentiAllY be adDed to $C$. tHe best Solution fOuND so far IS stored IN $c_{mAx}$. ThE important part, for THiS Discussion, is thAt we haVe A $\kwsTY{foR}$ loOp, and that tHe $\funcSTY{expand}$ FUnCTIOn cALls itself recuRsively insiDE thIs loop. soMe oF These rEcursIvE CalLs are then avOideD, by using a Greedy GRaph colOUring to Prove tHat The CurrENt PoRtiOn OF thE SeArcH SpaCe could nOt PoSsiblY leaD TO A BettEr sOlutIon thAn the one in $C_{maX}$ beIng fOUnd. (graph $g$, Set $C$, set $P$, seT $C_{max}$) (graph $G$) $\RightArRow$ Set We may view The rEcursive cAllS mAde By A branCH and boUnd AlgOrithm aS forminG A trEe, AS IN. HEre, each node in the trEe REPrEsents a rEcursiVE cAlL; TriangleS rEprEsenT LArge sUbtrEEs. we mark by $\Star$ thE LoCaTion of aN oPtimal SoLutIon. nodes SHown In mediUm blue arE thosE Which cannot be eLIminated by the BOuND, ReGArdlEss Of the strengTh of THe inCumbENt—We sAY such Nodes ArE *InELiminable*. Nodes in ligHt Blue arE thosE which could be Eliminated BY THe bound, iF $\staR$ HaS Been found—such nOdes aRe *eliminabLE*. The tree Is traVersed in A manner siMILar to a dePth-FirSt sEarCH, ExPloring subtreES From LeFt to rigHt. NOte that In a SeqUenTiaL rUn, the leftMost lighT bLuE sUbTreE will *NOt* be elimInAteD bY thE bounD, Since tHe seaRch wIlL nOT yeT have foUNd $\STAr$. In OtHeR worDs, nOt All elIminABle Nodes arE necessarIly ELimiNaTeD. We call Eliminable nodEs That are not ElImiNated iN THe sequenTial run *avoidable*. In maximUM clique TerMs, $\staR$ is tHe locatioN of A maximUm cLIque, whOse sizE is deNoTed $\OMEga$. ThE MEdIum BlUe nodes are THEn tHe nodEs WhicH must be Explored to prove thaT TheRe is no clique oF siZe $\omEGA + 1$ iN thE GrAPh. LIgHT blUE Nodes are those thAt would be eLiMInAted by the bOUnd, If The algoRithm weRe to bE InitialIsed with aN incumbenT sIze oF $\OMegA$ rather thaN $0$. For now wE assume thERe is

vertex ordering can makea substant ial d iff ere nc e in the size of the s e arch space, but that there is n o“ best ” o rderi ng. Se q ue n t ial B ra nch a n dBound -- ------- ---------- --- -- --- In we g i ve the under lyi ng sequentia l p rocedu re . T h e var iab le $C $ is a growin g clique, a n d $P$c ontains u nd ecid ed vertices which co u ld potentially be ad de d t o $C$ . T he best so lu tionf ound so fa r i s s t ored in $C_{m ax}$. The i m por tant p ar t,f or thi s dis cu s sio n, is thatwe h ave a $\K wSty{f o r}$ loo p , and t hat th e $ \Fu ncSt y {e xp and }$ fun c ti onc all s itself r ec ursiv elyi n s i de t his loo p. So me of these r ecu rsiv e ca lls a re th en a vo ided, by us ing a g reedy graph col ouri ng to pro veth atth e cur r ent po rti onof thesearchs pac ec o u ld not possibly lead t o abetter s olutio n t ha n the one i n $ C_{m a x }$ be ingf ou nd. (Gr aph $G $ ,Se t $C$,Se t $P$, S et$C_ {max} $ ) (G raph $ G$) $\ri ghtar r ow$ Set We ma y view the rec u rs i v ec alls ma de by a bra ncha nd b ound al gor i thm a s for mi n ga tree, as in. Here, e ach no de in the tree rep resents ar e c ursive c all; tr i angles represe nt la rge subtre e s. We ma rk by $\star$ the loca t i on of an op tim alsol u t io n. Nodes show n in m ed ium blu e a re thos e w hic h c ann ot be elimi nated by t he b ou nd, rega r dless of t hest ren gth o f the i ncumb ent— we s a y s uch nod e sa r e *i ne li mina ble *. Node s in lig ht blue are thos e w h ichco ul d be el iminated by t he bound, if $ \st ar$ ha s been fou nd—such nodes are *elim i nable*. Th e tre e is traverse d i n a ma nne r simil ar toa dep th -fi r s t sea r c h, ex pl oring subt r e esfromle ft t o right . Note that in a s e que ntial run, th e l eftm o s tlig h tb lue s u btr e e will *not* beeliminated b y t he bound,s inc ethe sea rch wil l not yet hav e found $ \star$. I nothe r wor ds, not al l elimin able node s aren ec essar ily elimi na ted . Wecall e l imi nable nodes t hat ar e not e liminate d in the sequential run *avoi dable *. In maxim umc liq ue terms, $\s tar$ is th e l oca tionofa maxi mumc li que , whos e si z e is deno t ed $\ o m eg a$. The med i u m bl ue no des are th en t he nodes which mu s t be exploredto p r o vetha t the re is no cliqueofsi z e $\omega + 1$ in thegraph. L ig h t blu e node s arethose t h a tw ould b e el imi nated bythe b o und, if t he algori thmwe re tobe ini t iali s e d with an incumb ent s i z e of$ \om ega$ra ther th a n $0 $. For now we assumethereis

vertex_ordering can_make a substantial difference_in the_size_of the_search_space, but that_there is no_“best” ordering. Sequential Branch and_Bound --------------------------- In we give_the_underlying sequential procedure. The variable $C$ is a growing clique, and $P$ contains undecided_vertices_which could_potentially_be_added to $C$. The best_solution found so far is_stored in_$C_{max}$. The important part, for this discussion, is_that_we have a_$\KwSty{for}$ loop, and that the $\FuncSty{expand}$ function calls itself_recursively inside this loop. Some of_these recursive calls_are_then_avoided, by using a_greedy graph colouring to prove that_the current portion of the search_space could not possibly lead to a_better solution than the one in_$C_{max}$ being found. (Graph $G$, Set_$C$, Set_$P$, Set $C_{max}$) (Graph $G$)_$\rightarrow$ Set We may_view the_recursive calls made_by a branch and bound algorithm_as forming a_tree, as in. Here, each node_in_the tree represents_a_recursive_call; triangles_represent large subtrees._We_mark by_$\star$_the location of an optimal solution._Nodes_shown in medium blue are those which_cannot be eliminated by_the_bound, regardless of the_strength of the incumbent—we say_such nodes are *ineliminable*. Nodes in_light blue_are those_which could be eliminated by the bound, if $\star$ has been_found—such nodes are *eliminable*. The tree_is traversed in a_manner similar_to_a depth-first search,_exploring_subtrees from_left to right. Note that in a_sequential run,_the leftmost light blue subtree will_*not* be eliminated by_the_bound, since the search will not_yet have found $\star$. In other_words, not all eliminable nodes_are_necessarily_eliminated. We call eliminable nodes_that are not eliminated in the_sequential run *avoidable*. In_maximum clique terms, $\star$ is the location_of_a maximum clique, whose size is_denoted_$\omega$. The medium blue nodes are_then_the_nodes which must be explored_to prove that there is no_clique of size $\omega + 1$ in the graph._Light blue nodes_are those that would be_eliminated_by_the bound, if the algorithm were to be initialised with_an incumbent_size of $\omega$_rather than $0$. For now we assume there is

12; H^1(\Omega'))\hookrightarrow BUC([-\tfrac12, \tfrac12];H^{\frac12}(\Omega'))$$ and $H^{\frac12}(\Omega')\hookrightarrow L^4(\Omega')$. Finally, the last embedding follows from $$\begin{aligned} \lefteqn{L^2(-\tfrac12,\tfrac12; H^{1+k}((-L,L)^{d-1}))\cap H^1(-\tfrac12,\tfrac12;H^1((-L,L)^{d-1}))}\\ && \hookrightarrow BUC([-\tfrac12,\tfrac12]; H^{1+\frac{k}2}((-L,L)^{d-1}))\hookrightarrow C^0({\overline}{\Omega}) \end{aligned}$$ where $k=d-2$ because of (\[eq:BUCEmbedding\]) and Sobolev embeddings. The spaces $H^{1,0}(\Omega)$ and $V(\Omega)$ are the fundamental spaces, which will be used to solve the evolution equation. We note that $$f\in V(\Omega)\quad \Leftrightarrow \quad f, \nabla f \in H^{1,0}(\Omega).$$ Most of the time we will estimate $f\in V(\Omega)$ by the $h$-dependent norm $$\|f\|_{V_h}:= \|(f,\nabla_h f)\|_{H^{1,0}(\Omega)}.$$ Because of the embedding $V(\Omega)\hookrightarrow L^\infty(\Omega)$, we are able to show that $V(\Omega)$ is an algebra with respect to point-wise multiplication. More precisely, we obtain: \[cor:Algebra\] Let $d=2,3$. Then there is some $C=C(\Omega)>0$ such that $$\label{eq:ProdEstim} \|(u_1\cdot u_2, \nabla_h (u_1\cdot u_2)\|_{H^{1,0}(\Omega)}\leq C\|(u_1, \nabla_h u_1)\|_{H^{1,0}(\Omega)}\|(u_2, \nabla_h u_2)\|_{H^{1,0}(\Omega)}$$ for all $u_1,u_2\in V(\Omega)$ uniformly in $0<h\leq 1$. Moreover, if $F\in C^{2

12; H^1(\Omega'))\hookrightarrow BUC([-\tfrac12, \tfrac12];H^{\frac12}(\Omega'))$$ and $ H^{\frac12}(\Omega')\hookrightarrow L^4(\Omega')$. Finally, the last embedding follows from $ $ \begin{aligned } \lefteqn{L^2(-\tfrac12,\tfrac12; H^{1+k}((-L, L)^{d-1}))\cap H^1(-\tfrac12,\tfrac12;H^1((-L, L)^{d-1}))}\\ & & \hookrightarrow BUC([-\tfrac12,\tfrac12 ]; H^{1+\frac{k}2}((-L, L)^{d-1}))\hookrightarrow C^0({\overline}{\Omega }) \end{aligned}$$ where $ k = d-2 $ because of (\[eq: BUCEmbedding\ ]) and Sobolev embeddings. The space $ H^{1,0}(\Omega)$ and $ V(\Omega)$ are the cardinal spaces, which will be used to resolve the development equation. We note that $ $ f\in V(\Omega)\quad \Leftrightarrow \quad f, \nabla f \in H^{1,0}(\Omega).$$ Most of the clock time we will estimate $ f\in V(\Omega)$ by the $ h$-dependent norm $ $ \|f\|_{V_h}:= \|(f,\nabla_h f)\|_{H^{1,0}(\Omega)}.$$ Because of the embedding $ V(\Omega)\hookrightarrow L^\infty(\Omega)$, we are able to prove that $ V(\Omega)$ is an algebra with respect to point - fresh multiplication. More precisely, we obtain: \[cor: Algebra\ ] lease $ d=2,3$. Then there is some $ C = C(\Omega)>0 $ such that $ $ \label{eq: ProdEstim } \|(u_1\cdot u_2, \nabla_h (u_1\cdot u_2)\|_{H^{1,0}(\Omega)}\leq C\|(u_1, \nabla_h u_1)\|_{H^{1,0}(\Omega)}\|(u_2, \nabla_h u_2)\|_{H^{1,0}(\Omega)}$$ for all $ u_1,u_2\in V(\Omega)$ uniformly in $ 0 < h\leq 1$. Moreover, if $ F\in C^{2

12; H^1(\Olega'))\hookrightarrow BUC([-\tfvac12, \tfrac12];H^{\frac12}(\Omggq'))$$ and $I^{\frac12}(\Omsga')\hookrkghtarrow L^4(\Omega')$. Finally, the lqst enbedding follows from $$\cegin{aligjed} \lwftewb{L^2(-\tfrac12,\tfred12; H^{1+k}((-L,L)^{d-1}))\gcp H^1(-\fnrac12,\tyrec12;H^1((-L,L)^{d-1}))}\\ && \hookrighjarrow BUC([-\tfsac12,\tfrac12]; H^{1+\fraw{k}2}((-U,L)^{b-1}))\hookrightarrow C^0({\overline}{\Omega}) \end{ajigned}$$ ehfre $k=d-2$ because of (\[tq:BTCEmgvdbing\]) and Sobolev embeddings. The apaces $I^{1,0}(\Omega)$ and $V(\Omeba)$ are the fundamental spafes, ahich will be used to solve tye edilution equagion. We noue that $$f\in B(\Omega)\quad \Leftrightarrow \quad w, \nabka f \in H^{1,0}(\Oneta).$$ Lmst of the vime wv will estimabv $f\in V(\Mmega)$ bu the $h$-dependekt nocm $$\|f\|_{C_h}:= \|(f,\nabla_h f)\|_{H^{1,0}(\Omega)}.$$ Berause of the embeddigg $V(\Omega)\vokkrightarrow L^\infry(\Imega)$, we dre xvle to siow that $G(\Omxga)$ is an amgebra with respect to point-wixe nultiplicatioh. More pwecisely, we obtain: \[cor:Algebra\] Let $d=2,3$. Then thsre is some $C=C(\Omega)>0$ sucy that $$\label{eq:ProdEstlm} \|(t_1\cdot u_2, \nabla_h (u_1\cdot u_2)\|_{H^{1,0}(\Omega)}\leq C\|(u_1, \nabla_h u_1)\|_{H^{1,0}(\Omega)}\|(g_2, \nabmx_h b_2)\|_{M^{1,0}(\Jowgw)}$$ for all $u_1,u_2\in V(\Omega)$ uniformly in $0<h\leq 1$. Morejbet, pf $F\in C^{2

12; H^1(\Omega'))\hookrightarrow BUC([-\tfrac12, \tfrac12];H^{\frac12}(\Omega'))$$ and $H^{\frac12}(\Omega')\hookrightarrow L^4(\Omega')$. last follows from \lefteqn{L^2(-\tfrac12,\tfrac12; H^{1+k}((-L,L)^{d-1}))\cap H^1(-\tfrac12,\tfrac12;H^1((-L,L)^{d-1}))}\\ \end{aligned}$$ $k=d-2$ because of and Sobolev embeddings. spaces $H^{1,0}(\Omega)$ and $V(\Omega)$ are the spaces, which will be used to solve the evolution equation. We note that V(\Omega)\quad \Leftrightarrow \quad f, \nabla f \in H^{1,0}(\Omega).$$ Most of the time we estimate V(\Omega)$ the norm $$\|f\|_{V_h}:= \|(f,\nabla_h f)\|_{H^{1,0}(\Omega)}.$$ Because of the embedding $V(\Omega)\hookrightarrow L^\infty(\Omega)$, we are able to show that is an algebra with respect to point-wise multiplication. precisely, we obtain: \[cor:Algebra\] $d=2,3$. Then there is some such $$\label{eq:ProdEstim} \|(u_1\cdot \nabla_h u_2)\|_{H^{1,0}(\Omega)}\leq \nabla_h u_1)\|_{H^{1,0}(\Omega)}\|(u_2, \nabla_h for all $u_1,u_2\in V(\Omega)$ uniformly in $0<h\leq 1$. Moreover, if $F\in C^{2

12; H^1(\Omega'))\hookrightarrow BUC([-\tfRac12, \tfrac12];H^{\fRac12}(\OmEga'))$$ And $h^{\fRac12}(\OMega')\Hookrightarrow l^4(\omegA')$. Finally, the last embeddiNg folLoWS froM $$\BeGin{alIgned} \leFTeQN{l^2(-\tfRaC12,\tFraC12; H^{1+K}((-l,L)^{D-1}))\cap H^1(-\TfrAc12,\tfrac12;h^1((-L,L)^{d-1}))}\\ && \hookriGhtArRow BUC([-\tfrac12,\tFRaC12]; H^{1+\frac{k}2}((-L,L)^{d-1}))\HooKrightarrow C^0({\OveRline}{\OMeGa}) \eND{aligNed}$$ Where $K=d-2$ becaUSe of (\[eq:bUCEmbeddInG\]) And SobOLev embeDDInGs. ThE spaces $H^{1,0}(\Omega)$ and $V(\oMeGA)$ are the fundameNtal spAcES, wHICh wIll Be used to soLvE the eVOlution EQuATIOn. WE Note that $$f\in V(\OMega)\quad \LefTRigHtarroW \qUad F, \Nabla f \In H^{1,0}(\OmEgA).$$ mosT of the time wE wilL estimate $F\in V(\OmEGa)$ by the $H$-DependeNt norm $$\|F\|_{V_h}:= \|(F,\naBla_h F)\|_{h^{1,0}(\OMeGa)}.$$ BEcAUse OF tHe eMBedDing $V(\OmeGa)\HoOkrigHtarROW l^\InftY(\OmEga)$, wE are aBle to show that $v(\OmEga)$ iS An aLgebrA with RespEcT to poInt-wisE multIpLication. More preCiseLy, we obtaiN: \[coR:ALgeBrA\] Let $d=2,3$. tHen theRe iS soMe $C=C(\OmeGa)>0$ such tHAt $$\lAbEL{EQ:PRodEstim} \|(u_1\cdot u_2, \nablA_h (U_1\CDoT u_2)\|_{H^{1,0}(\Omega)}\Leq C\|(u_1, \nABlA_h U_1)\|_{h^{1,0}(\Omega)}\|(u_2, \nAbLa_h U_2)\|_{H^{1,0}(\OmEGA)}$$ for aLl $u_1,u_2\IN V(\omega)$ uniFormly IN $0<h\LeQ 1$. MoreovEr, If $F\in C^{2

12; H^1(\Omega'))\hookrigh tarrow BUC ([-\t fra c12 ,\tfr ac12 ];H^{\frac12}( \ Omeg a'))$$ and $H^{\frac12 }(\Om eg a ')\h o ok right arrow L ^ 4( \ O meg a' )$ . F in a ll y, th e l ast emb edding fol low sfrom $$\begi n {a ligned} \ lefteqn{L^2( -\t frac12 ,\ tfr a c12;H^{ 1+k}( (-L,L) ^ {d-1}) )\cap H ^1 ( -\tfra c 12,\tfr a c 12 ;H^1 ((-L,L)^{d-1}))}\ \ & & \hookrightarr ow BUC ([ - \t f r ac1 2,\ tfrac12]; H^{1+ \ frac{k} 2 }( ( - L ,L) ^ {d-1}))\hookr ightarrow C ^ 0({ \overl in e}{ \ Omega} ) \ en d {al igned}$$ wh ere$k=d-2$ b ecause of (\[e q :BUCEmb edding \]) an d So b ol ev em be d din g s. T h e s paces $H ^{ 1, 0}(\O mega ) $ a nd $ V(\ Omeg a)$ a re the fundam ent al s p ace s, wh ich w illbe used to so lve t he evolution equa tion . We note th at $$ f\ in V( \ Omega) \qu ad\Leftri ghtarro w \q ua d f ,\nabla f \in H^{1, 0} ( \ Om ega).$$Most o f t he time wewi llesti m a te $f \inV (\ Omega)$by the $h $- depende nt norm$$ \|f \|_ {V_h} : = \ |(f,\n abla_h f )\|_{ H ^{1,0}(\Omega) } .$$ Becauseo ft h ee mbed din g $V(\Omega )\ho o krig htar r ow L^ \ infty (\Ome ga ) $, we are able to show t hat $V (\Ome ga)$ is an al gebra with r e spect to poi n t- w ise multiplica tion. More prec i sely, we obta in: \[c or:Algebr a \ ] Let $d =2, 3$. Th ent h er e is some $C= C ( \Ome ga )>0$ su chthat $$ \la bel {eq :Pr od Estim} \|( u_ 1\ cd ot u_ 2, \n a bla_h (u _1 \cd ot u_ 2)\|_ { H^{1,0 }(\Om ega) }\ le q C\ |(u_1,\ na b l a_hu_ 1) \|_{ H^{ 1, 0}(\O mega ) }\| (u_2, \ nabla_h u _2) \ |_{H ^{ 1, 0}(\Ome ga)}$$ for al l$u_1,u_2\i nV(\ Omega) $ uniforml y in $0<h\leq 1$. Moreo v er, if$F\ in C^ {2

12; H^1(\Omega'))\hookrightarrow_BUC([-\tfrac12, \tfrac12];H^{\frac12}(\Omega'))$$_and $H^{\frac12}(\Omega')\hookrightarrow L^4(\Omega')$. Finally,_the last_embedding_follows from_$$\begin{aligned} _ _\lefteqn{L^2(-\tfrac12,\tfrac12; H^{1+k}((-L,L)^{d-1}))\cap _H^1(-\tfrac12,\tfrac12;H^1((-L,L)^{d-1}))}\\ && \hookrightarrow BUC([-\tfrac12,\tfrac12]; _H^{1+\frac{k}2}((-L,L)^{d-1}))\hookrightarrow C^0({\overline}{\Omega}) _\end{aligned}$$_where $k=d-2$ because of (\[eq:BUCEmbedding\]) and Sobolev embeddings. The spaces $H^{1,0}(\Omega)$ and $V(\Omega)$ are the_fundamental_spaces, which_will_be_used to solve the evolution_equation. We note that $$f\in_V(\Omega)\quad \Leftrightarrow_\quad f, \nabla f \in H^{1,0}(\Omega).$$ Most of_the_time we will_estimate $f\in V(\Omega)$ by the $h$-dependent norm $$\|f\|_{V_h}:= _\|(f,\nabla_h f)\|_{H^{1,0}(\Omega)}.$$ Because of the embedding $V(\Omega)\hookrightarrow_L^\infty(\Omega)$, we are_able_to_show that $V(\Omega)$ is_an algebra with respect to point-wise_multiplication. More precisely, we obtain: \[cor:Algebra\] Let_$d=2,3$. Then there is some $C=C(\Omega)>0$ such_that $$\label{eq:ProdEstim} _ \|(u_1\cdot u_2, \nabla_h_(u_1\cdot u_2)\|_{H^{1,0}(\Omega)}\leq_C\|(u_1, \nabla_h u_1)\|_{H^{1,0}(\Omega)}\|(u_2, \nabla_h u_2)\|_{H^{1,0}(\Omega)}$$_for all $u_1,u_2\in_V(\Omega)$ uniformly_in $0<h\leq 1$._Moreover, if $F\in C^{2

m_1\;\forall m_2\; \forall b\; (\; \hspace{-5mm} & ( & \hspace{-5mm} instance(b,Brood) \wedge member(m_1,b) \wedge member(m_2,b) ) \\ & \;\rightarrow \; & \label{axiom:BroodSibling} sibling(m_1,m_2) \;)\end{aligned}$$ According to the type information in axioms (\[axiom:domainSibling1\]-\[axiom:domainSibling2\]), both arguments of are restricted to be instance of. Consequently, by translation (see [@ALR12] for more details), axiom (\[axiom:BroodSibling\]) gives raise to the following rule-axiom in v2.4: $$\label{formula:oneSortedBroodSibling} \begin{array}{rll} \forall m_1 \; \forall m_2 \; \forall b \;( \hspace{-7pt} & ( \hspace{-10pt} & instance( m_1,Organism) \; \wedge \\ & & instance( m_2,Organism) \; \wedge \\ & & instance( b,Brood) \; \wedge \\ & & member( m_1, b) \; \wedge \\ & & member( m_2, b) \; ) \; \rightarrow\\ & \multicolumn{2}{l}{sibling( m_1, m_2) \; )} \end{array}$$ Since its inner subformula is logically equivalent to the following disjunction (in negation normal form): $$\label{subformula:oneSortedBroodSiblingDisjunction} \begin{array}{rl} ( \;\hspace{-10pt} & \neg instance( m_1,Organism) \; \vee\\ & \neg instance( m_2,Organism) \; \vee \\ & \neg instance( b,Brood) \; \vee\\ & \neg member( m_1, b) \; \vee\\ & \neg member( m_2, b) \; ) \end{array} \begin{array}{c} ~ \\ ~ \\

m_1\;\forall m_2\; \forall b\; (\; \hspace{-5 mm } & (& \hspace{-5 mm } instance(b, Brood) \wedge member(m_1,b) \wedge member(m_2,b) ) \\ & \;\rightarrow \; & \label{axiom: BroodSibling } sibling(m_1,m_2) \;)\end{aligned}$$ According to the type information in axioms (\[axiom: domainSibling1\]-\[axiom: domainSibling2\ ]), both argument of are restrict to be instance of. Consequently, by translation (see [ @ALR12 ] for more detail), axiom (\[axiom: BroodSibling\ ]) gives raise to the following principle - maxim in v2.4: $ $ \label{formula: oneSortedBroodSibling } \begin{array}{rll } \forall m_1 \; \forall m_2 \; \forall b \; (\hspace{-7pt } & (\hspace{-10pt } & instance (m_1,Organism) \; \wedge \\ & & instance (m_2,Organism) \; \wedge \\ & & instance (b, Brood) \; \wedge \\ & & member (m_1, b) \; \wedge \\ & & member (m_2, b) \ ;) \; \rightarrow\\ & \multicolumn{2}{l}{sibling (m_1, m_2) \ ;) } \end{array}$$ Since its inner subformula is logically equivalent to the following disjunction (in negation normal shape ): $ $ \label{subformula: oneSortedBroodSiblingDisjunction } \begin{array}{rl } (\;\hspace{-10pt } & \neg instance (m_1,Organism) \; \vee\\ & \neg instance (m_2,Organism) \; \vee \\ & \neg example (b, Brood) \; \vee\\ & \neg member (m_1, b) \; \vee\\ & \neg member (m_2, b) \ ;) \end{array } \begin{array}{c } ~ \\ ~ \\

m_1\;\flrall m_2\; \forall b\; (\; \hspace{-5om} & ( & \hspace{-5mm} nbstancx(b,Brood) \wedge mdmber(m_1,b) \wedge member(m_2,b) ) \\ & \;\righterroq \; & \lqbel{axiom:BroodSibling} skbling(m_1,m_2) \;)\ene{alijned}$$ According to the tyiz infkvmatimi in axioms (\[axipm:domainSitling1\]-\[axiom:domahnRiyling2\]), both arguments of are restrictqd to br lnstance of. Cogseqlegtly, by translation (see [@ALR12] for more dstails), exiom (\[axiom:BroocSibling\]) gives raise to thf foplowing rule-axiom ln v2.4: $$\label{firmujq:oneSortedBruodSibling} \begin{array}{rlm} \forall m_1 \; \forall m_2 \; \forall c \;( \hs'ace{-7pt} & ( \hwpafg{-10pt} & instanre( m_1,Ordanism) \; \wedgc \\ & & itstance( m_2,Organism) \; \wecge \\ & & instance( b,Brood) \; \wedje \\ & & member( m_1, b) \; \weqge \\ & & mekbzr( m_2, b) \; ) \; \rightarrow\\ & \nultiwolukn{2}{l}{skvlivg( j_1, k_2) \; )} \end{argay}$$ Since its inner subfirmula is logically ezlovalent to tge foljoring disjunction (in negation normal fork): $$\lzbel{subformula:oneSortedVroodSiblingDisjunctiln} \begin{arway}{rl} ( \;\hspace{-10pt} & \neg instance( m_1,Organism) \; \vee\\ & \neg instenze( n_2,Ovgankwm) \; \vee \\ & \neg instance( b,Brood) \; \vee\\ & \neg member( i_1, b) \; \nee\\ & \neg member( m_2, b) \; ) \end{arrau} \bfgog{array}{c} ~ \\ ~ \\

m_1\;\forall m_2\; \forall b\; (\; \hspace{-5mm} & \hspace{-5mm} \wedge member(m_1,b) member(m_2,b) ) \\ sibling(m_1,m_2) According to the information in axioms both arguments of are restricted to instance of. Consequently, by translation (see [@ALR12] for more details), axiom (\[axiom:BroodSibling\]) gives to the following rule-axiom in v2.4: $$\label{formula:oneSortedBroodSibling} \begin{array}{rll} \forall m_1 \; \forall m_2 \forall \;( & \hspace{-10pt} & instance( m_1,Organism) \; \wedge \\ & & instance( m_2,Organism) \; \wedge \\ & & b,Brood) \; \wedge \\ & & member( m_1, \; \wedge \\ & member( m_2, b) \; ) \rightarrow\\ \multicolumn{2}{l}{sibling( m_1, \; \end{array}$$ its inner subformula logically equivalent to the following disjunction (in negation normal form): $$\label{subformula:oneSortedBroodSiblingDisjunction} \begin{array}{rl} ( \;\hspace{-10pt} & \neg instance( \; \vee\\ instance( m_2,Organism) \vee & instance( b,Brood) \; \neg member( m_1, b) \; \vee\\ m_2, b) \; ) \end{array} \begin{array}{c} ~ \\ \\

m_1\;\forall m_2\; \forall b\; (\; \hspace{-5mm} & ( & \hsPace{-5mm} instAnce(b,broOd) \wEdGe meMber(M_1,b) \wedge member(m_2,B) ) \\ & \;\RighTarrow \; & \label{axiom:BroodSIblinG} sIBlinG(M_1,m_2) \;)\End{alIgned}$$ AcCOrDINg tO tHe TypE iNFoRmatiOn iN axioms (\[Axiom:domaiNSiBlIng1\]-\[axiom:domaINSIbling2\]), both ArgUments of are rEstRicted To Be iNStancE of. conseQuentlY, By tranSlation (seE [@Alr12] for moRE detailS), AXiOm (\[axIom:BroodSibling\]) giVEs RAise to the folloWing ruLe-AXiOM In v2.4: $$\LabEl{formula:oNesorteDbroodSiBLiNG} \BEgiN{Array}{rll} \foralL m_1 \; \forall m_2 \; \foRAll B \;( \hspacE{-7pT} & ( \hsPAce{-10pt} & iNstanCe( M_1,orgAnism) \; \wedge \\ & & iNstaNce( m_2,OrganIsm) \; \wedGE \\ & & instanCE( b,Brood) \; \Wedge \\ & & mEmbEr( m_1, B) \; \wedGE \\ & & mEmBer( M_2, b) \; ) \; \RIghTArRow\\ & \MUltIcolumn{2}{l}{SiBlIng( m_1, m_2) \; )} \End{aRRAY}$$ sincE itS innEr subFormula is logiCalLy eqUIvaLent tO the fOlloWiNg disJunctiOn (in nEgAtion normal form): $$\LabeL{subformuLa:oNesorTeDBrooDsiblinGDiSjuNction} \bEgin{arrAY}{rl} ( \;\HsPACE{-10pT} & \neg instance( m_1,OrganIsM) \; \VEe\\ & \Neg instaNce( m_2,OrGAnIsM) \; \Vee \\ & \neg inStAncE( b,BrOOD) \; \vee\\ & \nEg meMBeR( m_1, b) \; \vee\\ & \neG membeR( M_2, b) \; ) \EnD{array} \bEgIn{arraY}{c} ~ \\ ~ \\

m_1\;\forall m_2\; \foral l b\; (\; \hs pac e{- 5m m} & ( & \hspace{-5mm} inst ance(b,Brood) \wedge membe r( m _1,b ) \ wedge member ( m_ 2 , b))\\ &\; \ ri ghtar row \; & \ label{axio m:B ro odSibling} s i bl ing(m_1,m_ 2) \;)\e nd{ aligne d} $$A ccord ing to t he typ e infor mation in a x ioms ( \ [axiom: d o ma inSi bling1\]-\[axiom: d om a inSibling2\]), bothar g um e n tsofare restri ct ed to be inst a nc e o f.C onsequently,by translat i on(see [ @A LR1 2 ] formorede t ail s), axiom ( \[ax iom:Brood Siblin g \]) giv e s raise to th e f oll owin g r ul e-a xi o m i n v 2.4 : $$ \label{f or mu la:on eSor t e d B rood Sib ling } \be gin{array}{rl l}\for a ll m_1\; \f oral l m_2\; \fo rall b \;( \hspace{-7 pt}& ( \h spa ce {-1 0p t} &instan ce( m_ 1,Organ ism) \; \we dg e \ \ & & instance( m_ 2, O r ga nism) \; \wedg e \ \& & inst an ce( b,B r o od) \ ; \w e dg e \\ && memb e r( m _1, b)\; \wedg e\\ && mem b er(m_2, b ) \; )\; \r i ghtarrow\\ &\ multicolumn{2 } {l } { si b ling ( m _1, m_2) \; )}\ end{ arra y }$ $ S i nce i ts in ne r s u bformula is logical ly equiv alent to the follo wing disju n c t ion (innega t io n normal form): $$\l abel{subfo r mula:one Sorte dBroodSi blingDisj u n ction} \ beg in{ arr ay} { r l} ( \;\hspace{ - 1 0pt} & \neg in stance( m_ 1,O rga nis m) \; \vee\ \ & \ne gin st an ce( m_2, O rganism) \ ; \ ve e \ \ &\ neg in stanc e( b ,B ro o d)\; \vee \ \& \ne gme mber ( m _1 , b)\; \ v ee\ \ & \n eg member ( m _ 2, b )\; ) \end {array} \begi n{ array}{c}~\\~ \\

m_1\;\forall_m_2\; \forall_b\; (\; \hspace{-5mm} &_( &_\hspace{-5mm}_instance(b,Brood) \wedge_ member(m_1,b)_\wedge member(m_2,b) ) \\ & \;\rightarrow_\; & \label{axiom:BroodSibling} sibling(m_1,m_2)_ _ \;)\end{aligned}$$_According_to the type information in axioms (\[axiom:domainSibling1\]-\[axiom:domainSibling2\]), both arguments of are restricted to be_instance_of. Consequently,_by_translation_(see [@ALR12] for more details),_axiom (\[axiom:BroodSibling\]) gives raise to_the following_rule-axiom in v2.4: $$\label{formula:oneSortedBroodSibling} \begin{array}{rll} \forall m_1 \; \forall__m_2 \; \forall_ b \;( \hspace{-7pt} & (_\hspace{-10pt} & instance( m_1,Organism) \;_\wedge \\ &_&__instance( m_2,Organism) \; \wedge_\\ & & instance( b,Brood) \;_\wedge \\ & & member( m_1,_b) \; \wedge \\ & & member(_m_2, b) \; ) \;_\rightarrow\\ & \multicolumn{2}{l}{sibling( m_1, m_2)_\; )} \end{array}$$_Since its inner subformula is_logically equivalent to_the following_disjunction (in negation_normal form): $$\label{subformula:oneSortedBroodSiblingDisjunction} \begin{array}{rl} ( \;\hspace{-10pt} & _\neg instance( m_1,Organism)_\; \vee\\ & \neg instance( m_2,Organism)_\;_\vee \\ &_\neg_instance(_b,Brood) \;_\vee\\ & \neg_member(_m_1, b)_\;_\vee\\ & \neg member( m_2, b)_\;_) \end{array} \begin{array}{c} ~ \\ ~ \\

aligned} \label{FModelUniformeSimu} {\mathscr{F}}= \left\{f_{\theta}, \, \theta \in [0.01, 10]\right\} \quad \text{where} \quad f_{\theta} = \theta^{-1} {\mathbbm{1}}_{[0, \theta]}.\end{aligned}$$ It is worthwhile to notice that $h^2(s,{\mathscr{F}}) = \mathcal{O} (n^{-1})$, which means that $s$ is close to ${\mathscr{F}}$ when $n$ is large, and that our estimator still satisfies ${\mathbb{E}}[h^2 (s,f_{\hat{\theta}})] = \mathcal{O} (n^{-1})$. Contrary to our estimator, the outliers make the m.l.e unstable as shown in the array below. $n = 10$ $n = 25$ $n = 50$ $n = 75$ $n = 100$ ------------------------------------------------ ---------- ---------- ---------- ---------- ----------- -- $\widehat{R}_N (\hat{\theta}) $ 0.20 0.06 0.03 0.02 0.015 $\widehat{R}_N (\tilde{\theta}_{\text{mle}}) $ 0.57 0.56 0.56 0.56 0.57 We now propose a second example based on the mixture of two uniform laws. We use the same statistical model ${\mathscr{F}}$ but we modify the distribution of the observations. We take $p \in (0,1)$ and define the true underlying density by $$s_{p} (x) = (1-p) f_1(x) + p f_2(x) \quad \text{for all $x \in {\mathbb{R}}$.}$$ Set $p_0 = 1-1/\sqrt{2}$. One can check that $$\begin{aligned} H^2(s_p,{\mathscr{F}}) &=& \begin{cases} H^2(s_p,f_1) & \text{if $p \leq p_0$} \\ H^2(s_p,f_2) & \text{if $p > p_0$,} \end{cases} \\ &=& \

aligned } \label{FModelUniformeSimu } { \mathscr{F}}= \left\{f_{\theta }, \, \theta \in [ 0.01, 10]\right\ } \quad \text{where } \quad f_{\theta } = \theta^{-1 } { \mathbbm{1}}_{[0, \theta]}.\end{aligned}$$ It is worthwhile to notice that $ h^2(s,{\mathscr{F } }) = \mathcal{O } (n^{-1})$, which means that $ s$ is close to $ { \mathscr{F}}$ when $ n$ is big, and that our calculator still satisfies $ { \mathbb{E}}[h^2 (s, f_{\hat{\theta } }) ] = \mathcal{O } (n^{-1})$. adverse to our estimator, the outlier make the m.l.e unstable as picture in the array below. $ n = 10 $ $ nitrogen = 25 $ $ n = 50 $ $ n = 75 $ $ normality = 100 $ ------------------------------------------------ ---------- ---------- ---------- ---------- ----------- -- $ \widehat{R}_N (\hat{\theta }) $ 0.20 0.06 0.03 0.02 0.015 $ \widehat{R}_N (\tilde{\theta}_{\text{mle } }) $ 0.57 0.56 0.56 0.56 0.57 We now propose a second exemplar free-base on the mixture of two uniform laws. We use the same statistical exemplar $ { \mathscr{F}}$ but we modify the distribution of the observations. We take $ phosphorus \in (0,1)$ and define the true implicit in density by $ $ s_{p } (x) = (1 - phosphorus) f_1(x) + p f_2(x) \quad \text{for all $ x \in { \mathbb{R}}$.}$$ fix $ p_0 = 1 - 1/\sqrt{2}$. One can check that $ $ \begin{aligned } H^2(s_p,{\mathscr{F } }) & = & \begin{cases } H^2(s_p, f_1) & \text{if $ p \leq p_0 $ } \\ H^2(s_p, f_2) & \text{if $ p > p_0 $, } \end{cases } \\ & = & \

alihned} \label{FModelUniformeRimu} {\mathscr{F}}= \leyr\{f_{\thete}, \, \thetz \in [0.01, 10]\rieht\} \quad \text{where} \quad f_{\theva} = \rheta^{-1} {\mathbbm{1}}_{[0, \theta]}.\end{aligved}$$ It is worthwhule uo notice that $h^2(s,{\mathscr{F}}) = \mathdwl{O} (u^{-1})$, xhich means thaj $s$ is close to ${\mathscr{F}}$ wvev $u$ is large, and that our estimator styll satosvies ${\mathbb{E}}[h^2 (f,f_{\hau{\thqta}})] = \mathcal{O} (n^{-1})$. Contrary to our estimafor, the outliers makr the m.l.e unstable as showj in the array below. $h = 10$ $n = 25$ $n = 50$ $n = 75$ $n = 100$ ------------------------------------------------ ---------- ---------- ---------- ---------- ----------- -- $\wudejdt{R}_N (\hat{\theva}) $ 0.20 0.06 0.03 0.02 0.015 $\wieehat{R}_N (\tilde{\theta}_{\texv{mle}}) $ 0.57 0.56 0.56 0.56 0.57 We noq propove a secubd dxajpke based on the mixtude of two ubiform laws. We use uhe wame statistidal moqej ${\mathscr{F}}$ but we modify the distributimn kf the observations. We rake $p \in (0,1)$ and define the true underlying density by $$s_{p} (x) = (1-p) f_1(x) + p f_2(x) \quad \texd{for elu $x \ik {\magybh{R}}$.}$$ Set $p_0 = 1-1/\sqrt{2}$. One can check that $$\begin{aligneq} G^2(s_k,{\manhscr{F}}) &=& \begin{cafes} H^2(s_p,f_1) & \tect{lf $k \leq p_0$} \\ H^2(s_p,f_2) & \text{iy $p > p_0$,} \end{cases} \\ &=& \

aligned} \label{FModelUniformeSimu} {\mathscr{F}}= \left\{f_{\theta}, \, \theta \in \quad \quad f_{\theta} \theta^{-1} {\mathbbm{1}}_{[0, \theta]}.\end{aligned}$$ that = \mathcal{O} (n^{-1})$, means that $s$ close to ${\mathscr{F}}$ when $n$ is and that our estimator still satisfies ${\mathbb{E}}[h^2 (s,f_{\hat{\theta}})] = \mathcal{O} (n^{-1})$. Contrary to estimator, the outliers make the m.l.e unstable as shown in the array below. = $n 25$ = 50$ $n = 75$ $n = 100$ ------------------------------------------------ ---------- ---------- ---------- ---------- ----------- -- $\widehat{R}_N $ 0.20 0.06 0.03 0.02 0.015 $\widehat{R}_N (\tilde{\theta}_{\text{mle}}) 0.57 0.56 0.56 0.56 We now propose a second based the mixture two laws. use the same model ${\mathscr{F}}$ but we modify the distribution of the observations. We take $p \in (0,1)$ and define true underlying $$s_{p} (x) (1-p) + f_2(x) \quad \text{for \in {\mathbb{R}}$.}$$ Set $p_0 = 1-1/\sqrt{2}$. that $$\begin{aligned} H^2(s_p,{\mathscr{F}}) &=& \begin{cases} H^2(s_p,f_1) & \text{if \leq p_0$} H^2(s_p,f_2) & \text{if $p > p_0$,} \\ &=& \

aligned} \label{FModelUniformESimu} {\mathsCr{F}}= \leFt\{f_{\TheTa}, \, \ThetA \in [0.01, 10]\rIght\} \quad \text{whERe} \quAd f_{\theta} = \theta^{-1} {\mathbbm{1}}_{[0, \thEta]}.\enD{aLIgneD}$$ it Is worThwhile TO nOTIce ThAt $H^2(s,{\mAtHScR{F}}) = \matHcaL{O} (n^{-1})$, whicH means that $S$ is ClOse to ${\mathscr{f}}$ WhEn $n$ is large, And That our estimAtoR still SaTisFIes ${\maThbB{E}}[h^2 (s,f_{\Hat{\theTA}})] = \mathcAl{O} (n^{-1})$. ContrArY To our eSTimator, THE oUtliErs make the m.l.e unstABlE As shown in the arRay belOw. $N = 10$ $N = 25$ $n = 50$ $N = 75$ $N = 100$ ------------------------------------------------ ---------- ---------- ---------- ---------- ----------- -- $\WidEhaT{R}_N (\hat{\thetA}) $ 0.20 0.06 0.03 0.02 0.015 $\wIdehaT{r}_N (\tilde{\THeTA}_{\TExt{MLe}}) $ 0.57 0.56 0.56 0.56 0.57 We now proposE a second exaMPle Based oN tHe mIXture oF two uNiFOrm Laws. We use thE samE statistiCal modEL ${\mathscR{f}}$ but we mOdify tHe dIstRibuTIoN oF thE oBSerVAtIonS. we tAke $p \in (0,1)$ anD dEfIne thE truE UNDErlyIng DensIty by $$S_{p} (x) = (1-p) f_1(x) + p f_2(x) \quad \TexT{for ALl $x \In {\matHbb{R}}$.}$$ SEt $p_0 = 1-1/\sQrT{2}$. One cAn checK that $$\BeGin{aligned} H^2(s_p,{\maThscR{F}}) &=& \begin{caSes} h^2(s_P,f_1) & \tExT{if $p \lEQ p_0$} \\ H^2(s_p,f_2) & \TexT{if $P > p_0$,} \end{caSes} \\ &=& \

aligned} \label{FModelUni formeSimu} {\ma ths cr{ F} }= \ left \{f_{\theta},\ , \t heta \in [0.01, 10]\ri ght\} \ q uad\ te xt{wh ere} \q u ad f _{\ th et a}=\ th eta^{ -1} {\mat hbbm{1}}_{ [0, \ theta]}.\end { al igned}$$ I t i s worthwhile to notic etha t $h^2 (s, {\mat hscr{F } }) = \ mathcal{O }( n^{-1} ) $, whic h me ansthat $s$ is close to ${\mathscr{F}} $ when $ n $i s la rge , and that o ur es t imators ti l l sat i sfies ${\math bb{E}}[h^2( s,f _{\hat {\ the t a}})]= \ma th c al{ O} (n^{-1}) $. C ontrary t o oure stimato r , the o utlier s m ake the m. l. e u ns t abl e a s s h own in thear ra y bel ow. $ n =10 $ $ n = 25 $ $ n= 50$ $n = 75 $ $n = 100$ --- -- ----- - ------ --- --- ------- ------- - --- -- - - - -- ---------- ------ -- - - - -------- - ---- - -- -- - ------- -- ---- $\ w id ehat{R}_ N (\ha t {\ th eta}) $ 0.2 0 0. 06 0.03 0.02 0.0 1 5 $\wi d eh a t {R } _N ( \ti lde{\theta} _{\t e xt{m le}} ) $ 0.57 0 .5 6 0.56 0 .5 6 0.5 7 We no w proposea s econd ex ampl e b a sed on the mix tureof two uni f orm laws . Weuse thesame stat i s tical mo del ${ \ma ths c r {F }}$ but we mo d i fy t he distri but ion ofthe ob ser vat io ns. We ta ke $p \i n(0 ,1 )$ an d def i ne the t ru e u nd erl yingd ensity by $ $s_{ p} ( x ) = (1-p) f_ 1 ( x) + p f _2(x ) \ qu ad \t ext{ f orall $x\in {\mat hbb { R}}$ .} $$ Set $p _0 = 1-1/\sqr t{ 2}$. One c an ch eck th a t $$\begi n{aligned} H^2(s_p,{\ma t hscr{F} })&=& \b egin{case s}H^2(s_ p,f _ 1) & \ text{i f $p\l eqp _ 0$} \ \ H ^2( s_ p,f_2) & \ t e xt{ if $p > p_0 $,} \ end{cases} \\ & = & \

aligned} \label{FModelUniformeSimu} {\mathscr{F}}=_\left\{f_{\theta}, \,_\theta \in [0.01, 10]\right\}_\quad \text{where}_\quad_f_{\theta} =_\theta^{-1}_ {\mathbbm{1}}_{[0, \theta]}.\end{aligned}$$_It is worthwhile_to notice that $h^2(s,{\mathscr{F}})_= \mathcal{O} (n^{-1})$,_which_means that $s$ is close to ${\mathscr{F}}$ when $n$ is large, and that our_estimator_still satisfies_${\mathbb{E}}[h^2_(s,f_{\hat{\theta}})]_= \mathcal{O} (n^{-1})$. Contrary to_our estimator, the outliers make_the m.l.e_unstable as shown in the array below. __ _ _ _ ___ _ _ _ _ $n = 10$_ $n = 25$_ _$n = 50$ _$n = 75$_ _$n = 100$_ ------------------------------------------------ ---------- ----------_---------- ---------- -----------_-- __ _$\widehat{R}_N_(\hat{\theta})_$ _ __ __ 0.20 __ 0.06 _ _0.03_ _ 0.02 _ 0.015 _ _ _ $\widehat{R}_N (\tilde{\theta}_{\text{mle}}) $ 0.57 _ 0.56 _ _ 0.56__ __ 0.56_ 0.57_ _ We now propose a_second example based on_the_mixture of two uniform laws. We_use the same statistical model ${\mathscr{F}}$_but we modify the distribution_of_the_observations. We take $p \in_(0,1)$ and define the true underlying_density by $$s_{p}_(x) = (1-p) f_1(x) + p_f_2(x)_\quad \text{for all $x \in {\mathbb{R}}$.}$$_Set_$p_0 = 1-1/\sqrt{2}$. One can check_that_$$\begin{aligned} H^2(s_p,{\mathscr{F}})_&=& \begin{cases} H^2(s_p,f_1) &_\text{if $p \leq p_0$} \\ H^2(s_p,f_2)_& \text{if $p > p_0$,} \end{cases} \\ _ &=& _ \

Gamma(M,B), L^2(M,B \otimes S), D_B)$. The conditions on $\gamma_B$ to be a grading operator for this spectral triple are easily checked. In the next section we show that the triple ${(\Gamma(M,B), L^2(M,B\otimes S),D_B,J,\gamma_B)}$ gives a non-trivial Yang–Mills theory over the manifold $M$. The Serre–Swan Theorem \[thm:serreswan2\] plays an essential role in the proof. First, we explore the form of this spectral triple in the context of Kasparov’s KK-theory. Relation with the unbounded Kasparov internal product {#sect:KK} ----------------------------------------------------- \[ssct:kkproduct\] In this section we establish that the spectral triple of Theorem \[thm:spectraltriple1\] is an unbounded Kasparov product of two unbounded KK-cycles [@KasparovKK; @BaajJulg]. Let us briefly recall some elementary notions from (unbounded) KK-theory. Denote by ${\mathcal}{B}(E)$ the bounded endomorphisms of a right Hilbert $B$-module $E$ and by ${\mathcal}{K}(E)$ the compact endomorphisms. Let $A$ and $B$ be $\mathbb{Z}_2$-graded $C^*$-algebras. A Kasparov $A$-$B$-module consists of a triple $(E, \phi, F)$ where $E$ is a countably generated $\mathbb{Z}_2$-graded Hilbert-$B$-module, $\phi$ is a graded $*$-homomorphism $A \rightarrow {\mathcal}{B}(E)$ and $F$ is a bounded operator of degree $1$, such that $[F, \phi(a)]$, $(F^2 -1)\phi(a)$, and $(F - F^*)\phi(a)$ are in ${\mathcal}{K}(E)$. There are the natural notions of unitary and homotopy equivalence and under the direct sum the set of equivalence classes of Kasparov $A-B$-modules forms an abelian group which is denoted by $KK(A,B)$ [@Kasparov]. One of the key properties of $KK$-theory is the existence of the internal Kasparov product. \[dfn:boundedkasparovproduct\] Let $E_1$

Gamma(M, B), L^2(M, B \otimes S), D_B)$. The conditions on $ \gamma_B$ to be a grading operator for this spectral trio are well check. In the next incision we testify that the triple $ { (\Gamma(M, B), L^2(M, B\otimes S),D_B, J,\gamma_B)}$ gives a non - superficial Yang – mill theory over the manifold $ M$. The Serre – Swan Theorem \[thm: serreswan2\ ] act an essential role in the validation. First, we explore the form of this spectral trio in the context of Kasparov ’s KK - theory. Relation with the unbounded Kasparov inner product { # sect: KK } ----------------------------------------------------- \[ssct: kkproduct\ ] In this part we establish that the spectral triple of Theorem \[thm: spectraltriple1\ ] is an boundless Kasparov product of two unbounded KK - cycle [ @KasparovKK; @BaajJulg ]. Let us concisely recall some elementary notion from (unbounded) KK - theory. Denote by $ { \mathcal}{B}(E)$ the restrict endomorphisms of a right Hilbert $ B$-module $ E$ and by $ { \mathcal}{K}(E)$ the compact endomorphisms. Let $ A$ and $ B$ be $ \mathbb{Z}_2$-graded $ C^*$-algebras. A Kasparov $ A$ -$B$-module consists of a triple $ (E, \phi, F)$ where $ E$ is a countably generated $ \mathbb{Z}_2$-graded Hilbert-$B$-module, $ \phi$ is a graded $ * $ -homomorphism $ A \rightarrow { \mathcal}{B}(E)$ and $ F$ is a bounded operator of degree $ 1 $, such that $ [ F, \phi(a)]$, $ (F^2 -1)\phi(a)$, and $ (F - F^*)\phi(a)$ are in $ { \mathcal}{K}(E)$. There are the natural notions of unitary and homotopy equivalence and under the direct kernel the set of equivalence class of Kasparov $ deoxyadenosine monophosphate - B$-modules forms an abelian group which is denoted by $ KK(A, B)$ [ @Kasparov ]. One of the key property of $ KK$-theory is the being of the internal Kasparov product. \[dfn: boundedkasparovproduct\ ] lease $ E_1 $

Gamla(M,B), L^2(M,B \otimes S), D_B)$. The conditions on $\yqmma_B$ vo be a grading operator for this spectral vripoe art easily checked. In ghe next dection qe siow that the tri'me ${(\Gamma(M,B), L^2(M,G\ltimzs S),D_B,J,\gamma_B)}$ giyes a non-trhvial Yang–Millv ghzory over the manifold $M$. The Serre–Swwn Theotel \[thm:serreswan2\] plajs an sssential role in the proof. First, se explmre the form pf this spectral triple in the context of Kasparlv’s KK-theori. Delwrion with thd unbounded Kasparov ihternal product {#sect:KK} ----------------------------------------------------- \[ssct:kkproauct\] Nn this secjnin ag establish vhat tre spectral bgiple ox Theorrm \[thm:spectralbriplx1\] is an unbounded Kasparot product of two unbjunded KK-wyeles [@KasparovKK; @BaajJylt]. Let us triewoy feczlk aome epemxntary notikns from (unvounded) KK-theory. Demoev by ${\mathcal}{B}(S)$ the fotnded endomorphisms of a right Hilbert $T$-mosule $E$ and by ${\mathcal}{K}(E)$ the compact endomorpjisms. Let $W$ and $B$ be $\mathbb{Z}_2$-graded $C^*$-algebras. A Kasparov $A$-$B$-mmdule zonwifgw lf a triple $(E, \phi, F)$ where $E$ is a countably gegsrstvd $\mathbb{Z}_2$-graded Milbert-$B$-module, $\phi$ id s graded $*$-homomurphism $A \rightarrow {\mathcap}{B}(E)$ and $F$ is a boundeq oprrator of degree $1$, such that $[F, \phi(a)]$, $(F^2 -1)\php(a)$, abd $(F - F^*)\phi(a)$ are in ${\mathcal}{K}(E)$. Tkere ate the natural notions of unicary ahd homotopy equivalehze and under the digect sum the set of equivalencq classes of Lasparox $A-B$-kodulef forms an abelldn group which is fenotgd by $NK(A,B)$ [@Kaspagov]. One of the key properties oh $KK$-theory is tve vxistence of tme internal Kas[arov product. \[dyn:boundebkaspafovproduct\] Let $E_1$

Gamma(M,B), L^2(M,B \otimes S), D_B)$. The conditions to a grading for this spectral the section we show the triple ${(\Gamma(M,B), S),D_B,J,\gamma_B)}$ gives a non-trivial Yang–Mills theory the manifold $M$. The Serre–Swan Theorem \[thm:serreswan2\] plays an essential role in the First, we explore the form of this spectral triple in the context of KK-theory. with unbounded internal product {#sect:KK} ----------------------------------------------------- \[ssct:kkproduct\] In this section we establish that the spectral triple of Theorem is an unbounded Kasparov product of two unbounded [@KasparovKK; @BaajJulg]. Let us recall some elementary notions from KK-theory. by ${\mathcal}{B}(E)$ bounded of right Hilbert $B$-module and by ${\mathcal}{K}(E)$ the compact endomorphisms. Let $A$ and $B$ be $\mathbb{Z}_2$-graded $C^*$-algebras. A Kasparov $A$-$B$-module consists a triple F)$ where is countably $\mathbb{Z}_2$-graded Hilbert-$B$-module, $\phi$ graded $*$-homomorphism $A \rightarrow {\mathcal}{B}(E)$ and bounded operator of degree $1$, such that $[F, $(F^2 -1)\phi(a)$, $(F - F^*)\phi(a)$ are in ${\mathcal}{K}(E)$. are the natural notions of unitary and homotopy and under the direct sum the set of equivalence classes of Kasparov $A-B$-modules forms an which is denoted by [@Kasparov]. One of key of is existence of internal Kasparov product. \[dfn:boundedkasparovproduct\] Let $E_1$

Gamma(M,B), L^2(M,B \otimes S), D_B)$. The conDitions on $\gAmma_B$ To bE a gRaDing OperAtor for this speCTral Triple are easily checked. in the NeXT secTIoN we shOw that tHE tRIPle ${(\gaMmA(M,B), l^2(M,b\OtImes S),d_B,J,\Gamma_B)}$ gIves a non-trIviAl yang–Mills theORy Over the manIfoLd $M$. The Serre–SWan theoreM \[tHm:sERreswAn2\] pLays aN essenTIal rolE in the proOf. fIrst, we EXplore tHE FoRm of This spectral triplE In THe context of KasParov’s kK-THeORY. ReLatIon with the UnBoundED KasparOV iNTERnaL Product {#sect:KK} ----------------------------------------------------- \[Ssct:kkproduCT\] In This seCtIon WE estabLish tHaT The Spectral triPle oF Theorem \[tHm:specTRaltripLE1\] is an unBoundeD KaSpaRov pROdUcT of TwO UnbOUnDed kk-cyCles [@KaspArOvkK; @BaaJJulG]. lET Us brIefLy reCall sOme elementary NotIons FRom (UnbouNded) Kk-theOrY. DenoTe by ${\maThcal}{b}(E)$ The bounded endomOrphIsms of a riGht hiLbeRt $b$-moduLE $E$ and bY ${\maThcAl}{K}(E)$ the Compact ENdoMoRPHIsMs. Let $A$ and $B$ be $\mathbb{z}_2$-gRADeD $C^*$-algebrAs. A KasPArOv $a$-$b$-module cOnSisTs of A TRiple $(e, \phi, f)$ WhEre $E$ is a cOuntabLY gEnErated $\mAtHbb{Z}_2$-grAdEd HIlbErt-$B$-mODule, $\Phi$ is a Graded $*$-hoMomorPHism $A \rightarroW {\Mathcal}{B}(E)$ and $F$ IS a BOUnDEd opEraTor of degree $1$, Such THat $[F, \Phi(a)]$, $(f^2 -1)\PhI(a)$, aND $(F - F^*)\phI(a)$ are In ${\MAtHCal}{K}(E)$. There are the natUrAl notiOns of Unitary and homOtopy equivALENce and unDer tHE dIRect sum the set oF equiValence claSSes of KasParov $a-B$-moduleS forms an aBELian grouP whIch Is dEnoTED bY $KK(A,B)$ [@Kasparov]. oNE of tHe Key propErtIes of $KK$-TheOry Is tHe eXiStence of tHe internAl kaSpArOv pRoducT. \[Dfn:boundEdKasPaRovProduCT\] Let $E_1$

Gamma(M,B), L^2(M,B \otime s S), D_B) $. Th e c ond it ions on$\gamma_B$ tob e agrading operator for t his s pe c tral tr ipleare eas i ly c hec ke d. I nt he next se ction w e show tha t t he triple ${(\ G am ma(M,B), L ^2( M,B\otimes S ),D _B,J,\ ga mma _ B)}$giv es anon-tr i vial Y ang–Mills t h eory o v er them a ni fold $M$. The Serre–S w an Theorem \[thm: serres wa n 2\ ] pla ysan essenti al role in thep ro o f . Fi r st, we explor e the formo f t his sp ec tra l tripl e inth e co ntext of Ka spar ov’s KK-t heory. Relati o n withthe un bou nde d Ka s pa ro v i nt e rna l p rod u ct{#sect:K K} - ----- ---- - - - - ---- --- ---- ----- ------------- --- ---- - -- \[ss ct:kk prod uc t\] I n this sect io n we establishthat the spec tra ltri pl e ofT heorem \[ thm :spectr altripl e 1\] i s a nunbounded Kasparov p r o du ct of tw o unbo u nd ed KK-cycle s[@K aspa r o vKK;@Baa j Ju lg]. Let us br i ef ly recall s ome el em ent ary noti o ns f rom (u nbounded ) KK- t heory. Denoteb y ${\mathcal} { B} ( E )$ thebou nded endomo rphi s ms o f ar ig htH ilber t $B$ -m o du l e $E$ and by ${\mat hc al}{K} (E)$the compact e ndomorphis m s . Let $A $ an d $ B $ be $\mathbb{ Z}_2$ -graded $C ^ *$-algeb ras.A Kaspar ov $A$-$B $ - module c ons ist s o f a t ri ple $(E, \phi , F)$wh ere $E$ is a coun tab lygen era te d $\mathb b{Z}_2$- gr ad ed H ilb ert-$ B $-module ,$\p hi $ i s a g r aded $ *$-ho momo rp hi s m $ A \righ t ar r o w {\ ma th cal} {B} (E )$ an d $F $ is a boun ded opera tor of d eg re e $1$,such that $[F ,\phi(a)]$, $ (F^ 2 -1)\ p h i(a)$, a nd $(F - F^*)\phi(a)$ a r e in ${ \ma thcal }{K} (E)$. Th ere are t hen atural notio ns of u nit a r y and h om oto py equivalen c e an d und er the direct sum the set of eq u iva lence classes of Kas p a ro v $ A -B $ -mo du l esf o rms an abeliangroup whic hi sdenoted by $KK (A ,B)$ [@ Kasparo v]. O n e of th e key pro perties o f$KK$ - t heo ry is theexistenc e of thei ntern a lKaspa rov produ ct . \[dfn :bound e dka sparo vprodu ct \] Let $E_1 $

Gamma(M,B), L^2(M,B_\otimes S),_D_B)$. The conditions on_$\gamma_B$ to_be_a grading_operator_for this spectral_triple are easily_checked. In the next section_we show that_the_triple ${(\Gamma(M,B), L^2(M,B\otimes S),D_B,J,\gamma_B)}$ gives a non-trivial Yang–Mills theory over the manifold $M$. The_Serre–Swan_Theorem \[thm:serreswan2\]_plays_an_essential role in the proof._First, we explore the form_of this_spectral triple in the context of Kasparov’s KK-theory. Relation_with_the unbounded Kasparov_internal product {#sect:KK} ----------------------------------------------------- \[ssct:kkproduct\] In this section we establish that_the spectral triple of Theorem \[thm:spectraltriple1\]_is an unbounded_Kasparov_product_of two unbounded KK-cycles_[@KasparovKK; @BaajJulg]. Let us briefly recall_some elementary notions from (unbounded) KK-theory._Denote by ${\mathcal}{B}(E)$ the bounded endomorphisms of_a right Hilbert $B$-module $E$ and_by ${\mathcal}{K}(E)$ the compact endomorphisms. Let_$A$ and_$B$ be $\mathbb{Z}_2$-graded $C^*$-algebras. A_Kasparov $A$-$B$-module consists_of a_triple $(E, \phi,_F)$ where $E$ is a countably_generated $\mathbb{Z}_2$-graded Hilbert-$B$-module,_$\phi$ is a graded $*$-homomorphism $A_\rightarrow_{\mathcal}{B}(E)$ and $F$_is_a_bounded operator_of degree $1$,_such_that $[F,_\phi(a)]$,_$(F^2 -1)\phi(a)$, and $(F - F^*)\phi(a)$_are_in ${\mathcal}{K}(E)$. There are the natural notions of_unitary and homotopy equivalence_and_under the direct sum_the set of equivalence classes_of Kasparov $A-B$-modules forms an abelian_group which_is denoted_by $KK(A,B)$ [@Kasparov]. One of the key properties of $KK$-theory is_the existence of the internal Kasparov_product. \[dfn:boundedkasparovproduct\] Let $E_1$

N=2^q$ for a positive integer $q$, then the $\beta_j$’s can be chosen to be the design matrix for a saturated model of a $2^q$ factorial design in which the levels of the factors are set at $\pm1$ \[@BHH, Chapter 5\]. In addition, assume that $X_1,\ldots,X_m$ are random diagonal matrices of size $N$ and $Y_{jk}$, $j=1,\ldots,N; k=1,\ldots,m$ are random variables such that all the diagonal elements of the $X_j$’s and all the $Y_{jk}$’s are i.i.d. symmetric Bernoulli random variables. Then define $$\label{Uj} U_j = \pmatrix{ Y_{j1}X_1 \cr \vdots \cr Y_{jm}X_m} \beta_j.$$ One can easily show that for any $Nm\times Nm$ matrix $M$, $E (\frac{1}{N}\sum_{j=1}^N U_j'MU_j ) = \operatorname{tr}(M)$. Thus, we can use this definition of the $U_j$’s in (\[ascore\]), and the resulting estimating equations are still unbiased. This design is closely related to a class of designs introduced by @avron, who propose selecting the $U_j$’s as follows. Suppose $H$ is a Hadamard matrix, that is, an $n\times n$ orthogonal matrix with elements $\pm1$. @avron actually consider $H$ a multiple of a unitary matrix, but the special case $H$ Hadamard makes their proposal most similar to ours. Then, using simple random sampling (with replacement), they choose $N$ columns from this matrix and multiply this $n\times N$ matrix by an $n\times n$ diagonal matrix with diagonal entries made up of independent symmetric Bernoulli random variables. The columns of this resulting matrix are the $U_j$’s. We are also multiplying a subset of the columns of a Hadamard matrix by a random diagonal matrix, but we do not select the columns by simple random sampling from some arbitrary Hadamard matrix. The extra structure we impose yields beneficial results in terms of the variance of the randomized trace approximation, as the following calculations show.

N=2^q$ for a positive integer $ q$, then the $ \beta_j$ ’s can be chosen to be the design matrix for a saturated model of a $ 2^q$ factorial invention in which the degree of the factors are set at $ \pm1 $ \[@BHH, Chapter 5\ ]. In addition, simulate that $ X_1,\ldots, X_m$ are random diagonal matrices of size $ N$ and $ Y_{jk}$, $ j=1,\ldots, N; k=1,\ldots, m$ are random variable such that all the aslant elements of the $ X_j$ ’s and all the $ Y_{jk}$ ’s are i.i.d. symmetric Bernoulli random variables. Then define $ $ \label{Uj } U_j = \pmatrix { Y_{j1}X_1 \cr \vdots \cr Y_{jm}X_m } \beta_j.$$ One can easily show that for any $ Nm\times Nm$ matrix $ M$, $ east (\frac{1}{N}\sum_{j=1}^N U_j'MU_j) = \operatorname{tr}(M)$. Thus, we can use this definition of the $ U_j$ ’s in (\[ascore\ ]), and the result estimating equations are still unbiased. This design is closely associate to a class of designs introduced by @avron, who nominate selecting the $ U_j$ ’s as follows. Suppose $ H$ is a Hadamard matrix, that is, an $ n\times n$ extraneous matrix with elements $ \pm1$. @avron actually regard $ H$ a multiple of a unitary matrix, but the special case $ H$ Hadamard make their proposal most similar to ours. Then, using simple random sample distribution (with surrogate), they choose $ N$ columns from this matrix and multiply this $ n\times N$ matrix by an $ n\times n$ diagonal matrix with diagonal entries made up of independent symmetric Bernoulli random variables. The column of this resulting matrix are the $ U_j$ ’s. We are also reproduce a subset of the column of a Hadamard matrix by a random diagonal matrix, but we do not select the columns by childlike random sampling from some arbitrary Hadamard matrix. The extra structure we impose yield beneficial results in terms of the variance of the randomized trace approximation, as the following calculation show.

N=2^q$ vor a positive integer $q$, then the $\beta_j$’s can bx choseh to be ghe design matrix for a satucatee modtj of a $2^q$ factorial design ij which rhe owvels of tis factovf ars set et $\pm1$ \[@BHH, Chaptgr 5\]. In addithon, assume thad $B_1,\lbots,X_m$ are random diagonal matrices jf size $N$ and $Y_{jk}$, $j=1,\ldotf,N; k=1,\kqots,j$ are random variables such that aml the viagonal elemenys of the $X_j$’s and all the $J_{jk}$’s are i.i.d. symmetric Bernoulli tzndjn variables. Ghen defint $$\nabel{Uj} U_j = \pmatrix{ Y_{j1}X_1 \cr \vdots \cr Y_{jm}X_m} \beta_j.$$ Une ccn easily syoq tjdt for any $Im\timef Nm$ matrix $M$, $E (\frac{1}{T}\sum_{j=1}^N I_j'MU_j ) = \operatprneme{te}(M)$. Thus, we can use thms definition of the $U_j$’s in (\[avckre\]), and the resulribg esjimathng dwuagiohs ade stipl nnbiased. Thia design is closely related to a bkass of desifns inerjduced by @avron, who propose selecting tve $H_j$’s as follows. Suppose $Y$ is a Hadamard matrid, that is, an $n\times n$ orthogonal matrix with elements $\pm1$. @aeron ecguaoln covwifer $H$ a multiple of a unitary matrix, but the flevisl case $H$ Hadaiard makes yhfit proposal mosj similcd fo ours. Then, using simple randim sampligg (woth replacement), they choose $N$ columns fgom rhis matrix and muptiply this $n\tikes N$ matrix by an $n\times n$ dnagonam matrix wihh diagonzu entries made uo on itdependent symmetric Bernotlli randim vcriables. The columgs of this resulting matrix are thf $U_j$’s. Wa are also multiplying a subset of the columns of a Hadsmdrd matrix yy a rsndom diagonaj matrix, but wg do not felecg the colujns by vimple randjm sampling fslm some arbivrary Hadwmare marrix. The dxtra structurr we impose yields veneficial results in ferms of the vaxnabce of the randpmixed tgacx appwmximation, as the foukowine calculatiivs snow.

N=2^q$ for a positive integer $q$, then can chosen to the design matrix a factorial design in the levels of factors are set at $\pm1$ \[@BHH, 5\]. In addition, assume that $X_1,\ldots,X_m$ are random diagonal matrices of size $N$ $Y_{jk}$, $j=1,\ldots,N; k=1,\ldots,m$ are random variables such that all the diagonal elements of $X_j$’s all $Y_{jk}$’s i.i.d. symmetric Bernoulli random variables. Then define $$\label{Uj} U_j = \pmatrix{ Y_{j1}X_1 \cr \vdots \cr Y_{jm}X_m} One can easily show that for any $Nm\times matrix $M$, $E (\frac{1}{N}\sum_{j=1}^N ) = \operatorname{tr}(M)$. Thus, we use definition of $U_j$’s (\[ascore\]), the resulting estimating are still unbiased. This design is closely related to a class of designs introduced by @avron, who selecting the follows. Suppose is Hadamard that is, an orthogonal matrix with elements $\pm1$. @avron a multiple of a unitary matrix, but the case $H$ makes their proposal most similar to Then, using simple random sampling (with replacement), they $N$ columns from this matrix and multiply this $n\times N$ matrix by an $n\times n$ with diagonal entries made of independent symmetric random The of resulting matrix the $U_j$’s. We are also multiplying a subset of the columns a Hadamard matrix by a random diagonal matrix, but we select columns by simple sampling from some arbitrary matrix. extra structure we impose results terms of randomized approximation, as the following show.

N=2^q$ for a positive integer $q$, theN the $\beta_j$’s Can be ChoSen To Be thE desIgn matrix for a sATuraTed model of a $2^q$ factorial dEsign In WHich THe LevelS of the fACtORS arE sEt At $\pM1$ \[@Bhh, CHapteR 5\]. In AdditioN, assume thaT $X_1,\lDoTs,X_m$ are randoM DiAgonal matrIceS of size $N$ and $Y_{Jk}$, $j=1,\Ldots,N; K=1,\lDotS,M$ are rAndOm varIables SUch thaT all the diAgONal eleMEnts of tHE $x_j$’S and All the $Y_{jk}$’s are i.i.d. sYMmETric Bernoulli rAndom vArIAbLES. ThEn dEfine $$\label{uj} u_j = \pmaTRix{ Y_{j1}X_1 \cR \VdOTS \Cr Y_{JM}X_m} \beta_j.$$ One caN easily show THat For any $nm\TimES Nm$ matRix $M$, $E (\FrAC{1}{N}\sUm_{j=1}^N U_j'MU_j ) = \opEratOrname{tr}(M)$. thus, we CAn use thIS definiTion of The $u_j$’s In (\[asCOrE\]), aNd tHe REsuLTiNg eSTimAting equAtIoNs are StilL UNBIaseD. ThIs deSign iS closely relatEd tO a clASs oF desiGns inTrodUcEd by @aVron, whO propOsE selecting the $U_j$’S as fOllows. SupPosE $H$ Is a haDamarD Matrix, ThaT is, An $n\timeS n$ orthoGOnaL mATRIx With elements $\pm1$. @avroN aCTUaLly consiDer $H$ a mULtIpLE of a unitArY maTrix, BUT the sPeciAL cAse $H$ HadaMard maKEs ThEir propOsAl most SiMilAr tO ours. tHen, uSing siMple randOm samPLing (with replacEMent), they choosE $n$ cOLUmNS froM thIs matrix and MultIPly tHis $n\TImEs N$ MAtrix By an $n\TiMEs N$ Diagonal matrix with dIaGonal eNtrieS made up of indePendent symMETRic BernoUlli RAnDOm variables. The ColumNs of this reSUlting maTrix aRe the $U_j$’s. we are also MULtiplyinG a sUbsEt oF thE COlUmns of a HadamaRD MatrIx By a randOm dIagonal MatRix, But We dO nOt select tHe columnS bY sImPlE raNdom sAMpling frOm SomE aRbiTrary hAdamarD matrIx. ThE eXtRA stRucture WE iMPOse yIeLdS benEfiCiAl resUlts IN teRms of thE variance Of tHE ranDoMiZed tracE approximatioN, aS the followInG caLculatIONs show.

N=2^q$ for a positive inte ger $q$, t hen t he$\b et a_j$ ’s c an be chosen t o bethe design matrix fora sat ur a tedm od el of a $2^q $ f a c tor ia ldes ig n i n whi chthe lev els of the fa ct ors are seta t$\pm1$ \[@ BHH , Chapter 5\ ].In add it ion , assu methat$X_1,\ l dots,X _m$ are r an d om dia g onal ma t r ic es o f size $N$ and $Y _ {j k }$, $j=1,\ldot s,N; k =1 , \l d o ts, m$are random v ariab l es such th a t all the diagonalelements of the $X_j$ ’s an d all t he $Y _{ j k}$ ’s are i.i. d. s ymmetricBernou l li rand o m varia bles.The n d efin e $ $\ lab el { Uj} U_ j = \pm atrix{ Y _{ j1 }X_1\cr\ v d o ts \ crY_{j m}X_m } \beta_j.$$One can eas ily s how t hatfo r any $Nm\t imesNm $ matrix $M$, $ E (\ frac{1}{N }\s um _{j =1 }^N U _ j'MU_j )= \ operato rname{t r }(M )$ . T hu s, we can use this d e f in ition of the $ U _j $’ s in (\[a sc ore \]), a nd th e re s ul ting est imatin g e qu ationsar e stil lunb ias ed. T hisdesign is clos ely r e lated to a cla s s of designsi nt r o du c ed b y @ avron, whoprop o se s elec t in g t h e $U_ j$’sas fo l lows. Suppose $H$ i sa Hada mardmatrix, thatis, an $n\ t i m es n$ or thog o na l matrix with e lemen ts $\pm1$. @avron a ctual ly consi der $H$ a m ultipleofa u nit ary m at rix, but thes p ecia lcase $H $ H adamard ma kes th eir p roposal m ost simi la rto o urs . The n , usingsi mpl eran dom s a mpling (wit h re pl ac e men t), the y c h o ose$N $colu mns f rom t hism atr ix andmultiplythi s $n\ ti me s N$ ma trix by an $n \t imes n$ di ag ona l matr i x with di agonal entries made upo f indep end ent s ymme tric Bern oul li ran dom variab les. T he co lu mns o f thi s re sul ti ng matrixa r e t he $U _j $’s. We are also multiplyinga su bset of the c olu mnso f a Ha d am a rdma t rix b y a random diag onal matri x, bu t we do no t se le ct thecolumns by s i mple ra ndom samp ling from s omea r bit rary Hadam ard matr ix. Thee xtras tr uctur e w e impo se yi eldsbenefi c ial resu lts in t erms o f the v arianceof the randomized trace appro ximat ion , as thefol l owi ng calcul atio ns show.

N=2^q$ for_a positive_integer $q$, then the_$\beta_j$’s can_be_chosen to_be_the design matrix_for a saturated_model of a $2^q$_factorial design in_which_the levels of the factors are set at $\pm1$ \[@BHH, Chapter 5\]. In addition,_assume_that $X_1,\ldots,X_m$_are_random_diagonal matrices of size $N$_and $Y_{jk}$, $j=1,\ldots,N; k=1,\ldots,m$ are_random variables_such that all the diagonal elements of the_$X_j$’s_and all the_$Y_{jk}$’s are i.i.d. symmetric Bernoulli random variables. Then define_$$\label{Uj} U_j = \pmatrix{ Y_{j1}X_1 \cr \vdots \cr Y_{jm}X_m} \beta_j.$$ One can easily_show that for_any_$Nm\times_Nm$ matrix $M$, $E_(\frac{1}{N}\sum_{j=1}^N U_j'MU_j ) = \operatorname{tr}(M)$. Thus,_we can use this definition of_the $U_j$’s in (\[ascore\]), and the resulting_estimating equations are still unbiased. This design_is closely related to a_class of_designs introduced by @avron, who_propose selecting the_$U_j$’s as_follows. Suppose $H$_is a Hadamard matrix, that is,_an $n\times n$_orthogonal matrix with elements $\pm1$. @avron_actually_consider $H$ a_multiple_of_a unitary_matrix, but the_special_case $H$_Hadamard_makes their proposal most similar to_ours._Then, using simple random sampling (with replacement),_they choose $N$ columns_from_this matrix and multiply_this $n\times N$ matrix by_an $n\times n$ diagonal matrix with diagonal_entries made_up of_independent symmetric Bernoulli random variables. The columns of this resulting matrix_are the $U_j$’s. We are also_multiplying a subset of_the columns_of_a Hadamard matrix_by_a random_diagonal matrix, but we do not select_the columns_by simple random sampling from some_arbitrary Hadamard matrix. The extra_structure_we impose yields beneficial results in_terms of the variance of the_randomized trace approximation, as the_following_calculations_show.

= \operatorname{Supp}G_{1}/E_{1} \cup \cdots \cup \operatorname{Supp}G_{n}/E_{n}.$$ Therefore $${\operatorname{grade}\,}G/E = \min_{1 \leq i \leq n} \{ {\operatorname{grade}\,}G_{i}/E_{i} \} \geq 2 \iff {\operatorname{grade}\,}G_{i}/E_{i} \geq 2, \; 1 \leq i \leq n,$$ proving the equivalence. We observe that a direct sum of ideals cannot be a [complete intersection]{} module. \[ds1\] Let $R$ be a [Noetherian ring]{} and $E = {{E}_1 {\oplus}\cdots {\oplus}{E}_{n}}$ with $E_{i}$ [finitely generated $R$-module]{}s having positive rank $e_{i}$, $1 \leq i \leq n$, $n \geq 2$. Then $F_{e}(E) = F_{e_{1}}(E_{1}) \cdots F_{e_{n}}(E_{n})$, $e= \operatorname{rank}E$. In particular ${\displaystyle}{{\operatorname{grade}\,}F_{e}(E) = \min_{1 \leq i \leq n} \{ {\operatorname{grade}\,}F_{e_{i}}(E_{i}) \}}$ and ${\displaystyle}{\operatorname{ht}F_{e}(E) = \min_{1 \leq i \leq n} \{ \operatorname{ht}F_{e_{i}}(E_{i}) \}}$. Since $E_{i}$ has rank $e_{i}>0$, then $F_{k}(E_{i}) = (0)$ for $k < e_{i}$. Now $$F_{e}(E) = F_{e}({{E}_1 {\oplus}\cdots {\oplus}{E}_{n}}) = \sum_{j_{1}+ \cdots +j_{n}=e} F_{j_{1}}(E_{1}) \cdots F_{j_{n}}(E_{n}) = F_{e_{1}}(E_{1}) \cdots F_{e_{n}}(E_{n}).$$ The other assertions follow. \[dsi2\] Let ${(R,{{\mathfrak{m}}}, {k})}$ be a [Noetherian local ring]{} and $

= \operatorname{Supp}G_{1}/E_{1 } \cup \cdots \cup \operatorname{Supp}G_{n}/E_{n}.$$ Therefore $ $ { \operatorname{grade}\,}G / E = \min_{1 \leq i \leq n } \ { { \operatorname{grade}\,}G_{i}/E_{i } \ } \geq 2 \iff { \operatorname{grade}\,}G_{i}/E_{i } \geq 2, \; 1 \leq i \leq n,$$ proving the equivalence. We note that a lineal sum of ideals cannot be a [ accomplished overlap ] { } module. \[ds1\ ] Let $ R$ be a [ Noetherian ring ] { } and $ vitamin e = { { E}_1 { \oplus}\cdots { \oplus}{E}_{n}}$ with $ E_{i}$ [ finitely generated $ R$-module]{}s having convinced rank $ e_{i}$, $ 1 \leq i \leq n$, $ n \geq 2$. Then $ F_{e}(E) = F_{e_{1}}(E_{1 }) \cdots F_{e_{n}}(E_{n})$, $ e= \operatorname{rank}E$. In finical $ { \displaystyle}{{\operatorname{grade}\,}F_{e}(E) = \min_{1 \leq i \leq n } \ { { \operatorname{grade}\,}F_{e_{i}}(E_{i }) \}}$ and $ { \displaystyle}{\operatorname{ht}F_{e}(E) = \min_{1 \leq i \leq n } \ { \operatorname{ht}F_{e_{i}}(E_{i }) \}}$. Since $ E_{i}$ has rank $ e_{i}>0 $, then $ F_{k}(E_{i }) = (0)$ for $ k < e_{i}$. immediately $ $ F_{e}(E) = F_{e}({{E}_1 { \oplus}\cdots { \oplus}{E}_{n } }) = \sum_{j_{1}+ \cdots + j_{n}=e } F_{j_{1}}(E_{1 }) \cdots F_{j_{n}}(E_{n }) = F_{e_{1}}(E_{1 }) \cdots F_{e_{n}}(E_{n}).$$ The early assertions follow. \[dsi2\ ] lease $ { (R,{{\mathfrak{m } } }, { k})}$ be a [ Noetherian local ring ] { } and $

= \operatorname{Supp}G_{1}/E_{1} \cup \cdots \cup \opercrornamx{Supp}G_{n}/S_{n}.$$ Therewore $${\operatorname{grade}\,}G/E = \mun_{1 \lew i \leq n} \{ {\operatornamd{grade}\,}G_{i}/E_{p} \} \geq 2 \idf {\okeratorname{grade}\,}G_{m}/S_{i} \geq 2, \; 1 \peq n \oeq n,$$ proving jhe equivaletce. We observe dhxt a direct sum of ideals cannot be a [compleye intersection]{} iodukq. \[ds1\] Mvt $R$ be a [Noetherian ring]{} and $E = {{E}_1 {\oplls}\cdots {\oplus}{E}_{n}}$ woth $E_{i}$ [finitely generated $G$-modkle]{}s having positige rank $e_{i}$, $1 \leq u \leq n$, $n \ged 2$. Then $F_{e}(T) = F_{e_{1}}(E_{1}) \cdots F_{e_{n}}(E_{n})$, $e= \operatorname{rank}E$. Iv parcicular ${\disklzydjyle}{{\operatoriame{grwde}\,}F_{e}(E) = \min_{1 \leq i \lex n} \{ {\oprratorname{gradc}\,}F_{e_{i}}(E_{m}) \}}$ abd ${\displaystyle}{\operatmrname{ht}F_{e}(E) = \min_{1 \lgq i \leq n} \{ \kperatorname{ht}F_{e_{i}}(W_{i}) \}}$. Sincg $E_{i}$ vas fqnk $e_{i}>0$, tien $F_{k}(E_{i}) = (0)$ fkr $k < e_{i}$. Nos $$F_{e}(E) = F_{e}({{E}_1 {\iplus}\cdots {\oplus}{E}_{n}}) = \stn_{j_{1}+ \cdots +j_{h}=e} F_{j_{1}}(E_{1}) \cqots F_{j_{n}}(E_{n}) = F_{e_{1}}(E_{1}) \cdots F_{e_{n}}(E_{n}).$$ The otheg asaertions follow. \[dsi2\] Let ${(E,{{\mathfrak{m}}}, {k})}$ be a [Noejherian losal ring]{} and $

= \operatorname{Supp}G_{1}/E_{1} \cup \cdots \cup \operatorname{Supp}G_{n}/E_{n}.$$ Therefore \min_{1 i \leq \{ {\operatorname{grade}\,}G_{i}/E_{i} \} 2, 1 \leq i n,$$ proving the We observe that a direct sum ideals cannot be a [complete intersection]{} module. \[ds1\] Let $R$ be a [Noetherian and $E = {{E}_1 {\oplus}\cdots {\oplus}{E}_{n}}$ with $E_{i}$ [finitely generated $R$-module]{}s having positive $e_{i}$, \leq \leq $n \geq 2$. Then $F_{e}(E) = F_{e_{1}}(E_{1}) \cdots F_{e_{n}}(E_{n})$, $e= \operatorname{rank}E$. In particular ${\displaystyle}{{\operatorname{grade}\,}F_{e}(E) = \min_{1 i \leq n} \{ {\operatorname{grade}\,}F_{e_{i}}(E_{i}) \}}$ and ${\displaystyle}{\operatorname{ht}F_{e}(E) \min_{1 \leq i \leq \{ \operatorname{ht}F_{e_{i}}(E_{i}) \}}$. Since $E_{i}$ rank then $F_{k}(E_{i}) (0)$ $k e_{i}$. Now $$F_{e}(E) F_{e}({{E}_1 {\oplus}\cdots {\oplus}{E}_{n}}) = \sum_{j_{1}+ \cdots +j_{n}=e} F_{j_{1}}(E_{1}) \cdots F_{j_{n}}(E_{n}) = F_{e_{1}}(E_{1}) \cdots F_{e_{n}}(E_{n}).$$ The other assertions \[dsi2\] Let be a local and

= \operatorname{Supp}G_{1}/E_{1} \cup \cdotS \cup \operatOrnamE{SuPp}G_{N}/E_{N}.$$ TheRefoRe $${\operatorname{GRade}\,}g/E = \min_{1 \leq i \leq n} \{ {\operatornAme{grAdE}\,}g_{i}/E_{i} \} \GEq 2 \Iff {\opEratornAMe{GRAde}\,}g_{i}/e_{i} \Geq 2, \; 1 \LeQ I \lEq n,$$ prOviNg the eqUivalence. WE obSeRve that a direCT sUm of ideals CanNot be a [compleTe iNterseCtIon]{} MOdule. \[Ds1\] LEt $R$ be A [NoethERian riNg]{} and $E = {{E}_1 {\opLuS}\Cdots {\oPLus}{E}_{n}}$ wiTH $e_{i}$ [FiniTely generated $R$-modULe]{}S Having positive Rank $e_{i}$, $1 \LeQ I \lEQ N$, $n \gEq 2$. THen $F_{e}(E) = F_{e_{1}}(E_{1}) \cDoTs F_{e_{n}}(e_{N})$, $e= \operaTOrNAME{raNK}E$. In particulaR ${\displaystyLE}{{\opEratorNaMe{gRAde}\,}F_{e}(E) = \Min_{1 \leQ i \LEq n} \{ {\OperatornamE{graDe}\,}F_{e_{i}}(E_{i}) \}}$ anD ${\displAYstyle}{\oPEratornAme{ht}F_{E}(E) = \mIn_{1 \lEq i \lEQ n} \{ \OpEraToRNamE{Ht}f_{e_{i}}(e_{I}) \}}$. SiNce $E_{i}$ has RaNk $E_{i}>0$, theN $F_{k}(E_{I}) = (0)$ FOR $K < e_{i}$. NOw $$F_{E}(E) = F_{e}({{e}_1 {\opluS}\cdots {\oplus}{E}_{n}}) = \Sum_{J_{1}+ \cdoTS +j_{n}=E} F_{j_{1}}(E_{1}) \cDots F_{J_{n}}(E_{n}) = f_{e_{1}}(e_{1}) \cdotS F_{e_{n}}(E_{n}).$$ the otHeR assertions follOw. \[dsI2\] Let ${(R,{{\mathFraK{m}}}, {K})}$ be A [NOetheRIan locAl rIng]{} And $

= \operatorname{Supp} G_{1}/E_{1 } \cu p \ cdo ts \cu p \o peratorname{Su p p}G_ {n}/E_{n}.$$ Therefore $${\ op e rato r na me{gr ade}\,} G /E = \m in_ {1 \l eq i\le q n} \{ {\operato rna me {grade}\,}G_ { i} /E_{i} \}\ge q 2 \iff {\o per atorna me {gr a de}\, }G_ {i}/E _{i} \ g eq 2, \; 1 \ l eq i \ l eq n,$$ p ro ving the equivalence. W e observe thata dire ct su m ofide als cannot b e a [ c omplete in t e r sec t ion]{} module . \[ds1\]L et$R$ be a [N o etheri an ri ng ] {}and $E = {{ E}_1 {\oplus} \cdots {\oplus } {E}_{n} }$ wit h $ E_{ i}$[ fi ni tel yg ene r at ed$ R$- module]{ }s h aving pos i t i v e ra nk$e_{ i}$,$1 \leq i \le q n $, $ n \g eq 2$ . The n $F _{ e}(E) = F_{ e_{1} }( E_{1}) \cdots F _{e_ {n}}(E_{n })$ ,$e= \o p erator nam e{r ank}E$. In par t icu la r $ {\ displaystyle}{{\op er a t or name{gra de}\,} F _{ e} ( E) = \mi n_ {1\leq i \leq n}\ {{\operat orname { gr ad e}\,}F_ {e _{i}}( E_ {i} ) \ }}$ a n d ${ \displ aystyle} {\ope r atorname{ht}F_ { e}(E) = \min_ { 1\ l eq i \l eqn} \{ \oper ator n ame{ ht}F _ {e _{i } }(E_{ i}) \ }} $ .Since $E_{i}$ has r an k $e_{ i}>0$ , then $F_{k} (E_{i}) =( 0 ) $ for $k < e _ {i } $. Now $$F_{e} (E) = F_{e}({{E } _1 {\opl us}\c dots {\o plus}{E}_ { n }}) = \s um_ {j_ {1} + \ c d ot s +j_{n}= e } F_{ j_ {1}}(E_ {1} ) \cdot s F _{j _{n }}( E_ {n}) = F_{e_{ 1} }( E_ {1 })\cdot s F_{e_{n }} (E_ {n }). $$ Th e other asse rtio ns f o llo w. \[d s i2 \ ] Let $ {( R,{{ \ma th frak{ m}}} , {k })}$ be a [Noeth eri a n lo ca lring]{} and $

= _ _ \operatorname{Supp}G_{1}/E_{1} \cup \cdots_\cup \operatorname{Supp}G_{n}/E_{n}.$$_Therefore_$${\operatorname{grade}\,}G/E = __ \min_{1_\leq i \leq_n} \{ {\operatorname{grade}\,}G_{i}/E_{i} \}_\geq 2 \iff_{\operatorname{grade}\,}G_{i}/E_{i}_\geq 2, \; 1 \leq i \leq n,$$ proving the_equivalence. We_observe that_a_direct_sum of ideals cannot be_a [complete intersection]{} module. \[ds1\] Let_$R$ be_a [Noetherian ring]{} and $E = {{E}_1 {\oplus}\cdots_{\oplus}{E}_{n}}$_with $E_{i}$ [finitely_generated $R$-module]{}s having positive rank $e_{i}$, $1 \leq i_\leq n$, $n \geq 2$. Then_$F_{e}(E) = F_{e_{1}}(E_{1})_\cdots_F_{e_{n}}(E_{n})$,_$e= _\operatorname{rank}E$. In particular ${\displaystyle}{{\operatorname{grade}\,}F_{e}(E) = \min_{1_\leq i \leq n} \{ {\operatorname{grade}\,}F_{e_{i}}(E_{i})_\}}$ and ${\displaystyle}{\operatorname{ht}F_{e}(E) = \min_{1 \leq i_\leq n} \{ \operatorname{ht}F_{e_{i}}(E_{i}) \}}$. Since $E_{i}$_has rank $e_{i}>0$, then $F_{k}(E_{i})_= (0)$_for $k < e_{i}$. Now $$F_{e}(E)_= F_{e}({{E}_1 {\oplus}\cdots_{\oplus}{E}_{n}}) =_\sum_{j_{1}+ \cdots _ +j_{n}=e} F_{j_{1}}(E_{1}) \cdots F_{j_{n}}(E_{n})_= _ F_{e_{1}}(E_{1}) \cdots F_{e_{n}}(E_{n}).$$ The other_assertions_follow. \[dsi2\] Let ${(R,{{\mathfrak{m}}},_{k})}$_be_a [Noetherian_local ring]{} and_$

\left[e^{2 {{\sigma_{\! \! x}}}^2k^2 r} (3 r^2 + 6r -25)-(3r^2-6r-25)\right] \\ \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad+ \frac{12r}{({{\sigma_{\! \! x}}}k)^4} \left[(3r^3-12r^2-19r-28)- e^{2{{\sigma_{\! \! x}}}^2k^2 r} (3 r^3+12r^2-19r+28)\right] \\ \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad+\frac{48}{({{\sigma_{\! \! x}}}k)^6} \left[e^{2{{\sigma_{\! \! x}}}^2k^2 r }(3r^2-12 r+7)-(3r^2+12r+7)\right]\\ \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad +\frac{576}{({{\sigma_{\! \! x}}}k)^8}\left(e^{2 {{\sigma_{\! \! x}}}^2k^2 r}-1\right)\Bigg\} \\ P_{\theta\theta,13}(k) &= \frac{k^3}{2\pi^2} \cdot \frac{1}{168} P_L(k)\int_0^\infty dr \ P_L(kr) \Bigg\{ \frac{12}{ r^2} -82 +4r^2 -6 r^4 -\frac{3}{ r^3} \left(r^2-1\right)^3 \left(r^2+2\right) \log \left|\frac{r-1}{r+1}\right| \Bigg\}\end{aligned}$$ $\mathrm{Ei}(x)$ denotes the exponential integral defined as $\mathrm{Ei}(x) = - \int_{-x}^\infty t^{-1} e^{-t}\ dt$. $$\begin{aligned} P_{\bar{{\bm{w} }}\bar{{\bm{w} }},22}(k) &= [(2\pi)^3 \delta_{\rm D}(0) ]^{-1}\langle \bar{{\bm{w} }}_2({\bm{k} }) \cdot \bar

\left[e^{2 { { \sigma_{\! \! x}}}^2k^2 r } (3 r^2 + 6r -25)-(3r^2 - 6r-25)\right ] \\ \notag & \qquad\qquad\qquad\qquad\qquad\qquad\qquad+ \frac{12r}{({{\sigma_{\! \! x}}}k)^4 } \left[(3r^3 - 12r^2 - 19r-28)- e^{2{{\sigma_{\! \! x}}}^2k^2 r } (3 r^3 + 12r^2 - 19r+28)\right ] \\ \notag & \qquad\qquad\qquad\qquad\qquad\qquad\qquad+\frac{48}{({{\sigma_{\! \! x}}}k)^6 } \left[e^{2{{\sigma_{\! \! x}}}^2k^2 r } (3r^2 - 12 r+7)-(3r^2 + 12r+7)\right]\\ \notag & \qquad\qquad\qquad\qquad\qquad\qquad\qquad + \frac{576}{({{\sigma_{\! \! x}}}k)^8}\left(e^{2 { { \sigma_{\! \! x}}}^2k^2 r}-1\right)\Bigg\ } \\ P_{\theta\theta,13}(k) & = \frac{k^3}{2\pi^2 } \cdot \frac{1}{168 } P_L(k)\int_0^\infty dr \ P_L(kr) \Bigg\ { \frac{12 } { r^2 } -82 +4r^2 -6 r^4 -\frac{3 } { r^3 } \left(r^2 - 1\right)^3 \left(r^2 + 2\right) \log \left|\frac{r-1}{r+1}\right| \Bigg\}\end{aligned}$$ $ \mathrm{Ei}(x)$ denotes the exponential integral defined as $ \mathrm{Ei}(x) = - \int_{-x}^\infty t^{-1 } e^{-t}\ dt$. $ $ \begin{aligned } P_{\bar{{\bm{w } } } \bar{{\bm{w } } }, 22}(k) & = [ (2\pi)^3 \delta_{\rm D}(0) ] ^{-1}\langle \bar{{\bm{w } } } _ 2({\bm{k } }) \cdot \bar

\levt[e^{2 {{\sigma_{\! \! x}}}^2k^2 r} (3 r^2 + 6r -25)-(3r^2-6r-25)\vight] \\ \notag &\qquab\wquad\qxuad\qqhad\qquad\dquad\qquad+ \frac{12r}{({{\sigma_{\! \! x}}}k)^4} \lxft[(3r^3-12e^2-19r-28)- e^{2{{\sugma_{\! \! x}}}^2k^2 r} (3 r^3+12r^2-19r+28)\right] \\ \votag &\qquwd\qquad\qwuad\wwuad\qquad\qquad\qquad+\frac{48}{({{\sjnma_{\! \! r}}}k)^6} \left[e^{2{{\sigma_{\! \! x}}}^2l^2 r }(3r^2-12 r+7)-(3r^2+12r+7)\rhght]\\ \notag &\qquag\qducd\qquad\qquad\qquad\qquad\qquad +\frac{576}{({{\sigmw_{\! \! x}}}k)^8}\legt(f^{2 {{\sigma_{\! \! x}}}^2k^2 r}-1\ryght)\Nygg\} \\ L_{\nhtta\theta,13}(k) &= \frac{k^3}{2\pi^2} \cdot \frac{1}{168} P_L(k)\jnt_0^\inftj dr \ P_L(kr) \Bigg\{ \ftac{12}{ r^2} -82 +4r^2 -6 r^4 -\frac{3}{ r^3} \left(r^2-1\rihht)^3 \peft(r^2+2\right) \log \lefh|\frac{r-1}{r+1}\righj| \Bidt\}\end{aligned}$$ $\mxthrm{Ei}(x)$ dtnmtes the esponential integral defined as $\oathrk{Ei}(x) = - \int_{-z}^\ibftj t^{-1} e^{-t}\ dt$. $$\begmn{aligged} P_{\bar{{\bm{w} }}\bar{{\bm{w} }},22}(k) &= [(2\pi)^3 \delya_{\rm D}(0) ]^{-1}\langle \nar{{\bm{x} }}_2({\bm{j} }) \cdot \bar

\left[e^{2 {{\sigma_{\! \! x}}}^2k^2 r} (3 r^2 -25)-(3r^2-6r-25)\right] \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad+ \! x}}}k)^4} \left[(3r^3-12r^2-19r-28)- r^3+12r^2-19r+28)\right] \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad+\frac{48}{({{\sigma_{\! \! \left[e^{2{{\sigma_{\! \! x}}}^2k^2 }(3r^2-12 r+7)-(3r^2+12r+7)\right]\\ \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad +\frac{576}{({{\sigma_{\! \! {{\sigma_{\! \! x}}}^2k^2 r}-1\right)\Bigg\} \\ P_{\theta\theta,13}(k) &= \frac{k^3}{2\pi^2} \cdot \frac{1}{168} P_L(k)\int_0^\infty dr \ \Bigg\{ \frac{12}{ r^2} -82 +4r^2 -6 r^4 -\frac{3}{ r^3} \left(r^2-1\right)^3 \left(r^2+2\right) \log \left|\frac{r-1}{r+1}\right| $\mathrm{Ei}(x)$ the integral as $\mathrm{Ei}(x) = - \int_{-x}^\infty t^{-1} e^{-t}\ dt$. $$\begin{aligned} P_{\bar{{\bm{w} }}\bar{{\bm{w} }},22}(k) &= [(2\pi)^3 \delta_{\rm D}(0) \bar{{\bm{w} }}_2({\bm{k} }) \cdot \bar

\left[e^{2 {{\sigma_{\! \! x}}}^2k^2 r} (3 r^2 + 6r -25)-(3r^2-6r-25)\right] \\ \notAg &\qquad\qquAd\qquAd\qQuaD\qQuad\QquaD\qquad+ \frac{12r}{({{\sigMA_{\! \! x}}}k)^4} \lEft[(3r^3-12r^2-19r-28)- e^{2{{\sigma_{\! \! x}}}^2k^2 r} (3 r^3+12r^2-19r+28)\righT] \\ \notaG &\qQUad\qQUaD\qquaD\qquad\qQUaD\QQuaD\qQuAd+\fRaC{48}{({{\SiGma_{\! \! x}}}k)^6} \LefT[e^{2{{\sigma_{\! \! X}}}^2k^2 r }(3r^2-12 r+7)-(3r^2+12r+7)\rigHt]\\ \nOtAg &\qquad\qquad\QQuAd\qquad\qquAd\qQuad\qquad +\fraC{576}{({{\siGma_{\! \! x}}}k)^8}\lEfT(e^{2 {{\sIGma_{\! \! x}}}^2k^2 R}-1\riGht)\BiGg\} \\ P_{\theTA\theta,13}(K) &= \frac{k^3}{2\pi^2} \cDoT \Frac{1}{168} P_L(K)\Int_0^\inftY DR \ P_l(kr) \BIgg\{ \frac{12}{ r^2} -82 +4r^2 -6 r^4 -\frac{3}{ r^3} \leFT(r^2-1\RIght)^3 \left(r^2+2\right) \Log \lefT|\fRAc{R-1}{R+1}\RigHt| \BIgg\}\end{aligNeD}$$ $\mathRM{Ei}(x)$ denOTeS THE exPOnential integRal defined aS $\MatHrm{Ei}(x) = - \InT_{-x}^\iNFty t^{-1} e^{-t}\ Dt$. $$\begIn{ALigNed} P_{\bar{{\bm{w} }}\bAr{{\bm{W} }},22}(k) &= [(2\pi)^3 \delta_{\Rm D}(0) ]^{-1}\lanGLe \bar{{\bm{W} }}_2({\Bm{k} }) \cdot \Bar

\left[e^{2 {{\sigma_{\! \ ! x}}}^2k^ 2 r}(3r^2 + 6r-25) -(3r^2-6r-25)\ r ight ] \\ \notag &\qquad\qq uad\q qu a d\qq u ad \qqua d\qquad \ qq u a d+\f ra c{1 2r } {( {{\si gma _{\! \! x}}}k)^4} \ le ft[(3r^3-12r ^ 2- 19r-28)- e ^{2 {{\sigma_{\! \! x}}}^ 2k ^2r } (3r^3 +12r^ 2-19r+ 2 8)\rig ht] \\ \ no t ag &\q q uad\qqu a d \q quad \qquad\qquad\qqua d \q q uad+\frac{48}{ ({{\si gm a _{ \ ! \! x} }}k)^6} \l ef t[e^{ 2 {{\sigm a _{ \ ! \!x }}}^2k^2 r }( 3r^2-12 r+7 ) -(3 r^2+12 r+ 7)\ r ight]\ \ \no ta g &\ qquad\qquad \qqu ad\qquad\ qquad\ q quad\qq u ad +\fr ac{576 }{( {{\ sigm a _{ \! \! x } }}k ) ^8 }\l e ft( e^{2 {{\ si gm a_{\! \!x } } } ^2k^ 2 r }-1\ right )\Bigg\} \\ P _{\ thet a \th eta,1 3}(k) &= \ frac{ k^3}{2 \pi^2 }\cdot \frac{1}{ 168} P_L(k)\i nt_ 0^ \in ft y dr\ P_L(k r)\Bi gg\{ \f rac{12} { r^ 2} - 8 2+4r^2 -6 r^4 -\fra c{ 3 } {r^3} \le ft(r^2 - 1\ ri g ht)^3 \l ef t(r ^2+2 \ r ight) \lo g \ left|\fr ac{r-1 } {r +1 }\right |\Bigg\ }\ end {al igned } $$ $\math rm{Ei}(x )$ de n otes the expon e ntial integra l d e f in e d as $\ mathrm{Ei}( x) = - \i nt_{ - x} ^\i n fty t ^{-1} e ^ {- t }\ dt$. $$\begin{a li gned}P_{\b ar{{\bm{w} }} \bar{{\bm{ w } }},22}(k ) &= [( 2 \pi)^3 \delta_ {\rmD}(0) ]^{- 1 }\langle \bar {{\bm{w} }}_2({\b m { k} }) \c dot \ bar

\left[e^{2_{{\sigma_{\! \!_x}}}^2k^2 r} (3 r^2_+ 6r_-25)-(3r^2-6r-25)\right]_\\ \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad+_\frac{12r}{({{\sigma_{\!_\! x}}}k)^4} _\left[(3r^3-12r^2-19r-28)- e^{2{{\sigma_{\! \!_x}}}^2k^2 r} (3 r^3+12r^2-19r+28)\right]_ \\ \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad+\frac{48}{({{\sigma_{\!_\!_x}}}k)^6} \left[e^{2{{\sigma_{\! \! x}}}^2k^2 r }(3r^2-12 r+7)-(3r^2+12r+7)\right]\\ \notag &\qquad\qquad\qquad\qquad\qquad\qquad\qquad +\frac{576}{({{\sigma_{\! \! x}}}k)^8}\left(e^{2 {{\sigma_{\! \! x}}}^2k^2_r}-1\right)\Bigg\}_\\ P_{\theta\theta,13}(k) _&=_\frac{k^3}{2\pi^2}_\cdot \frac{1}{168} P_L(k)\int_0^\infty dr \_P_L(kr) \Bigg\{ \frac{12}{ r^2} -82_+4r^2 -6_r^4 -\frac{3}{ r^3} \left(r^2-1\right)^3 \left(r^2+2\right) \log \left|\frac{r-1}{r+1}\right| \Bigg\}\end{aligned}$$ $\mathrm{Ei}(x)$_denotes_the exponential integral_defined as $\mathrm{Ei}(x) = - \int_{-x}^\infty t^{-1} e^{-t}\ dt$. $$\begin{aligned} P_{\bar{{\bm{w}_}}\bar{{\bm{w} }},22}(k) &= [(2\pi)^3 \delta_{\rm D}(0)_]^{-1}\langle \bar{{\bm{w} }}_2({\bm{k}_})_\cdot_ \bar

in time polynomial in $n$. Furthermore, if $K$ is any field possessing $D_P$ distinct ${D_P}{^{\text{\underline{th}}}}$ roots of unity, then a ${\mathtt{3CNFSAT}}$ instance $B(y)\!:=C_1(y)\wedge \cdots \wedge C_k(y)$ has a satisfying assignment iff the univariate polynomial system $F_B\!:=\!({{\mathcal{P}}}_P(C_1), \ldots,{{\mathcal{P}}}_P(C_k))$ has a root $\zeta\!\in\!K$ satisfying $\zeta^{D_P}-1$. [$\blacksquare$]{} Plaisted actually proved the special case $K\!=\!{\mathbb{C}}$ of the above lemma, in slightly different language, in [@plaisted]. However, his proof extends verbatim to the more general family of fields detailed above. A simple consequence of the resultant is that vanishing at a $D{^{\text{\underline{th}}}}$ root of unity is algebraically the same thing over ${\mathbb{C}}$ or ${\mathbb{Q}}_p$, provided $p$ lies in the right arithmetic progression. \[lemma:uni\] Suppose $D\!\in\!{\mathbb{N}}$, $f\!\in\!{\mathbb{Z}}[x]$, and $p$ is any prime congruent to $1$ mod $D$. Then $f$ vanishes at a complex $D{^{\text{\underline{th}}}}$ root of unity $\Longleftrightarrow f$ vanishes at a $D{^{\text{\underline{th}}}}$ root of unity in ${\mathbb{Q}}_p$. Note that $x^2+x+1$ vanishes at a $3{^{\text{\underline{rd}}}}$ root of unity in ${\mathbb{C}}$, but has [**no**]{} roots at all in ${\mathbb{F}}_5$ or ${\mathbb{Q}}_5$. So our congruence assumption on $p$ is necessary. [$\diamond$]{} [**Proof of Lemma \[lemma:uni\]:**]{} First note that by our assumption on $p$, ${\mathbb{Q}}_p$ has $D$ distinct $D{^{\text{\underline{th}}}}$ roots of unity: This follows easily from Hensel’s Lemma and ${\mathbb{F}}_p$ having $D$ distinct $D{^{\text{\underline{th}}}}$ roots of unity. Since ${\mathbb{

in time polynomial in $ n$. Furthermore, if $ K$ is any field possessing $ D_P$ distinct $ { D_P}{^{\text{\underline{th}}}}$ root of oneness, then a $ { \mathtt{3CNFSAT}}$ instance $ B(y)\!:=C_1(y)\wedge \cdots \wedge C_k(y)$ has a satisfying appointment iff the univariate polynomial system $ F_B\!:=\!({{\mathcal{P}}}_P(C_1), \ldots,{{\mathcal{P}}}_P(C_k))$ has a etymon $ \zeta\!\in\!K$ satisfy $ \zeta^{D_P}-1$. [ $ \blacksquare$ ] { } Plaisted actually proved the special case $ K\!=\!{\mathbb{C}}$ of the above lemma, in slightly unlike language, in [ @plaisted ]. However, his proof extends verbatim to the more cosmopolitan family of fields detail above. A simple consequence of the resultant is that disappear at a $ D{^{\text{\underline{th}}}}$ root of unity is algebraically the same matter over $ { \mathbb{C}}$ or $ { \mathbb{Q}}_p$, provided $ p$ lies in the good arithmetic progress. \[lemma: uni\ ] Suppose $ D\!\in\!{\mathbb{N}}$, $ f\!\in\!{\mathbb{Z}}[x]$, and $ p$ is any prime congruent to $ 1 $ mod $ D$. Then $ f$ vanishes at a complex $ D{^{\text{\underline{th}}}}$ root of unity $ \Longleftrightarrow f$ vanishes at a $ D{^{\text{\underline{th}}}}$ root of oneness in $ { \mathbb{Q}}_p$. Note that $ x^2+x+1 $ vanishes at a $ 3{^{\text{\underline{rd}}}}$ root of unity in $ { \mathbb{C}}$, but has [ * * no * * ] { } roots at all in $ { \mathbb{F}}_5 $ or $ { \mathbb{Q}}_5$. So our congruence assumption on $ p$ is necessary. [ $ \diamond$ ] { } [ * * Proof of Lemma \[lemma: uni\ ] :* * ] { } First note that by our assumption on $ p$, $ { \mathbb{Q}}_p$ has $ D$ distinct $ D{^{\text{\underline{th}}}}$ roots of unity: This follow easily from Hensel ’s Lemma and $ { \mathbb{F}}_p$ having $ D$ distinct $ D{^{\text{\underline{th}}}}$ roots of oneness. Since $ { \mathbb {

in time polynomial in $n$. Fuvthermore, if $K$ is any fmeld poasessing $D_P$ distinct ${D_P}{^{\text{\underline{vh}}}}$ riots if unity, then a ${\mathtt{3ZNFSAT}}$ indtance $B(t)\!:=C_1(y)\wtdge \cdots \wedge R_i(y)$ has a satianying essignment iff jhe univariade polynomial vyrtzm $F_B\!:=\!({{\mathcal{P}}}_P(C_1), \ldots,{{\mathcal{P}}}_P(C_k))$ has a root $\zfta\!\in\!K$ satisfyyng $\eetw^{D_P}-1$. [$\glacksquare$]{} Plaisted actually proves the skecial case $K\!=\!{\mathbn{C}}$ of the above lemma, in spighhly different langkage, in [@plausteq]. However, his proof exttnbs verbatim to the more general family of wieldx detailed avovf. D simple coisequegce of the rcxultand is thst vanishing ab a $D{^{\vext{\ynderline{th}}}}$ root of uiity is algebraicalli the same tking over ${\mathbb{C}}$ or ${\narhbb{Q}}_k$, proeidea $p$ uiea mn fhe rihht arithmetid progressiin. \[lemma:uni\] Suppose $C\!\ig\!{\nathbb{N}}$, $f\!\in\!{\mafhbb{Z}}[x]$, agd $p$ is any prime congruent to $1$ mod $D$. Tven $f$ vanishes at a complez $D{^{\text{\underline{th}}}}$ rolt of uniey $\Longleftrightarrow f$ vanishes at a $D{^{\text{\underlite{th}}}}$ cout if jbihy in ${\mathbb{Q}}_p$. Note that $x^2+x+1$ vanishes at a $3{^{\text{\thdtrlpne{rd}}}}$ root of uniby in ${\mathbb{C}}$, but nad [**mj**]{} roots at alu in ${\mcfhgb{F}}_5$ or ${\mathbb{Q}}_5$. So lur condruenxe assumpuion pn $p$ is necessary. [$\diamond$]{} [**Priof of Lemma \[oemma:uni\]:**]{} First notz that by oux assukptiom on $p$, ${\mathbb{Q}}_p$ has $D$ dnstincf $D{^{\text{\undegline{th}}}}$ rkuts of unity: Thir fplnows easily from Hensel’s Lqmma and ${\nathyb{F}}_p$ havkng $C$ distynct $D{^{\text{\knderline{th}}}}$ roots of unihy. Siuce ${\mdthbb{

in time polynomial in $n$. Furthermore, if any possessing $D_P$ ${D_P}{^{\text{\underline{th}}}}$ roots of $B(y)\!:=C_1(y)\wedge \wedge C_k(y)$ has satisfying assignment iff univariate polynomial system $F_B\!:=\!({{\mathcal{P}}}_P(C_1), \ldots,{{\mathcal{P}}}_P(C_k))$ has root $\zeta\!\in\!K$ satisfying $\zeta^{D_P}-1$. [$\blacksquare$]{} Plaisted actually proved the special case $K\!=\!{\mathbb{C}}$ of above lemma, in slightly different language, in [@plaisted]. However, his proof extends verbatim the general of detailed above. A simple consequence of the resultant is that vanishing at a $D{^{\text{\underline{th}}}}$ root of is algebraically the same thing over ${\mathbb{C}}$ or provided $p$ lies in right arithmetic progression. \[lemma:uni\] Suppose $f\!\in\!{\mathbb{Z}}[x]$, $p$ is prime to mod $D$. Then vanishes at a complex $D{^{\text{\underline{th}}}}$ root of unity $\Longleftrightarrow f$ vanishes at a $D{^{\text{\underline{th}}}}$ root of unity ${\mathbb{Q}}_p$. Note vanishes at $3{^{\text{\underline{rd}}}}$ of in ${\mathbb{C}}$, but roots at all in ${\mathbb{F}}_5$ or congruence assumption on $p$ is necessary. [$\diamond$]{} [**Proof Lemma \[lemma:uni\]:**]{} note that by our assumption on ${\mathbb{Q}}_p$ has $D$ distinct $D{^{\text{\underline{th}}}}$ roots of unity: follows easily from Hensel’s Lemma and ${\mathbb{F}}_p$ having $D$ distinct $D{^{\text{\underline{th}}}}$ roots of unity. Since

in time polynomial in $n$. FurtheRmore, if $K$ is Any fiEld PosSeSsinG $D_P$ dIstinct ${D_P}{^{\text{\uNDerlIne{th}}}}$ roots of unity, then a ${\MathtT{3CnfSAT}}$ INsTance $b(y)\!:=C_1(y)\wedGE \cDOTs \wEdGe c_k(y)$ HaS A sAtisfYinG assignMent iff the UniVaRiate polynomIAl System $F_B\!:=\!({{\maThcAl{P}}}_P(C_1), \ldots,{{\maThcAl{P}}}_P(C_k))$ HaS a rOOt $\zetA\!\in\!k$ satiSfying $\ZEta^{D_P}-1$. [$\bLacksquarE$]{} PLAisted ACtually PROvEd thE special case $K\!=\!{\mathBB{C}}$ OF the above lemma, In sligHtLY dIFFerEnt Language, in [@PlAisteD]. however, HIs PROOf eXTends verbatim To the more geNEraL familY oF fiELds detAiled AbOVe. A Simple conseQuenCe of the reSultanT Is that vANishing At a $D{^{\teXt{\uNdeRlinE{Th}}}}$ RoOt oF uNIty IS aLgeBRaiCally the SaMe Thing Over ${\MATHBb{C}}$ oR ${\maThbb{q}}_p$, proVided $p$ lies in tHe rIght ARitHmetiC progRessIoN. \[lemmA:uni\] SuPpose $d\!\iN\!{\mathbb{N}}$, $f\!\in\!{\mathBb{Z}}[x]$, And $p$ is any PriMe ConGrUent tO $1$ Mod $D$. ThEn $f$ VanIshes at A compleX $d{^{\teXt{\UNDErLine{th}}}}$ root of unity $\LOnGLEfTrightarRow f$ vaNIsHeS At a $D{^{\text{\UnDerLine{TH}}}}$ Root oF uniTY iN ${\mathbb{Q}}_P$. Note tHAt $X^2+x+1$ VanisheS aT a $3{^{\text{\UnDerLinE{rd}}}}$ roOT of uNity in ${\Mathbb{C}}$, bUt has [**NO**]{} roots at all in ${\mAThbb{F}}_5$ or ${\mathbb{q}}_5$. so OUR cONgruEncE assumption On $p$ iS NeceSsarY. [$\DiAmoND$]{} [**ProoF of LeMmA \[LeMMa:uni\]:**]{} First note that bY oUr assuMptioN on $p$, ${\mathbb{Q}}_p$ hAs $D$ distincT $d{^{\TExt{\underLine{TH}}}}$ rOOts of unity: This FolloWs easily frOM Hensel’s lemma And ${\mathbB{F}}_p$ having $d$ DIstinct $D{^{\TexT{\unDerLinE{TH}}}}$ rOots of unity. SiNCE ${\matHbB{

in time polynomial in $n$ . Furtherm ore,if$K$ i s an y fi eld possessing $D_P $ distinct ${D_P}{^{\t ext{\ un d erli n e{ th}}} }$ root s o f uni ty ,the na $ {\mat htt {3CNFSA T}}$ insta nce $ B(y)\!:=C_1( y )\ wedge \cdo ts\wedge C_k( y)$ has a s ati s fying as signm ent if f the u nivariate p o lynomi a l syste m $F _B\! :=\!({{\mathcal{P } }} _ P(C_1), \ldots ,{{\ma th c al { P }}} _P( C_k))$ has a root $\zeta\ ! \i n \ ! K$s atisfying $\z eta^{D_P}-1 $ . [ $\blac ks qua r e$]{} Plai st e d a ctually pro vedthe speci al cas e $K\!=\ ! {\mathb b{C}}$ of th e ab o ve l emm a, ins li ght l y d ifferent l an guage , in [ @ p lais ted ]. H oweve r, his proofext ends ver batim to t he m or e gen eral f amily o f fields detail ed a bove. Asim pl e c on seque n ce ofthe re sultant is tha t va ni s h i ng at a $D{^{\text{\ un d e rl ine{th}} }}$ ro o tof unity is a lge brai c a lly t he s a me thing o ver ${ \ ma th bb{C}}$ o r ${\m at hbb {Q} }_p$, prov ided $ p$ liesin th e right arithme t ic progressio n .\ [l e mma: uni \] Suppose$D\! \ in\! {\ma t hb b{N } }$, $ f\!\i n\ ! {\ m athbb{Z}}[x]$, and$p $ is a ny pr ime congruent to $1$ mo d $ D$. Then $f$ va n ishes at a com plex$D{^{\text { \underli ne{th }}}}$ ro ot of uni t y $\Longl eft rig hta rro w f$ vanishes ata $D{^ {\ text{\u nde rline{t h}} }}$ ro otof unity in ${\math bb {Q }} _p $. Note that $x^ 2+ x+1 $van ishes at a $ 3{^{\ text {\ un d erl ine{rd} } }} $ root o funit y i n${\ma thbb { C}} $, buthas [**no **] { } ro ot sat allin ${\mathbb{ F} }_5$ or ${ \m ath bb{Q}} _ 5 $. So ou r congruence assumption on $p$isneces sary . [$\diam ond $]{} [** P roof o f Lemm a \[l em ma: u n i\]:* * ] {} Fi rs t note tha t byour a ss umpt ion on$p$, ${\mathbb{Q}} _ p$has $D$ disti nct $D{ ^ { \t ext { \u n der li n e{t h } }}}$ roots of u nity: This f o ll ows easily fro mHensel’ s Lemma and$ {\mathb b{F}}_p$having $D $dist i n ct$D{^{\text {\underl ine{th}}} } $ roo t sof un ity . Sinc e${\ mathb b{

in_time polynomial_in $n$. Furthermore, if_$K$ is_any_field possessing_$D_P$_distinct ${D_P}{^{\text{\underline{th}}}}$ roots_of unity, then_a ${\mathtt{3CNFSAT}}$ instance $B(y)\!:=C_1(y)\wedge_\cdots \wedge C_k(y)$_has_a satisfying assignment iff the univariate polynomial system $F_B\!:=\!({{\mathcal{P}}}_P(C_1), \ldots,{{\mathcal{P}}}_P(C_k))$ has a root $\zeta\!\in\!K$_satisfying_$\zeta^{D_P}-1$. [$\blacksquare$]{} Plaisted_actually_proved_the special case $K\!=\!{\mathbb{C}}$ of_the above lemma, in slightly_different language,_in [@plaisted]. However, his proof extends verbatim to_the_more general family_of fields detailed above. A simple consequence of the resultant_is that vanishing at a $D{^{\text{\underline{th}}}}$_root of unity_is_algebraically_the same thing over_${\mathbb{C}}$ or ${\mathbb{Q}}_p$, provided $p$ lies_in the right arithmetic progression. \[lemma:uni\] Suppose_$D\!\in\!{\mathbb{N}}$, $f\!\in\!{\mathbb{Z}}[x]$, and $p$ is any prime_congruent to $1$ mod $D$. Then_$f$ vanishes at a complex_$D{^{\text{\underline{th}}}}$ root_of unity $\Longleftrightarrow f$ vanishes at_a $D{^{\text{\underline{th}}}}$ root_of unity_in ${\mathbb{Q}}_p$. Note that_$x^2+x+1$ vanishes at a $3{^{\text{\underline{rd}}}}$ root_of unity in_${\mathbb{C}}$, but has [**no**]{} roots at_all_in ${\mathbb{F}}_5$ or_${\mathbb{Q}}_5$._So_our congruence_assumption on $p$_is_necessary. [$\diamond$]{} [**Proof_of_Lemma \[lemma:uni\]:**]{} First note that by_our_assumption on $p$, ${\mathbb{Q}}_p$ has $D$ distinct_$D{^{\text{\underline{th}}}}$ roots of unity:_This_follows easily from Hensel’s_Lemma and ${\mathbb{F}}_p$ having $D$_distinct $D{^{\text{\underline{th}}}}$ roots of unity. Since_${\mathbb{

n |c_i|^p\right)^{1/p}\leq \left(\sum_{k=0}^\infty \lambda_k^p\right)^{1/p}\leq \left(\frac{1}{1-\delta^{p}}\right)^{1/p}<1+\e$$ for $\delta$ sufficiently small. This “greedy algorithm” especifies a unique decomposition for each $f$ in the unit sphere of $F$. However it does not guarantee any kind of homogeneity in these decompositions. To amend this, let $S_0$ be a maximal subset of the unit sphere of $F$ with the property that any two points of $S_0$ are linearly independent (of course, when the ground field is ${\mathbb{R}}$ this just means that $S_0$ does not contain “antipodal” points). Equivalently, $S_0$ is a subset of the sphere such that every nonzero $f\in F$ can be written in a unique way as $f=cx$, with $c\in\mathbb K$ and $x\in S_0$. Now we define $\Phi':F{\longrightarrow}Y$ as follows: if $f\in S_0$, then we put $$\Phi'(f)= \sum_{i=1}^n c_i\Phi(f_i)$$ where $f=\sum_{i=1}^n c_if_i$ is the decomposition provided by the algorithm. We extend the map to the whole of $F$ by homogeneity: that is, for arbitrary $f\in F$ we write $x=cf$, with $c\in \mathbb K$ and $f\in S_0$, in the only way that this can be done, and we set $\Phi'(f)=c\Phi'(f)$. It is clear that the resulting map is homogeneous. Let us check that $\Phi'$ works properly. Let $K$ denote the $p$-linearity constant of the starting map $\Phi$. For $f\in S_0$, one has $$\|\Phi(f)-\Phi'(f)\|= \left\|\Phi(f)- \sum_{i=1}^n c_i\Phi(f_i) \right\|\leq Q^{(p)} \left(\sum_{i=1}^n |c_i|^p\right)^{1/p}<

n |c_i|^p\right)^{1 / p}\leq \left(\sum_{k=0}^\infty \lambda_k^p\right)^{1 / p}\leq \left(\frac{1}{1-\delta^{p}}\right)^{1 / p}<1+\e$$ for $ \delta$ sufficiently small. This “ greedy algorithm ” especifies a unique decomposition for each $ f$ in the unit of measurement sector of $ F$. However it does not guarantee any kind of homogeneity in these decompositions. To amend this, get $ S_0 $ be a maximal subset of the unit sphere of $ F$ with the place that any two points of $ S_0 $ are linearly independent (of course, when the land field is $ { \mathbb{R}}$ this just intend that $ S_0 $ does not contain “ antipodal ” points). Equivalently, $ S_0 $ is a subset of the celestial sphere such that every nonzero $ f\in F$ can be write in a unique way as $ f = cx$, with $ c\in\mathbb K$ and $ x\in S_0$. Now we define $ \Phi':F{\longrightarrow}Y$ as follows: if $ f\in S_0 $, then we frame $ $ \Phi'(f)= \sum_{i=1}^n c_i\Phi(f_i)$$ where $ f=\sum_{i=1}^n c_if_i$ is the decomposition provided by the algorithm. We extend the function to the whole of $ F$ by homogeneity: that is, for arbitrary $ f\in F$ we write $ x = cf$, with $ c\in \mathbb K$ and $ f\in S_0 $, in the lone way that this can be make, and we set $ \Phi'(f)=c\Phi'(f)$. It is clear that the resulting map is homogeneous. Let us control that $ \Phi'$ works properly. Let $ K$ denote the $ p$-linearity constant of the starting function $ \Phi$. For $ f\in S_0 $, one hold $ $ \|\Phi(f)-\Phi'(f)\|= \left\|\Phi(f)- \sum_{i=1}^n c_i\Phi(f_i) \right\|\leq Q^{(p) } \left(\sum_{i=1}^n |c_i|^p\right)^{1 / p } <

n |c_l|^p\right)^{1/p}\leq \left(\sum_{k=0}^\infuy \lambda_k^p\right)^{1/k}\lwq \lefv(\frac{1}{1-\demta^{p}}\righg)^{1/p}<1+\e$$ for $\delta$ sufficiently slaol. Thiw “greedy algorithm” espdcifies a unique eeconposition for each $n$ in fme unnt sphere of $F$. Hpwever it goes not guarattde any kind of homogeneity in these dqcomposotlons. To amend tris, kqt $S_0$ be a maximal subset of the unit slhere oh $F$ with the prpperty that any two points of $D_0$ are linearly indfpendent (of couwwe, when the eround field is ${\mathbb{T}}$ this just means that $S_0$ does nog concain “antipoeao” plhnts). Equivaoentlj, $S_0$ is a subscn of tha spherr such that evcry nmnzwro $f\in F$ can be writven in a unique way ws $f=cx$, widh $c\in\mathbb K$ and $z\ib S_0$. Nof we defkbe $\Ohi':R{\lpnfrightwrrkw}Y$ as folmows: if $f\in S_0$, then we put $$\Phi'(f)= \stn_{i=1}^n c_i\Phi(f_i)$$ wgere $f=\fui_{i=1}^n c_if_i$ is the decomposition provided ty fhe algorithm. We extend the map to the whole of $F$ by romogeneity: that is, for arbitrary $f\in F$ we write $f=cf$, wmtf $c\nk \oqtjbb K$ and $f\in S_0$, in the only way that this can ge dpne, and we set $\Phi'(f)=c\Phi'(f)$. Iy ls slear that thg resulcjnf map is homogeneoks. Let os chexk that $\Pri'$ wprks properly. Let $K$ denote rhe $p$-lineariny cinstant of the staxting map $\Phn$. For $g\in S_0$, one has $$\|\Phi(f)-\Phi'(f)\|= \left\|\Pki(f)- \shm_{i=1}^n c_i\Phi(f_l) \right\|\les Q^{(p)} \left(\sum_{i=1}^n |c_k|^p\rpght)^{1/[}<

n |c_i|^p\right)^{1/p}\leq \left(\sum_{k=0}^\infty \lambda_k^p\right)^{1/p}\leq \left(\frac{1}{1-\delta^{p}}\right)^{1/p}<1+\e$$ for $\delta$ This algorithm” especifies unique decomposition for sphere $F$. However it not guarantee any of homogeneity in these decompositions. To this, let $S_0$ be a maximal subset of the unit sphere of $F$ the property that any two points of $S_0$ are linearly independent (of course, the field ${\mathbb{R}}$ just means that $S_0$ does not contain “antipodal” points). Equivalently, $S_0$ is a subset of the such that every nonzero $f\in F$ can be in a unique way $f=cx$, with $c\in\mathbb K$ and S_0$. we define as if S_0$, then we $$\Phi'(f)= \sum_{i=1}^n c_i\Phi(f_i)$$ where $f=\sum_{i=1}^n c_if_i$ is the decomposition provided by the algorithm. We extend the map the whole by homogeneity: is, arbitrary F$ we write $c\in \mathbb K$ and $f\in S_0$, way that this can be done, and we $\Phi'(f)=c\Phi'(f)$. It clear that the resulting map is Let us check that $\Phi'$ works properly. Let denote the $p$-linearity constant of the starting map $\Phi$. For $f\in S_0$, one has $$\|\Phi(f)-\Phi'(f)\|= c_i\Phi(f_i) \right\|\leq Q^{(p)} \left(\sum_{i=1}^n

n |c_i|^p\right)^{1/p}\leq \left(\sum_{k=0}^\inftY \lambda_k^p\rIght)^{1/p}\Leq \LefT(\fRac{1}{1-\dElta^{P}}\right)^{1/p}<1+\e$$ for $\delTA$ sufFiciently small. This “greeDy algOrIThm” eSPeCifieS a uniquE DeCOMpoSiTiOn fOr EAcH $f$ in tHe uNit spheRe of $F$. HowevEr iT dOes not guaranTEe Any kind of hOmoGeneity in theSe dEcompoSiTioNS. To amEnd This, lEt $S_0$ be a MAximal Subset of tHe UNit sphERe of $F$ wiTH ThE proPerty that any two poINtS Of $S_0$ are linearly IndepeNdENt (OF CouRse, When the groUnD fielD Is ${\mathbB{r}}$ tHIS JusT Means that $S_0$ doeS not contain “ANtiPodal” pOiNts). eQuivalEntly, $s_0$ iS A suBset of the spHere Such that eVery noNZero $f\in f$ Can be wrItten iN a uNiqUe waY As $F=cX$, wiTh $C\In\mAThBb K$ ANd $x\In S_0$. Now we DeFiNe $\Phi':f{\lonGRIGHtarRow}y$ as fOllowS: if $f\in S_0$, then we Put $$\phi'(f)= \SUm_{i=1}^N c_i\PhI(f_i)$$ whEre $f=\SuM_{i=1}^n c_iF_i$ is thE decoMpOsition provided By thE algorithM. We ExTenD tHe map TO the whOle Of $F$ By homogEneity: tHAt iS, fOR ARbItrary $f\in F$ we write $x=Cf$, WITh $C\in \mathbB K$ and $f\IN S_0$, In THe only waY tHat This CAN be doNe, anD We Set $\Phi'(f)=c\phi'(f)$. It IS cLeAr that tHe ResultInG maP is HomogENeouS. Let us Check thaT $\Phi'$ wORks properly. Let $k$ Denote the $p$-linEArITY cONstaNt oF the startinG map $\pHi$. FoR $f\in s_0$, OnE haS $$\|\phi(f)-\PHi'(f)\|= \leFt\|\pHi(F)- \Sum_{i=1}^n c_i\Phi(f_i) \right\|\leQ Q^{(P)} \left(\sUm_{i=1}^n |c_I|^p\right)^{1/p}<

n |c_i|^p\right)^{1/p}\leq \left(\s um_{k =0} ^\i nf ty \lam bda_k^p\right) ^ {1/p }\leq \left(\frac{1}{1 -\del ta ^ {p}} \ ri ght)^ {1/p}<1 + \e $ $ fo r$\ del ta $ s uffic ien tly sma ll. This“gr ee dy algorithm ” e specifiesa u nique decomp osi tion f or ea c h $f$ in theunit s p here o f $F$. Ho we v er itd oes not g ua rant ee any kind of ho m og e neity in these decom po s it i o ns. T o amend th is , let $S_0$ b e a m a xim a l subset of t he unit sph e reof $F$ w ith the pr opert yt hat any two po ints of $S_0$ are l i nearlyi ndepend ent (o f c our se,w he nthe g r oun d f iel d is ${\math bb {R }}$ t hisj u s t mea nsthat $S_0 $ does not co nta in “ a nti podal ” poi nts) .Equiv alentl y, $S _0 $ is a subset o f th e spheresuc htha tevery nonzer o $ f\i n F$ ca n be wr i tte ni n aunique way as $f=c x$ , wi th $c\in \mathb b K $a nd $x\in S _0$ . N o w we d efin e $ \Phi':F{ \longr i gh ta rrow}Y$ a s foll ow s:if$f\in S_0$ , then we put$$\Ph i '(f)= \sum_{i= 1 }^n c_i\Phi(f _ i) $ $ w h ere$f= \sum_{i=1}^ n c_ i f_i$ ist he de c ompos ition p r ov i ded by the algorith m. We ex tendthe map to th e whole of $ F $ by hom ogen e it y : that is, for arbi trary $f\i n F$ we w rite$x=cf$,with $c\i n \mathbbK$and $f \in S _0 $, in the onl y wayth at this ca n be do ne, an d w e s et $\Phi'(f )=c\Phi' (f )$ . I t i s cle a r that t he re su lti ng ma p is ho mogen eous .Le t us checkt ha t $\Ph i' $work s p ro perly . Le t $K $ denot e the $p$ -li n eari ty c onstant of the start in g map $\Ph i$ . F or $f\ i n S_0$, o ne has $$\|\Phi(f)-\Phi ' (f)\|=\le ft\|\ Phi( f)- \sum _{i =1}^nc_i \ Phi(f_ i) \ri ght\| \l eqQ ^ {(p)} \l eft (\ sum_{i=1}^ n |c_ i|^p\ ri ght) ^{1/p}<

n |c_i|^p\right)^{1/p}\leq_ \left(\sum_{k=0}^\infty_ \lambda_k^p\right)^{1/p}\leq \left(\frac{1}{1-\delta^{p}}\right)^{1/p}<1+\e$$ for_$\delta$ sufficiently_small. This_“greedy algorithm”_especifies_a unique decomposition_for each $f$_in the unit sphere_of $F$. However_it_does not guarantee any kind of homogeneity in these decompositions. To amend this, let $S_0$_be_a maximal_subset_of_the unit sphere of $F$_with the property that any_two points_of $S_0$ are linearly independent (of course, when_the_ground field is_${\mathbb{R}}$ this just means that $S_0$ does not contain_“antipodal” points). Equivalently, $S_0$ is a_subset of the_sphere_such_that every nonzero $f\in_F$ can be written in a_unique way as $f=cx$, with $c\in\mathbb_K$ and $x\in S_0$. Now we define $\Phi':F{\longrightarrow}Y$_as follows: if $f\in S_0$, then_we put $$\Phi'(f)= \sum_{i=1}^n c_i\Phi(f_i)$$_where $f=\sum_{i=1}^n_c_if_i$ is the decomposition provided_by the algorithm._We extend_the map to_the whole of $F$ by homogeneity:_that is, for_arbitrary $f\in F$ we write $x=cf$,_with_$c\in \mathbb K$_and_$f\in_S_0$, in_the only way_that_this can_be_done, and we set $\Phi'(f)=c\Phi'(f)$. It is_clear_that the resulting map is homogeneous. Let_us check that $\Phi'$_works_properly. Let $K$ denote_the $p$-linearity constant of the_starting map $\Phi$. For $f\in S_0$,_one has_$$\|\Phi(f)-\Phi'(f)\|= \left\|\Phi(f)-_ \sum_{i=1}^n c_i\Phi(f_i) \right\|\leq Q^{(p)} \left(\sum_{i=1}^n |c_i|^p\right)^{1/p}<

between vertices of the first type. As a consequence, every vertex of the first type admits [*at most*]{} one incoming edge, [*i.e.*]{} $\Gamma_A$ can be only of the second type in Figure 15; any of the vertices of $\Gamma_A$ may further have [*at most*]{} one outgoing edge. We briefly discuss the coloring of an admissible graph. We choose a system of coordinates on $\mathfrak g$ which is adapted to the coisotropic submanifolds $\mathfrak k^\perp$, $f+\mathfrak b_1$ and $f+\mathfrak b_2$: in other words, we assume there is a partition of $\left\{1,\dots,d\right\}$, where $d$ is the dimension of $\mathfrak g$ of the form $$\begin{aligned} \{1,\dots,d\}=&(I_1\cap I_2\cap I_3)\sqcup(I_1^c\cap I_2\cap I_3)\sqcup(I_1\cap I_2^c\cap I_3)\sqcup(I_1\cap I_2\cap I_3^c)\sqcup\\ &(I_1^c\cap I_2^c\cap I_3)\sqcup(I_1^c\cap I_2\cap I_3^c)\sqcup(I_1\cap I_2^c\cap I_3^c)\sqcup(I_1^c\cap I_2^c\cap I_3^c), \end{aligned}$$ labeling a basis of $\mathfrak g$. This means that [*e.g.*]{} the elements of the basis indexed by $i$ in $I_1\cap I_2\cap I_3$ constitute a coordinate system for the intersection of $\mathfrak k^\perp$, $f+\mathfrak b_1$ and $f+\mathfrak b_2$, the elements indexed by $i$ in $I_1^c\cap I_2\cap I_3$ a coordinate system for the intersection of $\mathfrak k^*$ with $f+\mathfrak b_1$ and $f+\mathfrak b_2$, [*et similiter*]{}. The choice of labeling the coordinates on $\mathfrak g^*$ with respect to the above partition yields an obvious coloring of an admissible graph $\Gamma$ in $\mathcal G_{

between vertices of the first type. As a consequence, every vertex of the first character admit [ * at most * ] { } one incoming edge, [ * i.e. * ] { } $ \Gamma_A$ can be only of the second character in Figure 15; any of the vertex of $ \Gamma_A$ may further have [ * at most * ] { } one extroverted boundary. We briefly discuss the coloring of an admissible graph. We choose a system of coordinates on $ \mathfrak g$ which is adjust to the coisotropic submanifolds $ \mathfrak k^\perp$, $ f+\mathfrak b_1 $ and $ f+\mathfrak b_2 $: in other words, we simulate there constitute a partition of $ \left\{1,\dots, d\right\}$, where $ d$ is the dimension of $ \mathfrak g$ of the form $ $ \begin{aligned } \{1,\dots, d\}=&(I_1\cap I_2\cap I_3)\sqcup(I_1^c\cap I_2\cap I_3)\sqcup(I_1\cap I_2^c\cap I_3)\sqcup(I_1\cap I_2\cap I_3^c)\sqcup\\ & (I_1^c\cap I_2^c\cap I_3)\sqcup(I_1^c\cap I_2\cap I_3^c)\sqcup(I_1\cap I_2^c\cap I_3^c)\sqcup(I_1^c\cap I_2^c\cap I_3^c), \end{aligned}$$ labeling a basis of $ \mathfrak g$. This mean that [ * e.g. * ] { } the elements of the basis index by $ i$ in $ I_1\cap I_2\cap I_3 $ constitute a coordinate system for the intersection of $ \mathfrak k^\perp$, $ f+\mathfrak b_1 $ and $ f+\mathfrak b_2 $, the component indexed by $ i$ in $ I_1^c\cap I_2\cap I_3 $ a coordinate system for the intersection of $ \mathfrak k^*$ with $ f+\mathfrak b_1 $ and $ f+\mathfrak b_2 $, [ * et similiter * ] { }. The choice of labeling the coordinate on $ \mathfrak g^*$ with respect to the above partition yields an obvious coloring of an admissible graph $ \Gamma$ in $ \mathcal G _ {

behween vertices of the fivst type. As a couwequenre, everg vertex of the first type admits [*at mist*]{} obe incoming edge, [*i.e.*]{} $\Gaoma_A$ can he only if tie second type ii Figure 15; any kn the tertices of $\Gamka_A$ may fusther have [*at kort*]{} one outgoing edge. We briefly discusf the cpllring of an adiisspbje gdaph. We choose a system of coordinztes on $\mathfrak g$ wnich is adapted to the coidotrlpic submanifolds $\lathfrak k^\pgdp$, $s+\nathfrak b_1$ avd $f+\mathfrak b_2$: in othet words, we assume there is a pargitiou of $\left\{1,\dojs,s\rlcht\}$, where $d$ is tre dimension of $\mathxrak g$ pf the form $$\benin{almgnee} \{1,\dots,d\}=&(I_1\cap I_2\cap I_3)\sqcu'(I_1^c\cap I_2\cap I_3)\sqcup(I_1\cwp I_2^c\cap H_3)\sscup(I_1\cap I_2\cap I_3^c)\swcyp\\ &(I_1^c\cdp I_2^w\cap U_3)\sqzup(J_1^c\rap I_2\cap L_3^c)\sscup(I_1\cap I_2^d\cap I_3^c)\sqcup(I_1^c\cap I_2^c\cap I_3^c), \end{akidbed}$$ labeling z basif jf $\mathfrak g$. This means that [*e.g.*]{} the eltmenta of the basis indexed vy $i$ in $I_1\cap I_2\cap I_3$ clnstitute a coordinate system for the intersection of $\mathxrak i^\oerk$, $n+\ixrhvrak b_1$ and $f+\mathfrak b_2$, the elements indexed br $i$ ik $I_1^c\cap I_2\cap I_3$ a coordinate sjsygm for the intgrsection kf $\mathfrak k^*$ with $f+\mathftak b_1$ qnd $f+\mathsrak b_2$, [*et similiter*]{}. The choice od labeling tke xoordinates on $\matkfrak g^*$ with reskect tp the above partition ynelds zn obvious foloring kw an admissible eraih $\Gdmma$ in $\mathcal G_{

between vertices of the first type. As every of the type admits [*at $\Gamma_A$ be only of second type in 15; any of the vertices of may further have [*at most*]{} one outgoing edge. We briefly discuss the coloring an admissible graph. We choose a system of coordinates on $\mathfrak g$ which adapted the submanifolds k^\perp$, $f+\mathfrak b_1$ and $f+\mathfrak b_2$: in other words, we assume there is a partition of where $d$ is the dimension of $\mathfrak g$ the form $$\begin{aligned} \{1,\dots,d\}=&(I_1\cap I_3)\sqcup(I_1^c\cap I_2\cap I_3)\sqcup(I_1\cap I_2^c\cap I_3)\sqcup(I_1\cap I_3^c)\sqcup\\ I_2^c\cap I_3)\sqcup(I_1^c\cap I_3^c)\sqcup(I_1\cap I_3^c)\sqcup(I_1^c\cap I_3^c), \end{aligned}$$ labeling basis of $\mathfrak g$. This means that [*e.g.*]{} the elements of the basis indexed by $i$ in I_2\cap I_3$ coordinate system the of k^\perp$, $f+\mathfrak b_1$ b_2$, the elements indexed by $i$ I_3$ a coordinate system for the intersection of k^*$ with b_1$ and $f+\mathfrak b_2$, [*et similiter*]{}. choice of labeling the coordinates on $\mathfrak g^*$ respect to the above partition yields an obvious coloring of an admissible graph $\Gamma$ in

between vertices of the first Type. As a conSequeNce, EveRy VertEx of The first type adMIts [*aT most*]{} one incoming edge, [*i.e.*]{} $\gamma_a$ cAN be oNLy Of the Second tYPe IN figUrE 15; aNy oF tHE vErticEs oF $\Gamma_A$ May further HavE [*aT most*]{} one outgOInG edge. We briEflY discuss the cOloRing of An AdmISsiblE grAph. We Choose A System Of coordinAtES on $\matHFrak g$ whICH iS adaPted to the coisotroPIc SUbmanifolds $\matHfrak k^\PeRP$, $f+\MAThfRak B_1$ and $f+\mathfRaK b_2$: in oTHer wordS, We ASSUme THere is a partitIon of $\left\{1,\doTS,d\rIght\}$, whErE $d$ iS The dimEnsioN oF $\MatHfrak g$ of the Form $$\Begin{aligNed} \{1,\dotS,D\}=&(I_1\cap I_2\cAP I_3)\sqcup(i_1^c\cap I_2\Cap i_3)\sqCup(I_1\CAp i_2^c\Cap i_3)\sQCup(i_1\CaP I_2\cAP I_3^c)\Sqcup\\ &(I_1^c\cAp i_2^c\Cap I_3)\sQcup(i_1^C\CAP I_2\caP I_3^c)\SqcuP(I_1\cap i_2^c\cap I_3^c)\sqcup(I_1^C\caP I_2^c\cAP I_3^c), \End{alIgned}$$ LabeLiNg a baSis of $\mAthfrAk G$. This means that [*e.G.*]{} the Elements oF thE bAsiS iNdexeD By $i$ in $I_1\Cap i_2\caP I_3$ constItute a cOOrdInATE SyStem for the intersecTiON Of $\Mathfrak K^\perp$, $f+\MAtHfRAk b_1$ and $f+\mAtHfrAk b_2$, tHE ElemeNts iNDeXed by $i$ in $i_1^c\cap I_2\CAp i_3$ a CoordinAtE systeM fOr tHe iNtersECtioN of $\matHfrak k^*$ wiTh $f+\maTHfrak b_1$ and $f+\mathFRak b_2$, [*et similitER*]{}. THE ChOIce oF laBeling the coOrdiNAtes On $\maTHfRak G^*$ With rEspecT tO ThE Above partition yieldS aN obvioUs colOring of an admiSsible grapH $\gAMma$ in $\matHcal g_{

between vertices of the f irst type. As a co nse qu ence , ev ery vertex oft he f irst type admits [*atmost* ]{ } one in comin g edge, [* i . e.* ]{ }$\G am m a_ A$ ca n b e onlyof the sec ond t ype in Figur e 1 5; any ofthe vertices of $\ Gamma_ A$ ma y furt her have [*atm ost*]{ } one out go i ng edg e . We b r i ef ly d iscuss the colori n go f an admissibl e grap h. We c hoo sea system o fcoord i nates o n $ \ m a thf r ak g$ which i s adapted t o th e cois ot rop i c subm anifo ld s $\ mathfrak k^ \per p$, $f+\m athfra k b_1$ a n d $f+\m athfra k b _2$ : in ot he r w or d s,w eass u methere is a p artit iono f $ \lef t\{ 1,\d ots,d \right\}$, wh ere $d$ isthe d imens ionof $\ma thfrak g$ o fthe form $$\beg in{a ligned} \ {1, \d ots ,d \}=&( I _1\cap I_ 2\c ap I_3) \sqcup( I _1^ c\ c a p I _2\cap I_3)\sqcup( I_ 1 \ ca p I_2^c\ cap I_ 3 )\ sq c up(I_1\c ap I_ 2\ca p I_3^c )\sq c up \\ &(I_1 ^c\cap I_ 2^ c\cap I _3 )\sqcu p( I_1 ^c\ cap I _ 2\ca p I_3^ c)\sqcup (I_1\ c ap I_2^c\cap I _ 3^c)\sqcup(I_ 1 ^c \ c ap I_2^ c\c ap I_3^c),\end { alig ned} $ $lab e linga bas is of $\mathfrak g$. This m eans t hat [ *e.g.*]{} the elementso f the basi s in d ex e d by $i$ in $I _1\ca p I_2\capI _3$ cons titut e a coor dinate sy s t em for t heint ers ect i o nof $\mathfrak k ^\pe rp $, $f+\ mat hfrak b _1$ an d $ f+\ ma thfrak b_ 2$, theel em en ts in dexed by $i$ i n$I_ 1^ c\c ap I_ 2 \cap I _3$ a coo rd in a tesystemf or t he i nt er sect ion o f $\m athf r akk^*$ wi th $f+\ma thf r ak b _1 $and $f+ \mathfrak b_2 $, [*et simi li ter *]{}.T he choic e of labeling the coord i nates o n $ \math frak g^*$ wit h r espect to the ab ove pa rtiti on yi e l ds an o bv iou scoloring o f anadmis si blegraph $ \Gamma$ in $\mathc a l G _{

between_vertices of_the first type. As_a consequence,_every_vertex of_the_first type admits_[*at most*]{} one_incoming edge, [*i.e.*]{} $\Gamma_A$_can be only_of_the second type in Figure 15; any of the vertices of $\Gamma_A$ may further_have_[*at most*]{}_one_outgoing_edge. We briefly discuss the coloring_of an admissible graph. We_choose a_system of coordinates on $\mathfrak g$ which is_adapted_to the coisotropic_submanifolds $\mathfrak k^\perp$, $f+\mathfrak b_1$ and $f+\mathfrak b_2$: in_other words, we assume there is_a partition of_$\left\{1,\dots,d\right\}$,_where_$d$ is the dimension_of $\mathfrak g$ of the form_$$\begin{aligned} \{1,\dots,d\}=&(I_1\cap I_2\cap I_3)\sqcup(I_1^c\cap I_2\cap I_3)\sqcup(I_1\cap I_2^c\cap_I_3)\sqcup(I_1\cap I_2\cap I_3^c)\sqcup\\ &(I_1^c\cap I_2^c\cap I_3)\sqcup(I_1^c\cap I_2\cap I_3^c)\sqcup(I_1\cap_I_2^c\cap I_3^c)\sqcup(I_1^c\cap I_2^c\cap I_3^c), \end{aligned}$$ labeling a_basis of $\mathfrak g$. This_means that_[*e.g.*]{} the elements of the_basis indexed by_$i$ in_$I_1\cap I_2\cap I_3$_constitute a coordinate system for the_intersection of $\mathfrak_k^\perp$, $f+\mathfrak b_1$ and $f+\mathfrak b_2$,_the_elements indexed by_$i$_in_$I_1^c\cap I_2\cap_I_3$ a coordinate_system_for the_intersection_of $\mathfrak k^*$ with $f+\mathfrak b_1$_and_$f+\mathfrak b_2$, [*et similiter*]{}. The choice of labeling_the coordinates on $\mathfrak_g^*$_with respect to the_above partition yields an obvious_coloring of an admissible graph $\Gamma$_in $\mathcal_G_{

Similarly, *meet-irreducible elements* cannot be written as an infimum of other elements, and are such that they are covered by a single element. We denote by $\mathcal{M}(L)$ the set of meet-irreducible elements of $L$. *Co-atoms* are meet-irreducible elements covered by $\top$. For any $x\in L$, we say that $x$ *has a complement in $L$* if there exists $x'\in L$ such that $x\wedge x'=\bot$ and $x\vee x'=\top$. The complement is unique if the lattice is distributive. $L$ is said to be *complemented* if any element has a complement. On Fig. \[fig:lat1\] (left), no element has a complement, except top and bottom, while the two others are complemented lattices. *Boolean lattices* are distributive and complemented lattices, and in a finite setting, they are of the type $2^N$ for some set $N$, i.e. they are isomorphic to the lattice of subsets of some set, ordered by inclusion (see Fig. \[fig:lat1\] (middle,right)). Boolean lattices are atomistic, and atoms correspond to singletons, while co-atoms are of the form $N\setminus\{i\}$ for some $i\in N$. An important property is that in a lower locally distributive lattice, any element $x$ can be written as an irredundant supremum of join-irreducible elements in a unique way (this is called the *minimal decomposition* of $x$). We denote by $\eta^*(x)$ the set of join-irreducible elements in the minimal decomposition of $x$, and we denote by $\eta(x)$ the *normal decomposition* of $x$, defined as the set of join-irreducible elements smaller or equal to $x$, i.e., $\eta(x):=\{j\in \mathcal{J}(L)\mid j\leq x\}$. Hence $\eta^*(x)\subseteq \eta(x)$, and $$x=\bigvee_{j\in \eta^*(x)} j = \bigvee_{j\in \eta(x)} j.$$ Put differently, the mapping $\eta$ is an isomorphism of $L$ onto $\mathcal{O}(\mathcal{J}(L))$ (Birkhoff’s theorem). Like

Similarly, * meet - irreducible elements * cannot be written as an infimum of other elements, and are such that they are cover by a individual element. We denote by $ \mathcal{M}(L)$ the set of meet - irreducible component of $ L$. * Co - atoms * are meet - irreducible element cover by $ \top$. For any $ x\in L$, we say that $ x$ * have a complement in $ L$ * if there exists $ x'\in L$ such that $ x\wedge x'=\bot$ and $ x\vee x'=\top$. The complement is unique if the wicket is distributive. $ L$ is said to be * complemented * if any component experience a complement. On Fig. \[fig: lat1\ ] (left), no element has a complement, except peak and bottom, while the two others are complemented lattices. * Boolean lattice * are distributive and complemented lattices, and in a finite setting, they are of the type $ 2^N$ for some fit $ N$, i.e. they are isomorphic to the lattice of subsets of some set, ordered by inclusion body (see Fig. \[fig: lat1\ ] (middle, right) ). Boolean lattices are atomistic, and atoms correspond to singleton, while co - atoms are of the form $ N\setminus\{i\}$ for some $ i\in N$. An important property is that in a lower locally distributive lattice, any element $ x$ can be written as an irredundant supremum of join - irreducible component in a unique way (this is call the * minimal decay * of $ x$). We announce by $ \eta^*(x)$ the set of join - irreducible elements in the minimal decomposition of $ x$, and we denote by $ \eta(x)$ the * normal decomposition * of $ x$, defined as the set of articulation - irreducible elements smaller or equal to $ x$, i.e., $ \eta(x):=\{j\in \mathcal{J}(L)\mid j\leq x\}$. Hence $ \eta^*(x)\subseteq \eta(x)$, and $ $ x=\bigvee_{j\in \eta^*(x) } j = \bigvee_{j\in \eta(x) } j.$$ Put differently, the function $ \eta$ is an isomorphism of $ L$ onto $ \mathcal{O}(\mathcal{J}(L))$ (Birkhoff ’s theorem). Like

Simllarly, *meet-irreducible euements* cannot yw writven as zn infimjm of other elements, and are sych tyat they are covered bh a singlv element. We venote by $\mathcal{M}(L)$ the set of leet-nrceducible elemekts of $L$. *Co-dtoms* are meet-hrfebucible elements covered by $\top$. For agy $x\in K$, ae say that $x$ *ras s comllement in $L$* if there exists $x'\in L$ such tiat $x\wedge x'=\bot$ and $x\vee x'=\top$. The complemfnt ls unique if the lwttice is dustryvutive. $L$ is raid to be *complementes* if any element has a complemevt. On Fig. \[fig:laj1\] (mevj), no element has w complement, except dop and bottom, while bhe txo orhers are complementev lattices. *Boolean lajtices* are dnstributive and complwmwnted latdicer, ana ih e fjnite detving, they ade of the ttpe $2^N$ for some set $M$, y.v. they are iskmorphyc to the lattice of subsets of some set, mrdsred by inclusion (see Fug. \[fig:lat1\] (middle,right)). Boolean jattices are atomistic, and atoms correspond to sitgletkvs, cmlle zi-ahoms are of the form $N\setminus\{i\}$ for some $i\in G$. Zn ikportant propevty is that in a lpwfr jocally distrkbutivz lzttice, any element $x$ can fe wrutten as wn itredundant supremum of join-urreducible vlemwnts in a unique wcy (this is cclled jhe *mimimal decomposition* of $r$). We dsnote by $\etw^*(x)$ the sef of join-irreducicle enements in the minimal decjmpositioi of $r$, and we denpte by $\eta(x)$ the *jormal decomposition* of $d$, defnned ds the set of join-irreducible elements smekler or equak do $q$, i.e., $\eta(x):=\{l\in \mabhcal{J}(L)\mid j\leq x\}$. Hence $\eta^*(x)\sobseteq \eca(x)$, ana $$x=\bigvee_{j\pn \eta^*(x)} j = \bigvee_{j\in \eta(x)} j.$$ Put dhvferently, thx mapping $\eta$ is qn isomufphism of $L$ onyo $\mathcal{O}(\mathcal{J}(O))$ (Birkhoff’s theorek). Liys

Similarly, *meet-irreducible elements* cannot be written as of elements, and such that they element. denote by $\mathcal{M}(L)$ set of meet-irreducible of $L$. *Co-atoms* are meet-irreducible elements by $\top$. For any $x\in L$, we say that $x$ *has a complement $L$* if there exists $x'\in L$ such that $x\wedge x'=\bot$ and $x\vee x'=\top$. complement unique the is distributive. $L$ is said to be *complemented* if any element has a complement. On Fig. (left), no element has a complement, except top bottom, while the two are complemented lattices. *Boolean lattices* distributive complemented lattices, in finite they are of type $2^N$ for some set $N$, i.e. they are isomorphic to the lattice of subsets of some ordered by Fig. \[fig:lat1\] Boolean are and atoms correspond while co-atoms are of the form $i\in N$. An important property is that in lower locally lattice, any element $x$ can be as an irredundant supremum of join-irreducible elements in unique way (this is called the *minimal decomposition* of $x$). We denote by $\eta^*(x)$ the join-irreducible elements in the decomposition of $x$, we by the decomposition* of defined as the set of join-irreducible elements smaller or equal to i.e., $\eta(x):=\{j\in \mathcal{J}(L)\mid j\leq x\}$. Hence $\eta^*(x)\subseteq \eta(x)$, and $$x=\bigvee_{j\in = \eta(x)} j.$$ Put the mapping $\eta$ is isomorphism $L$ onto $\mathcal{O}(\mathcal{J}(L))$ (Birkhoff’s

Similarly, *meet-irreducible eLements* canNot be WriTteN aS an iNfimUm of other elemeNTs, anD are such that they are covEred bY a SInglE ElEment. we denotE By $\MAThcAl{m}(L)$ The SeT Of Meet-iRreDucible Elements of $l$. *Co-AtOms* are meet-irREdUcible elemEntS covered by $\toP$. FoR any $x\iN L$, We sAY that $X$ *haS a comPlemenT In $L$* if tHere existS $x'\IN L$ such THat $x\wedGE X'=\bOt$ anD $x\vee x'=\top$. The complEMeNT is unique if the LatticE iS DiSTRibUtiVe. $L$ is said tO bE *compLEmented* IF aNY ELemENt has a complemEnt. On Fig. \[fig:LAt1\] (lEft), no eLeMenT Has a coMplemEnT, ExcEpt top and boTtom, While the tWo otheRS are comPLementeD lattiCes. *booLean LAtTiCes* ArE DisTRiButIVe aNd compleMeNtEd latTiceS, AND In a fIniTe seTting, They are of the tYpe $2^n$ for SOme Set $N$, i.E. they Are iSoMorphIc to thE lattIcE of subsets of somE set, Ordered by IncLuSioN (sEe Fig. \[FIg:lat1\] (mIddLe,rIght)). BooLean latTIceS aRE AToMistic, and atoms corrEsPONd To singleTons, whILe Co-AToms are oF tHe fOrm $N\SETminuS\{i\}$ foR SoMe $i\in N$. An ImportANt PrOperty iS tHat in a LoWer LocAlly dIStriButive Lattice, aNy eleMEnt $x$ can be writtEN as an irredundANt SUPrEMum oF joIn-irreducibLe elEMentS in a UNiQue WAy (thiS is caLlED tHE *minimal decompositiOn* Of $x$). We dEnote By $\eta^*(x)$ the set oF join-irredUCIBle elemeNts iN ThE Minimal decompoSitioN of $x$, and we dENote by $\etA(x)$ the *Normal deCompositiON* Of $x$, definEd aS thE seT of JOIn-Irreducible elEMEnts SmAller or EquAl to $x$, i.e., $\Eta(X):=\{j\iN \maThcAl{j}(L)\mid j\leq X\}$. Hence $\etA^*(x)\SuBsEtEq \eTa(x)$, anD $$X=\bigvee_{j\In \Eta^*(X)} j = \BigVee_{j\iN \Eta(x)} j.$$ PUt difFereNtLy, THe mApping $\eTA$ iS AN isoMoRpHism Of $L$ OnTo $\matHcal{o}(\MatHcal{J}(L))$ (BIrkhoff’s tHeoREm). LiKe

Similarly, *meet-irreducib le element s* ca nno t b ewrit tenas an infimumo f ot her elements, and aresuchth a t th e yare c overedb ya sin gl eele me n t. We d eno te by $ \mathcal{M }(L )$ the set ofm ee t-irreduci ble elements of $L $. *Co -a tom s * are me et-ir reduci b le ele ments cov er e d by $ \ top$. F o rany$x\in L$, we sayt ha t $x$ *has a co mpleme nt in $ L$* if there exi st s $x' \ in L$ s u ch t h at$ x\wedge x'=\b ot$ and $x\ v eex'=\to p$ . T h e comp lemen ti s u nique if th e la ttice isdistri b utive.$ L$ is s aid to be *c ompl e me nt ed* i f an y e lem e nthas a co mp le ment. OnF i g . \[f ig: lat1 \] (l eft), no elem ent has a c omple ment, exc ep t top and b ottom ,while the two o ther s are com ple me nte dlatti c es. * Boo lea n latti ces* ar e di st r i b ut ive and complement ed l at tices, a nd ina f in i te setti ng , t heya r e ofthet yp e $2^N$for so m ese t $N$,i. e. the yare is omorp h ic t o thelatticeof su b sets of some s e t, ordered by in c l us i on ( see Fig. \[fig :lat 1 \] ( midd l e, rig h t)).Boole an la t tices are atomistic ,and at oms c orrespond tosingletons , w hile co- atom s a r e of the form$N\se tminus\{i\ } $ for so me $i \in N$. An impor t a nt prope rty is th ati n a lower locall y dist ri butivelat tice, a nyele men t $ x$ can be w ritten a san i rr edu ndant supremum o f j oi n-i rredu c ible e lemen ts i nau niq ue way( th i s isca ll ed t he*m inima l de c omp osition * of $x$) . W e den ot eby $\et a^*(x)$ the s et of join-i rr edu ciblee l ements i n the minimal decomposi t ion of$x$ , and wedenote by $\ eta(x) $ t h e *nor mal de compo si tio n * of $ x $ ,def in ed as thes e t o f joi n- irre ducible elements smallero r e qual to $x$,i.e ., $ \ e ta (x) : =\ { j\i n\ mat h c al{J}(L)\mid j\ leq x\}$.He n ce $\eta^*(x ) \su bs eteq \e ta(x)$, and$ $x=\big vee_{j\in \eta^*(x )} j = \ big vee_{j\in\eta(x)} j.$$ Put diffe r en tly,the mappi ng $\ eta$is ani som orphi sm of$L $ onto $\ma th cal{O}(\ mathcal{J}(L))$ (Birkho ff’s t heore m). Like

Similarly, *meet-irreducible_elements* cannot_be written as an_infimum of_other_elements, and_are_such that they_are covered by_a single element. We_denote by $\mathcal{M}(L)$_the_set of meet-irreducible elements of $L$. *Co-atoms* are meet-irreducible elements covered by $\top$. For any_$x\in_L$, we_say_that_$x$ *has a complement in_$L$* if there exists $x'\in_L$ such_that $x\wedge x'=\bot$ and $x\vee x'=\top$. The complement_is_unique if the_lattice is distributive. $L$ is said to be *complemented*_if any element has a complement._On Fig. \[fig:lat1\]_(left),_no_element has a complement,_except top and bottom, while the_two others are complemented lattices. *Boolean lattices*_are distributive and complemented lattices, and in_a finite setting, they are of_the type $2^N$ for some_set $N$,_i.e. they are isomorphic to_the lattice of_subsets of_some set, ordered_by inclusion (see Fig. \[fig:lat1\] (middle,right))._Boolean lattices are_atomistic, and atoms correspond to singletons,_while_co-atoms are of_the_form_$N\setminus\{i\}$ for_some $i\in N$. An_important_property is_that_in a lower locally distributive lattice,_any_element $x$ can be written as an_irredundant supremum of join-irreducible_elements_in a unique way_(this is called the *minimal_decomposition* of $x$). We denote by_$\eta^*(x)$ the_set of_join-irreducible elements in the minimal decomposition of $x$, and we denote_by $\eta(x)$ the *normal decomposition* of_$x$, defined as the_set of_join-irreducible_elements smaller or_equal_to $x$,_i.e., $\eta(x):=\{j\in \mathcal{J}(L)\mid j\leq x\}$. Hence $\eta^*(x)\subseteq_\eta(x)$, and_$$x=\bigvee_{j\in \eta^*(x)} j = \bigvee_{j\in \eta(x)}_j.$$ Put differently, the_mapping_$\eta$ is an isomorphism of $L$_onto $\mathcal{O}(\mathcal{J}(L))$ (Birkhoff’s theorem). Like

[Bilder/Observations/cont08.pdf]{} (3,89) [Bilder/Observations/dop08.pdf]{} (3,89) [Bilder/dopmean\_spot/dopmean2010\_07\_06.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmean2010\_07\_06\_mask.pdf]{} (3,71) [Bilder/dopmean\_moat/dopmean2010\_09\_25.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmean2010\_10\_17.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmeanAR11131\_1.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmean2011\_01\_02.pdf]{} (3,71) Data selection and preprocessing: {#sec:selection} --------------------------------- The moat flow is coupled to the presence of a penumbra and follows the direction of the penumbral filaments. Thus, we searched for sunspots with a fully-fledged, circular penumbra and found 31 applicable, single sunspots in a slight decay[^4] in the time between June 2010 and January 2012. The selected sunspots vary in size between 9Mm and 22Mm and are listed in Table \[apptab\] of the appendix. A contour of the outer penumbra for each spot is determined by means of an intensity threshold for a spatially averaged image. The center of a sunspot is determined as the center-of-gravity of all points inside its penumbral contour. To [[analyze]{}]{} predominantly horizontal flows with [[LOS Doppler]{}]{} maps, we selected the observing position of the sunspot center at heliocentric angles [[of some]{}]{} $\theta\!\approx\!50^\circ$. There, the LOS component of the horizontal flow is significant, while the calibration uncertainties that increase towards the solar limb are still small. Each sunspot FOV from the size of 301x301px is tracked by its averaged center, $(x_{\rm c},y_{\rm c})$, for 15 successive [[720s intensity]{}]{} maps [[(Fig. \[fig:dopcont\]a)]{}]{}, while the respective Doppler maps

[ Bilder / Observations / cont08.pdf ] { } (3,89) [ Bilder / Observations / dop08.pdf ] { } (3,89) [ Bilder / dopmean\_spot / dopmean2010\_07\_06.pdf ] { } (3,89) [ Bilder / dopmean\_moat / dopmean2010\_07\_06\_mask.pdf ] { } (3,71) [ Bilder / dopmean\_moat / dopmean2010\_09\_25.pdf ] { } (3,89) [ Bilder / dopmean\_moat / dopmean2010\_10\_17.pdf ] { } (3,89) [ Bilder / dopmean\_moat / dopmeanAR11131\_1.pdf ] { } (3,89) [ Bilder / dopmean\_moat / dopmean2011\_01\_02.pdf ] { } (3,71) Data selection and preprocessing: { # sec: selection } --------------------------------- The moat flow is coupled to the bearing of a penumbra and take after the direction of the penumbral filaments. therefore, we searched for sunspot with a fully - fledged, round penumbra and found 31 applicable, single sunspots in a little decay[^4 ] in the time between June 2010 and January 2012. The selected sunspot vary in size between 9Mm and 22Mm and are listed in Table   \[apptab\ ] of the appendix. A contour of the outer penumbra for each blot is determined by means of an intensity threshold for a spatially averaged persona. The center of a sunspot is determined as the center - of - graveness of all points inside its penumbral contour. To [ [ analyze ] { } ] { } predominantly horizontal stream with [ [ LOS Doppler ] { } ] { } maps, we choose the observing situation of the sunspot center at heliocentric angles [ [ of some ] { } ] { } $ \theta\!\approx\!50^\circ$. There, the LOS component of the horizontal menstruation is significant, while the calibration uncertainties that increase towards the solar limb are still small. Each sunspot FOV from the size of 301x301px is tracked by its averaged center, $ (x_{\rm c},y_{\rm c})$, for 15 successive [ [ 720s intensity ] { } ] { } maps [ [ (Fig.   \[fig: dopcont\]a) ] { } ] { }, while the respective Doppler maps

[Bilfer/Observations/cont08.pdf]{} (3,89) [Bllder/Observations/dop08.pdf]{} (3,89) [Bilded/dopmean\_rpot/dopmean2010\_07\_06.pdf]{} (3,89) [Bilder/dopmean\_loqt/dopnean2010\_07\_06\_mask.pdf]{} (3,71) [Bilder/dopmdan\_moat/doimean2010\_09\_25.pdf]{} (3,89) [Vildtr/dopmean\_moat/dopmxzn2010\_10\_17.pdf]{} (3,89) [Bljder/slpmecn\_noat/dopmeanAR11131\_1.pcf]{} (3,89) [Bilder/dmpmean\_moat/dopmaav2011\_01\_02.pbf]{} (3,71) Data selection and preprocessing: {#fec:selevtlon} --------------------------------- The moat fljw ix coulled to the presence of a penumbra and fonlows the dirrction of the penumbral fipamejts. Thus, we searchfd for sunskkts qith a fully-wledged, cigeular penumgra and found 31 applicable, singld sunxpots in a soigjj decay[^4] in tie timv between Junc 2010 and Banuary 2012. The selected suispors vary in size betwexn 9Mm and 22Mm and are listed it Cable \[apptab\] of the appebdix. D cottouf of tht onted penulbre for each apot is detwrmined by means of ag intensity thdesholq sor a spatially averaged image. The centtr of a sunspot is determinee as the center-of-gravlty of alj points inside its penumbral contour. To [[analyze]{}]{} psedommnxntoy firlzontal flows with [[LOS Doppler]{}]{} maps, we selectqs uhe observing posltion of the sunsppt crgter at heliozentrie ahgles [[of some]{}]{} $\thetw\!\approx\!50^\sirc$. Rhere, the LOS component of the horizontao flow is siynidicant, while the cclibration uucertaontiex that increase towards the solar limb are stilm small. Each sunsput NOV from tht size of 301x301px is twacked by its averagdd cgnter, $(x_{\wm c},y_{\rm c})$, vor 15 successive [[720s intenslty]{}]{} mcps [[(Fhg. \[fig:dopcojt\]a)]{}]{}, while the respective Dopplec maps

[Bilder/Observations/cont08.pdf]{} (3,89) [Bilder/Observations/dop08.pdf]{} (3,89) [Bilder/dopmean\_spot/dopmean2010\_07\_06.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmean2010\_07\_06\_mask.pdf]{} (3,89) (3,89) [Bilder/dopmean\_moat/dopmeanAR11131\_1.pdf]{} [Bilder/dopmean\_moat/dopmean2011\_01\_02.pdf]{} (3,71) Data The flow is coupled the presence of penumbra and follows the direction of penumbral filaments. Thus, we searched for sunspots with a fully-fledged, circular penumbra and 31 applicable, single sunspots in a slight decay[^4] in the time between June and 2012. selected vary in size between 9Mm and 22Mm and are listed in Table \[apptab\] of the appendix. contour of the outer penumbra for each spot determined by means of intensity threshold for a spatially image. center of sunspot determined the center-of-gravity of points inside its penumbral contour. To [[analyze]{}]{} predominantly horizontal flows with [[LOS Doppler]{}]{} maps, we selected the position of center at angles some]{}]{} There, the LOS the horizontal flow is significant, while that increase towards the solar limb are still Each sunspot from the size of 301x301px is by its averaged center, $(x_{\rm c},y_{\rm c})$, for successive [[720s intensity]{}]{} maps [[(Fig. \[fig:dopcont\]a)]{}]{}, while the respective Doppler maps

[Bilder/Observations/cont08.pdf]{} (3,89) [bilder/ObseRvatiOns/Dop08.PdF]{} (3,89) [BilDer/dOpmean\_spot/dopmEAn2010\_07\_06.pdF]{} (3,89) [Bilder/dopmean\_moat/dopmEan2010\_07\_06\_maSk.PDf]{} (3,71) [BiLDeR/dopmEan\_moat/DOpMEAn2010\_09\_25.pDf]{} (3,89) [biLdeR/dOPmEan\_moAt/dOpmean2010\_10\_17.pDf]{} (3,89) [Bilder/doPmeAn\_Moat/dopmeanAr11131\_1.PdF]{} (3,89) [Bilder/dopMeaN\_moat/dopmean2011\_01\_02.Pdf]{} (3,71) data seLeCtiON and pRepRocesSing: {#seC:SelectIon} --------------------------------- The moaT fLOw is coUPled to tHE PrEsenCe of a penumbra and fOLlOWs the direction Of the pEnUMbRAL fiLamEnts. Thus, we SeArcheD For sunsPOtS WITh a FUlly-fledged, ciRcular penumBRa aNd founD 31 aPplICable, sIngle SuNSpoTs in a slight DecaY[^4] in the timE betweEN June 2010 anD january 2012. the selEctEd sUnspOTs VaRy iN sIZe bETwEen 9mM anD 22Mm and arE lIsTed in tablE \[APPTab\] oF thE appEndix. a contour of the OutEr peNUmbRa for Each sPot iS dEtermIned by Means Of An intensity threSholD for a spatIalLy AveRaGed imAGe. The cEntEr oF a sunspOt is detERmiNeD AS ThE center-of-gravity of AlL POiNts insidE its peNUmBrAL contour. to [[AnaLyze]{}]{} PREdomiNantLY hOrizontaL flows WItH [[LoS DopplEr]{}]{} Maps, we SeLecTed The obSErviNg posiTion of thE sunsPOt center at heliOCentric angles [[OF sOME]{}]{} $\tHEta\!\aPprOx\!50^\circ$. There, The Los comPoneNT oF thE HorizOntal FlOW iS Significant, while the CaLibratIon unCertainties thAt increase TOWArds the sOlar LImB Are still small. EAch suNspot FOV frOM the size Of 301x301px Is trackeD by its aveRAGed centeR, $(x_{\rM c},y_{\Rm c})$, For 15 SUCcEssive [[720s intensITY]{}]{} mapS [[(FIg. \[fig:doPcoNt\]a)]{}]{}, whilE thE reSpeCtiVe doppler maPs

[Bilder/Observations/con t08.pdf]{} (3,8 9) [B il der/ Obse rvations/dop08 . pdf] {} (3,89) [Bilder/dop mean\ _s p ot/d o pm ean20 10\_07\ _ 06 . p df] {} ( 3,8 9) [ Bilde r/d opmean\ _moat/dopm ean 20 10\_07\_06\_ m as k.pdf]{} ( 3,7 1) [Bilder/ dop mean\_ mo at/ d opmea n20 10\_0 9\_25. p df]{}(3,89) [ Bi l der/do p mean\_m o a t/ dopm ean2010\_10\_17.p d f] { } (3,89) [Bil der/do pm e an \ _ moa t/d opmeanAR11 13 1\_1. p df]{} ( 3 ,8 9 ) [B i lder/dopmean\ _moat/dopme a n20 11\_01 \_ 02. p df]{}(3,71 )Dat a selection and preproce ssing: {#sec:s e lection } ---- --- --- ---- - -- -- --- -- - --- - -- --The moat fl ow i s cou pled t o thepre senc e ofa penumbra an d f ollo w s t he di recti on o fthe p enumbr al fi la ments. Thus, we sea rched for su ns pot switha fully -fl edg ed, cir cular p e num br a a nd found 31 applicab le , si ngle sun spotsi nas light de ca y[^ 4] i n the t imeb et ween Jun e 2010 an dJanuary 2 012. T he se lec ted s u nspo ts var y in siz e bet w een 9Mm and 22 M m and are lis t ed i nT able  \[ apptab\] of the appe ndix . A co n tourof th eo ut e r penumbra for each s pot is dete rmined by mea ns of an i n t e nsity th resh o ld for a spatiall y ave raged imag e . The ce nterof a sun spot is d e t erminedasthe ce nte r - of -gravity of a l l poi nt s insid e i ts penu mbr alcon tou r. To [[an alyze]{} ]{ }pr ed omi nantl y horizon ta l f lo wswith[ [LOS D opple r]{} ]{ }m aps , we se l ec t e d th eob serv ing p ositi on o f th e sunsp ot center at heli oc en tric an gles [[of som e] {}]{} $\th et a\! \appro x \ !50^\cir c$. There, the LOS comp o nent of th e hor izon tal flowissignif ica n t, whi le the cali br ati o n unce r t ai nti es that incr e a setowar ds the solarlimb are still sma l l. Each sunspot FO V fr o m t hes iz e of 3 0 1x3 0 1 px is tracked b y its aver ag e dcenter, $( x _{\ rm c},y_{ \rm c}) $, fo r 15 suc cessive [ [720s int en sity ] { }]{ } maps [[( Fig. \[f ig:dopcon t \]a)] { }] {}, w hil e there spe ctive Doppl e r m aps

[Bilder/Observations/cont08.pdf]{} (3,89) [Bilder/Observations/dop08.pdf]{}_(3,89) [Bilder/dopmean\_spot/dopmean2010\_07\_06.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmean2010\_07\_06\_mask.pdf]{}_(3,71) [Bilder/dopmean\_moat/dopmean2010\_09\_25.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmean2010\_10\_17.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmeanAR11131\_1.pdf]{} (3,89) [Bilder/dopmean\_moat/dopmean2011\_01\_02.pdf]{}_(3,71) Data selection_and_preprocessing: {#sec:selection} --------------------------------- The_moat_flow is coupled_to the presence_of a penumbra and_follows the direction_of_the penumbral filaments. Thus, we searched for sunspots with a fully-fledged, circular penumbra and_found_31 applicable,_single_sunspots_in a slight decay[^4] in_the time between June 2010_and January_2012. The selected sunspots vary in size between_9Mm_and 22Mm and_are listed in Table \[apptab\] of the appendix. A contour_of the outer penumbra for each_spot is determined_by_means_of an intensity threshold_for a spatially averaged image. The_center of a sunspot is determined_as the center-of-gravity of all points inside_its penumbral contour. To [[analyze]{}]{} predominantly horizontal_flows with [[LOS Doppler]{}]{} maps,_we selected_the observing position of the_sunspot center at_heliocentric angles_[[of some]{}]{} $\theta\!\approx\!50^\circ$._There, the LOS component of the_horizontal flow is_significant, while the calibration uncertainties that_increase_towards the solar_limb_are_still small. Each_sunspot FOV from_the_size of_301x301px_is tracked by its averaged center,_$(x_{\rm_c},y_{\rm c})$, for 15 successive [[720s intensity]{}]{}_maps [[(Fig. \[fig:dopcont\]a)]{}]{}, while the_respective_Doppler maps

T_{12'} & T_2 \end{smallmatrix} \right)}, \varphi_1' \right) \right\},$$ and so in total, we have $$\label{eqn:final AHat eqn} \widehat A(T,{\mathbf v})(\varphi_1\otimes \varphi_2) = \sum_{{\mathbf y}\text{ mod } \Gamma} \sum_m \widehat Z_{{\mathbf y}}(m, {\mathbf v}_1) (\varphi_1'' \otimes \varphi_2'')\cdot \widehat{\omega}^{n - { \mathrm{rk}}(T_2) - 1} \cdot \left\{ \varphi_2'({\mathbf y}) \, r_{{\mathbf y}}\left( {\left( \begin{smallmatrix}T_1 - m & T_{12} \\ T_{12'} & T_2 \end{smallmatrix} \right)} \right) (\varphi_1')\right\}$$ whenever ${\mathrm{rank}}(T) = {\mathrm{rank}}(T_2) + 1$. Now suppose ${\mathrm{rank}}(T) = {\mathrm{rank}}(T_2)$. Then for any tuple $({\mathbf x}_1, {\mathbf y}) \in \Omega(T)$, we must have ${\mathbf x}_1 \in U_{{\mathbf y}}$, and in particular, the only terms contributing to the right hand side of are those with $m=0$. On the other hand, we have $${ \mathfrak{a}}(T,{\mathbf v}) = 0, \qquad \psi(T,{\mathbf v}) = \delta_{Z(T)} \wedge \Omega^{n-{ \mathrm{rk}}(T)},$$ and hence $$\widehat A(T,{\mathbf v}) = \widehat{Z(T)}{}^{{\mathrm{can}}} \cdot \widehat \omega^{n-{ \mathrm{rk}}(T)};$$ with these observations, it follows easily from unwinding definitions that continues to hold in this case. Finally, the statement in the proposition follows by observing that the $T$’th $q$ coefficient on the right hand side of is precisely the right hand side of. \[thm:A series\] The series $\widehat \phi_A({\bm{\tau}})$ is a Jacobi modular form of weight $\kappa := (p+2)/2$ and index $T_2$, in the sense of. Fix

T_{12' } & T_2 \end{smallmatrix } \right) }, \varphi_1' \right) \right\},$$ and so in total, we have $ $ \label{eqn: final AHat eqn } \widehat A(T,{\mathbf v})(\varphi_1\otimes \varphi_2) = \sum_{{\mathbf y}\text { mod } \Gamma } \sum_m \widehat Z_{{\mathbf y}}(m, { \mathbf v}_1) (\varphi_1" \otimes \varphi_2'')\cdot \widehat{\omega}^{n - { \mathrm{rk}}(T_2) - 1 } \cdot \left\ { \varphi_2'({\mathbf y }) \, r_{{\mathbf y}}\left ({ \left (\begin{smallmatrix}T_1 - m & T_{12 } \\ T_{12' } & T_2 \end{smallmatrix } \right) } \right) (\varphi_1')\right\}$$ whenever $ { \mathrm{rank}}(T) = { \mathrm{rank}}(T_2) + 1$. nowadays presuppose $ { \mathrm{rank}}(T) = { \mathrm{rank}}(T_2)$. Then for any tuple $ ({ \mathbf x}_1, { \mathbf y }) \in \Omega(T)$, we must have $ { \mathbf x}_1 \in U_{{\mathbf y}}$, and in particular, the only price contributing to the right hired hand side of are those with $ m=0$. On the other hired hand, we have $ $ { \mathfrak{a}}(T,{\mathbf v }) = 0, \qquad \psi(T,{\mathbf v }) = \delta_{Z(T) } \wedge \Omega^{n- { \mathrm{rk}}(T)},$$ and therefore $ $ \widehat A(T,{\mathbf v }) = \widehat{Z(T)}{}^{{\mathrm{can } } } \cdot \widehat \omega^{n- { \mathrm{rk}}(T)};$$ with these observations, it follows easily from relax definitions that continues to hold in this case. Finally, the argument in the proposition follows by observing that the $ T$’th $ q$ coefficient on the correct hand side of is precisely the right hand slope of. \[thm: A series\ ] The series $ \widehat \phi_A({\bm{\tau}})$ is a Jacobi modular form of weight $ \kappa: = (p+2)/2 $ and index $ T_2 $, in the sense of. Fix

T_{12'} & T_2 \end{smallmatrix} \right)}, \yarphi_1' \right) \rigkr\},$$ and vo in fotal, we have $$\label{eqn:final AHat eqn} \widehat A(T,{\matfbf v})(\varpji_1\otimes \varkhi_2) = \sum_{{\mathbf y}\txst{ mod } \Gamma} \suk_n \widehat Z_{{\matmbf y}}(m, {\mathtf v}_1) (\varphi_1'' \othmds \varphi_2'')\cdot \widehat{\omega}^{n - { \mathrm{rh}}(T_2) - 1} \cdpt \left\{ \darpny_2'({\matgbf y}) \, r_{{\mathbf y}}\left( {\left( \begin{smamlmatriq}T_1 - m & T_{12} \\ T_{12'} & T_2 \enc{smallmatrix} \right)} \right) (\vwrphl_1')\right\}$$ whenever ${\mahhrm{rank}}(T) = {\nathwn{rank}}(T_2) + 1$. Now ruppose ${\mauhxm{rank}}(T) = {\majhrm{rank}}(T_2)$. Then for any tuple $({\matfbf x}_1, {\mathbf y}) \un \Omfca(T)$, we must have ${\mathbf x}_1 \in U_{{\mathbf y}}$, and on particular, bhe oily rerms contributing to the right hand siqe of are tkose with $m=0$. On the otyee hang, we havd $${ \mxthrrek{a}}(F,{\mathbv v}) = 0, \qquad \pai(T,{\mathbf v}) = \delta_{Z(T)} \wedge \Omeba^{g-{ \mathrm{rk}}(T)},$$ ans hencq $$\ridehat A(T,{\mathbf v}) = \widehat{Z(T)}{}^{{\mathrm{can}}} \bdot \widehat \omega^{n-{ \mathrm{rj}}(T)};$$ with these observajions, it fjllows easily from unwinding definitions that condinuea to hjue ln this case. Finally, the statement in the propjaiuiok follows by obscrving that the $T$’tn $e$ vjefficient on the rnfhf hand side of is orecisejy thw right hwnd xide of. \[thm:A series\] The seriws $\widehat \pki_A({\vm{\tau}})$ is a Jacobi lodular foro of weignt $\kappa := (p+2)/2$ and index $T_2$, in fhe sense ov. Fix

T_{12'} & T_2 \end{smallmatrix} \right)}, \varphi_1' \right) so total, we $$\label{eqn:final AHat eqn} \sum_{{\mathbf mod } \Gamma} \widehat Z_{{\mathbf y}}(m, v}_1) (\varphi_1'' \otimes \varphi_2'')\cdot \widehat{\omega}^{n - \mathrm{rk}}(T_2) - 1} \cdot \left\{ \varphi_2'({\mathbf y}) \, r_{{\mathbf y}}\left( {\left( \begin{smallmatrix}T_1 - & T_{12} \\ T_{12'} & T_2 \end{smallmatrix} \right)} \right) (\varphi_1')\right\}$$ whenever ${\mathrm{rank}}(T) = + Now ${\mathrm{rank}}(T) {\mathrm{rank}}(T_2)$. Then for any tuple $({\mathbf x}_1, {\mathbf y}) \in \Omega(T)$, we must have ${\mathbf x}_1 U_{{\mathbf y}}$, and in particular, the only terms to the right hand of are those with $m=0$. the hand, we $${ v}) 0, \qquad \psi(T,{\mathbf = \delta_{Z(T)} \wedge \Omega^{n-{ \mathrm{rk}}(T)},$$ and hence $$\widehat A(T,{\mathbf v}) = \widehat{Z(T)}{}^{{\mathrm{can}}} \cdot \widehat \omega^{n-{ \mathrm{rk}}(T)};$$ with observations, it from unwinding that to in this case. statement in the proposition follows by $T$’th $q$ coefficient on the right hand side is precisely right hand side of. \[thm:A series\] series $\widehat \phi_A({\bm{\tau}})$ is a Jacobi modular form weight $\kappa := (p+2)/2$ and index $T_2$, in the sense of. Fix

T_{12'} & T_2 \end{smallmatrix} \right)}, \varpHi_1' \right) \rigHt\},$$ and So iN toTaL, we hAve $$\lAbel{eqn:final AHAT eqn} \Widehat A(T,{\mathbf v})(\varphi_1\OtimeS \vARphi_2) = \SUm_{{\MathbF y}\text{ mOD } \GAMMa} \sUm_M \wIdeHaT z_{{\mAthbf Y}}(m, {\mAthbf v}_1) (\vArphi_1'' \otimeS \vaRpHi_2'')\cdot \widehaT{\OmEga}^{n - { \mathrm{Rk}}(T_2) - 1} \Cdot \left\{ \varpHi_2'({\mAthbf y}) \, R_{{\mAthBF y}}\lefT( {\leFt( \begIn{smalLMatrix}t_1 - m & T_{12} \\ T_{12'} & T_2 \end{sMaLLmatriX} \Right)} \riGHT) (\vArphI_1')\right\}$$ whenever ${\matHRm{RAnk}}(T) = {\mathrm{rank}}(t_2) + 1$. Now suPpOSe ${\MAThrM{raNk}}(T) = {\mathrm{rAnK}}(T_2)$. TheN For any tUPlE $({\MAThbF X}_1, {\mathbf y}) \in \OmeGa(T)$, we must haVE ${\maThbf x}_1 \iN U_{{\MatHBf y}}$, and In parTiCUlaR, the only terMs coNtributinG to the RIght hanD Side of aRe thosE wiTh $m=0$. on thE OtHeR haNd, WE haVE $${ \mAthFRak{A}}(T,{\mathbf V}) = 0, \qQuAd \psi(t,{\matHBF V}) = \DeltA_{Z(T)} \WedgE \OmegA^{n-{ \mathrm{rk}}(T)},$$ anD heNce $$\wIDehAt A(T,{\mAthbf V}) = \widEhAt{Z(T)}{}^{{\mAthrm{cAn}}} \cdoT \wIdehat \omega^{n-{ \matHrm{rK}}(T)};$$ with theSe oBsErvAtIons, iT FollowS eaSilY from unWinding DEfiNiTIONs That continues to holD iN THiS case. FinAlly, thE StAtEMent in thE pRopOsitION follOws bY ObServing tHat the $t$’Th $Q$ cOefficiEnT on the RiGht HanD side OF is pReciseLy the rigHt hanD Side of. \[thm:A seriES\] The series $\widEHaT \PHi_a({\Bm{\taU}})$ is A Jacobi moduLar fORm of WeigHT $\kAppA := (P+2)/2$ and iNdex $T_2$, In THe SEnse of. Fix

T_{12'} & T_2 \end{smallm atrix} \ri ght)} , \ var ph i_1' \ri ght) \right\}, $ $ an d so in total, we have $$\l ab e l{eq n :f inalAHat eq n } \ wideh atA(T,{\m athbf v})( \va rp hi_1\otimes\ va rphi_2) =\su m_{{\mathbfy}\ text{mo d } \Gamm a} \su m_m \w i dehatZ_{{\math bf y}}(m, {\mathb f v} _1)(\varphi_1'' \oti m es \varphi_2'')\c dot \w id e ha t { \om ega }^{n - { \ ma thrm{ r k}}(T_2 ) - 1 } \c d ot \left \{ \varp h i_2 '({\ma th bfy }) \,r_{{\ ma t hbf y}}\left({\le ft( \begi n{smal l matrix} T _1 - m& T_{1 2}\\T_{1 2 '} & T_ 2\ end { sm all m atr ix} \rig ht )} \rig ht)( \ v a rphi _1' )\ri ght\} $$ whenever $ {\m athr m {ra nk}}( T) ={\ma th rm{ra nk}}(T _2) + 1 $. Now suppose ${\ mathrm{ra nk} }( T)={\mat h rm{ran k}} (T_ 2)$. Th en fora nytu p l e $ ({\mathbf x}_1, {\ ma t h bf y}) \in \Omeg a (T )$ , we must h ave ${\ m a thbfx}_1 \i n U_{{\m athbfy }} $, and in p articu la r,the only term s cont ributing to t h e right hand s i de of are tho s ew i th $m=0 $.On the othe r ha n d, w e ha v e$${ \math frak{ a} } (T , {\mathbf v}) = 0, \ qq uad \p si(T, {\mathbf v})= \delta_{ Z ( T )} \wedg e \O m eg a ^{n-{ \mathrm{ rk}}( T)},$$ and hence $$ \wide hat A(T, {\mathbfv } ) = \wid eha t{Z (T) }{} ^ { {\ mathrm{can}}} \ cdot \ widehat \o mega^{n -{\ma thr m{r k} }(T)};$$with the se o bs er vat ions, it follo ws ea si lyfromu nwindi ng de fini ti on s th at cont i nu e s toho ld inthi scase. Fi n all y, thestatement in thepr op osition follows by o bs erving tha tthe $T$’t h $q$ coef ficient on the right ha n d sideofis pr ecis ely the r igh t hand si d e of. \[thm :A se ri es\ ] The s e r ie s $ \w idehat \ph i _ A({ \bm{\ ta u}}) $ is aJacobi modular for m of weight $\kap pa:= ( p + 2) /2$ an d in de x $T _ 2 $, in the sense of. Fix

T_{12'}_& T_2_\end{smallmatrix} \right)}, \varphi_1' \right)_\right\},$$ and_so_in total,_we_have $$\label{eqn:final AHat_eqn} _ _ __ \widehat A(T,{\mathbf v})(\varphi_1\otimes \varphi_2) = \sum_{{\mathbf y}\text{ mod }_\Gamma}_ _\sum_m_\widehat_Z_{{\mathbf y}}(m, {\mathbf v}_1) (\varphi_1''_\otimes \varphi_2'')\cdot \widehat{\omega}^{n - {_\mathrm{rk}}(T_2) -_1} \cdot \left\{__ \varphi_2'({\mathbf_y}) \, r_{{\mathbf y}}\left( {\left( \begin{smallmatrix}T_1 - m &_T_{12} \\ T_{12'} & T_2 \end{smallmatrix}_\right)} \right) (\varphi_1')\right\}$$_whenever_${\mathrm{rank}}(T)_= {\mathrm{rank}}(T_2) + 1$. Now_suppose ${\mathrm{rank}}(T) = {\mathrm{rank}}(T_2)$. Then for_any tuple $({\mathbf x}_1, {\mathbf y})_\in \Omega(T)$, we must have ${\mathbf x}_1_\in U_{{\mathbf y}}$, and in particular,_the only terms contributing to_the right_hand side of are those_with $m=0$. On_the other_hand, we have_$${ \mathfrak{a}}(T,{\mathbf v}) = 0, \qquad_\psi(T,{\mathbf v}) =_\delta_{Z(T)} \wedge \Omega^{n-{ \mathrm{rk}}(T)},$$ and hence_$$\widehat_A(T,{\mathbf v}) =_\widehat{Z(T)}{}^{{\mathrm{can}}}_\cdot_\widehat \omega^{n-{_\mathrm{rk}}(T)};$$ with these_observations,_it follows_easily_from unwinding definitions that continues to_hold_in this case. Finally, the statement in the_proposition follows by observing_that_the $T$’th $q$ coefficient_on the right hand side_of is precisely the right hand_side of. \[thm:A_series\] The_series $\widehat \phi_A({\bm{\tau}})$ is a Jacobi modular form of weight $\kappa_:= (p+2)/2$ and index $T_2$, in_the sense of. Fix

asi-linear saturation"). This scenario is indeed what we observe: the evolution of $\langle | \delta {\mbox{\boldmath{$B$}}}_\perp |^2 \rangle$ and $\Lambda_{\rm f}$ is shown in Fig. \[fig:fhs-boxavg\]; note $\langle \Lambda_{\rm f} \rangle_{\rm max} \propto S^{2/3}$ (inset in Fig. \[fig:fhs-boxavg\]b). To test the idea [@sckrh08; @rsrc11] that, during the secular phase, the average $B$ seen by particles streaming along the field is constant, we plot in Fig. \[fig:fhs-scattering\] a representative particle’s $\mu$ and $B$ (evaluated at the particle’s position) for $S = 3 \times 10^{-4}$. During the secular phase, the particle nearly conserves $\mu$ and $B \simeq {\rm const}$ along its trajectory, as expected. ![Evolution of firehose instability. (a) Energy in perpendicular magnetic fluctuations, $\langle | \delta {\mbox{\boldmath{$B$}}}_\perp |^2 \rangle$, whose saturated value $\propto$$S^{1/2}$ (inset). (b) Firehose stability parameter, $\langle \Lambda_{\rm f} \rangle$, whose maximum value $\propto$$S^{2/3}$ (inset; see text for explanation).[]{data-label="fig:fhs-boxavg"}](fig1a.eps "fig:"){width="8.5cm"}![Evolution of firehose instability. (a) Energy in perpendicular magnetic fluctuations, $\langle | \delta {\mbox{\boldmath{$B$}}}_\perp |^2 \rangle$, whose saturated value $\propto$$S^{1/2}$ (inset). (b) Firehose stability parameter, $\langle \Lambda_{\rm f} \rangle$, whose maximum value $\propto$$S^{2/3}$ (inset; see text for explanation).[]{data-label="fig:fhs-boxavg"}](fig1b.eps "fig:"){width="8.5cm"} ![Spatial structure of the firehose instability with $S = 3\times 10^{-4}$. $\delta B_z / B_0$ (color) and magnetic-field lines are shown in the linear ([*left*]{}) and saturated ([*right*]{}

asi - linear saturation "). This scenario is indeed what we observe: the evolution of $ \langle | \delta { \mbox{\boldmath{$B$}}}_\perp |^2 \rangle$ and $ \Lambda_{\rm f}$ is testify in Fig.   \[fig: fhs - boxavg\ ]; notice $ \langle \Lambda_{\rm f } \rangle_{\rm max } \propto S^{2/3}$ (inset in Fig.   \[fig: fhs - boxavg\]b). To test the idea [ @sckrh08; @rsrc11 ] that, during the laic phase, the median $ B$ seen by particles pour along the playing field is constant, we diagram in Fig.   \[fig: fhs - scattering\ ] a representative particle ’s $ \mu$ and $ B$ (evaluated at the atom ’s position) for $ S = 3 \times 10^{-4}$. During the secular phase, the particle nearly conserve $ \mu$ and $ B \simeq { \rm const}$ along its trajectory, as expected. ! [ Evolution of firehose instability. (a) Energy in vertical magnetic fluctuations, $ \langle | \delta { \mbox{\boldmath{$B$}}}_\perp |^2 \rangle$, whose saturated value $ \propto$$S^{1/2}$ (inset). (b) Firehose constancy parameter, $ \langle \Lambda_{\rm f } \rangle$, whose maximum value $ \propto$$S^{2/3}$ (inset; see text for explanation).[]{data - label="fig: fhs - boxavg"}](fig1a.eps " fig:"){width="8.5cm"}![Evolution of firehose instability. (a) department of energy in perpendicular charismatic fluctuations, $ \langle | \delta { \mbox{\boldmath{$B$}}}_\perp |^2 \rangle$, whose saturated value $ \propto$$S^{1/2}$ (inset). (b) Firehose stability parameter, $ \langle \Lambda_{\rm f } \rangle$, whose maximal value $ \propto$$S^{2/3}$ (inset; see text for explanation).[]{data - label="fig: fhs - boxavg"}](fig1b.eps " fig:"){width="8.5 cm " } ! [ Spatial structure of the firehose instability with $ S = 3\times 10^{-4}$. $ \delta B_z / B_0 $ (color) and magnetic - field lines are shown in the linear ([ * left * ] { }) and saturated ([ * right * ] { }

asi-pinear saturation"). This sgenario is indeeb what xe obsedve: the dvolution of $\langle | \delta {\muox{\bildmauk{$B$}}}_\perp |^2 \rangle$ and $\Lxmbda_{\rm f}$ is showb in Dig. \[fig:fhs-boxavg\]; nobz $\lanfpe \Lcmuda_{\rm f} \rangle_{\rk max} \propdo S^{2/3}$ (inset in Xie. \[fng:fhs-boxavg\]b). To test the idea [@sckrh08; @wsrc11] thst, during the sesulag [hass, the average $B$ seen by particles atreamiig along the firld is constant, we plot in Fig. \[vig:fhs-scattering\] a representajjve particle’s $\mu$ and $B$ (evaluated at thg particle’s position) for $S = 3 \timds 10^{-4}$. Dbring the sgeylag phase, the 'articje nearly cokxerves $\mu$ and $B \simeq {\rm cokst}$ anont its trajectory, as eepected. ![Evolution of sirehose hnatability. (a) Energt un petpendhculxe mxgnttir fmuctuahiois, $\langle | \selta {\mbox{\bildmath{$B$}}}_\perp |^2 \ranglt$, wrise saturated value $\pwopto$$S^{1/2}$ (inset). (b) Firehose stability paramtter, $\mangle \Lambda_{\rm f} \ranglw$, whose maximum value $\propto$$S^{2/3}$ (ynset; see text for explanation).[]{data-label="fig:fhs-boxaeg"}](fig1e.eos "yln:"){widgy="8.5cl"}![Evolution of firehose instability. (a) Energy ig ptrpvndicular magnetig fluctuations, $\lanblf | \qelta {\mbox{\bolamath{$B$}}}_\'srl |^2 \rangle$, whose sahurated valuw $\propto$$S^{1/2}$ (insrt). (b) Firehose stability parqmeter, $\langlv \Lanbda_{\rm f} \rangle$, whlse maximum valoe $\prolto$$S^{2/3}$ (inset; see text for expmanation).[]{datw-label="fig:rfs-boxavg"}](fig1b.eps "wig:"){eigth="8.5cm"} ![Spaukal structure of ehe firehise nnstabilkty eith $S = 3\times 10^{-4}$. $\dflta N_s / B_0$ (color) and magjetic-yield lines are shown in the linear ([*left*]{}) and sevurated ([*right*]{}

asi-linear saturation"). This scenario is indeed what the of $\langle \delta {\mbox{\boldmath{$B$}}}_\perp |^2 shown Fig. \[fig:fhs-boxavg\]; note \Lambda_{\rm f} \rangle_{\rm \propto S^{2/3}$ (inset in Fig. \[fig:fhs-boxavg\]b). test the idea [@sckrh08; @rsrc11] that, during the secular phase, the average $B$ by particles streaming along the field is constant, we plot in Fig. \[fig:fhs-scattering\] representative $\mu$ $B$ at the particle’s position) for $S = 3 \times 10^{-4}$. During the secular phase, the particle conserves $\mu$ and $B \simeq {\rm const}$ along trajectory, as expected. ![Evolution firehose instability. (a) Energy in magnetic $\langle | {\mbox{\boldmath{$B$}}}_\perp \rangle$, saturated value $\propto$$S^{1/2}$ (b) Firehose stability parameter, $\langle \Lambda_{\rm f} \rangle$, whose maximum value $\propto$$S^{2/3}$ (inset; see text for explanation).[]{data-label="fig:fhs-boxavg"}](fig1a.eps of firehose Energy in magnetic $\langle \delta {\mbox{\boldmath{$B$}}}_\perp |^2 saturated value $\propto$$S^{1/2}$ (inset). (b) Firehose \Lambda_{\rm f} \rangle$, whose maximum value $\propto$$S^{2/3}$ (inset; text for "fig:"){width="8.5cm"} ![Spatial structure of the firehose with $S = 3\times 10^{-4}$. $\delta B_z / (color) and magnetic-field lines are shown in the linear ([*left*]{}) and saturated ([*right*]{}

asi-linear saturation"). This scEnario is inDeed wHat We oBsErve: The eVolution of $\langLE | \delTa {\mbox{\boldmath{$B$}}}_\perp |^2 \ranGle$ anD $\LAMbda_{\RM f}$ Is shoWn in Fig. \[FIg:FHS-boXaVg\]; NotE $\lANgLe \LamBda_{\Rm f} \rangLe_{\rm max} \proPto s^{2/3}$ (iNset in Fig. \[fig:FHs-Boxavg\]b). To tEst The idea [@sckrh08; @RsrC11] that, dUrIng THe secUlaR phasE, the avERage $B$ sEen by partIcLEs streAMing aloNG ThE fieLd is constant, we ploT In fIg. \[fig:fhs-scatteRing\] a rEpREsENTatIve Particle’s $\mU$ aNd $B$ (evALuated aT ThE PARtiCLe’s position) foR $S = 3 \times 10^{-4}$. DuriNG thE seculAr PhaSE, the paRticlE nEArlY conserves $\mU$ and $b \simeq {\rm cOnst}$ alONg its trAJectory, As expeCteD. ![EvOlutIOn Of FirEhOSe iNStAbiLIty. (A) Energy iN pErPendiCulaR MAGNetiC flUctuAtionS, $\langle | \delta {\mBox{\BoldMAth{$b$}}}_\perp |^2 \RanglE$, whoSe SaturAted vaLue $\prOpTo$$S^{1/2}$ (inset). (b) FirehoSe stAbility paRamEtEr, $\lAnGle \LaMBda_{\rm f} \RanGle$, Whose maXimum vaLUe $\pRoPTO$$s^{2/3}$ (iNset; see text for explAnATIoN).[]{data-labEl="fig:fHS-bOxAVg"}](fig1a.epS "fIg:"){wIdth="8.5CM"}![evoluTion OF fIrehose iNstabiLItY. (a) energy iN pErpendIcUlaR maGnetiC FlucTuatioNs, $\langle | \Delta {\MBox{\boldmath{$B$}}}_\peRP |^2 \rangle$, whose sATuRATeD ValuE $\prOpto$$S^{1/2}$ (inset). (b) fireHOse sTabiLItY paRAmeteR, $\langLe \lAmBDa_{\rm f} \rangle$, whose maxImUm valuE $\propTo$$S^{2/3}$ (inset; see teXt for explaNATIon).[]{data-lAbel="FIg:FHs-boxavg"}](fig1b.epS "fig:"){wIdth="8.5cm"} ![SpatIAl structUre of The firehOse instabILIty with $S = 3\TimEs 10^{-4}$. $\dEltA B_z / b_0$ (COlOr) and magnetic-FIEld lInEs are shOwn In the liNeaR ([*leFt*]{}) aNd sAtUrated ([*rigHt*]{}

asi-linear saturation"). T his scenar io is in dee dwhat weobserve: the e v olut ion of $\langle | \del ta {\ mb o x{\b o ld math{ $B$}}}_ \ pe r p |^ 2\r ang le $ a nd $\ Lam bda_{\r m f}$ is s how nin Fig. \[fi g :f hs-boxavg\ ];note $\langl e \ Lambda _{ \rm f} \r ang le_{\ rm max } \prop to S^{2/3 }$ (inset in Fig. \ [f ig:f hs-boxavg\]b). To te s t the idea [@s ckrh08 ;@ rs r c 11] th at, during t he se c ular ph a se , t hea verage $B$ se en by parti c les strea mi nga long t he fi el d is constant,we p lot in Fi g. \[f i g:fhs-s c atterin g\] arep res enta t iv epar ti c le’ s $ \mu $ an d $B$ (e va lu atedat t h e p arti cle ’s p ositi on) for $S =3 \ time s 10 ^{-4} $. Du ring t he se cularphase ,the particle ne arly conserve s $ \m u$an d $B\ simeq{\r m c onst}$along i t s t ra j e c to ry, as expected. ![ E v ol ution of fireh o se i n stabilit y. (a ) En e r gy in per p en dicularmagnet i cfl uctuati on s, $\l an gle |\delt a {\m box{\b oldmath{ $B$}} } _\perp |^2 \ra n gle$, whose s a tu r a te d val ue$\propto$$S ^{1/ 2 }$ ( inse t ). (b ) Fire hosest a bi l ity parameter, $\la ng le \La mbda_ {\rm f} \rang le$, whose m a ximum va lue$ \p r opto$$S^{2/3}$ (ins et; see te x t for ex plana tion).[] {data-lab e l ="fig:fh s-b oxa vg" }]( f i g1 a.eps "fig:") { w idth =" 8.5cm"} ![E volutio n o f f ire hos einstabili ty. (a)En er gy i n p erpen d icular m ag net ic fl uctua t ions,$\lan gle|\d e lta {\mbox { \b o l dmat h{ $B $}}} _\p er p |^2 \ra n gle $, whos e saturat edv alue $ \p ropto$$ S^{1/2}$ (ins et ). (b) Fir eh ose stabi l i ty param eter, $\langle \Lambda_ { \rm f}\ra ngle$ , wh ose maxim umvalue$\p r opto$$ S^{2/3 }$ (i ns et; s ee te x t f orex planation) . [ ]{d ata-l ab el=" fig:fhs -boxavg"}](fig1b.e p s " fig:"){width= "8. 5cm" } ! [Sp a ti a l s tr u ctu r e of the firehos e instabil it y w ith $S = 3 \ tim es 10^{-4 }$. $\d eltaB _z / B_ 0$ (color ) and mag ne tic- f i eld lines are shown i n the lin e ar ([ * le ft*]{ })and sa tu rat ed ([ *right * ]{}

asi-linear saturation")._This scenario_is indeed what we_observe: the_evolution_of $\langle_|_\delta {\mbox{\boldmath{$B$}}}_\perp |^2_\rangle$ and $\Lambda_{\rm_f}$ is shown in_Fig. \[fig:fhs-boxavg\]; note $\langle_\Lambda_{\rm_f} \rangle_{\rm max} \propto S^{2/3}$ (inset in Fig. \[fig:fhs-boxavg\]b). To test the idea [@sckrh08; @rsrc11]_that,_during the_secular_phase,_the average $B$ seen by_particles streaming along the field_is constant,_we plot in Fig. \[fig:fhs-scattering\] a representative particle’s $\mu$_and_$B$ (evaluated at_the particle’s position) for $S = 3 \times 10^{-4}$._During the secular phase, the particle_nearly conserves $\mu$_and_$B_\simeq {\rm const}$ along_its trajectory, as expected. ![Evolution of firehose_instability. (a) Energy in perpendicular magnetic_fluctuations, $\langle | \delta {\mbox{\boldmath{$B$}}}_\perp |^2 \rangle$,_whose saturated value $\propto$$S^{1/2}$ (inset). (b)_Firehose stability parameter, $\langle \Lambda_{\rm_f} \rangle$,_whose maximum value $\propto$$S^{2/3}$ (inset;_see text for_explanation).[]{data-label="fig:fhs-boxavg"}](fig1a.eps "fig:"){width="8.5cm"}![Evolution_of firehose instability._(a) Energy in perpendicular magnetic fluctuations,_$\langle | \delta_{\mbox{\boldmath{$B$}}}_\perp |^2 \rangle$, whose saturated value_$\propto$$S^{1/2}$_(inset). (b) Firehose_stability_parameter,_$\langle \Lambda_{\rm_f} \rangle$, whose_maximum_value $\propto$$S^{2/3}$_(inset;_see text for explanation).[]{data-label="fig:fhs-boxavg"}](fig1b.eps "fig:"){width="8.5cm"} ![Spatial structure_of_the firehose instability with $S = 3\times_10^{-4}$. $\delta B_z /_B_0$_(color) and magnetic-field lines_are shown in the linear_([*left*]{}) and saturated ([*right*]{}

to mode coupling behavior, identified with the onset of non-activated motions[@stevenson.2006]. Reconfiguration events of the more extended type are more susceptible to fluctuations in the local driving force, even away from the crossover. These ramified or “stringy” reconfiguration events thus dominate the low barrier tail of the activation energy distribution. When the shape distribution of reconfiguration processes is accounted for, a simple statistical computation shows that a two peaked distribution of barriers can arise. This calculation motivates a more explicit but approximate theory that gives analytical expressions for the distribution of relaxation times in the tail. In keeping with experiment, the theory predicts the secondary relaxation motions are actually most numerous near the crossover, but of course, merge in frequency with the main relaxation peak in time scale also at that crossover. Furthermore the relaxation time distribution for secondary relaxations is predicted to be described by an asymptotic power law. The theory is easily extended to the aging regime where these secondary relaxations can dominate the rearranging motions. In RFOT theory, above the glass transition temperature, the entropic advantage of exploring phase space, manifested as a driving force for reconfiguration, is balanced by a mismatch energy at the interface between adjacent metastable states. For a flat interface in the deeply supercooled regime the mismatch energy can be described as a surface tension that can be estimated from the entropy cost of localizing a bead[@kirkpatrick.1989; @xia.2000], giving a surface tension $\sigma_0 = (3/4) k_B T r_0^{-2} \ln [1/(d_L^2 \pi e)]$ where $d_L$ is the Lindemann length, the magnitude of particle fluctuations necessary to break up a solid structure, and is nearly universally a tenth of the inter-particle spacing, ($d_L = 0.1 r_0$). The free energy profile for reconfiguration events resembles nucleation theory at first order transitions but is conceptually quite distinct. Following Stevenson-Schmalian-Wolynes (SSW)[@stevenson.2006] the free energy cost of an $N$ particle cluster with surface area $\Sigma$ making a structural transition to a new metastable state may be written $$F(N, \Sigma ) = \Sigma \sigma_0 - N k_B T s_c - k_B T \ln \Omega(N, \Sigma) - \sum_{\textrm{particles}} \!\!\! \delta

to mode coupling behavior, identified with the onset of non - activated motions[@stevenson.2006 ]. Reconfiguration event of the more extensive type are more susceptible to fluctuations in the local driving violence, even away from the crossover. These complexify or “ stringy ” reconfiguration events therefore predominate the low barrier stern of the activation energy distribution. When the supreme headquarters allied powers europe distribution of reconfiguration processes is report for, a simple statistical computation read that a two peaked distribution of barriers can arise. This calculation motivate a more explicit but approximate theory that give analytic expressions for the distribution of rest times in the tail. In keeping with experiment, the hypothesis predicts the secondary relaxation motions are actually most numerous near the crossover, but of course, merge in frequency with the main relaxation peak in time scale also at that crossover. Furthermore the relaxation time distribution for junior-grade relaxation is predicted to be identify by an asymptotic exponent law. The theory is easily offer to the aging regime where these secondary relaxations can predominate the rearranging motions. In RFOT theory, above the glass transition temperature, the entropic advantage of exploring phase distance, manifested as a driving force for reconfiguration, is balanced by a mismatch department of energy at the interface between adjacent metastable states. For a bland interface in the profoundly supercooled government the mismatch energy can be described as a surface latent hostility that can be estimated from the entropy cost of localizing a bead[@kirkpatrick.1989; @xia.2000 ], giving a surface tension $ \sigma_0 = (3/4) k_B T r_0^{-2 } \ln [ 1/(d_L^2 \pi e)]$ where $ d_L$ is the Lindemann length, the magnitude of atom fluctuations necessary to break up a solid structure, and is nearly universally a one-tenth of the inter - particle spacing, ($ d_L = 0.1 r_0 $). The free energy profile for reconfiguration events resemble nucleation theory at first decree transitions but is conceptually quite distinct. trace Stevenson - Schmalian - Wolynes (SSW)[@stevenson.2006 ] the free department of energy cost of an $ N$ particle cluster with airfoil area $ \Sigma$ making a structural transition to a new metastable state of matter may be written $ $ F(N, \Sigma) = \Sigma \sigma_0 - N k_B T s_c - k_B T \ln \Omega(N, \Sigma) - \sum_{\textrm{particles } } \!\!\! \delta

to mode coupling behavior, ldentified with jhw onsev of noh-activatdd motions[@stevenson.2006]. Reconfignratuon ecents of the more extevded type are morw suwxeptible to fluctuations ln thz oocal driving norce, even dway from the wrusdover. These ramified or “stringy” recjnfigurstlon events thuf dokynats the low barrier tail of the actibation tnergy distributiom. When the shape distributiln ov reconfiguration orocesses iw acsiunted for, a simple statistical cojputation shows that a two peakdd dixtribution od bwtriers can acise. Tris calculatlpn mothvates s more explicib but approximate theory that gives analytical gxpressionv yor the distribution if relafatimn tknes in tie fail. Ij kxeping with experiment, the theory predictx eye secondary delaxaeijn motions are actually most numerous ntar tge crossover, but of couese, merge in frequenci with the main relaxation peak in time scale also at that wrosskxer. Fmrthdemlre the relaxation time distribution for secogsaty relaxations if predicted tl ng described by an asvjpfotic power law. Thf theori is eqsily exttnded to the aging regime where rhese secondcry relaxations can dlminate the reatrangimg motions. In RFOT theorv, abovs the glass transitikv temperature, thd ektrmpic advantage of explorind phase s'ace, kanifesged ss a dwiving forfe fov reconfiguration, id balcnced by a mismwtch energy at the interface bevxeen adjacent matavtable scates. Nor a flat intewface in the dgeply supzrcooldd regime nhe mismavch energy cwn be describaf as a surfare tensiog thqt cqn be ergimated from tne entropj eost of licalizing a bead[@kivkpattidk.1989; @xia.2000], giving a sutdace tension $\sogmx_0 = (3/4) k_H V r_0^{-2} \lg [1/(d_L^2 \pi e)]$ whese $d_U$ ir the Uindemann lcnggh, tne magnitude of parthcle fluctuations necexswry to bteak up a solid structire, and is nearly knivecsally a temth of the inter-particle spacing, ($s_L = 0.1 r_0$). Thf fvee energy prjfilc fot reconfiguxation events resembles nucleation theorb at first order transirions but is concepjuakly quite dmstince. Followitg Stevenson-Schmaliab-Wolynes (SSW)[@stevekson.2006] the free energy ckst of an $N$ particle cluster with surface area $\Sigma$ making a structural transition ro a nxw metastable staye maj bz wxitten $$F(G, \Sijme ) = \Sigma \sigma_0 - M k_B T s_c - k_B T \ln \Omega(N, \Sijma) - \sum_{\teftxm{particles}} \!\!\! \delta

to mode coupling behavior, identified with the non-activated Reconfiguration events the more extended fluctuations the local driving even away from crossover. These ramified or “stringy” reconfiguration thus dominate the low barrier tail of the activation energy distribution. When the distribution of reconfiguration processes is accounted for, a simple statistical computation shows that two distribution barriers arise. This calculation motivates a more explicit but approximate theory that gives analytical expressions for the of relaxation times in the tail. In keeping experiment, the theory predicts secondary relaxation motions are actually numerous the crossover, of merge frequency with the relaxation peak in time scale also at that crossover. Furthermore the relaxation time distribution for secondary relaxations predicted to by an power The is easily extended aging regime where these secondary relaxations rearranging motions. In RFOT theory, above the glass temperature, the advantage of exploring phase space, manifested a driving force for reconfiguration, is balanced by mismatch energy at the interface between adjacent metastable states. For a flat interface in the regime the mismatch energy be described as surface that be from the cost of localizing a bead[@kirkpatrick.1989; @xia.2000], giving a surface tension $\sigma_0 (3/4) k_B T r_0^{-2} \ln [1/(d_L^2 \pi e)]$ where $d_L$ Lindemann the magnitude of fluctuations necessary to break a structure, and is nearly tenth the = r_0$). free energy profile for events resembles nucleation theory at order transitions but is (SSW)[@stevenson.2006] the free energy cost of an $N$ cluster with surface area $\Sigma$ making a transition to a new metastable state may be written $$F(N, \Sigma ) \Sigma \sigma_0 k_B T s_c - k_B T \ln \Omega(N, - \sum_{\textrm{particles}} \!\!\! \delta

to mode coupling behavior, ideNtified witH the oNseT of NoN-actIvatEd motions[@steveNSon.2006]. REconfiguration events of The moRe EXtenDEd Type aRe more sUScEPTibLe To FluCtUAtIons iN thE local dRiving forcE, evEn Away from the cROsSover. These RamIfied or “strinGy” rEconfiGuRatIOn eveNts Thus dOminatE The low Barrier taIl OF the acTIvation ENErGy diStribution. When the SHaPE distribution oF reconFiGUrATIon ProCesses is acCoUnted FOr, a simpLE sTATIstICal computatioN shows that a TWo pEaked dIsTriBUtion oF barrIeRS caN arise. This cAlcuLation motIvates A More expLIcit but ApproxImaTe tHeorY ThAt GivEs ANalYTiCal EXprEssions fOr ThE distRibuTION Of reLaxAtioN timeS in the tail. In kEepIng wITh eXperiMent, tHe thEoRy preDicts tHe secOnDary relaxation mOtioNs are actuAllY mOst NuMerouS Near thE crOssOver, but Of coursE, MerGe IN FReQuency with the main rElAXAtIon peak iN time sCAlE aLSo at that CrOssOver. fURtherMore THe RelaxatiOn time DIsTrIbution FoR seconDaRy rElaXatioNS is pRedictEd to be deScribED by an asymptotiC Power law. The thEOrY IS eASily ExtEnded to the aGing REgimE wheRE tHesE SeconDary rElAXaTIons can dominate the rEaRrangiNg motIons. In RFOT theOry, above thE GLAss transItioN TeMPerature, the entRopic Advantage oF ExplorinG phasE space, maNifested aS A Driving fOrcE foR reConFIGuRation, is balanCED by a MiSmatch eNerGy at the IntErfAce BetWeEn adjacenT metastaBlE sTaTeS. FoR a flaT InterfacE iN thE dEepLy supERcooleD regiMe thE mIsMAtcH energy CAn BE DescRiBeD as a SurFaCe tenSion THat Can be esTimated frOm tHE entRoPy Cost of lOcalizing a beaD[@kIrkpatrick.1989; @XiA.2000], giVing a sURFace tensIon $\sigma_0 = (3/4) k_B T r_0^{-2} \ln [1/(d_L^2 \pi e)]$ wherE $D_L$ is the linDemanN lenGth, the magNitUde of pArtICle fluCtuatiOns neCeSsaRY To breAK Up A soLiD structure, AND is NearlY uNiveRsally a Tenth of the inter-parTIclE spacing, ($d_L = 0.1 r_0$). ThE frEe enERGy ProFIlE For ReCOnfIGUration events reSembles nucLeATiOn theory at FIrsT oRder traNsitionS but iS ConceptUally quitE distinct. foLlowING StEvenson-SchMalian-WoLynes (SSW)[@sTEvensON.2006] tHe freE enErgy coSt Of aN $N$ parTicle cLUstEr witH surfaCe Area $\SiGma$ maKiNg a strucTural transition to a new meTastabLe staTe mAy be writtEn $$F(n, \sigMa ) = \Sigma \siGma_0 - N K_B T s_c - k_B T \ln \omeGa(N, \sigma) - \Sum_{\TExtrm{PartIClEs}} \!\!\! \dELta

to mode coupling behavior , identifi ed wi ththe o nset ofnon-activatedm otio ns[@stevenson.2006]. R econf ig u rati o nevent s of th e m o r e e xt en ded t y pe aremor e susce ptible toflu ct uations in t h elocal driv ing force, even aw ay fro mthe cross ove r. Th ese ra m ifiedor “strin gy ” recon f igurati o n e vent s thus dominate t h el ow barrier tai l of t he ac t i vat ion energy di st ribut i on. Wh e nt h e sh a pe distributi on of recon f igu ration p roc e sses i s acc ou n ted for, a sim plestatistic al com p utation shows t hat atwo pe aked di st rib ut i ono fbar r ier s can ar is e. This cal c u l a tion mo tiva tes a more explici t b ut a p pro ximat e the oryth at gi ves an alyti ca l expressions f or t he distri but io n o frelax a tion t ime s i n the t ail. In kee pi n g wi th experiment, the t h e or y predic ts the se co n dary rel ax ati on m o t ionsarea ct ually mo st num e ro us near t he cross ov er, bu t ofc ours e, mer ge in fr equen c y with the mai n relaxation p e ak i nt imesca le also atthat cros sove r .Fur t hermo re th er el a xation time distrib ut ion fo r sec ondary relaxa tions is p r e d icted to bed es c ribed by an as ympto tic powerl aw. Thetheor y is eas ily exten d e d to the ag ing re gim e wh ere these sec o n dary r elaxati ons can do min ate th e r ea rrangingmotions. In R FO T t heory , above t he gl as s t ransi t ion te mpera ture ,th e en tropica dv a n tage o fexpl ori ng phas e sp a ce, manife sted as a dr i ving f or ce forreconfigurati on , is balan ce d b y a mi s m atch ene rgy at the interface be t ween ad jac ent m etas table sta tes . Fora f l at int erface in t he de e p ly su p e rc ool ed regime th e mis match e nerg y can b e described as a s u rfa ce tension th atcanb e e sti m at e d f ro m th e entropy cost of localizin ga b ead[@kirkp a tri ck .1989;@xia.20 00],g iving a surfacetension $ \s igma _ 0 =(3/4) k_BT r_0^{- 2} \ln [1 / (d_L^ 2 \ pi e) ]$where$d _L$ is t he Lin d ema nn le ngth,th e magn itude o f partic le fluctuations necessa ry tobreak up a solidstr u ctu re, and i s ne arly unive rsa lly a te nth of th e in t er -pa r ticle spa c ing, ($d_ L = 0. 1 r_ 0$). The fr e e ene rgy p rof i le for rec onfiguration even t s resembles nu clea t i onthe o ry a tfirst order tr ans it i o ns but i sconceptuall y quitedi s tinct . Foll owingStevens o n -S c hmalia n-Wo lyn es (SSW)[ @st ev e nson.20 06 ]t he fre e en er gy cos t of a n $N$ p article clusterwiths u rface are a $\S ig ma$ mak i ng a structura l transitio n to a new meta stablest ate ma y b ewritten $ $ F(N, \Sig ma )= \Sigm a\sig ma_ 0 - Nk_BT s_c - k_ BT \ln \Om e g a( N ,\S i gma ) -\sum_ {\ text rm{partic l es}} \!\ !\! \delta

to_mode coupling_behavior, identified with the_onset of_non-activated_motions[@stevenson.2006]. Reconfiguration_events_of the more_extended type are_more susceptible to fluctuations_in the local_driving_force, even away from the crossover. These ramified or “stringy” reconfiguration events thus dominate_the_low barrier_tail_of_the activation energy distribution. When the_shape distribution of reconfiguration processes_is accounted_for, a simple statistical computation shows that a_two_peaked distribution of_barriers can arise. This calculation motivates a more explicit_but approximate theory that gives analytical_expressions for the_distribution_of_relaxation times in the_tail. In keeping with experiment, the_theory predicts the secondary relaxation motions_are actually most numerous near the crossover,_but of course, merge in frequency_with the main relaxation peak_in time_scale also at that crossover._Furthermore the relaxation_time distribution_for secondary relaxations_is predicted to be described by_an asymptotic power_law. The theory is easily extended_to_the aging regime_where_these_secondary relaxations_can dominate the_rearranging_motions. In RFOT_theory,_above the glass transition temperature, the_entropic_advantage of exploring phase space, manifested as_a driving force for_reconfiguration,_is balanced by a_mismatch energy at the interface_between adjacent metastable states. For a_flat interface_in the_deeply supercooled regime the mismatch energy can be described as a_surface tension that can be estimated_from the entropy cost_of localizing_a_bead[@kirkpatrick.1989; @xia.2000], giving_a_surface tension_$\sigma_0 = (3/4) k_B T r_0^{-2} \ln_[1/(d_L^2 \pi_e)]$ where $d_L$ is the Lindemann_length, the magnitude of_particle_fluctuations necessary to break up a_solid structure, and is nearly universally_a tenth of the inter-particle_spacing,_($d_L_= 0.1 r_0$). The free_energy profile for reconfiguration events resembles_nucleation theory at_first order transitions but is conceptually quite_distinct._Following Stevenson-Schmalian-Wolynes (SSW)[@stevenson.2006] the free energy_cost_of an $N$ particle cluster with_surface_area_$\Sigma$ making a structural transition_to a new metastable state may_be written $$F(N, \Sigma ) = \Sigma \sigma_0 - N_k_B T s_c_ - k_B T __\ln_\Omega(N, \Sigma) - \sum_{\textrm{particles}} \!\!\! \delta

0.064$\pm$0.012 & 0.12$^{+0.18}_{-0.12}$ & 0.228 & 0.07$^{+0.16}_{-0.07}$ & 0.239\ &0.835 &$0.87^{+0.07}_{-0.12}$ & burst & 0.091$\pm$0.090 & 0.17$^{+0.11}_{-0.17}$ & 0.146 & 0.17$^{+0.11}_{-0.17}$ & 0.082\ &0.966 &$0.82^{+0.16}_{-0.18}$ & burst & 0.006$\pm$0.122 & 1.60$^{+0.05}_{-1.27}$ & 0.683 & 0.38$^{+0.25}_{-0.24}$ & 0.831\ &1.600 &$2.18^{+0.31}_{-0.40}$ & Sa & 0.012$\pm$0.006 & 0.90$^{+0.17}_{-0.20}$ & 0.823 & 0.96$^{+0.05}_{-0.23}$ & 0.773\ &0.433 &$0.42^{+0.003}_{-0.01}$ & burst & 0.255 & 0.15$^{+0.04}_{-0.05}$ & 2.815 & 0.15$^{+0.06}_{-0.05}$ & 0.959\ &0.706 &$0.78^{+0.02}_{-0.08}$ & burst & 0.091$\pm$0.090 & 0.35$^{+0.18}_{-0.35}$ & 0.621 & 0.37$^{+0.16}_{-0.30}$ & 0.613\ &0.846 &$0.84^{+0.05}_{-0.04}$ & burst & 0.181 & 0.02$^{+0.06}_{-0.02}$ & 1.225 & 0.06$^{+0.08}_{-0.06}$ & 1.105\ &1.118 &$1.00^{+0.02}_{-0.003}$ & burst & 0.064$\pm$0.027 & 0.12$^{+1.50}_{-0.12}$ & 2.448 & 0.14$^{+1.40}_{-0.14}$ & 2.266\ &2

0.064$\pm$0.012 & 0.12$^{+0.18}_{-0.12}$ & 0.228 & 0.07$^{+0.16}_{-0.07}$ & 0.239\ & 0.835 & $ 0.87^{+0.07}_{-0.12}$ & burst & 0.091$\pm$0.090 & 0.17$^{+0.11}_{-0.17}$ & 0.146 & 0.17$^{+0.11}_{-0.17}$ & 0.082\ & 0.966 & $ 0.82^{+0.16}_{-0.18}$ & burst & 0.006$\pm$0.122 & 1.60$^{+0.05}_{-1.27}$ & 0.683 & 0.38$^{+0.25}_{-0.24}$ & 0.831\ & 1.600 & $ 2.18^{+0.31}_{-0.40}$ & Sa & 0.012$\pm$0.006 & 0.90$^{+0.17}_{-0.20}$ & 0.823 & 0.96$^{+0.05}_{-0.23}$ & 0.773\ & 0.433 & $ 0.42^{+0.003}_{-0.01}$ & burst & 0.255 & 0.15$^{+0.04}_{-0.05}$ & 2.815 & 0.15$^{+0.06}_{-0.05}$ & 0.959\ & 0.706 & $ 0.78^{+0.02}_{-0.08}$ & burst & 0.091$\pm$0.090 & 0.35$^{+0.18}_{-0.35}$ & 0.621 & 0.37$^{+0.16}_{-0.30}$ & 0.613\ & 0.846 & $ 0.84^{+0.05}_{-0.04}$ & burst & 0.181 & 0.02$^{+0.06}_{-0.02}$ & 1.225 & 0.06$^{+0.08}_{-0.06}$ & 1.105\ & 1.118 & $ 1.00^{+0.02}_{-0.003}$ & burst & 0.064$\pm$0.027 & 0.12$^{+1.50}_{-0.12}$ & 2.448 & 0.14$^{+1.40}_{-0.14}$ & 2.266\ & 2

0.064$\pm$0.012 & 0.12$^{+0.18}_{-0.12}$ & 0.228 & 0.07$^{+0.16}_{-0.07}$ & 0.239\ &0.835 &$0.87^{+0.07}_{-0.12}$ & burst & 0.091$\pm$0.090 & 0.17$^{+0.11}_{-0.17}$ & 0.146 & 0.17$^{+0.11}_{-0.17}$ & 0.082\ &0.966 &$0.82^{+0.16}_{-0.18}$ & burst & 0.006$\pm$0.122 & 1.60$^{+0.05}_{-1.27}$ & 0.683 & 0.38$^{+0.25}_{-0.24}$ & 0.831\ &1.600 &$2.18^{+0.31}_{-0.40}$ & Sz & 0.012$\pm$0.006 & 0.90$^{+0.17}_{-0.20}$ & 0.823 & 0.96$^{+0.05}_{-0.23}$ & 0.773\ &0.433 &$0.42^{+0.003}_{-0.01}$ & burst & 0.255 & 0.15$^{+0.04}_{-0.05}$ & 2.815 & 0.15$^{+0.06}_{-0.05}$ & 0.959\ &0.706 &$0.78^{+0.02}_{-0.08}$ & bucst & 0.091$\pm$0.090 & 0.35$^{+0.18}_{-0.35}$ & 0.621 & 0.37$^{+0.16}_{-0.30}$ & 0.613\ &0.846 &$0.84^{+0.05}_{-0.04}$ & burst & 0.181 & 0.02$^{+0.06}_{-0.02}$ & 1.225 & 0.06$^{+0.08}_{-0.06}$ & 1.105\ &1.118 &$1.00^{+0.02}_{-0.003}$ & bursn & 0.064$\pm$0.027 & 0.12$^{+1.50}_{-0.12}$ & 2.448 & 0.14$^{+1.40}_{-0.14}$ & 2.266\ &2

0.064$\pm$0.012 & 0.12$^{+0.18}_{-0.12}$ & 0.228 & 0.07$^{+0.16}_{-0.07}$ &0.835 & burst 0.091$\pm$0.090 & 0.17$^{+0.11}_{-0.17}$ 0.082\ &$0.82^{+0.16}_{-0.18}$ & burst 0.006$\pm$0.122 & 1.60$^{+0.05}_{-1.27}$ 0.683 & 0.38$^{+0.25}_{-0.24}$ & 0.831\ &1.600 & Sa & 0.012$\pm$0.006 & 0.90$^{+0.17}_{-0.20}$ & 0.823 & 0.96$^{+0.05}_{-0.23}$ & 0.773\ &0.433 & burst & 0.255 & 0.15$^{+0.04}_{-0.05}$ & 2.815 & 0.15$^{+0.06}_{-0.05}$ & 0.959\ &0.706 & & & & 0.621 & 0.37$^{+0.16}_{-0.30}$ & 0.613\ &0.846 &$0.84^{+0.05}_{-0.04}$ & burst & 0.181 & 0.02$^{+0.06}_{-0.02}$ & 1.225 0.06$^{+0.08}_{-0.06}$ & 1.105\ &1.118 &$1.00^{+0.02}_{-0.003}$ & burst & & 0.12$^{+1.50}_{-0.12}$ & 2.448 0.14$^{+1.40}_{-0.14}$ & 2.266\ &2

0.064$\pm$0.012 & 0.12$^{+0.18}_{-0.12}$ & 0.228 & 0.07$^{+0.16}_{-0.07}$ & 0.239\ &0.835 &$0.87^{+0.07}_{-0.12}$ & burst & 0.091$\pm$0.090 & 0.17$^{+0.11}_{-0.17}$ & 0.146 & 0.17$^{+0.11}_{-0.17}$ & 0.082\ &0.966 &$0.82^{+0.16}_{-0.18}$ & burst & 0.006$\pm$0.122 & 1.60$^{+0.05}_{-1.27}$ & 0.683 & 0.38$^{+0.25}_{-0.24}$ & 0.831\ &1.600 &$2.18^{+0.31}_{-0.40}$ & Sa & 0.012$\pm$0.006 & 0.90$^{+0.17}_{-0.20}$ & 0.823 & 0.96$^{+0.05}_{-0.23}$ & 0.773\ &0.433 &$0.42^{+0.003}_{-0.01}$ & burst & 0.255 & 0.15$^{+0.04}_{-0.05}$ & 2.815 & 0.15$^{+0.06}_{-0.05}$ & 0.959\ &0.706 &$0.78^{+0.02}_{-0.08}$ & Burst & 0.091$\pm$0.090 & 0.35$^{+0.18}_{-0.35}$ & 0.621 & 0.37$^{+0.16}_{-0.30}$ & 0.613\ &0.846 &$0.84^{+0.05}_{-0.04}$ & burSt & 0.181 & 0.02$^{+0.06}_{-0.02}$ & 1.225 & 0.06$^{+0.08}_{-0.06}$ & 1.105\ &1.118 &$1.00^{+0.02}_{-0.003}$ & burSt & 0.064$\pM$0.027 & 0.12$^{+1.50}_{-0.12}$ & 2.448 & 0.14$^{+1.40}_{-0.14}$ & 2.266\ &2

0.064$\pm$0.012 & 0.12$^{ +0.18}_{-0 .12}$ &0.2 28 & 0 .07$ ^{+0.16}_{-0.0 7 }$ & 0.239\ &0.835 &$0.87^ {+0.0 7} _ {-0. 1 2} $ & b urst &0 .0 9 1 $\p m$ 0. 090 & 0. 17$^{ +0. 11}_{-0 .17}$ & 0. 146 & 0.17$^{+0.1 1 }_ {-0.17}$ & 0. 082\ &0.966&$0 .82^{+ 0. 16} _ {-0.1 8}$ & bu rst &0 .006$\ pm$0.122&1 .60$^{ + 0.05}_{ - 1 .2 7}$& 0.683 & 0.38$^{ + 0. 2 5}_{-0.24}$ &0.831\ & 1 .6 0 0 &$ 2.1 8^{+0.31}_ {- 0.40} $ & Sa & 0. 0 1 2 $\p m $0.006 & 0.90 $^{+0.17}_{ - 0.2 0}$ &0. 823 & 0.96 $^{+0 .0 5 }_{ -0.23}$ & 0 .773 \ &0.433&$0.42 ^ {+0.003 } _{-0.01 }$ & b urs t & 0.2 5 5&0.1 5$ ^ {+0 . 04 }_{ - 0.0 5}$ & 2. 81 5& 0.1 5$^{ + 0 . 0 6}_{ -0. 05}$ & 0. 959\ &0.706 & $0. 78^{ + 0.0 2}_{- 0.08} $ &bu rst & 0.091 $\pm$ 0. 090 & 0.35$^{+0 .18} _{-0.35}$ &0. 621 & 0.37 $ ^{+0.1 6}_ {-0 .30}$ & 0.613\ &0. 84 6 & $0 .84^{+0.05}_{-0.04 }$ & b urst & 0 .181 & 0. 02 $ ^{+0.06} _{ -0. 02}$ & 1.22 5 &0 .0 6$^{+0.0 8}_{-0 . 06 }$ & 1.10 5\ &1.11 8&$1 .00 ^{+0. 0 2}_{ -0.003 }$ & bur st &0 .064$\pm$0.027 & 0.12$^{+1.5 0 }_ { - 0. 1 2}$& 2 .448 & 0.14 $^{+ 1 .40} _{-0 . 14 }$& 2.26 6\ &2

0.064$\pm$0.012_& 0.12$^{+0.18}_{-0.12}$_& 0.228 & 0.07$^{+0.16}_{-0.07}$_& 0.239\ &0.835_&$0.87^{+0.07}_{-0.12}$_& burst_&_0.091$\pm$0.090 & 0.17$^{+0.11}_{-0.17}$_& 0.146 &_0.17$^{+0.11}_{-0.17}$ & 0.082\ &0.966 &$0.82^{+0.16}_{-0.18}$_& burst &_0.006$\pm$0.122_& 1.60$^{+0.05}_{-1.27}$ & 0.683 & 0.38$^{+0.25}_{-0.24}$ & 0.831\ &1.600 &$2.18^{+0.31}_{-0.40}$ & Sa & 0.012$\pm$0.006 &_0.90$^{+0.17}_{-0.20}$_& 0.823_&_0.96$^{+0.05}_{-0.23}$_& 0.773\ &0.433 &$0.42^{+0.003}_{-0.01}$ & burst_& 0.255 & 0.15$^{+0.04}_{-0.05}$ &_2.815 &_0.15$^{+0.06}_{-0.05}$ & 0.959\ &0.706 &$0.78^{+0.02}_{-0.08}$ & burst & 0.091$\pm$0.090_&_0.35$^{+0.18}_{-0.35}$ & 0.621_& 0.37$^{+0.16}_{-0.30}$ & 0.613\ &0.846 &$0.84^{+0.05}_{-0.04}$ & burst & 0.181_& 0.02$^{+0.06}_{-0.02}$ & 1.225 & 0.06$^{+0.08}_{-0.06}$_& 1.105\ &1.118 &$1.00^{+0.02}_{-0.003}$_&_burst_& 0.064$\pm$0.027 & 0.12$^{+1.50}_{-0.12}$_& 2.448 & 0.14$^{+1.40}_{-0.14}$ & 2.266\ &2

a look on the general form for the scattering matrix. The general form for diagrams contributing to the scattering matrix is $$\begin{split} {\cal M} &\sim \Big(A+B q^2 + \ldots + (\alpha_1 \kappa^2 + \alpha_2 e^2) \frac{1}{q^2} \\ &+\beta_1 e^2\kappa^2\ln(-q^2) + \beta_2 e^2\kappa^2 \frac{m}{\sqrt{-q^2}} + \ldots\Big) \end{split}$$ where $A, B, \ldots$ correspond to the local analytical interactions and $\alpha_1, \alpha_2$ and $\beta_1, \beta_2, \ldots$ correspond to the leading non-analytical, non-local, long range interactions. The space parts of the non-analytical terms Fourier transform as, $$\begin{split} \int \frac {d^3 k}{(2\pi)^3} \frac 1{|{\bf k}|^2}e^{i\bf{k}\cdot\bf{r}}& = \frac {1}{4\pi r}\\ \int \frac {d^3 k}{(2\pi)^3} \frac 1{|{\bf k}|}e^{i\bf{k}\cdot\bf{r}}& = \frac {1}{2\pi^2 r^2}\\ \int \frac {d^3 k}{(2\pi)^3} \ln({\bf k}^2)e^{i\bf{k}\cdot\bf{r}}& = \frac {-1}{2\pi r^3}\\ \end{split}$$ so clearly these terms will contribute to the long range corrections. The non-analytical contribution, corresponding to the $\frac{1}{q^2}$ part, gives as seen the Newtonian and Coulomb potentials respectively. The other non-analytical contributions generate the leading quantum and classical corrections to the Coulomb and Newtonian potentials in powers of $\frac1{r}$. It is necessary to have non-analytic contributions in the matrix element, to ensure that the S-matrix is unitary. The analytic contributions will not be considered in this work. As noted previously these corrections correspond to local interactions, and are thus only needed for the high energy manifestation of the

a look on the general form for the scattering matrix. The cosmopolitan human body for diagrams contributing to the scatter matrix is $ $ \begin{split } { \cal M } & \sim \Big(A+B q^2 + \ldots + (\alpha_1 \kappa^2 + \alpha_2 e^2) \frac{1}{q^2 } \\ & + \beta_1 e^2\kappa^2\ln(-q^2) + \beta_2 e^2\kappa^2 \frac{m}{\sqrt{-q^2 } } + \ldots\Big) \end{split}$$ where $ A, b-complex vitamin, \ldots$ correspond to the local analytical interaction and $ \alpha_1, \alpha_2 $ and $ \beta_1, \beta_2, \ldots$ equate to the leading non - analytical, non - local, long range interaction. The space parts of the non - analytic term Fourier transform as, $ $ \begin{split } \int \frac { d^3 k}{(2\pi)^3 } \frac 1{|{\bf k}|^2}e^{i\bf{k}\cdot\bf{r } } & = \frac { 1}{4\pi r}\\ \int \frac { d^3 k}{(2\pi)^3 } \frac 1{|{\bf k}|}e^{i\bf{k}\cdot\bf{r } } & = \frac { 1}{2\pi^2 r^2}\\ \int \frac { d^3 k}{(2\pi)^3 } \ln({\bf k}^2)e^{i\bf{k}\cdot\bf{r } } & = \frac { -1}{2\pi r^3}\\ \end{split}$$ so clearly these terms will contribute to the long scope corrections. The non - analytical contribution, corresponding to the $ \frac{1}{q^2}$ share, gives as seen the Newtonian and Coulomb potentials respectively. The other non - analytical contribution generate the leading quantum and classical corrections to the Coulomb and Newtonian potentials in power of $ \frac1{r}$. It is necessary to have non - analytic contributions in the matrix element, to ensure that the S - matrix is unitary. The analytic contribution will not be view in this work. As noted previously these corrections correspond to local interactions, and are thus only needed for the eminent energy manifestation of the

a pook on the general form for the scattetibg matcix. The feneral worm for diagrams contributiig ti the scattering matrix is $$\cegin{splin} {\cal M} &\sin \Bij(A+B q^2 + \ldots + (\al'ga_1 \kappa^2 + \alpgw_2 e^2) \yrec{1}{q^2} \\ &+\beta_1 e^2\kappa^2\kn(-q^2) + \beta_2 a^2\kappa^2 \frac{m}{\sqst{-d^2}} + \ldots\Big) \end{split}$$ where $A, B, \ldots$ cjrrespomd to the local wnaljtycal pnueractions and $\alpha_1, \alpha_2$ and $\befa_1, \beta_2, \ldots$ correslond to the leading non-anapytifal, non-local, long gange interqctijbs. The space oarts of the non-analytjcal terms Fourier transform as, $$\begiu{split} \int \ftcx {d^3 n}{(2\pi)^3} \frac 1{|{\bf k}|^2}e^{i\bs{k}\cdot\bf{r}}& = \fvsc {1}{4\pi s}\\ \int \frsc {d^3 k}{(2\pi)^3} \frac 1{|{\nf k}|}e^{m\bf{k}\xdot\bf{r}}& = \frac {1}{2\pi^2 r^2}\\ \int \frac {d^3 k}{(2\pi)^3} \ln({\bf k}^2)g^{i\bf{k}\cdot\bx{r}}& = \frac {-1}{2\pi r^3}\\ \end{splut}$$ so cnearny tfwse tedmx sill clntcibute to tge long rante corrections. The npn-wbalytical confributyog, corresponding to the $\frac{1}{q^2}$ part, gives as seen the Newtonian and Coulomb potentials rgspectivelr. The other non-analytical contributions generate dhe lxaainy quangym and classical corrections to the Coulomb and Heetpnian potentiajs in powerx lf $\srac1{r}$. It is ngcessarv tk have non-analytic contrifutiobs in the mattix element, to ensure that rhe S-matrix ps ubitary. The analytic contributiuns eill mot be considered in thns wori. As noted oreviouslg these correctiovs borraspond to local interactiogs, and arx thux only veedgd for ehe high ejergy manifestation of tje

a look on the general form for matrix. general form diagrams contributing to {\cal &\sim \Big(A+B q^2 \ldots + (\alpha_1 + \alpha_2 e^2) \frac{1}{q^2} \\ &+\beta_1 + \beta_2 e^2\kappa^2 \frac{m}{\sqrt{-q^2}} + \ldots\Big) \end{split}$$ where $A, B, \ldots$ correspond to local analytical interactions and $\alpha_1, \alpha_2$ and $\beta_1, \beta_2, \ldots$ correspond to the non-analytical, long interactions. space parts of the non-analytical terms Fourier transform as, $$\begin{split} \int \frac {d^3 k}{(2\pi)^3} \frac 1{|{\bf = \frac {1}{4\pi r}\\ \int \frac {d^3 k}{(2\pi)^3} 1{|{\bf k}|}e^{i\bf{k}\cdot\bf{r}}& = \frac r^2}\\ \int \frac {d^3 k}{(2\pi)^3} k}^2)e^{i\bf{k}\cdot\bf{r}}& \frac {-1}{2\pi \end{split}$$ clearly terms will contribute the long range corrections. The non-analytical contribution, corresponding to the $\frac{1}{q^2}$ part, gives as seen the Newtonian Coulomb potentials other non-analytical generate leading and classical corrections Coulomb and Newtonian potentials in powers is necessary to have non-analytic contributions in the element, to that the S-matrix is unitary. The contributions will not be considered in this work. noted previously these corrections correspond to local interactions, and are thus only needed for the manifestation of the

a look on the general form for tHe scatteriNg matRix. the GeNeraL forM for diagrams coNTribUting to the scattering maTrix iS $$\bEGin{sPLiT} {\cal M} &\Sim \Big(A+b Q^2 + \lDOTs + (\aLpHa_1 \KapPa^2 + \ALpHa_2 e^2) \frAc{1}{q^2} \\ &+\Beta_1 e^2\kaPpa^2\ln(-q^2) + \beta_2 E^2\kaPpA^2 \frac{m}{\sqrt{-q^2}} + \lDOtS\Big) \end{splIt}$$ wHere $A, B, \ldots$ cOrrEspond To The LOcal aNalYticaL interACtions And $\alpha_1, \aLpHA_2$ and $\beTA_1, \beta_2, \ldOTS$ cOrreSpond to the leading NOn-ANalytical, non-loCal, lonG rANgE INteRacTions. The spAcE partS Of the noN-AnALYTicAL terms Fourier Transform as, $$\BEgiN{split} \InT \frAC {d^3 k}{(2\pi)^3} \fRac 1{|{\bf K}|^2}e^{I\Bf{k}\Cdot\bf{r}}& = \frac {1}{4\Pi r}\\ \iNt \frac {d^3 k}{(2\pI)^3} \frac 1{|{\bF K}|}e^{i\bf{k}\cDOt\bf{r}}& = \frAc {1}{2\pi^2 r^2}\\ \iNt \fRac {D^3 k}{(2\pi)^3} \LN({\bF k}^2)E^{i\bF{k}\CDot\BF{r}}& = \FraC {-1}{2\Pi r^3}\\ \End{split}$$ So ClEarly ThesE TERMs wiLl cOntrIbute To the long rangE coRrecTIonS. The nOn-anaLytiCaL contRibutiOn, corReSponding to the $\frAc{1}{q^2}$ pArt, gives aS seEn The neWtoniAN and CoUloMb pOtentiaLs respeCTivElY. tHE oTher non-analytical cOnTRIbUtions geNerate THe LeADing quanTuM anD claSSIcal cOrreCTiOns to the coulomB AnD NEwtoniaN pOtentiAlS in PowErs of $\FRac1{r}$. it is neCessary tO have NOn-analytic contRIbutions in the MAtRIX eLEmenT, to Ensure that tHe S-mATrix Is unITaRy. THE analYtic cOnTRiBUtions will not be consIdEred in This wOrk. As noted preViously theSE COrrectioNs coRReSPond to local intEractIons, and are THus only nEeded For the hiGh energy mANIfestatiOn oF thE

a look on the general for m for thescatt eri ngma trix . T he general for m for diagrams contributing to t he scat t er ing m atrix i s $ $ \ beg in {s pli t} {\ cal M } & \sim \B ig(A+B q^2 +\l dots + (\alp h a_ 1 \kappa^2 +\alpha_2 e^2 ) \ frac{1 }{ q^2 } \\ & +\b eta_1 e^2\k a ppa^2\ ln(-q^2)+\ beta_2 e^2\kap p a ^2 \fr ac{m}{\sqrt{-q^2} } + \ldots\Big) \e nd{spl it } $$ w her e $ A, B, \ldo ts $ cor r espondt ot h e lo c al analytical interactio n s a nd $\a lp ha_ 1 , \alp ha_2$ a n d $ \beta_1, \b eta_ 2, \ldots $ corr e spond t o the le adingnon -an alyt i ca l, no n- l oca l ,lon g ra nge inte ra ct ions. Th e s p acepar ts o f the non-analytic alterm s Fo urier tran sfor mas, $ $\begi n{spl it } \int \frac {d ^3 k }{(2\pi)^ 3}\f rac 1 {|{\b f k}|^2 }e^ {i\ bf{k}\c dot\bf{ r }}& = \ f ra c {1}{4\pi r}\\ \i nt \ fr ac {d^3k}{(2\ p i) ^3 } \frac 1 {| {\b f k} | } e^{i\ bf{k } \c dot\bf{r }}& =\ fr ac {1}{2\ pi ^2 r^2 }\ \ \ int \fra c {d^ 3 k}{( 2\pi)^3} \ln( { \bf k}^2)e^{i\ b f{k}\cdot\bf{ r }} & =\ frac {- 1}{2\pi r^3 }\\\ end{ spli t }$ $ s o clea rly t he s et erms will contribut eto the long range correc tions. Th e n on-analy tica l c o ntribution, co rresp onding tot he $\fra c{1}{ q^2}$ pa rt, gives a s seen t heNew ton ian a nd Coulomb pote n t ials r especti vel y. Theoth ernon -an al ytical co ntributi on sge ne rat e the leadingqu ant um an d cla s sicalcorre ctio ns t o th e Coulo m ba n d Ne wt on ianpot en tials inp owe rs of $ \frac1{r} $.I t is n ec essaryto have non-a na lytic cont ri but ions i n the matr ix element, to ensure t h at theS-m atrix isunitary. Th e anal yti c contr ibutio ns wi ll no t be co n s id ere din this wo r k . A s not ed pre viously these corrections cor respond to lo cal int e r ac tio n s, and a r e t h u s only needed f or the hig he ne rgy manife s tat io n of th e

a_look on_the general form for_the scattering_matrix. The_general form_for_diagrams contributing to_the scattering matrix_is $$\begin{split} {\cal M} &\sim_\Big(A+B q^2 +_\ldots_+ (\alpha_1 \kappa^2 + \alpha_2 e^2) \frac{1}{q^2} \\ &+\beta_1 e^2\kappa^2\ln(-q^2) + \beta_2 e^2\kappa^2 \frac{m}{\sqrt{-q^2}} +_\ldots\Big) \end{split}$$_where $A,_B,_\ldots$_correspond to the local analytical_interactions and $\alpha_1, \alpha_2$ and_$\beta_1, \beta_2,_\ldots$ correspond to the leading non-analytical, non-local, long_range_interactions. The space parts_of the non-analytical terms Fourier transform as, $$\begin{split} \int \frac_{d^3 k}{(2\pi)^3} \frac 1{|{\bf k}|^2}e^{i\bf{k}\cdot\bf{r}}& =_\frac {1}{4\pi r}\\ \int_\frac_{d^3_k}{(2\pi)^3} \frac 1{|{\bf k}|}e^{i\bf{k}\cdot\bf{r}}&_= \frac {1}{2\pi^2 r^2}\\ \int \frac {d^3_k}{(2\pi)^3} \ln({\bf k}^2)e^{i\bf{k}\cdot\bf{r}}& = \frac {-1}{2\pi_r^3}\\ \end{split}$$ so clearly these terms will contribute_to the long range corrections. The non-analytical_contribution, corresponding to the $\frac{1}{q^2}$_part, gives_as seen the Newtonian and_Coulomb potentials respectively._The other_non-analytical contributions generate_the leading quantum and classical corrections_to the Coulomb_and Newtonian potentials in powers of_$\frac1{r}$._It is necessary_to_have_non-analytic contributions_in the matrix_element,_to ensure_that_the S-matrix is unitary. The analytic contributions_will_not be considered in this work. As_noted previously these corrections_correspond_to local interactions, and_are thus only needed for_the high energy manifestation of the

$. For $k=1$, formula (\[eq:pieruno\]) is trivially true. Let us prove it directly for $k=2$. For each $h\geq 0$, let us split sum (\[eq:prepier\]) as: $$\begin{aligned} D_h(\epsilon^{i_1}\wedge\epsilon^{i_2})=\sum_{h_1+h_2=h}\epsilon^{i_1+h_1}\wedge\ep^{i_2+h_2} ={\cal P}+\overline{\cal P}.\label{eq:piercorr1}\end{aligned}$$ where $${\cal P}=\sum_{\matrix{_{i_1+h_1< i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2} \quad {\rm and}\quad \overline{\cal P}= \sum_{\matrix{_{i_1+h_1\geq i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2}.$$ One contends that $\overline{\cal P}$ vanishes. In fact, on the finite set of all integers $i_2-i_1\leq a\leq i_2-i_1+h$, define the bijection $\rho(a)=i_2-i_1+h-a$. Then: $$\begin{aligned} &{}&2\overline{\cal P}=\sum_{h_1=i_2-i_1}^h\ep^{i_1+h_1}\wedge\ep^{i_2+h-h_1}+ \sum_{h_1=i_2-i_1}^h\ep^{i_1+\rho(h_1)}\wedge\ep^{i_2+h-\rho(h_1)}=\\ &=&\sum_{h_1=i_2-i_1}^h\ep^{i_2+h-h_1}\wedge\ep^{i_1+h_1}-\sum_{h_1=i_2-i_1}^h\ep^{i_1+h_1}\wedge\ep^{i_2+h_2}=0,\

$. For $ k=1 $, formula   (\[eq: pieruno\ ]) is trivially true. Let us prove it immediately for $ k=2$. For each $ h\geq 0 $, lease us split sum   (\[eq: prepier\ ]) as: $ $ \begin{aligned } D_h(\epsilon^{i_1}\wedge\epsilon^{i_2})=\sum_{h_1+h_2 = h}\epsilon^{i_1+h_1}\wedge\ep^{i_2+h_2 } = { \cal P}+\overline{\cal P}.\label{eq: piercorr1}\end{aligned}$$ where $ $ { \cal P}=\sum_{\matrix{_{i_1+h_1 < i_2}\cr _ { h_1+h_2 = h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2 } \quad { \rm and}\quad \overline{\cal P}= \sum_{\matrix{_{i_1+h_1\geq i_2}\cr _ { h_1+h_2 = h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2}.$$ One contest that $ \overline{\cal P}$ vanishes. In fact, on the finite bent of all integers $ i_2 - i_1\leq a\leq i_2 - i_1+h$, define the bijection $ \rho(a)=i_2 - i_1+h - a$. Then: $ $ \begin{aligned } & { } & 2\overline{\cal P}=\sum_{h_1 = i_2 - i_1}^h\ep^{i_1+h_1}\wedge\ep^{i_2+h - h_1}+ \sum_{h_1 = i_2 - i_1}^h\ep^{i_1+\rho(h_1)}\wedge\ep^{i_2+h-\rho(h_1)}=\\ & = & \sum_{h_1 = i_2 - i_1}^h\ep^{i_2+h - h_1}\wedge\ep^{i_1+h_1}-\sum_{h_1 = i_2 - i_1}^h\ep^{i_1+h_1}\wedge\ep^{i_2+h_2}=0,\

$. Fog $k=1$, formula (\[eq:pieruno\]) is urivially true. Lej ys prote it djrectly wor $k=2$. For each $h\geq 0$, let us dpoit sym (\[eq:prepier\]) as: $$\begin{alkgned} D_h(\epdilon^{i_1}\weege\eksilon^{i_2})=\sum_{h_1+h_2=h}\epsilon^{i_1+h_1}\wednz\ep^{i_2+h_2} ={\dwl P}+\mterline{\cal P}.\labgl{eq:piercorr1}\and{aligned}$$ whese $${\ccl P}=\sum_{\matrix{_{i_1+h_1< i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2} \quad {\rk wnd}\quad \overjine{\baj P}= \sum_{\matrix{_{i_1+h_1\geq i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{j_2+h_2}.$$ One cmntends that $\pverline{\cal P}$ vanishes. In vact, on the finite set of all intgferf $i_2-i_1\leq a\leq k_2-i_1+h$, define the bijectikn $\rho(a)=i_2-i_1+h-a$. Then: $$\begin{aligned} &{}&2\ovdrlinz{\cal P}=\sum_{h_1=i_2-u_1}^h\wp^{i_1+j_1}\fedge\ep^{i_2+h-h_1}+ \snm_{h_1=i_2-i_1}^h\vp^{i_1+\rho(h_1)}\wedge\ei^{p_2+h-\rho(h_1)}=\\ &=&\vum_{h_1=i_2-i_1}^h\rp^{i_2+h-h_1}\wedge\ep^{i_1+h_1}-\xum_{i_1=i_2-i_1}^h\wp^{i_1+h_1}\wedge\ep^{i_2+h_2}=0,\

$. For $k=1$, formula (\[eq:pieruno\]) is trivially us it directly $k=2$. For each sum as: $$\begin{aligned} D_h(\epsilon^{i_1}\wedge\epsilon^{i_2})=\sum_{h_1+h_2=h}\epsilon^{i_1+h_1}\wedge\ep^{i_2+h_2} P}+\overline{\cal P}.\label{eq:piercorr1}\end{aligned}$$ where P}=\sum_{\matrix{_{i_1+h_1< i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2} \quad {\rm and}\quad P}= \sum_{\matrix{_{i_1+h_1\geq i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2}.$$ One contends that $\overline{\cal P}$ vanishes. In fact, on finite set of all integers $i_2-i_1\leq a\leq i_2-i_1+h$, define the bijection $\rho(a)=i_2-i_1+h-a$. Then: &{}&2\overline{\cal \sum_{h_1=i_2-i_1}^h\ep^{i_1+\rho(h_1)}\wedge\ep^{i_2+h-\rho(h_1)}=\\

$. For $k=1$, formula (\[eq:pieruno\]) is triVially true. let us ProVe iT dIrecTly fOr $k=2$. For each $h\geq 0$, LEt us Split sum (\[eq:prepier\]) as: $$\begIn{aliGnED} D_h(\ePSiLon^{i_1}\wEdge\epsILoN^{I_2})=\Sum_{H_1+h_2=H}\ePsiLoN^{I_1+h_1}\Wedge\Ep^{i_2+H_2} ={\cal P}+\ovErline{\cal P}.\LabEl{Eq:piercorr1}\enD{AlIgned}$$ where $${\Cal p}=\sum_{\matrix{_{i_1+h_1< I_2}\cr _{H_1+h_2=h}}}\ep^{i_1+H_1}\wEdgE\Ep^{i_2+h_2} \qUad {\Rm and}\Quad \ovERline{\cAl P}= \sum_{\matRiX{_{I_1+h_1\geq i_2}\CR _{h_1+h_2=h}}}\ep^{i_1+H_1}\WEdGe\ep^{I_2+h_2}.$$ One contends that $\OVeRLine{\cal P}$ vanishEs. In faCt, ON tHE FinIte Set of all inTeGers $i_2-I_1\Leq a\leq I_2-I_1+h$, DEFIne THe bijection $\rhO(a)=i_2-i_1+h-a$. Then: $$\bEGin{AligneD} &{}&2\oVerLIne{\cal p}=\sum_{h_1=I_2-i_1}^H\Ep^{i_1+H_1}\wedge\ep^{i_2+h-h_1}+ \Sum_{h_1=I_2-i_1}^h\ep^{i_1+\rho(H_1)}\wedge\EP^{i_2+h-\rho(h_1)}=\\ &=&\SUm_{h_1=i_2-i_1}^h\eP^{i_2+h-h_1}\weDge\Ep^{i_1+H_1}-\sum_{H_1=I_2-i_1}^H\eP^{i_1+h_1}\WeDGe\eP^{I_2+h_2}=0,\

$. For $k=1$, formula (\[e q:pieruno\ ]) is tr ivi al ly t rue. Let us provei t di rectly for $k=2$. Foreach$h \ geq0 $, letus spli t s u m  (\ [e q: pre pi e r\ ]) as : $ $\begin {aligned}D_h (\ epsilon^{i_1 } \w edge\epsil on^ {i_2})=\sum_ {h_ 1+h_2= h} \ep s ilon^ {i_ 1+h_1 }\wedg e \ep^{i _2+h_2} = {\ c al P}+ \ overlin e { \c al P }.\label{eq:pierc o rr 1 }\end{aligned} $$ whe re $$ { \ cal P} =\sum_{\ma tr ix{_{ i _1+h_1< i_ 2 } \ cr_ {h_1+h_2=h}}} \ep^{i_1+h_ 1 }\w edge\e p^ {i_ 2 +h_2}\quad { \ rmand}\quad \o verline{\ cal P} = \sum_ { \matrix {_{i_1 +h_ 1\g eq i _ 2} \c r _ {h _ 1+h _ 2= h}} } \ep ^{i_1+h_ 1} \w edge\ ep^{ i _ 2 + h_2} .$$ On e con tends that $\ ove rlin e {\c al P} $ van ishe s. In f act, o n the f inite set of al l in tegers $i _2- i_ 1\l eq a\le q i_2-i _1+ h$, define the bi j ect io n $ \r ho(a)=i_2-i_1+h-a$ .T h en : $$\beg in{ali g ne d} &{}&2\ov er lin e{\c a l P}=\ sum_ { h_ 1=i_2-i_ 1}^h\e p ^{ i_ 1+h_1}\ we dge\ep ^{ i_2 +h- h_1}+ \sum _{h_1= i_2-i_1} ^h\ep ^ {i_1+\rho(h_1) } \wedge\ep^{i_ 2 +h - \ rh o (h_1 )}= \\ &=&\sum_ {h_1 = i_2- i_1} ^ h\ ep^ { i_2+h -h_1} \w e dg e \ep^{i_1+h_1}-\sum_ {h _1=i_2 -i_1} ^h\ep^{i_1+h_ 1}\wedge\e p ^ { i_2+h_2} =0,\

$. For_$k=1$, formula (\[eq:pieruno\])_is trivially true. Let_us prove_it_directly for_$k=2$._For each $h\geq_0$, let us_split sum (\[eq:prepier\]) as: $$\begin{aligned} D_h(\epsilon^{i_1}\wedge\epsilon^{i_2})=\sum_{h_1+h_2=h}\epsilon^{i_1+h_1}\wedge\ep^{i_2+h_2} ={\cal_P}+\overline{\cal P}.\label{eq:piercorr1}\end{aligned}$$ where_$${\cal_P}=\sum_{\matrix{_{i_1+h_1< i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2} \quad {\rm and}\quad \overline{\cal P}= \sum_{\matrix{_{i_1+h_1\geq i_2}\cr _{h_1+h_2=h}}}\ep^{i_1+h_1}\wedge\ep^{i_2+h_2}.$$ One_contends_that $\overline{\cal_P}$_vanishes._In fact, on the finite_set of all integers $i_2-i_1\leq_a\leq i_2-i_1+h$,_define the bijection $\rho(a)=i_2-i_1+h-a$. Then: $$\begin{aligned} &{}&2\overline{\cal P}=\sum_{h_1=i_2-i_1}^h\ep^{i_1+h_1}\wedge\ep^{i_2+h-h_1}+ \sum_{h_1=i_2-i_1}^h\ep^{i_1+\rho(h_1)}\wedge\ep^{i_2+h-\rho(h_1)}=\\ &=&\sum_{h_1=i_2-i_1}^h\ep^{i_2+h-h_1}\wedge\ep^{i_1+h_1}-\sum_{h_1=i_2-i_1}^h\ep^{i_1+h_1}\wedge\ep^{i_2+h_2}=0,\

Tutukov]{}, A. V. 1984,, 54, 335 , A. & [Vennes]{}, S. 2009,, 506, L25 , M., [Brown]{}, W. R., [Allende Prieto]{}, C., & [Kenyon]{}, S. J. 2010,, 716, 122 , H. A., [Charbonneau]{}, D., [Noyes]{}, R. W., [Brown]{}, T. M., & [Gilliland]{}, R. L. 2007,, 655, 564 , S. R. & [van Kerkwijk]{}, M. H. 2010,, in press, arXiv:1003.2169 , C., [Camilo]{}, F., [Wex]{}, N., [Kramer]{}, M., [Backer]{}, D. C., [Lyne]{}, A. G., & [Doroshenko]{}, O. 2001,, 326, 274 , J. W., [Kim]{}, S., [Kim]{}, C., [Koch]{}, R. H., [Lee]{}, C., [Kim]{}, H., & [Park]{}, J. 2009,, 137, 3181 , J., [Bergeron]{}, P., & [Holberg]{}, J. B. 2005,, 156, 47 , A. 2005,, 623, L45 , D. R. & [Kramer]{}, M. 2004, [Handbook of Pulsar Astronomy]{} (Cambridge: Cambridge Univ. Press) , T. R., [Dhillon]{}, V. S., & [Duck]{}, S. R. 1995,, 275, 828 , T. R., [Gaensicke]{}, B. T., [Steeghs]{}, D., [Southworth]{}, J., [Koester]{}, D., [Harris]{}, V., & [Merry]{}, L. 2010,, in press, arXiv:1002.4677 , J. 2002,, 564, 1019 , F., [Badenes]{}, C., [Thompson]{}, S. E., & [Lupton]{}, R. 2009,, 707, L51 , G. 2009, Class. Quantum Grav., 26, 094030 , G., [et al.]{}

Tutukov ] { }, A.   V. 1984, , 54, 335 , A. & [ Vennes ] { }, S. 2009, , 506, L25 , M., [ Brown ] { }, W.   R., [ Allende Prieto ] { }, C., & [ Kenyon ] { }, S.   J. 2010, , 716, 122 , H.   A., [ Charbonneau ] { }, D., [ Noyes ] { }, R.   W., [ Brown ] { }, T.   M., & [ Gilliland ] { }, R.   L. 2007, , 655, 564 , S.   R. & [ van Kerkwijk ] { }, M.   H. 2010, , in press, arXiv:1003.2169 , C., [ Camilo ] { }, F., [ Wex ] { }, N., [ Kramer ] { }, M., [ Backer ] { }, D.   C., [ Lyne ] { }, A.   G., & [ Doroshenko ] { }, O. 2001, , 326, 274 , J.   W., [ Kim ] { }, S., [ Kim ] { }, C., [ Koch ] { }, R.   H., [ Lee ] { }, C., [ Kim ] { }, H., & [ Park ] { }, J. 2009, , 137, 3181 , J., [ Bergeron ] { }, P., & [ Holberg ] { }, J.   B. 2005, , 156, 47 , A. 2005, , 623, L45 , D.   R. & [ Kramer ] { }, M. 2004, [ Handbook of Pulsar Astronomy ] { } (Cambridge: Cambridge Univ.   Press) , T.   R., [ Dhillon ] { }, V.   S., & [ Duck ] { }, S.   R. 1995, , 275, 828 , T.   R., [ Gaensicke ] { }, B.   T., [ Steeghs ] { }, D., [ Southworth ] { }, J., [ Koester ] { }, D., [ Harris ] { }, V., & [ Merry ] { }, L. 2010, , in press, arXiv:1002.4677 , J. 2002, , 564, 1019 , F., [ Badenes ] { }, C., [ Thompson ] { }, S.   E., & [ Lupton ] { }, R. 2009, , 707, L51 , G. 2009, Class.   Quantum Grav. , 26, 094030 , G., [ et al. ] { }

Tutkkov]{}, A. V. 1984,, 54, 335 , A. & [Vennes]{}, S. 2009,, 506, L25 , M., [Brown]{}, W. R., [Allende 'rieto]{}, D., & [Kenyov]{}, S. J. 2010,, 716, 122 , H. A., [Charbonneau]{}, D., [Noyxs]{}, R. Q., [Broqn]{}, T. M., & [Gilliland]{}, R. L. 2007,, 655, 564 , S. R. & [van Nerkwijk]{}, M. H. 2010,, un press, acSiv:1003.2169 , C., [Camilo]{}, R., [Wex]{}, I., [Kramer]{}, M., [Backgr]{}, D. C., [Lyne]{}, A. C., & [Doroshenko]{}, M. 2001,, 326, 274 , J. W., [Kim]{}, S., [Kim]{}, C., [Koch]{}, R. H., [Lee]{}, C., [Kim]{}, H., & [Park]{}, K. 2009,, 137, 3181 , J., [Bergeron]{}, K., & [Hokferg]{}, J. B. 2005,, 156, 47 , A. 2005,, 623, L45 , D. R. & [Kramer]{}, M. 2004, [Handbooi of Punsar Astronomu]{} (Cambridge: Cambridge Univ. Oresd) , T. R., [Dhillon]{}, V. S., & [Dkck]{}, S. R. 1995,, 275, 828 , T. T., [Gaqbsicke]{}, B. T., [Stdeghs]{}, D., [Solchworth]{}, J., [Kkester]{}, D., [Harris]{}, V., & [Merry]{}, L. 2010,, in oress, arXiv:1002.4677 , J. 2002,, 564, 1019 , D., [Bwgenes]{}, C., [Thonpson]{}, S. E., & [Lupton]{}, V. 2009,, 707, L51 , G. 2009, Class. Auantum Grav., 26, 094030 , G., [xt ao.]{}

Tutukov]{}, A. V. 1984,, 54, 335 , [Vennes]{}, 2009,, 506, , M., [Brown]{}, & S. J. 2010,, 122 , H. [Charbonneau]{}, D., [Noyes]{}, R. W., [Brown]{}, M., & [Gilliland]{}, R. L. 2007,, 655, 564 , S. R. & [van M. H. 2010,, in press, arXiv:1003.2169 , C., [Camilo]{}, F., [Wex]{}, N., [Kramer]{}, [Backer]{}, C., A. & [Doroshenko]{}, O. 2001,, 326, 274 , J. W., [Kim]{}, S., [Kim]{}, C., [Koch]{}, R. H., C., [Kim]{}, H., & [Park]{}, J. 2009,, 137, , J., [Bergeron]{}, P., [Holberg]{}, J. B. 2005,, 156, , 2005,, 623, , R. [Kramer]{}, M. 2004, of Pulsar Astronomy]{} (Cambridge: Cambridge Univ. Press) , T. R., [Dhillon]{}, V. S., & [Duck]{}, S. R. 275, 828 R., [Gaensicke]{}, T., D., J., [Koester]{}, D., & [Merry]{}, L. 2010,, in press, 2002,, 564, 1019 , F., [Badenes]{}, C., [Thompson]{}, E., & R. 2009,, 707, L51 , G. Class. Quantum Grav., 26, 094030 , G., [et

Tutukov]{}, A. V. 1984,, 54, 335 , A. & [Vennes]{}, S. 2009,, 506, L25 , M., [Brown]{}, W. r., [Allende PrIeto]{}, C., & [kenYon]{}, s. J. 2010,, 716, 122 , h. A., [ChArboNneau]{}, D., [Noyes]{}, R. W., [BROwn]{}, T. m., & [Gilliland]{}, R. L. 2007,, 655, 564 , S. R. & [van KerkwIjk]{}, M. H. 2010,, In PRess, ARXIv:1003.2169 , C., [CaMilo]{}, F., [WeX]{}, n., [KRAMer]{}, m., [BAcKer]{}, d. C., [lYnE]{}, A. G., & [DoRosHenko]{}, O. 2001,, 326, 274 , J. w., [Kim]{}, S., [Kim]{}, C., [KOch]{}, r. H., [lee]{}, C., [Kim]{}, H., & [Park]{}, j. 2009,, 137, 3181 , j., [BErgeron]{}, P., & [HoLbeRg]{}, J. B. 2005,, 156, 47 , A. 2005,, 623, L45 , D. R. & [KramEr]{}, M. 2004, [handboOk Of PULsar AStrOnomy]{} (cambriDGe: CambRidge Univ. prESs) , T. R., [DhILlon]{}, V. S., & [DUCK]{}, S. r. 1995,, 275, 828 , T. R., [GAensicke]{}, B. T., [Steeghs]{}, d., [soUThworth]{}, J., [KoesteR]{}, D., [HarrIs]{}, v., & [meRRY]{}, L. 2010,, iN prEss, arXiv:1002.4677 , J. 2002,, 564, 1019 , F., [baDenes]{}, c., [thompsoN]{}, s. E., & [lUPTon]{}, r. 2009,, 707, l51 , G. 2009, Class. QuantuM Grav., 26, 094030 , G., [et al.]{}

Tutukov]{}, A. V. 1984,, 5 4, 335 ,A. &[Ve nne s] {},S. 2 009,, 506, L25 , M ., [Brown]{}, W. R., [ Allen de Prie t o] {}, C ., & [K e ny o n ]{} ,S.  J. 2 0 10 ,, 71 6,122 ,H. A., [Ch arb on neau]{}, D., [N oyes]{}, R . W ., [Brown]{} , T . M.,&[Gi l lilan d]{ }, R.  L. 20 0 7,, 65 5, 564 , S .  R. &[ van Ker k w ij k]{} , M. H. 2010,, in pr e ss, arXiv:1003 .2169 , C. , [Ca mil o]{}, F.,[W ex]{} , N., [K r am e r ] {}, M., [Backer]{ }, D. C., [ L yne ]{}, A .G., & [Dor oshen ko ] {}, O. 2001,,326, 274 , J . W.,[ Kim]{}, S., [Ki m]{},C., [K och] { }, R . H ., [Le e ]{ },C .,[Kim]{}, H ., & [P ark] { } , J. 2 009 ,, 1 37, 3 181 , J., [B erg eron ] {}, P.,& [Ho lber g] {}, J . B. 2 005,, 1 56, 47 , A. 20 05,, 623, L45 , D . R .& [Kr a mer]{} , M . 2 004, [H andbook ofPu l s a rAstronomy]{} (Camb ri d g e: Cambrid ge Uni v .Pr e ss) , T .R., [Dh i l lon]{ }, V .  S ., & [Du ck]{}, S.  R . 1995, ,275, 8 28 , T.  R.,[ Gaen sicke] {}, B. T ., [S t eeghs]{}, D.,[ Southworth]{} , J . , [ K oest er] {}, D., [Ha rris ] {},V.,& [ Mer r y]{}, L. 2 01 0 ,, in press, arXiv:100 2. 4677 , J.2002,, 564, 1 019 , F., [ B adenes]{ }, C . ,[ Thompson]{}, S . E., & [Lupton ] {}, R. 2 009,, 707, L5 1 , G. 2 0 0 9, Class . Q uan tum Gr a v ., 26, 094030 , G.,[e t al.]{ }

Tutukov]{}, A. V._1984,, 54,_335 , A. & [Vennes]{},_S. 2009,,_506,_L25 , M.,_[Brown]{},_W. R., [Allende Prieto]{},_C., & [Kenyon]{},_S. J. 2010,, 716, 122 ,_H. A., [Charbonneau]{}, D.,_[Noyes]{},_R. W., [Brown]{}, T. M., & [Gilliland]{}, R. L. 2007,, 655, 564 , S. R. & [van Kerkwijk]{}, M. H._2010,,_in press,_arXiv:1003.2169 ,_C.,_[Camilo]{}, F., [Wex]{}, N., [Kramer]{},_M., [Backer]{}, D. C., [Lyne]{}, A. G.,_& [Doroshenko]{},_O. 2001,, 326, 274 , J. W., [Kim]{}, S., [Kim]{},_C.,_[Koch]{}, R. H., [Lee]{},_C., [Kim]{}, H., & [Park]{}, J. 2009,, 137, 3181 ,_J., [Bergeron]{}, P., & [Holberg]{}, J. B._2005,, 156, 47 ,_A._2005,,_623, L45 , D. R. &_[Kramer]{}, M. 2004, [Handbook of Pulsar_Astronomy]{} (Cambridge: Cambridge Univ. Press) , T. R., [Dhillon]{},_V. S., & [Duck]{}, S. R. 1995,, 275, 828 ,_T. R., [Gaensicke]{}, B. T., [Steeghs]{}, D., [Southworth]{},_J., [Koester]{}, D., [Harris]{}, V.,_& [Merry]{},_L. 2010,, in press, arXiv:1002.4677 ,_J. 2002,, 564,_1019 , F.,_[Badenes]{}, C., [Thompson]{},_S. E., & [Lupton]{}, R. 2009,, 707,_L51 , G. 2009,_Class. Quantum Grav., 26, 094030 , G., [et_al.]{}

steps that were done to the invention of SUSY in Kharkov team headed by D.Volkov. This paper is dedicated to the memory of Prof. Yu. Gol’fand, whose ideas of SUSY inspired the most active developments in High Energy Physics over thirty years. author: - 'V. Akulov [^1]' title: 'Non-linear way to Supersymmetry and N-extended SUSY [^2] ' --- [*”Geometry of space is associated with mathematical group”*]{} Felix Klein, ”Erlagen Program” 1872 This year, the science community celebrates the 30th anniversary of SUSY. I have been asked to give a short historical introduction to the first steps in this direction that were done by our group headed by Prof. Dmitry Volkov. Rochester’s High Energy Conference was held in Kiev (Ukraine, Soviet Union) in 1970. Prof. Yu. Golfand announced 2 reports for this conference, but the Org. Committee gave him time only for one. Prof. Yu. Gol’fand preferred to discuss the problem of vacuum in QED, because the problem of Superalgebra Poincaré, which was obtained already, seemed to him very complicated for a first discussion. But the abstract of this report, with Superalgebra Poincaré, was published in the Rotaprint edition of the Proceeding [@/1/]. It was the first publication about the superalgebra Poincaré. Unfortunately, in the final version of Proceedings did not appear this abstracts. But the first attempts of the introduction of superalgebras in physics was given in paper by G.Stavraki [@/2/] in 1966 and H.Miyazava [@/3/] in 1968. The mathematical background for supersymmetry was constructed in 1970. Felix Berezin(Moscow) and Gregory Kats (Kiev) [@/4/] published the paper about the groups with a commuting and anticommuting parameters in the Russian journal ”Mathematicheskiy Zbornik” – that is how the supergroups appeared in mathematics. Before that Prof. D. Volkov investigated the connection between spin and statistics and rediscovered (after M.Green,1953) [@/5/] the parastatistics in 1959. Later he considered fermionic Regge trajectory. The success of the application of Goldstone’s Theorem [@/6

steps that were done to the invention of SUSY in Kharkov team headed by D.Volkov. This paper is dedicate to the memory of Prof. Yu. Gol’fand, whose idea of SUSY inspired the most active developments in High Energy Physics over thirty long time. author: -' V. Akulov [ ^1 ]' title:' Non - analogue way to Supersymmetry and N - extended SUSY [ ^2 ]' --- [ * ” Geometry of space is associate with mathematical group ” * ] { } Felix Klein, ” Erlagen Program ” 1872 This year, the skill community celebrates the 30th anniversary of SUSY. I have been asked to pass a short historical introduction to the first steps in this guidance that were done by our group headed by Prof. Dmitry Volkov. Rochester ’s High Energy Conference was held in Kiev (Ukraine, Soviet Union) in 1970. Prof. Yu. Golfand announce 2 reports for this conference, but the Org. Committee gave him time merely for one. Prof. Yu. Gol’fand preferred to discuss the problem of vacuum in QED, because the problem of Superalgebra Poincaré, which was prevail already, seemed to him very complicated for a first discussion. But the abstract of this report, with Superalgebra Poincaré, was published in the Rotaprint edition of the Proceeding   [ @/1/ ]. It was the first publication about the superalgebra Poincaré. Unfortunately, in the final version of Proceedings did not appear this abstracts. But the first attempts of the introduction of superalgebras in physics was given in composition by G.Stavraki [ @/2/ ] in 1966 and H.Miyazava [ @/3/ ] in 1968. The mathematical background for supersymmetry was construct in 1970. Felix Berezin(Moscow) and Gregory Kats (Kiev) [ @/4/ ] publish the paper about the groups with a commuting and anticommuting parameters in the Russian journal ” Mathematicheskiy Zbornik ” – that is how the supergroups appeared in mathematics. Before that Prof. D. Volkov investigated the association between spin and statistics and rediscovered (after M.Green,1953) [ @/5/ ] the parastatistics in 1959. Later he considered fermionic Regge trajectory. The success of the application of Goldstone ’s Theorem [ @/6

stfps that were done to tht invention of SUSY in Kiarkov feam heaaed by D.Volkov. This paper iw deducated to the memory ow Prof. Yu. Gol’fand, whoww ideas of SUSY inspired bhe mmwt active devekopments it High Energy [hhsncs over thirty years. author: - 'V. Akulov [^1]' title: 'Moj-linear way to Suptrsrmmefgy and N-extended SUSY [^2] ' --- [*”Geometry of spabe is associated eith mathematical group”*]{} Fellx Kpein, ”Erlagen Progrwm” 1872 This yeat, thq science comounity celtbxates the 30tg anniversary of SUSY. I have bedn asled to givg z dvort histormcal igtroduction bp the xirst syeps in this dlrectmon rhat were done by our group headed by Ptof. Dmitry Vklkov. Rochester’s Hugy Enetgy Cmnfefwncd wzs hsld in Kixv (Ukraine, Aoviet Uniob) in 1970. Prof. Yu. Golfamd qnnounced 2 relorts sow this conference, but the Org. Committee gabe him time only for onw. Prof. Yu. Gol’fand preverred to discuss the problem of vacuum in QED, because the probmdm if Rypfralgebra Poincaré, which was obtained already, fsekec to him very gomplicated for a gigsy discussion. Bot the cgsfract of this repogt, with Supeealgebra Koincsré, was published in the Rotqprint editijb of the Proceediny [@/1/]. It was the yirst kublicstion about the superalyebra Loincaré. Unflrtunatelg, in the final vefsipn of Proctddings did not ap[ear this abscracts. Bug thg first attempts lf thc introduction of skperapgabras in pjysics was given in paper by G.Svevraki [@/2/] in 1966 amd H.Mpyazava [@/3/] nn 1968. The mathematical background fot supersyimetrh was consnructed ii 1970. Felix Berqzin(Moscow) ang Gregory Katv (Kiev) [@/4/] [ublushee the pxoer about the broups winh a cimmuting and anticpmmotjng parameters nu uhw Russian jourmal ”Maehvmavichefniy Zbornik” – thag ir how ghe suptxgrijps sppeared in mathemathcs. Bsfore that Prof. D. Fojkov invgstigated the connectipn between spin anf stavisticr and reqiscovered (after M.Green,1953) [@/5/] the pzrastatishicf in 1959. Later re cinsidered fexmionic Regge trajectory. The success of vhe application of Goldwtone’s Theorem [@/6

steps that were done to the invention in team headed D.Volkov. This paper of Yu. Gol’fand, whose of SUSY inspired most active developments in High Energy over thirty years. author: - 'V. Akulov [^1]' title: 'Non-linear way to Supersymmetry N-extended SUSY [^2] ' --- [*”Geometry of space is associated with mathematical group”*]{} Klein, Program” This the science community celebrates the 30th anniversary of SUSY. I have been asked to give a historical introduction to the first steps in this that were done by group headed by Prof. Dmitry Rochester’s Energy Conference held Kiev Soviet Union) in Prof. Yu. Golfand announced 2 reports for this conference, but the Org. Committee gave him time only one. Prof. preferred to the of in QED, because of Superalgebra Poincaré, which was obtained him very complicated for a first discussion. But abstract of report, with Superalgebra Poincaré, was published the Rotaprint edition of the Proceeding [@/1/]. It the first publication about the superalgebra Poincaré. Unfortunately, in the final version of Proceedings did this abstracts. But the attempts of the of in was in paper G.Stavraki [@/2/] in 1966 and H.Miyazava [@/3/] in 1968. The mathematical for supersymmetry was constructed in 1970. Felix Berezin(Moscow) and Gregory [@/4/] the paper about groups with a commuting anticommuting in the Russian journal – is appeared mathematics. that Prof. D. Volkov the connection between spin and and rediscovered (after M.Green,1953) Later he considered fermionic Regge trajectory. The success the application of Goldstone’s Theorem [@/6

steps that were done to the invEntion of SUsY in KHarKov TeAm heAded By D.Volkov. This pAPer iS dedicated to the memory oF Prof. yu. gOl’faND, wHose iDeas of SusY INSpiReD tHe mOsT AcTive dEveLopmentS in High EneRgy phYsics over thiRTy Years. authoR: - 'V. AKulov [^1]' title: 'NoN-liNear waY tO SuPErsymMetRy and n-extenDEd SUSY [^2] ' --- [*”geometry oF sPAce is aSSociateD WItH matHematical group”*]{} FelIX KLEin, ”Erlagen ProgRam” 1872 ThiS yEAr, THE scIenCe communitY cElebrATes the 30tH AnNIVErsARy of SUSY. I have Been asked to GIve A short HiStoRIcal inTroduCtIOn tO the first stEps iN this direCtion tHAt were dONe by our Group hEadEd bY ProF. dmItRy VOlKOv. ROChEstER’s HIgh EnergY COnFerenCe waS HELD in KIev (ukraIne, SoViet Union) in 1970. PrOf. YU. GolFAnd AnnouNced 2 rEporTs For thIs confErencE, bUt the Org. CommittEe gaVe him time OnlY fOr oNe. prof. YU. gol’fanD prEfeRred to dIscuss tHE prObLEM Of Vacuum in QED, because ThE PRoBlem of SuPeralgEBrA POIncaré, whIcH waS obtAINed alReadY, SeEmed to hiM very cOMpLiCated foR a First dIsCusSioN. But tHE absTract oF this repOrt, wiTH Superalgebra POIncaré, was publIShED In THe RoTapRint edition Of thE procEediNG [@/1/]. IT waS The fiRst puBlICaTIon about the superalgEbRa PoinCaré. UNfortunately, iN the final vERSIon of ProCeedINgS Did not appear thIs absTracts. But tHE first atTemptS of the inTroductioN OF superalGebRas In pHysICS wAs given in papeR BY G.StAvRaki [@/2/] in 1966 aNd H.miyazavA [@/3/] in 1968. the MatHemAtIcal backgRound for SuPeRsYmMetRy was COnstructEd In 1970. FElIx BEreziN(moscow) And GrEgorY KAtS (kieV) [@/4/] publisHEd THE papEr AbOut tHe gRoUps wiTh a cOMmuTing and AnticommuTinG ParaMeTeRs in the russian journaL ”MAthematichEsKiy zborniK” – THat is how The supergroups appeared iN MathemaTicS. BefoRe thAt Prof. D. VoLkoV invesTigATed the ConnecTion bEtWeeN SPin anD STaTisTiCs and redisCOVerEd (aftEr m.GreEn,1953) [@/5/] the paRastatistics in 1959. LateR He cOnsidered fermIonIc ReGGE tRajECtORy. THe SUccESS of the applicatiOn of GoldstOnE’S THeorem [@/6

steps that were done to t he inventi on of SU SYin Kha rkov team headed b y D.V olkov. This paper is d ed i cate d t o the memory of P rof .Yu . G ol ’ fa nd, w hos e ideas of SUSY i nsp ir ed the mosta ct ive develo pme nts in HighEne rgy Ph ys ics overthi rty y ears.a uthor: - 'V. Ak ul o v [^1] ' title: ' No n-li near way to Super s ym m etry and N-ext endedSU S Y[ ^ 2]' - -- [*”Geo me try o f spacei sa s s oci a ted with math ematical gr o up” *]{} Fe lix Klein, ”Erl ag e n P rogram” 187 2 T his year, the s c ience c o mmunity celeb rat esthe3 0t hann iv e rsa r yofS USY . I have b ee n ask ed t o g i ve a sh orthisto rical introdu cti on t o th e fir st st epsin this direc tionth at were done by our group he ade dbyPr of. D m itry V olk ov. Roche ster’sH igh E n e r gy Conference was he ld i nKiev (Uk raine, So vi e t Union) i n 1 970. P rof.Yu.G ol fand ann ounced 2re ports f or thisco nfe ren ce, b u t th e Org. Committ ee ga v e him time onl y for one. Pro f .Y u .G ol’f and preferredto d i scus s th e p rob l em of vacu um in QED, because the pr ob lem of Supe ralgebra Poin caré, whic h w as obtai neda lr e ady, seemed to himvery compl i cated fo r a f irst dis cussion.B u t the ab str act of th i s r eport, with S u p eral ge bra Poi nca ré, was pu bli she d i nthe Rotap rint edi ti on o fthe Proc e eding [@ /1 /]. Itwas t h e firs t pub lica ti on abo ut thes up e r alge br aPoin car é. Unfo rtun a tel y, in t he finalver s ionof P roceedi ngs did not a pp ear this a bs tra cts. B u t the fi rst attempts of the int r oductio n o f sup eral gebras in ph ysicswas givenin pap er by G .St a v raki[ @ /2 /]in 1966 andH . Miy azava [ @/3/ ] in 19 68. The mathemati c albackground fo r s uper s y mm etr y w a s c on s tru c t ed in 1970. Fel ix Berezin (M o sc ow) and Gr e gor yKats (K iev) [@ /4/]p ublishe d the pap er aboutth e gr o u pswith a com muting a nd antico m mutin g p arame ter s in t he Ru ssian journ a l ” Mathe matich es kiy Zb ornik ”– that i s how the supergroups a ppeare d inmat hematics. B e for e that Pr of.D. Volkovinv est igate d t h e con nect i on be t weenspin and stati s ti csa n drediscovere d ( aft er M. Gre e n,1953 ) [@ /5/] the parastat i stics in 1959. Lat e r he co n side re d fermionic Re gge t r a jectory. T he successof the a pp l icati on ofGoldst one’s T h e or e m [@/6

steps_that were_done to the invention_of SUSY_in_Kharkov team_headed_by D.Volkov. _ This_paper is dedicated to_the memory of_Prof._Yu. Gol’fand, whose ideas of SUSY inspired the most active developments in High Energy_Physics_over thirty_years. author: -_'V._Akulov [^1]' title: 'Non-linear way to_Supersymmetry and N-extended SUSY [^2]_' --- [*”Geometry of_space is associated with mathematical group”*]{} Felix Klein, ”Erlagen_Program”_1872 This year, the_science community celebrates the 30th anniversary of SUSY. I_have been asked to give a_short historical introduction_to_the_first steps in this_direction that were done by our_group headed by Prof. Dmitry Volkov. Rochester’s_High Energy Conference was held in Kiev_(Ukraine, Soviet Union) in 1970. Prof._Yu. Golfand announced 2 reports_for this_conference, but the Org. Committee_gave him time_only for_one. Prof. Yu._Gol’fand preferred to discuss the problem_of vacuum in_QED, because the problem of Superalgebra_Poincaré,_which was obtained_already,_seemed_to him_very complicated for_a_first discussion._But_the abstract of this report, with_Superalgebra_Poincaré, was published in the Rotaprint edition_of the Proceeding [@/1/]. It was_the_first publication about the_superalgebra Poincaré. Unfortunately, in the_final version of Proceedings did not_appear this_abstracts. But the_first attempts of the introduction of superalgebras in physics was given_in paper by G.Stavraki [@/2/] in_1966 and H.Miyazava [@/3/]_in 1968. The_mathematical_background for supersymmetry_was_constructed in_1970. Felix Berezin(Moscow) and Gregory Kats (Kiev)_[@/4/] published_the paper about the groups with_a commuting and anticommuting_parameters_in the Russian journal ”Mathematicheskiy Zbornik”_– that is how the supergroups_appeared in mathematics. Before that Prof._D._Volkov_investigated the connection between spin_and statistics and rediscovered (after M.Green,1953)_[@/5/] the parastatistics_in 1959. Later he considered fermionic Regge_trajectory._The success of the application of_Goldstone’s_Theorem [@/6

y CDGAs -------------------------------------------------- Given a DG Artin $N$-hypergroupoid $X$, and an $O(X)$-module $M$, we may pull $M$ back along the unit $\eta \co O(X) \to DD^*O(X)$ of the adjunction $D^*\dashv D$, and then apply Lemma \[denormmod\] to obtain a $D^*O(X)$-module $N_c\eta^*M$. As for instance in [@stacks2 Definition \[stacks2-delta\*\]], define almost cosimplicial diagrams to be functors on the subcategory $\Delta_*$ of the ordinal number category $\Delta$ containing only those morphisms $f$ with $f(0)=0$; define almost simplicial diagrams dually. Thus an almost simplicial diagram $X_*$ in $\C$ consists of objects $X_n \in \C$, with all of the operations $\pd_i, \sigma_i$ of a simplicial diagram except $\pd_0$, satisfying the usual relations. Given a simplicial (resp. cosimplicial) diagram $X$, we write $X_{\#}$ (resp. $X^{\#}$) for the underlying almost simplicial diagram (resp. almost cosimplicial) diagram. The denormalisation functor $D$ descends to a functor from graded-commutative algebras to almost cosimplicial algebras, with $D^*$ thus descending to a functor in the opposite direction. In other words, $(D^*B)^{\#}$ does not depend on $\pd^0_B$, and $\pd^0_{DA}$ is the only part of the structure on $DA$ to depend on $\pd_A$. From this, it can be seen that for any DG Artin $N$-hypergroupoid $X$, the graded-commutative algebra $\H_0D^*O(X)^{\#}$ is freely generated over $\H_0O(X)^0= \H_0D^*O(X)^0$ by a graded projective module, and that $$\H_i^*O(X)^{\#}\cong \H_0D^*O(X)^{\#}\ten_{\H_0O(X)^0}\H_iO(X)^0.$$ If $M$ is a homotopy-Cartesian $O(X)$-module, the map $$(\eta^*M)^0\ten_{D^0

y CDGAs -------------------------------------------------- Given a DG Artin $ N$-hypergroupoid $ X$, and an $ O(X)$-module $ M$, we may pull $ M$ back along the unit $ \eta \co O(X) \to DD^*O(X)$ of the adjunction $ D^*\dashv D$, and then lend oneself Lemma \[denormmod\ ] to receive a $ D^*O(X)$-module $ N_c\eta^*M$. As for instance in [ @stacks2 Definition \[stacks2 - delta\*\ ] ], define almost cosimplicial diagram to be functors on the subcategory $ \Delta_*$ of the ordinal number category $ \Delta$ containing merely those morphisms $ f$ with $ f(0)=0 $; define almost simplicial diagram dually. therefore an almost simplicial diagram $ X_*$ in $ \C$ consists of object $ X_n \in \C$, with all of the operations $ \pd_i, \sigma_i$ of a simplicial diagram except $ \pd_0 $, satisfying the usual relation. Given a simplicial (resp. cosimplicial) diagram $ X$, we spell $ X_{\#}$ (resp. $ X^{\#}$) for the underlying almost simplicial diagram (resp. about cosimplicial) diagram. The denormalisation functor $ D$ descends to a functor from graded - commutative algebras to about cosimplicial algebras, with $ D^*$ thus descending to a functor in the opposite direction. In early words, $ (D^*B)^{\#}$ does not count on $ \pd^0_B$, and $ \pd^0_{DA}$ is the only part of the structure on $ DA$ to depend on $ \pd_A$. From this, it can be see that for any DG Artin $ N$-hypergroupoid $ X$, the graded - commutative algebra $ \H_0D^*O(X)^{\#}$ is freely generated over $ \H_0O(X)^0= \H_0D^*O(X)^0 $ by a graded projective module, and that $ $ \H_i^*O(X)^{\#}\cong \H_0D^*O(X)^{\#}\ten_{\H_0O(X)^0}\H_iO(X)^0.$$ If $ M$ is a homotopy - Cartesian $ O(X)$-module, the map $ $ (\eta^*M)^0\ten_{D^0

y CFGAs -------------------------------------------------- Given a DG Artin $N$-hykergroupoid $X$, and an $O(X)$-mmdule $J$, we may pull $M$ back along the unit $\xta \xo O(X) \to DD^*O(X)$ of the adjuncgion $D^*\dasjv D$, and thei apply Lemma \[deikrmmod\] bj obfwin c $V^*O(X)$-module $N_c\eta^*K$. As for invtance in [@stacns2 Dzfinition \[stacks2-delta\*\]], define almost sosimploclal diagrams tj be suncfors on the subcategory $\Delta_*$ of tge ordiial number catebory $\Delta$ containing only thode morphisms $f$ witj $f(0)=0$; define qlmofr simplicial diagrams dually. Thus zn almost simplicial diagram $X_*$ kn $\C$ eonsists of ovjefjs $X_n \in \C$, wmth alj of the opevstions $\pd_i, \sibma_i$ of a simpkicmal eiagram except $\pd_0$, satmsfying the usual rejations. Gieeu a simplicial (resp. cisumpliwial) diaeeam $X$, se wdite $X_{\#}$ (reap. $X^{\#}$) for tge underlyibg almost simpliciak qpsgram (resp. ammost sofimplicial) diagram. The denormalisation flnctkr $D$ descends to a funcror from graded-commutwtive algqbras to almost cosimplicial algebras, with $D^*$ thus descxnainy to a dujctor in the opposite direction. In other wordf, $(D^*N)^{\#}$ coes not depenb on $\pd^0_B$, and $\pd^0_{DS}$ ls jhe only part uf the stducture on $DA$ to dfpend og $\pd_A$. Drom this, it van be seen that for any DG Artin $N$-hypeggroypoid $X$, the graded-eommutative clgebrs $\H_0D^*O(C)^{\#}$ is freely generated orer $\H_0O(S)^0= \H_0D^*O(X)^0$ by a graded pdujective module, xnd tvat $$\H_i^*O(X)^{\#}\cong \H_0D^*O(X)^{\#}\ten_{\H_0O(X)^0}\H_iO(V)^0.$$ If $M$ is e homptopy-Caftesoan $O(X)$-iodule, the map $$(\cda^*M)^0\ten_{D^0

y CDGAs -------------------------------------------------- Given a DG Artin and $O(X)$-module $M$, may pull $M$ \co \to DD^*O(X)$ of adjunction $D^*\dashv D$, then apply Lemma \[denormmod\] to obtain $D^*O(X)$-module $N_c\eta^*M$. As for instance in [@stacks2 Definition \[stacks2-delta\*\]], define almost cosimplicial diagrams be functors on the subcategory $\Delta_*$ of the ordinal number category $\Delta$ containing those $f$ $f(0)=0$; almost simplicial diagrams dually. Thus an almost simplicial diagram $X_*$ in $\C$ consists of objects $X_n \C$, with all of the operations $\pd_i, \sigma_i$ a simplicial diagram except satisfying the usual relations. Given simplicial cosimplicial) diagram we $X_{\#}$ $X^{\#}$) for the almost simplicial diagram (resp. almost cosimplicial) diagram. The denormalisation functor $D$ descends to a functor from graded-commutative to almost with $D^*$ descending a in the opposite other words, $(D^*B)^{\#}$ does not depend $\pd^0_{DA}$ is the only part of the structure $DA$ to on $\pd_A$. From this, it can seen that for any DG Artin $N$-hypergroupoid $X$, graded-commutative algebra $\H_0D^*O(X)^{\#}$ is freely generated over $\H_0O(X)^0= \H_0D^*O(X)^0$ by a graded projective module, and \H_0D^*O(X)^{\#}\ten_{\H_0O(X)^0}\H_iO(X)^0.$$ If $M$ is homotopy-Cartesian $O(X)$-module, the $$(\eta^*M)^0\ten_{D^0

y CDGAs -------------------------------------------------- Given a DG Artin $N$-hyperGroupoid $X$, aNd an $O(x)$-moDulE $M$, We maY pulL $M$ back along the UNit $\eTa \co O(X) \to DD^*O(X)$ of the adjunCtion $d^*\dAShv D$, ANd Then aPply LemMA \[dENOrmMoD\] tO obTaIN a $d^*O(X)$-moDulE $N_c\eta^*M$. as for instaNce In [@Stacks2 DefiniTIoN \[stacks2-delTa\*\]], dEfine almost cOsiMpliciAl DiaGRams tO be FunctOrs on tHE subcaTegory $\DelTa_*$ OF the orDInal numBER cAtegOry $\Delta$ containinG OnLY those morphismS $f$ with $F(0)=0$; dEFiNE AlmOst Simplicial DiAgramS Dually. THUs AN ALmoST simplicial diAgram $X_*$ in $\C$ coNSisTs of obJeCts $x_N \in \C$, wiTh all Of THe oPerations $\pd_I, \sigMa_i$ of a simPliciaL Diagram EXcept $\pd_0$, SatisfYinG thE usuAL rElAtiOnS. givEN a SimPLicIal (resp. cOsImPliciAl) diAGRAM $X$, we WriTe $X_{\#}$ (rEsp. $X^{\#}$) fOr the underlyiNg aLmosT SimPliciAl diaGram (ReSp. almOst cosImpliCiAl) diagram. The denOrmaLisation fUncToR $D$ dEsCends TO a funcTor FroM graded-CommutaTIve AlGEBRaS to almost cosimplicIaL ALgEbras, witH $D^*$ thus DEsCeNDing to a fUnCtoR in tHE OpposIte dIReCtion. In oTher woRDs, $(d^*B)^{\#}$ Does not DePend on $\Pd^0_b$, anD $\pd^0_{dA}$ is tHE onlY part oF the struCture ON $DA$ to depend on $\pD_a$. From this, it caN Be SEEn THat fOr aNy DG Artin $N$-hYperGRoupOid $X$, THe GraDEd-comMutatIvE AlGEbra $\H_0D^*O(X)^{\#}$ is freely genErAted ovEr $\H_0O(X)^0= \h_0D^*O(X)^0$ by a graded Projective MODUle, and thAt $$\H_i^*o(x)^{\#}\cONg \H_0D^*O(X)^{\#}\ten_{\H_0O(X)^0}\H_io(X)^0.$$ If $M$ Is a homotopY-cartesiaN $O(X)$-moDule, the mAp $$(\eta^*M)^0\ten_{d^0

y CDGAs ------------------ ---------- ----- --- --- -- ---- ---- - Given a DGA rtin $N$-hypergroupoid $X$ , and a n $O( X )$ -modu le $M$, we m aypu ll $M $b ac k alo ngthe uni t $\eta \c o O (X ) \to DD^*O( X )$ of the ad jun ction $D^*\d ash v D$,an d t h en ap ply Lemm a \[de n ormmod \] to obt ai n a $D^ * O(X)$-m o d ul e $N _c\eta^*M$. As f o ri nstance in [@s tacks2 D e fi n i tio n \ [stacks2-d el ta\*\ ] ], defi n ea l m ost cosimplicialdiagrams to befuncto rs on the su bcate go r y $ \Delta_*$ o f th e ordinal numbe r catego r y $\Del ta$ co nta ini ng o n ly t hos em orp h is ms$ f$with $f( 0) =0 $; de fine a l m ostsim plic ial d iagrams duall y.Thus analmos t sim plic ia l dia gram $ X_*$in $\C$ consistsof o bjects $X _n\i n \ C$ , wit h all o f t heoperati ons $\p d _i, \ s i g ma _i$ of a simplicia ld i ag ram exce pt $\p d _0 $, satisfyi ng th e us u a l rel atio n s. Givena simp l ic ia l (resp .cosimp li cia l)diagr a m $X $, wewrite $X _{\#} $ (resp. $X^{\# } $) for the un d er l y in g alm ost simplicial dia g ram(res p .alm o st co simpl ic i al ) diagram. The deno rm alisat ion f unctor $D$ de scends toa f unctor f romg ra d ed-commutative alge bras to al m ost cosi mplic ial alge bras, wit h $D^*$ th usdes cen din g to a functor in t he o pp osite d ire ction.Inoth erwor ds , $(D^*B) ^{\#}$ d oe sno tdep end o n $\pd^0_ B$ , a nd $\ pd^0_ { DA}$ i s the onl ypa r t o f the s t ru c t ureon $ DA$tode pendon $ \ pd_ A$. Fr om this,itc an b ese en that for any DG A rt in $N$-hyp er gro upoid$ X $, the g raded-commutative algeb r a $\H_0 D^* O(X)^ {\#} $ is free lygenera ted over $ \H_0O( X)^0= \ H_0 D ^ *O(X) ^ 0 $byagraded pro j e cti ve mo du le,and tha t $$\H_i^*O(X)^{\# } \co ng \H_0D^*O(X )^{ \#}\ t e n_ {\H _ 0O ( X)^ 0} \ H_i O ( X)^0.$$ If $M$ is a homo to p y- Cartesian$ O(X )$ -module , the m ap $$ ( \eta^*M )^0\ten_{ D^0

y CDGAs -------------------------------------------------- Given_a DG_Artin $N$-hypergroupoid $X$, and_an $O(X)$-module_$M$,_we may_pull_$M$ back along_the unit $\eta_\co O(X) \to DD^*O(X)$_of the adjunction_$D^*\dashv_D$, and then apply Lemma \[denormmod\] to obtain a $D^*O(X)$-module $N_c\eta^*M$. As for instance in_[@stacks2_Definition \[stacks2-delta\*\]],_define_almost_cosimplicial diagrams to be functors_on the subcategory $\Delta_*$ of_the ordinal_number category $\Delta$ containing only those morphisms $f$_with_$f(0)=0$; define almost_simplicial diagrams dually. Thus an almost simplicial diagram $X_*$_in $\C$ consists of objects $X_n_\in \C$, with_all_of_the operations $\pd_i, \sigma_i$_of a simplicial diagram except $\pd_0$,_satisfying the usual relations. Given a simplicial_(resp. cosimplicial) diagram $X$, we write $X_{\#}$_(resp. $X^{\#}$) for the underlying almost_simplicial diagram (resp. almost cosimplicial)_diagram. The denormalisation_functor $D$ descends to a_functor from graded-commutative_algebras to_almost cosimplicial algebras,_with $D^*$ thus descending to a_functor in the_opposite direction. In other words, $(D^*B)^{\#}$_does_not depend on_$\pd^0_B$,_and_$\pd^0_{DA}$ is_the only part_of_the structure_on_$DA$ to depend on $\pd_A$. From this,_it_can be seen that for any DG_Artin $N$-hypergroupoid $X$, the_graded-commutative_algebra $\H_0D^*O(X)^{\#}$ is freely_generated over $\H_0O(X)^0= \H_0D^*O(X)^0$ by_a graded projective module, and that_$$\H_i^*O(X)^{\#}\cong \H_0D^*O(X)^{\#}\ten_{\H_0O(X)^0}\H_iO(X)^0.$$ If_$M$ is_a homotopy-Cartesian $O(X)$-module, the map $$(\eta^*M)^0\ten_{D^0

a tuple of multiplicities $m = (m_j)$ and let $\Omega_m \subset {\mathrm{Hom}}$ be the closure of the corresponding $GL$-orbit. Based on $m$ we define a parabolic $P_m\subset GL$ and a linear subspace $Z_m\subset{\mathrm{Hom}}$ as follows. For each vertex $i\in Q_0$ we divide the sets of row and column indices of $GL({\mathbb{C}}^{d(i)})$ into contiguous subsets of sizes $d_j(i)$ as $j$ runs from $1$ to $N$. This defines a standard parabolic subgroup $P_m\subset GL$ whose $i$th component (for $i\in Q_0$) is the block lower triangular subgroup of $GL({\mathbb{C}}^{d(i)})$ with the given diagonal block sizes. The decompositions $d(i) = \sum_{j=1}^N d_j(i)$ also induce a block structure on each component ${\mathrm{Hom}}({\mathbb{C}}^{d(ta)},{\mathbb{C}}^{d(ha)})$ of ${\mathrm{Hom}}$, whose $(j,j')$ block is a $d_j(ta) \times d_{j'}(ha)$ rectangle. Define the linear subspace $Z_m\subset{\mathrm{Hom}}$ to be those elements with zeroes in all blocks strictly above the “block diagonal". This $Z_m$ is easily seen to be $P_m$-invariant. Reineke proves that [*for certain choices of ordering*]{} (built using reduced words for $w_0$ adapted to the quiver, as spelled out in Section \[sec:finitequivers\]) on $R_+$, the Kempf collapsing $GL\times^{P_m} Z_m \to {\mathrm{Hom}}$ is birational to $\Omega_m$. Let $\{x_k^{(i)}\mid i\in Q_0,\, k \in \{1,\ldots,d(i)\} \}$ be a basis for the weight lattice $T^*$ of the standard maximal torus $T$ given by the tuples of diagonal matrices in $GL$. Then the $(k,k')$th matrix entry in the $a$th component of ${\mathrm{Hom}}$ has weight $x_k^{(ta)}-x_{k'}^{(ha)}$. \[thm:

a tuple of multiplicities $ m = (m_j)$ and let $ \Omega_m \subset { \mathrm{Hom}}$ be the closure of the corresponding $ GL$-orbit. free-base on $ m$ we specify a parabolic $ P_m\subset GL$ and a linear subspace $ Z_m\subset{\mathrm{Hom}}$ as follows. For each vertex $ i\in Q_0 $ we divide the set of row and column indices of $ GL({\mathbb{C}}^{d(i)})$ into adjacent subset of sizes $ d_j(i)$ as $ j$ runs from $ 1 $ to $ N$. This defines a standard parabolic subgroup $ P_m\subset GL$ whose $ i$th component (for $ i\in Q_0 $) is the block low triangular subgroup of $ GL({\mathbb{C}}^{d(i)})$ with the given diagonal block sizes. The decompositions $ d(i) = \sum_{j=1}^N d_j(i)$ also induce a pulley structure on each component $ { \mathrm{Hom}}({\mathbb{C}}^{d(ta)},{\mathbb{C}}^{d(ha)})$ of $ { \mathrm{Hom}}$, whose $ (j, j')$ block is a $ d_j(ta) \times d_{j'}(ha)$ rectangle. specify the linear subspace $ Z_m\subset{\mathrm{Hom}}$ to be those elements with zeroes in all blocks rigorously above the “ block aslant ". This $ Z_m$ is easily seen to be $ P_m$-invariant. Reineke proves that [ * for certain choice of ordering * ] { } (built using reduced words for $ w_0 $ adapt to the quiver, as spelled out in Section \[sec: finitequivers\ ]) on $ R_+$, the Kempf collapsing $ GL\times^{P_m } Z_m \to { \mathrm{Hom}}$ is birational to $ \Omega_m$. Let $ \{x_k^{(i)}\mid i\in Q_0,\, k \in \{1,\ldots, d(i)\ } \}$ be a basis for the weight lattice $ T^*$ of the standard maximal torus $ T$ given by the tuples of diagonal matrices in $ GL$. Then the $ (kelvin, k')$th matrix entry in the $ a$th component of $ { \mathrm{Hom}}$ has weight unit $ x_k^{(ta)}-x_{k'}^{(ha)}$. \[thm:

a huple of multiplicities $o = (m_j)$ and let $\Omega_m \suuset {\mafhrm{Hom}}$ ce the closure of the corres'ondung $GO$-orbit. Based on $m$ we dewine a pagabolic $P_n\subwwt GL$ and e linear subspzge $Z_m\vnbset{\mathrm{Hom}}$ ss follows. For each vertax $i\nn Q_0$ we divide the sets of row and cjlumn imdlces of $GL({\mathfb{C}}^{d(p)})$ ynto boktiguous subsets of sizes $d_j(i)$ as $j$ runs from $1$ to $N$. Tnis defines a standard parwbollc subgroup $P_m\subsft GL$ whose $i$th xomponent (fof $i\in Q_0$) is the block lkwer triangular subgroup of $GL({\mxthbb{E}}^{d(i)})$ with thg figgn diagonal ulock fizes. The decompositimns $d(i) = \sum_{j=1}^N d_j(i)$ alsp iiducw a block structure oi each component ${\matrrm{Hom}}({\matvby{C}}^{d(ta)},{\mathbb{C}}^{d(ha)})$ of ${\marhem{Hom}}$, whove $(j,g')$ bluck ix z $d_j(ta) \tijes d_{j'}(ha)$ rsctangle. Dedine the linear subxpwbr $Z_m\subset{\mafhrm{Hoi}}$ eo be those elements with zeroes in all blkcks strictly above the “block diagonal". This $E_m$ is easijy seen to be $P_m$-invariant. Reineke proves that [*for certeiv ckiices if ordering*]{} (built using reduced words for $w_0$ ada[fec no the quiver, as fpelled out ij Xgction \[sec:finijequivexa\]) kn $R_+$, the Kempf colpapsing $GL\tines^{P_m} Z_m \uo {\mayhrm{Hom}}$ is birational to $\Omwga_m$. Let $\{x_k^{(i)}\mpd i\un Q_0,\, k \in \{1,\ldots,d(i)\} \}$ be a basis for the eeight lattice $T^*$ of the stahdard maximwl torus $F$ given by the tuolex mf diagonal matrices in $GL$. Then the $(k,k')$tk matrix entty in tre $a$th comoonenb of ${\mathrm{Hom}}$ has aeighj $x_k^{(ta)}-f_{k'}^{(ha)}$. \[thm:

a tuple of multiplicities $m = (m_j)$ $\Omega_m {\mathrm{Hom}}$ be closure of the we a parabolic $P_m\subset and a linear $Z_m\subset{\mathrm{Hom}}$ as follows. For each vertex Q_0$ we divide the sets of row and column indices of $GL({\mathbb{C}}^{d(i)})$ into subsets of sizes $d_j(i)$ as $j$ runs from $1$ to $N$. This defines standard subgroup GL$ $i$th component (for $i\in Q_0$) is the block lower triangular subgroup of $GL({\mathbb{C}}^{d(i)})$ with the given block sizes. The decompositions $d(i) = \sum_{j=1}^N d_j(i)$ induce a block structure each component ${\mathrm{Hom}}({\mathbb{C}}^{d(ta)},{\mathbb{C}}^{d(ha)})$ of ${\mathrm{Hom}}$, $(j,j')$ is a \times rectangle. the linear subspace to be those elements with zeroes in all blocks strictly above the “block diagonal". This $Z_m$ is seen to Reineke proves [*for choices ordering*]{} (built using for $w_0$ adapted to the quiver, in Section \[sec:finitequivers\]) on $R_+$, the Kempf collapsing Z_m \to is birational to $\Omega_m$. Let $\{x_k^{(i)}\mid Q_0,\, k \in \{1,\ldots,d(i)\} \}$ be a basis the weight lattice $T^*$ of the standard maximal torus $T$ given by the tuples of in $GL$. Then the matrix entry in $a$th of has $x_k^{(ta)}-x_{k'}^{(ha)}$. \[thm:

a tuple of multiplicities $m = (m_j)$ And let $\OmegA_m \subSet {\MatHrM{Hom}}$ Be thE closure of the cORresPonding $GL$-orbit. Based on $m$ We defInE A parABoLic $P_m\Subset Gl$ AnD A LinEaR sUbsPaCE $Z_M\subsEt{\mAthrm{HoM}}$ as follows. for EaCh vertex $i\in Q_0$ WE dIvide the seTs oF row and columN inDices oF $Gl({\maTHbb{C}}^{d(I)})$ inTo conTiguouS SubsetS of sizes $d_J(i)$ AS $j$ runs FRom $1$ to $N$. THIS dEfinEs a standard paraboLIc SUbgroup $P_m\subseT GL$ whoSe $I$Th COMpoNenT (for $i\in Q_0$) is ThE blocK Lower trIAnGULAr sUBgroup of $GL({\matHbb{C}}^{d(i)})$ with tHE giVen diaGoNal BLock siZes. ThE dEComPositions $d(i) = \Sum_{j=1}^n d_j(i)$ also iNduce a BLock strUCture on Each coMpoNenT ${\matHRm{hoM}}({\maThBB{C}}^{d(TA)},{\mAthBB{C}}^{d(Ha)})$ of ${\mathRm{hoM}}$, whosE $(j,j')$ bLOCK Is a $d_J(ta) \TimeS d_{j'}(ha)$ Rectangle. DefiNe tHe liNEar SubspAce $Z_m\SubsEt{\MathrM{Hom}}$ to Be thoSe Elements with zerOes iN all blockS stRiCtlY aBove tHE “block DiaGonAl". This $Z_M$ is easiLY seEn TO BE $P_M$-invariant. Reineke pRoVES tHat [*for ceRtain cHOiCeS Of orderiNg*]{} (BuiLt usING reduCed wORdS for $w_0$ adaPted to THe QuIver, as sPeLled ouT iN SeCtiOn \[sec:FInitEquiveRs\]) on $R_+$, the kempf COllapsing $GL\timES^{P_m} Z_m \to {\mathrm{hOm}}$ IS BiRAtioNal To $\Omega_m$. Let $\{X_k^{(i)}\mID i\in q_0,\, k \in \{1,\LDoTs,d(I)\} \}$ Be a baSis foR tHE wEIght lattice $T^*$ of the stAnDard maXimal Torus $T$ given by The tuples oF DIAgonal maTricES iN $gL$. Then the $(k,k')$th mAtrix Entry in the $A$Th componEnt of ${\Mathrm{HoM}}$ has weighT $X_K^{(ta)}-x_{k'}^{(ha)}$. \[tHm:

a tuple of multiplicities $m = (m_j )$ an d l et$\ Omeg a_m\subset {\math r m{Ho m}}$ be the closure of theco r resp o nd ing $ GL$-orb i t. Bas ed o n $ m$ we defi nea parab olic $P_m\ sub se t GL$ and al in ear subspa ce$Z_m\subset{ \ma thrm{H om }}$ as fo llo ws. F or eac h verte x $i\in Q _0 $ we di v ide the s et s of row and column i n di c es of $GL({\ma thbb{C }} ^ {d ( i )}) $ i nto contig uo us su b sets of si z e s $d _ j(i)$ as $j$runs from $ 1 $ t o $N$. T his define s a s ta n dar d parabolic sub group $P_ m\subs e t GL$ w h ose $i$ th com pon ent (fo r $ i\ inQ_ 0 $)i sthe blo ck lower t ri angul ar s u b g r oupof$GL( {\mat hbb{C}}^{d(i) })$ wit h th e giv en di agon al bloc k size s. T he decompositions $d( i) = \sum _{j =1 }^N d _j(i) $ alsoind uce a bloc k struc t ure o n e ac h component ${\mat hr m { Ho m}}({\ma thbb{C } }^ {d ( ta)},{\m at hbb {C}} ^ { d(ha) })$o f${\mathr m{Hom} } $, w hose $( j, j')$ b lo ckisa $d_ j (ta) \time s d_{j'} (ha)$ rectangle. Def i ne the linear su b s pa c e $Z _m\ subset{\mat hrm{ H om}} $ to be th o se el ement sw it h zeroes in all bloc ks stric tly a bove the “blo ck diagona l " . This $Z _m$i se asily seen tobe $P _m$-invari a nt. Rein eke p roves th at [*forc e rtain ch oic esoford e r in g*]{} (builtu s ingre duced w ord s for $ w_0 $ a dap ted t o the qui ver, assp el le dout in S e ction \[ se c:f in ite quive r s\]) o n $R_ +$,th eK emp f colla p si n g $GL \t im es^{ P_m }Z_m \ to { \ mat hrm{Hom }}$ is bi rat i onal t o$\Omega _m$. Let $\{ x_ k^{(i)}\mi di\i n Q_0, \ , k \in \ {1,\ldots,d(i)\} \}$ be a basis fo r the wei ght latti ce$T^*$oft he sta ndardmaxim al to r u s $T$ g iv enby the tuple s ofdiago na l ma tricesin $GL$. Then the$ (k, k')$th matrix en tryi n t he$ a$ t h c om p one n t of ${\mathrm{H om}}$ haswe i gh t $x_k^{(t a )}- x_ {k'}^{( ha)}$. \[th m :

a_tuple of_multiplicities $m = (m_j)$_and let_$\Omega_m \subset_{\mathrm{Hom}}$ be_the_closure of the_corresponding $GL$-orbit. Based on_$m$ we define a_parabolic $P_m\subset GL$_and_a linear subspace $Z_m\subset{\mathrm{Hom}}$ as follows. For each vertex $i\in Q_0$ we divide the_sets_of row_and_column_indices of $GL({\mathbb{C}}^{d(i)})$ into contiguous_subsets of sizes $d_j(i)$ as_$j$ runs_from $1$ to $N$. This defines a standard_parabolic_subgroup $P_m\subset GL$_whose $i$th component (for $i\in Q_0$) is the block_lower triangular subgroup of $GL({\mathbb{C}}^{d(i)})$ with_the given diagonal_block_sizes. The_decompositions $d(i) = \sum_{j=1}^N_d_j(i)$ also induce a block structure_on each component ${\mathrm{Hom}}({\mathbb{C}}^{d(ta)},{\mathbb{C}}^{d(ha)})$ of ${\mathrm{Hom}}$,_whose $(j,j')$ block is a $d_j(ta) \times_d_{j'}(ha)$ rectangle. Define the linear subspace_$Z_m\subset{\mathrm{Hom}}$ to be those elements_with zeroes_in all blocks strictly above_the “block diagonal"._This $Z_m$_is easily seen_to be $P_m$-invariant. Reineke proves that_[*for certain choices_of ordering*]{} (built using reduced words_for_$w_0$ adapted to_the_quiver,_as spelled_out in Section_\[sec:finitequivers\])_on $R_+$,_the_Kempf collapsing $GL\times^{P_m} Z_m \to {\mathrm{Hom}}$_is_birational to $\Omega_m$. Let $\{x_k^{(i)}\mid i\in Q_0,\, k_\in \{1,\ldots,d(i)\} \}$ be_a_basis for the weight_lattice $T^*$ of the standard_maximal torus $T$ given by the_tuples of_diagonal matrices_in $GL$. Then the $(k,k')$th matrix entry in the $a$th component_of ${\mathrm{Hom}}$ has weight $x_k^{(ta)}-x_{k'}^{(ha)}$. \[thm:

{x}}^T {\bm{s}}$. If $d({\bm{x}}, {\bm{s}}, \mu) \leq \eta \mu$, then $(1+\eta) \norm{{\bm{s}}^{-1}}_2 \geq \norm{\mu^{-1} {\bm{x}}}_2$. For part 1, let $\{\lambda_i\}_{i=1}^{2r}$ be the eigenvalues of $T_{{\bm{x}}}{\bm{s}}$, note that $T_{x}$ is invertible as $x \in {\mathcal{L}}$. Then using part 1 of Claim \[claim:Properties of Qx\], we have $${\bm{x}}^T{\bm{s}} = {\bm{x}}^T T_{{\bm{x}}}^{-1} T_{{\bm{x}}} {\bm{s}} = (T_{{\bm{x}}^{-1}} {\bm{x}})^T T_{{\bm{x}}} {\bm{s}} = {\bm{e}}^T T_{{\bm{x}}} {\bm{s}} = \frac12 \sum_{i=1}^{2r} \lambda_i,$$ We can therefore bound the duality gap $x^{T}s$ as follows, $$\begin{aligned} {\bm{x}}^T {\bm{s}} = \frac12 \sum_{i=1}^{2r} \lambda_i &\leq r\nu + \frac12 \sum_{i=1}^{2r} |\lambda_i - \nu| \\ &\leq r\nu + \sqrt{\frac{r}{2}}\sqrt{\sum_{i=1}^{2r} (\lambda_i - \nu)^2} \\ &= r\nu + \sqrt{\frac{r}{2}} \cdot d({\bm{x}}, {\bm{s}}, \nu). \end{aligned}$$ The second step used the Cauchy-Schwarz inequality while the third follows from the definition $d({\bm{x}}, {\bm{s}}, \nu)^2 = \sum_{i=1}^{2r} (\lambda_i - \nu)^2$. The proof of the lower bound is similar, but starts instead with the inequality $$\frac12\sum_{i=1}^{2r} \lambda_i \geq r\nu - \frac12 \sum_{i=1}^{2r} |\nu - \lambda_i|.$$

{ x}}^T { \bm{s}}$. If $ d({\bm{x } }, { \bm{s } }, \mu) \leq \eta \mu$, then $ (1+\eta) \norm{{\bm{s}}^{-1}}_2 \geq \norm{\mu^{-1 } { \bm{x}}}_2$. For part 1, let $ \{\lambda_i\}_{i=1}^{2r}$ be the eigenvalues of $ T_{{\bm{x}}}{\bm{s}}$, note that $ T_{x}$ is invertible as $ x \in { \mathcal{L}}$. Then use character 1 of Claim \[claim: Properties of Qx\ ], we have $ $ { \bm{x}}^T{\bm{s } } = { \bm{x}}^T T_{{\bm{x}}}^{-1 } T_{{\bm{x } } } { \bm{s } } = (T_{{\bm{x}}^{-1 } } { \bm{x}})^T T_{{\bm{x } } } { \bm{s } } = { \bm{e}}^T T_{{\bm{x } } } { \bm{s } } = \frac12 \sum_{i=1}^{2r } \lambda_i,$$ We can therefore restrict the duality gap $ x^{T}s$ as surveil, $ $ \begin{aligned } { \bm{x}}^T { \bm{s } } = \frac12 \sum_{i=1}^{2r } \lambda_i & \leq r\nu + \frac12 \sum_{i=1}^{2r } |\lambda_i - \nu| \\ & \leq r\nu + \sqrt{\frac{r}{2}}\sqrt{\sum_{i=1}^{2r } (\lambda_i - \nu)^2 } \\ & = r\nu + \sqrt{\frac{r}{2 } } \cdot d({\bm{x } }, { \bm{s } }, \nu). \end{aligned}$$ The second step used the Cauchy - Schwarz inequality while the one-third follows from the definition $ d({\bm{x } }, { \bm{s } }, \nu)^2 = \sum_{i=1}^{2r } (\lambda_i - \nu)^2$. The proof of the lower bind is similar, but starts rather with the inequality $ $ \frac12\sum_{i=1}^{2r } \lambda_i \geq r\nu - \frac12 \sum_{i=1}^{2r } |\nu - \lambda_i|.$$

{x}}^T {\hm{s}}$. If $d({\bm{x}}, {\bm{s}}, \mu) \leq \eua \mu$, then $(1+\eta) \notm{{\vm{s}}^{-1}}_2 \gex \norm{\ju^{-1} {\bm{x}}}_2$. Fof part 1, let $\{\lambda_i\}_{i=1}^{2r}$ be the eugenvqlues of $T_{{\bm{x}}}{\bm{s}}$, note ghat $T_{x}$ id invertuble qs $x \in {\mavgcal{L}}$. Tmzn usjkg paxt 1 of Claim \[clalm:Propertiev of Qx\], we hava $${\cm{r}}^T{\bm{s}} = {\bm{x}}^T T_{{\bm{x}}}^{-1} T_{{\bm{x}}} {\bm{s}} = (T_{{\bm{x}}^{-1}} {\bm{x}})^E T_{{\bm{x}}} {\nm{d}} = {\bm{e}}^T T_{{\bm{x}}} {\bi{s}} = \gwac12 \alm_{l=1}^{2r} \lambda_i,$$ We can therefore bouns the dlality gap $x^{T}s$ as follows, $$\begin{aligned} {\bl{x}}^T {\bm{s}} = \frac12 \sum_{i=1}^{2g} \lambda_i &\lgs r\gy + \frac12 \sum_{i=1}^{2f} |\lambda_i - \nu| \\ &\meq r\nu + \sqrt{\frac{r}{2}}\sqrt{\sum_{i=1}^{2r} (\lamcda_i - \nu)^2} \\ &= r\nk + \sqrt{\frac{r}{2}} \cdot d({\bm{x}}, {\bm{s}}, \nu). \end{anigned}$$ Yhe second stei usev thw Cauchy-Schwarz inequelity while the thirq follows fxom the definition $d({\bn{x}}, {\bm{s}}, \tu)^2 = \vum_{i=1}^{2f} (\laobdz_i - \hu)^2$. The prkof of the lower boune is similar, but stsrew instead witg the ynqquality $$\frac12\sum_{i=1}^{2r} \lambda_i \geq r\nu - \frab12 \suj_{i=1}^{2r} |\nu - \lambda_i|.$$

{x}}^T {\bm{s}}$. If $d({\bm{x}}, {\bm{s}}, \mu) \leq then \norm{{\bm{s}}^{-1}}_2 \geq {\bm{x}}}_2$. For part eigenvalues $T_{{\bm{x}}}{\bm{s}}$, note that is invertible as \in {\mathcal{L}}$. Then using part 1 Claim \[claim:Properties of Qx\], we have $${\bm{x}}^T{\bm{s}} = {\bm{x}}^T T_{{\bm{x}}}^{-1} T_{{\bm{x}}} {\bm{s}} = {\bm{x}})^T T_{{\bm{x}}} {\bm{s}} = {\bm{e}}^T T_{{\bm{x}}} {\bm{s}} = \frac12 \sum_{i=1}^{2r} \lambda_i,$$ We can bound duality $x^{T}s$ follows, $$\begin{aligned} {\bm{x}}^T {\bm{s}} = \frac12 \sum_{i=1}^{2r} \lambda_i &\leq r\nu + \frac12 \sum_{i=1}^{2r} |\lambda_i - \nu| &\leq r\nu + \sqrt{\frac{r}{2}}\sqrt{\sum_{i=1}^{2r} (\lambda_i - \nu)^2} \\ r\nu + \sqrt{\frac{r}{2}} \cdot {\bm{s}}, \nu). \end{aligned}$$ The second used Cauchy-Schwarz inequality the follows the definition $d({\bm{x}}, \nu)^2 = \sum_{i=1}^{2r} (\lambda_i - \nu)^2$. The proof of the lower bound is similar, but starts instead the inequality \geq r\nu \frac12 |\nu \lambda_i|.$$

{x}}^T {\bm{s}}$. If $d({\bm{x}}, {\bm{s}}, \mu) \leq \eta \mu$, tHen $(1+\eta) \norm{{\Bm{s}}^{-1}}_2 \geQ \noRm{\mU^{-1} {\bM{x}}}_2$. FoR parT 1, let $\{\lambda_i\}_{i=1}^{2r}$ bE The eIgenvalues of $T_{{\bm{x}}}{\bm{s}}$, notE that $t_{x}$ IS invERtIble aS $x \in {\matHCaL{l}}$. theN uSiNg pArT 1 Of claim \[ClaIm:PropeRties of Qx\], wE haVe $${\Bm{x}}^T{\bm{s}} = {\bm{x}}^T T_{{\BM{x}}}^{-1} t_{{\bm{x}}} {\bm{s}} = (T_{{\bm{X}}^{-1}} {\bm{X}})^T T_{{\bm{x}}} {\bm{s}} = {\bm{e}}^t T_{{\bM{x}}} {\bm{s}} = \fRaC12 \suM_{I=1}^{2r} \lamBda_I,$$ We caN thereFOre bouNd the dualItY Gap $x^{T}s$ AS followS, $$\BEgIn{alIgned} {\bm{x}}^T {\bm{s}} = \frac12 \sUM_{i=1}^{2R} \Lambda_i &\leq r\nu + \fRac12 \sum_{I=1}^{2r} |\LAmBDA_i - \nU| \\ &\leQ r\nu + \sqrt{\frAc{R}{2}}\sqrt{\SUm_{i=1}^{2r} (\lamBDa_I - \NU)^2} \\ &= R\nu + \SQrt{\frac{r}{2}} \cdot d({\Bm{x}}, {\bm{s}}, \nu). \end{ALigNed}$$ The SeConD Step usEd the caUChy-schwarz ineqUaliTy while thE third FOllows fROm the deFinitiOn $d({\Bm{x}}, {\Bm{s}}, \nU)^2 = \SuM_{i=1}^{2R} (\laMbDA_i - \nU)^2$. thE prOOf oF the loweR bOuNd is sImilAR, BUT staRts InstEad wiTh the inequaliTy $$\fRac12\sUM_{i=1}^{2r} \LambdA_i \geq R\nu - \fRaC12 \sum_{i=1}^{2R} |\nu - \lamBda_i|.$$

{x}}^T {\bm{s}}$. If $d({\ bm{x}}, {\ bm{s} },\mu )\leq \et a \mu$, then $ ( 1+\e ta) \norm{{\bm{s}}^{-1 }}_2\g e q \n o rm {\mu^ {-1} {\ b m{ x } }}_ 2$ . Fo rp ar t 1,let $\{\la mbda_i\}_{ i=1 }^ {2r}$ be the ei genvaluesof$T_{{\bm{x}} }{\ bm{s}} $, no t e tha t $ T_{x} $ is i n vertib le as $x\i n {\mat h cal{L}} $ . T henusing part 1 of C l ai m \[claim:Prope rtiesof Qx \ ] , w e h ave $${\bm {x }}^T{ \ bm{s}}= { \ b m {x} } ^T T_{{\bm{x} }}^{-1} T_{ { \bm {x}}}{\ bm{ s }} = (T_{{ \b m {x} }^{-1}} {\b m{x} })^T T_{{ \bm{x} } } {\bm{ s }} = {\ bm{e}} ^TT_{ {\bm { x} }} {\ bm { s}} =\fr a c12 \sum_{i =1 }^ {2r}\lam b d a _ i,$$ We can ther efore bound t hedual i tygap $ x^{T} s$ a sfollo ws, $$ \begi n{ aligned} {\ bm{x}}^T{\b m{ s}} = \fra c 12 \su m_{ i=1 }^{2r}\lambda _ i & \l e q r\ nu + \frac12 \sum_ {i = 1 }^ {2r} |\l ambda_ i - \ n u| \\ &\l e q r\nu + \ s qr t{\frac{ r}{2}} \ sq rt {\sum_{ i= 1}^{2r }(\l amb da_i- \nu )^2} \ \ &=r \nu + \sqrt{\f r ac{r}{2}} \cd o td ( {\ b m{x} },{\bm{s}}, \ nu). \end { al ign e d}$$The s ec o nd step used the Cauch y- Schwar z ine quality while the third f o llows fr om t h ed efinition $d({ \bm{x }}, {\bm{s } }, \nu)^ 2 = \ sum_{i=1 }^{2r} (\ l a mbda_i - \n u)^ 2$. Th e pr oof of the lo w e r bo un d is si mil ar, but st art s i nst ea d with th e inequa li ty $ $\ fra c12\s u m_{i=1}^ {2 r}\l amb da_i\ geq r\ nu -\fra c1 2\ sum _{i=1}^ { 2r } |\nu - \ lamb da_ i| .$$

{x}}^T {\bm{s}}$._If $d({\bm{x}},_{\bm{s}}, \mu) \leq \eta_\mu$, then_$(1+\eta)_\norm{{\bm{s}}^{-1}}_2 \geq_\norm{\mu^{-1}_{\bm{x}}}_2$. For part 1,_let $\{\lambda_i\}_{i=1}^{2r}$ be_the eigenvalues of $T_{{\bm{x}}}{\bm{s}}$,_note that $T_{x}$_is_invertible as $x \in {\mathcal{L}}$. Then using part 1 of Claim \[claim:Properties of Qx\],_we_have $${\bm{x}}^T{\bm{s}}_=_{\bm{x}}^T_T_{{\bm{x}}}^{-1} T_{{\bm{x}}} {\bm{s}} = _(T_{{\bm{x}}^{-1}} {\bm{x}})^T T_{{\bm{x}}} {\bm{s}} =_{\bm{e}}^T T_{{\bm{x}}}_{\bm{s}} = \frac12 \sum_{i=1}^{2r} \lambda_i,$$ We can therefore_bound_the duality gap_$x^{T}s$ as follows, $$\begin{aligned} _ {\bm{x}}^T {\bm{s}} = \frac12_\sum_{i=1}^{2r} \lambda_i &\leq_r\nu_+_\frac12 \sum_{i=1}^{2r} |\lambda_i -_\nu| \\ _ &\leq r\nu +_\sqrt{\frac{r}{2}}\sqrt{\sum_{i=1}^{2r} (\lambda_i - \nu)^2} \\ _ &=_r\nu + \sqrt{\frac{r}{2}} \cdot d({\bm{x}},_{\bm{s}}, \nu). _ \end{aligned}$$ The_second step used_the Cauchy-Schwarz_inequality while the_third follows from the definition $d({\bm{x}},_{\bm{s}}, \nu)^2 =_\sum_{i=1}^{2r} (\lambda_i - \nu)^2$. The proof_of_the lower bound_is_similar,_but starts_instead with the_inequality_$$\frac12\sum_{i=1}^{2r} \lambda_i_\geq_r\nu - \frac12 \sum_{i=1}^{2r} |\nu -_\lambda_i|.$$

profile is parameterized by an extension of the Hernquist model [@HaloProfile2] $$\rho_{\mathrm{DM}} (r) = \frac{\rho_0} {\left ( \delta + {r \over r_s} \right )^\gamma \cdot \left [ 1 + \left ( {r \over r_s} \right )^\alpha \right ]^{ (\beta - \gamma)/\alpha } }, \label{eq:rho_profile}$$ where $(\alpha, \beta, \gamma, \delta)$ are dimensionless parameters. $r_s$ is a scaling radius and $\rho_0$ is the normalization density. Both have to be determined for each galaxy. In this paper the halo profile of Navarro, Frenk and White (NFW) [@NFW; @navarro_universal_1997] with $(1,3,1,0)$ is used as baseline. For the Milky Way $r_s=16.1^{+17.0}_{-7.8}\,\mathrm{kpc}$ and $\rho(r_s)=0.47^{+0.05}_{-0.06}\,{\mathrm{GeV}}/\mathrm{cm}^3$ are used [@nesti_dark_2013]. A currently favored model is the Burkert profile, that was obtained by the observation of dark matter dominated dwarf galaxies. The Burkert profile is described by $(2,3,1,1)$ [@Burkert], $r_s=9.26^{+5.6}_{-4.2}\,\mathrm{kpc}$ and $\rho(r_s)=0.49^{+0.07}_{-0.09}\,{\mathrm{GeV}}/\mathrm{cm}^3$ [@nesti_dark_2013]. While for the central part of the galaxy the models differ by orders of magnitude, the outer profiles are rather similar. The expected differential neutrino flux $\mathrm{d}\phi_{\nu}/\mathrm{d}E$ at Earth depends on the annihilation rate $\Gamma_A= {\langle \sigma_{\mathrm{A}} v\rangle}\rho(r)^2/2$ along the line of sight $l$, the muon neutrino yield per annihilation $\mathrm{d}N_\nu/\mathrm{d}E$, and the self-annihilation cross-section of dark matter averaged over the

profile is parameterized by an extension of the Hernquist model   [ @HaloProfile2 ] $ $ \rho_{\mathrm{DM } } (r) = \frac{\rho_0 } { \left (\delta + { r \over r_s } \right) ^\gamma \cdot \left [ 1 + \left ({ radius \over r_s } \right) ^\alpha \right ] ^ { (\beta - \gamma)/\alpha } }, \label{eq: rho_profile}$$ where $ (\alpha, \beta, \gamma, \delta)$ are dimensionless argument. $ r_s$ is a scaling radius and $ \rho_0 $ is the normalization density. Both own to be determined for each galaxy. In this newspaper the ring profile of Navarro, Frenk and White (NFW)   [ @NFW; @navarro_universal_1997 ] with $ (1,3,1,0)$ is use as baseline. For the Milky Way $ r_s=16.1^{+17.0}_{-7.8}\,\mathrm{kpc}$ and $ \rho(r_s)=0.47^{+0.05}_{-0.06}\,{\mathrm{GeV}}/\mathrm{cm}^3 $ are used   [ @nesti_dark_2013 ]. A presently favored model is the Burkert profile, that was obtained by the notice of dark matter dominate dwarf galaxies. The Burkert visibility is described by $ (2,3,1,1)$   [ @Burkert ], $ r_s=9.26^{+5.6}_{-4.2}\,\mathrm{kpc}$ and $ \rho(r_s)=0.49^{+0.07}_{-0.09}\,{\mathrm{GeV}}/\mathrm{cm}^3 $   [ @nesti_dark_2013 ]. While for the central part of the galax the models differ by orders of magnitude, the outer profile are rather similar. The expected differential neutrino flux $ \mathrm{d}\phi_{\nu}/\mathrm{d}E$ at Earth depends on the annihilation pace $ \Gamma_A= { \langle \sigma_{\mathrm{A } } v\rangle}\rho(r)^2/2 $ along the line of sight $ l$, the muon neutrino yield per annihilation $ \mathrm{d}N_\nu/\mathrm{d}E$, and the self - annihilation cross - section of dark topic average over the

prlfile is parameterized bn an extension oy the Hxrnquisf model [@HxloProfile2] $$\rho_{\mathrm{DM}} (r) = \frqc{\rho_0} {\left ( \delta + {g \over r_s} \rigit )^\gamma \cdot \leff [ 1 + \keft ( {r \ovar r_s} \right )^\al[hx \right ]^{ (\beta - \gamma)/\alpra } }, \lwbel{tq:rro_prkfile}$$ where $(\alpha, \beta, \gamma, \delta)$ are dikensionless psrameters. $r_s$ is a scaling gadiks and $\rho_0$ is the jormalizatiin dqbsity. Both hxve to be determined fkr each galaxy. In this paper the halo profile od Bavwtro, Frenk anv Whitv (NFW) [@NFW; @navavgo_univessal_1997] wiyh $(1,3,1,0)$ is used as bavelune. For the Milky Way $r_s=16.1^{+17.0}_{-7.8}\,\mathrm{kpc}$ and $\rro(r_s)=0.47^{+0.05}_{-0.06}\,{\mathrk{GzV}}/\mathrm{cm}^3$ are used [@newtu_dark_2013]. A cgrrevrly faboced model is the Burkedt profile, rhat was obtained bu eye observatioh of dwrh matter dominated dwarf galaxies. The Blrkedt profile is described by $(2,3,1,1)$ [@Burkert], $r_s=9.26^{+5.6}_{-4.2}\,\mathrm{kkc}$ and $\rho(w_s)=0.49^{+0.07}_{-0.09}\,{\mathrm{GeV}}/\mathrm{cm}^3$ [@nesti_dark_2013]. While for the centran parv uf uhc eqlwxy the models differ by orders of magnitude, ege olter profiles are rather simolwr. Yre expected dkffereufizl neutrino flux $\mwthrm{d}\pri_{\nu}/\mqthrm{d}E$ au Earyh depends on the annihilatuon rate $\Gamiq_A= {\langle \sigma_{\matkrm{A}} v\rangle}\xho(r)^2/2$ akong yhe line of sight $l$, the muoh neutrino jield per xnnihilation $\matfrm{c}N_\tu/\mathrm{d}E$, and the self-annyhilation croxs-sectiun og dark matter avfraged over the

profile is parameterized by an extension of model $$\rho_{\mathrm{DM}} (r) \frac{\rho_0} {\left ( \right \cdot \left [ + \left ( \over r_s} \right )^\alpha \right ]^{ - \gamma)/\alpha } }, \label{eq:rho_profile}$$ where $(\alpha, \beta, \gamma, \delta)$ are dimensionless parameters. is a scaling radius and $\rho_0$ is the normalization density. Both have to determined each In paper the halo profile of Navarro, Frenk and White (NFW) [@NFW; @navarro_universal_1997] with $(1,3,1,0)$ is used baseline. For the Milky Way $r_s=16.1^{+17.0}_{-7.8}\,\mathrm{kpc}$ and $\rho(r_s)=0.47^{+0.05}_{-0.06}\,{\mathrm{GeV}}/\mathrm{cm}^3$ used [@nesti_dark_2013]. A currently model is the Burkert profile, was by the of matter dwarf galaxies. The profile is described by $(2,3,1,1)$ [@Burkert], $r_s=9.26^{+5.6}_{-4.2}\,\mathrm{kpc}$ and $\rho(r_s)=0.49^{+0.07}_{-0.09}\,{\mathrm{GeV}}/\mathrm{cm}^3$ [@nesti_dark_2013]. While for the central part of the the models orders of the profiles rather similar. The neutrino flux $\mathrm{d}\phi_{\nu}/\mathrm{d}E$ at Earth depends rate $\Gamma_A= {\langle \sigma_{\mathrm{A}} v\rangle}\rho(r)^2/2$ along the line sight $l$, muon neutrino yield per annihilation $\mathrm{d}N_\nu/\mathrm{d}E$, the self-annihilation cross-section of dark matter averaged over

profile is parameterized by aN extension Of the herNquIsT modEl [@HaLoProfile2] $$\rho_{\maTHrm{Dm}} (r) = \frac{\rho_0} {\left ( \delta + {r \oveR r_s} \riGhT )^\GammA \CdOt \lefT [ 1 + \left ( {r \oVEr R_S} \RigHt )^\AlPha \RiGHt ]^{ (\Beta - \gAmmA)/\alpha } }, \lAbel{eq:rho_pRofIlE}$$ where $(\alpha, \bETa, \Gamma, \delta)$ Are DimensionlesS paRameteRs. $R_s$ iS A scalIng RadiuS and $\rhO_0$ Is the nOrmalizatIoN DensitY. both havE TO bE detErmined for each galAXy. iN this paper the hAlo proFiLE oF nAvaRro, frenk and WhItE (NFW) [@Nfw; @navarrO_UnIVERsaL_1997] With $(1,3,1,0)$ is used as bAseline. For tHE MiLky Way $R_s=16.1^{+17.0}_{-7.8}\,\MatHRm{kpc}$ aNd $\rho(R_s)=0.47^{+0.05}_{-0.06}\,{\MAthRm{GeV}}/\mathrm{Cm}^3$ arE used [@nestI_dark_2013]. A CUrrentlY Favored Model iS thE BuRkerT PrOfIle, ThAT waS ObTaiNEd bY the obseRvAtIon of Dark MATTEr doMinAted Dwarf Galaxies. The BuRkeRt prOFilE is deScribEd by $(2,3,1,1)$ [@buRkert], $R_s=9.26^{+5.6}_{-4.2}\,\mathRm{kpc}$ AnD $\rho(r_s)=0.49^{+0.07}_{-0.09}\,{\mathrm{GeV}}/\MathRm{cm}^3$ [@nesti_DarK_2013]. WHilE fOr the CEntral ParT of The galaXy the moDEls DiFFER bY orders of magnitude, ThE OUtEr profilEs are rAThEr SImilar. ThE eXpeCted DIFfereNtiaL NeUtrino flUx $\mathRM{d}\PhI_{\nu}/\mathRm{D}E$ at EaRtH dePenDs on tHE annIhilatIon rate $\GAmma_A= {\LAngle \sigma_{\mathRM{A}} v\rangle}\rho(r)^2/2$ ALoNG ThE Line Of sIght $l$, the muoN neuTRino YielD PeR anNIhilaTion $\mAtHRm{D}n_\nu/\mathrm{d}E$, and the seLf-AnnihiLatioN cross-section Of dark mattER AVeraged oVer tHE

profile is parameterizedby an exte nsion of th eHern quis t model [@Halo P rofi le2] $$\rho_{\mathrm{D M}} ( r) = \f rac{\rh o _0 } {\ left( \ delta + {r \overr_s }\right )^\ga m ma \cdot \lef t [ 1 + \lef t ( {r \overr_s} \rig ht )^\alp h a \right ]^{ (\bet a - \gamma)/\alph a } }, \la be l{eq: r ho_prof i le } $ $ wh e re $(\alpha,\beta, \gam m a,\delta )$ ar e dimen sionl es s pa rameters. $ r_s$ is a sca ling r a dius an d $\rho_ 0$ isthe no rmal i za ti onde n sit y .Bot h ha ve to be d et ermin ed f o r e achgal axy. Inthis paper th e h alop rof ile o f Nav arro ,Frenk and W hite(N FW) [@NFW; @nav arro _universa l_1 99 7]wi th $( 1 ,3,1,0 )$isused as baseli n e.Fo r t he Milky Way $r_s=16 .1 ^ { +1 7.0}_{-7 .8}\,\ m at hr m {kpc}$ a nd $\ rho( r _ s)=0. 47^{ + 0. 05}_{-0. 06}\,{ \ ma th rm{GeV} }/ \mathr m{ cm} ^3$ areu sed[@nest i_dark_2 013]. A currently fa v ored model is th e Bu r kert pr ofile, that was obta ined by th e obse rvati on of dark matter dominat ed dwarf gala xies. The Bur kert profi l e is descr ibed by $(2,3,1,1)$ [@ Burke rt], $r_s= 9 .26^{+5. 6}_{- 4.2}\,\m athrm{kpc } $ and $\r ho( r_s )=0 .49 ^ { +0 .07}_{-0.09}\ , { \mat hr m{GeV}} /\m athrm{c m}^ 3$[@n est i_ dark_2013 ]. While f or t he ce ntral part ofth e g al axy them odelsdiffe r by o rd e rsof magn i tu d e , th eou terpro fi les a re r a the r simil ar. Theexp e cted d if ferenti al neutrino f lu x $\mathrm {d }\p hi_{\n u } /\mathrm {d}E$ at Earth dependso n the a nni hilat ionrate $\Ga mma _A= {\ lan g le \si gma_{\ mathr m{ A}} v \rang l e }\ rho (r )^2/2$ alo n g th e lin eof s ight $l $, the muon neutri n o y ield per anni hil atio n $\ mat h rm { d}N _\ n u/\ m a thrm{d}E$, andthe self-a nn i hi lation cro s s-s ec tion of dark m atter average d over th e

profile_is parameterized_by an extension of_the Hernquist_model [@HaloProfile2]_$$\rho_{\mathrm{DM}} (r)_= _ _ _ \frac{\rho_0} _ __ {\left ( \delta + {r \over r_s} \right_)^\gamma_\cdot ___ _ _\left [_1 + __ _ \left ( {r \over_r_s} \right )^\alpha _ ___ _ \right ]^{ (\beta -_\gamma)/\alpha } _ _ }, _ \label{eq:rho_profile}$$ where $(\alpha, \beta,_\gamma, \delta)$_are dimensionless parameters. $r_s$ is_a scaling radius_and $\rho_0$_is the normalization_density. Both have to be determined_for each galaxy. In_this paper the halo profile of_Navarro,_Frenk and White_(NFW) [@NFW;_@navarro_universal_1997]_with $(1,3,1,0)$_is used as_baseline._For the_Milky_Way $r_s=16.1^{+17.0}_{-7.8}\,\mathrm{kpc}$ and $\rho(r_s)=0.47^{+0.05}_{-0.06}\,{\mathrm{GeV}}/\mathrm{cm}^3$ are used [@nesti_dark_2013]._A_currently favored model is the Burkert profile,_that was obtained by_the_observation of dark matter_dominated dwarf galaxies. The Burkert_profile is described by $(2,3,1,1)$ [@Burkert], $r_s=9.26^{+5.6}_{-4.2}\,\mathrm{kpc}$_and $\rho(r_s)=0.49^{+0.07}_{-0.09}\,{\mathrm{GeV}}/\mathrm{cm}^3$ [@nesti_dark_2013]._While for_the central part of the galaxy the models differ by orders_of magnitude, the outer profiles are_rather similar. The expected differential_neutrino flux_$\mathrm{d}\phi_{\nu}/\mathrm{d}E$_at Earth depends_on_the annihilation_rate $\Gamma_A= {\langle \sigma_{\mathrm{A}} v\rangle}\rho(r)^2/2$ along the_line of_sight $l$, the muon neutrino yield_per annihilation $\mathrm{d}N_\nu/\mathrm{d}E$, and_the_self-annihilation cross-section of dark matter averaged_over the

$, $\chi_3$.. are nonlinear susceptibilities. For an atomic spin-glass, $M_0$ and the coefficients of even power of $H$ i.e $\chi_2$, $\chi_4$.. are zero while coefficients of the odd power of $H$ i.e $\chi_3$, $\chi_5$.. diverges as $T$ approaches $T_g$ in the critical regime.[@Suzuki; @Wada; @Fujiki] For cluster-glass, if the nonlinear response of isolated ferromagnetic clusters is small, coefficients of the odd power of $H$ will also diverge in the critical regime similar to that of atomic spin-glass.[@TJonsson] The overall nonlinear susceptibility $\chi_{NL}$ which diverges in the critical temperature regime in a spin-glass system can be written as $$\chi_{NL}=\chi_1-M/H= \chi_3H^2+\chi_5H^4+.....\label{eq:dc scaling}$$ The phenomenological theory of spin-glass by Suziki predicts that the $\chi_{NL}$ should follow the static scaling relation[@Suzuki] $$\begin{aligned} \chi_{NL}&=\epsilon^{\beta}F(H^2/\epsilon^{\beta+\gamma})\label{second}\\ \text{or}\;\;\;\; \chi_{NL}&=H^{2\beta/(\beta+\gamma)}G(H^2/\epsilon^{\beta+\gamma}),\label{third}\end{aligned}$$ where $\epsilon$=(T-T$_g$)/T$_g$ is the reduced temperature, $\beta$ is the critical exponent for spin-glass order parameter, $\gamma$ is the critical exponent for spin-glass susceptibility, and $F(x)$ and $G(x)$ are the scaling functions. The scaling is achieved by plotting $\chi_{NL}/H^{2\beta/(\beta+\gamma)}$ versus $H^2/\epsilon^{\beta+\gamma}$ for $\chi_{NL}$ at different fields and varying the parameters $T_g$, $\beta$ and $\gamma$ such that all data collapse on a master curve. In the limit of $\epsilon\rightarrow$0 the abscissa and ordinate have a span of many decades, and therefore, are plotted on log scales. The log scale plotting gives equal weightage to all data points irrespective of their

$, $ \chi_3 $.. are nonlinear susceptibilities. For an atomic spin - glass, $ M_0 $ and the coefficient of even exponent of $ H$ i.e $ \chi_2 $, $ \chi_4 $.. are zero while coefficients of the odd exponent of $ H$ i.e $ \chi_3 $, $ \chi_5 $.. diverge as $ T$ approaches $ T_g$ in the critical regime.[@Suzuki; @Wada; @Fujiki ] For cluster - field glass, if the nonlinear response of isolated ferromagnetic bunch is humble, coefficients of the leftover power of $ H$ will also deviate in the critical regime similar to that of atomic spin - glass.[@TJonsson ] The overall nonlinear susceptibility $ \chi_{NL}$ which diverges in the critical temperature regime in a spin - field glass system can be written as $ $ \chi_{NL}=\chi_1 - M / H= \chi_3H^2+\chi_5H^4+..... \label{eq: dc scaling}$$ The phenomenological theory of tailspin - glass by Suziki predicts that the $ \chi_{NL}$ should pursue the static scaling relation[@Suzuki ] $ $ \begin{aligned } \chi_{NL}&=\epsilon^{\beta}F(H^2/\epsilon^{\beta+\gamma})\label{second}\\ \text{or}\;\;\;\; \chi_{NL}&=H^{2\beta/(\beta+\gamma)}G(H^2/\epsilon^{\beta+\gamma}),\label{third}\end{aligned}$$ where $ \epsilon$=(T - T$_g$)/T$_g$ is the boil down temperature, $ \beta$ is the critical exponent for spin - glass order parameter, $ \gamma$ is the critical advocate for spin - glass susceptibility, and $ F(x)$ and $ G(x)$ are the scaling functions. The scaling is achieved by plotting $ \chi_{NL}/H^{2\beta/(\beta+\gamma)}$ versus $ H^2/\epsilon^{\beta+\gamma}$ for $ \chi_{NL}$ at different playing field and varying the parameters $ T_g$, $ \beta$ and $ \gamma$ such that all data collapse on a master curve. In the limit of $ \epsilon\rightarrow$0 the abscissa and ordinate have a couple of many decades, and therefore, are plot on logarithm scale. The log scale plotting give equal weightage to all data points irrespective of their

$, $\chl_3$.. are nonlinear susceptinilities. For an cromic vpin-glzss, $M_0$ ana the coefficients of even plwwr of $H$ i.e $\chi_2$, $\chi_4$.. are zero while covfficientw of rhe odd poxsr of $H$ i.e $\chj_3$, $\chi_5$.. viverges as $T$ akproaches $T_g$ in the criticdl rzgime.[@Suzuki; @Wada; @Fujiki] For cluster-gjass, if tje nonlinear rgsponxq of psilated ferromagnetic clusters is smanl, coefficienys of the odd power of $H$ wlll wlso diverge in thf critical tsgiiw similar to that of auokic spin-glzss.[@TJonsson] The overall nonlinexr suxceptibilijv $\chl_{TL}$ which diterges in the critlbal tem[eraturr regime in a xpii-glaws system can be writven as $$\chi_{NL}=\chi_1-M/H= \chi_3R^2+\chi_5H^4+.....\laben{es:dc scaling}$$ The phwnimenonogiwal gyeofy kf slin-glads uy Suziki pdedicts thar the $\chi_{NL}$ should gojoow the statid scalynd relation[@Suzuki] $$\begin{aligned} \chi_{NL}&=\epsilot^{\befa}F(H^2/\epsilon^{\beta+\gamma})\labwl{second}\\ \text{or}\;\;\;\; \chi_{NL}&=H^{2\bgta/(\beta+\gamia)}G(H^2/\epsilon^{\beta+\gamma}),\label{third}\end{aligned}$$ where $\epshlon$=(T-V$_g$)/G$_g$ nw the eefuced temperature, $\beta$ is the critical exponegf gog spin-glass order parameter, $\balms$ is the critizal ex'knsnt for spin-glass dusceptybiliry, and $F(x)$ and $G(x)$ are the scaling functiobs. The scaliug us achieved by plocting $\chi_{NL}/H^{2\yeta/(\beja+\gamms)}$ versus $H^2/\epsilon^{\beta+\gaoma}$ ror $\chi_{NL}$ ah differehg fields and varhinn tve paramtgers $T_g$, $\beta$ and $\damma$ suci thac all daga cpllapsq on a masher cmsve. In the limit ov $\epsnlon\rhghtarrow$0 hhe abscissa and ordinate have e span of mani dacages, and cherefpre, are plottqd on log scalgs. The loy scald plotting gives xqual weightwge to all dadw points irrxspective of rheie

$, $\chi_3$.. are nonlinear susceptibilities. For an $M_0$ the coefficients even power of zero coefficients of the power of $H$ $\chi_3$, $\chi_5$.. diverges as $T$ approaches in the critical regime.[@Suzuki; @Wada; @Fujiki] For cluster-glass, if the nonlinear response of ferromagnetic clusters is small, coefficients of the odd power of $H$ will also in critical similar that of atomic spin-glass.[@TJonsson] The overall nonlinear susceptibility $\chi_{NL}$ which diverges in the critical temperature regime a spin-glass system can be written as $$\chi_{NL}=\chi_1-M/H= scaling}$$ The phenomenological theory spin-glass by Suziki predicts that $\chi_{NL}$ follow the scaling $$\begin{aligned} \text{or}\;\;\;\; \chi_{NL}&=H^{2\beta/(\beta+\gamma)}G(H^2/\epsilon^{\beta+\gamma}),\label{third}\end{aligned}$$ where is the reduced temperature, $\beta$ is the critical exponent for spin-glass order parameter, $\gamma$ is the critical for spin-glass $F(x)$ and are scaling The scaling is plotting $\chi_{NL}/H^{2\beta/(\beta+\gamma)}$ versus $H^2/\epsilon^{\beta+\gamma}$ for $\chi_{NL}$ and varying the parameters $T_g$, $\beta$ and $\gamma$ that all collapse on a master curve. In limit of $\epsilon\rightarrow$0 the abscissa and ordinate have span of many decades, and therefore, are plotted on log scales. The log scale plotting weightage to all data irrespective of their

$, $\chi_3$.. are nonlinear susceptibiLities. For aN atomIc sPin-GlAss, $M_0$ And tHe coefficients OF eveN power of $H$ i.e $\chi_2$, $\chi_4$.. are zeRo whiLe COeffICiEnts oF the odd POwER Of $H$ I.e $\ChI_3$, $\chI_5$.. dIVeRges aS $T$ aPproachEs $T_g$ in the cRitIcAl regime.[@SuzuKI; @WAda; @Fujiki] FOr cLuster-glass, iF thE nonliNeAr rESponsE of IsolaTed ferROmagneTic clusteRs IS small, COefficiENTs Of thE odd power of $H$ will aLSo DIverge in the criTical rEgIMe SIMilAr tO that of atoMiC spin-GLass.[@TJoNSsON] tHe oVErall nonlineaR susceptibiLIty $\Chi_{NL}$ wHiCh dIVerges In the CrITicAl temperatuRe reGime in a spIn-glasS System cAN be writTen as $$\cHi_{Nl}=\chI_1-M/H= \cHI_3H^2+\ChI_5H^4+.....\lAbEL{eq:DC sCalINg}$$ THe phenomEnOlOgicaL theORY OF spiN-glAss bY SuziKi predicts thaT thE $\chi_{nl}$ shOuld fOllow The sTaTic scAling rElatiOn[@suzuki] $$\begin{aligNed} \cHi_{NL}&=\epsilOn^{\bEtA}F(H^2/\EpSilon^{\BEta+\gamMa})\lAbeL{second}\\ \Text{or}\;\;\;\; \cHI_{NL}&=h^{2\bETA/(\BeTa+\gamma)}G(H^2/\epsilon^{\beTa+\GAMmA}),\label{thIrd}\end{ALiGnED}$$ where $\epSiLon$=(t-T$_g$)/T$_G$ IS the rEducED tEmperatuRe, $\beta$ IS tHe CriticaL eXponenT fOr sPin-Glass ORder ParameTer, $\gamma$ Is the CRitical exponenT For spin-glass sUScEPTiBIlitY, anD $F(x)$ and $G(x)$ are The sCAlinG funCTiOns. tHe scaLing iS aCHiEVed by plotting $\chi_{NL}/H^{2\BeTa/(\beta+\Gamma)}$ Versus $H^2/\epsiloN^{\beta+\gamma}$ FOR $\Chi_{NL}$ at dIffeREnT Fields and varyiNg the Parameters $t_G$, $\beta$ and $\Gamma$ Such that All data coLLApse on a mAstEr cUrvE. In THE lImit of $\epsilon\RIGhtaRrOw$0 the abSciSsa and oRdiNatE haVe a SpAn of many dEcades, anD tHeReFoRe, aRe ploTTed on log ScAleS. THe lOg scaLE plottIng giVes eQuAl WEigHtage to ALl DATa poInTs IrreSpeCtIve of TheiR

$, $\chi_3$.. are nonlinea r suscepti bilit ies . F or anatom ic spin-glass, $M_0 $ and the coefficients of e ve n pow e rof $H $ i.e $ \ ch i _ 2$, $ \c hi_ 4$ . .are z ero whilecoefficien tsof the odd pow e rof $H$ i.e $\ chi_3$, $\ch i_5 $.. di ve rge s as $ T$appro aches$ T_g$ i n the cri ti c al reg i me.[@Su z u ki ; @W ada; @Fujiki] For cl u ster-glass, if the n on l in e a r r esp onse of is ol atedf erromag n et i c clu s ters is small , coefficie n tsof the o ddp ower o f $H$ w i llalso diverg e in the crit ical r e gime si m ilar to thatofato mics pi n- gla ss . [@T J on sso n ] T he overa ll n onlin ears u s c epti bil ity$\chi _{NL}$ whichdiv erge s in thecriti calte mpera ture r egime i n a spin-glasssyst em can be wr it ten a s $$ \ chi_{N L}= \ch i_1-M/H = \chi_ 3 H^2 +\ c h i _5 H^4+.....\label{eq :d c sc aling}$$ Thep he no m enologic al th eory o f spi n-gl a ss by Suzi ki pre d ic ts that t he $\chi _{ NL} $ s hould foll ow the staticscali n g relation[@Su z uki] $$\begi n {a l i gn e d} \ chi _{NL}&=\eps ilon ^ {\be ta}F ( H^ 2/\ e psilo n^{\b et a +\ g amma})\label{second }\ \ \tex t{or} \;\;\;\; \chi _{NL}&=H^{ 2 \ b eta/(\be ta+\ g am m a)}G(H^2/\epsi lon^{ \beta+\gam m a}),\lab el{th ird}\end {aligned} $ $ where$\e psi lon $=( T - T$ _g$)/T$_g$ is t he r ed uced te mpe rature, $\ bet a$isth e critica l expone nt f or s pin -glas s order p ar ame te r,$\gam m a$ isthe c riti ca le xpo nent fo r s p i n-gl as ssusc ept ib ility , an d $F (x)$ an d $G(x)$are thesc al ing fun ctions. The s ca ling is ac hi eve d by p l o tting $\ chi_{NL}/H^{2\beta/(\be t a+\gamm a)} $ ver sus$H^2/\eps ilo n^{\be ta+ \ gamma} $ for$\chi _{ NL} $ at di f f er ent f ields andv a ryi ng th epara meters$T_g$, $\beta$ and $\g amma$ such th atalld a ta co l la p seon a m a s ter curve. In t he limit o f$ \e psilon\rig h tar ro w$0 the abscis sa an d ordina te have a span ofma ny d e c ade s, and the refore,are plott e d onl og scal es. The l og sc ale p lottin g gi ves e qual w ei ghtage to a ll data po ints irrespective of th eir

$, $\chi_3$.._are nonlinear_susceptibilities. For an atomic_spin-glass, $M_0$_and_the coefficients_of_even power of_$H$ i.e $\chi_2$,_$\chi_4$.. are zero while_coefficients of the_odd_power of $H$ i.e $\chi_3$, $\chi_5$.. diverges as $T$ approaches $T_g$ in the critical_regime.[@Suzuki;_@Wada; @Fujiki]_For_cluster-glass,_if the nonlinear response of_isolated ferromagnetic clusters is small,_coefficients of_the odd power of $H$ will also diverge_in_the critical regime_similar to that of atomic spin-glass.[@TJonsson] The overall nonlinear_susceptibility $\chi_{NL}$ which diverges in the_critical temperature regime_in_a_spin-glass system can be_written as $$\chi_{NL}=\chi_1-M/H= \chi_3H^2+\chi_5H^4+.....\label{eq:dc scaling}$$ The phenomenological theory_of spin-glass by Suziki predicts that_the $\chi_{NL}$ should follow the static scaling_relation[@Suzuki] $$\begin{aligned} \chi_{NL}&=\epsilon^{\beta}F(H^2/\epsilon^{\beta+\gamma})\label{second}\\ \text{or}\;\;\;\; \chi_{NL}&=H^{2\beta/(\beta+\gamma)}G(H^2/\epsilon^{\beta+\gamma}),\label{third}\end{aligned}$$ where $\epsilon$=(T-T$_g$)/T$_g$ is the reduced temperature,_$\beta$ is the critical exponent_for spin-glass_order parameter, $\gamma$ is the_critical exponent for_spin-glass susceptibility,_and $F(x)$ and_$G(x)$ are the scaling functions. The_scaling is achieved_by plotting $\chi_{NL}/H^{2\beta/(\beta+\gamma)}$ versus $H^2/\epsilon^{\beta+\gamma}$ for_$\chi_{NL}$_at different fields_and_varying_the parameters_$T_g$, $\beta$ and_$\gamma$_such that_all_data collapse on a master curve._In_the limit of $\epsilon\rightarrow$0 the abscissa and_ordinate have a span_of_many decades, and therefore,_are plotted on log scales._The log scale plotting gives equal_weightage to_all data_points irrespective of their

C_0^2\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^d(X)} \\ &\leq 2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^d(X)}.\end{aligned}$$ Third, for the case $d<p<\infty$, $$\begin{aligned} \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} &\leq 2\|\nabla_{A_0}u_0\|_{L^\infty(X)}\|A-A_0\|_{L^p(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^p(X)} \\ &\leq 2C_0\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{L^p(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^p(X)} \\ &\leq 2C_0\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{L^p(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^p(X)}.\end{aligned}$$ Hence, the combination of the preceding three cases gives $$\begin{gathered} \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} \leq 2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,p}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^p(X)}, \\ \quad\text{for } p<\infty \text{ and } d/2 \leq p \leq q.\end{gathered}$$ and

C_0 ^ 2\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A - A_0\|_{W_{A_0}^{1,d}(X) } + \|\nabla_{A_0}(A - A_0)\|_{L^d(X) } \\ & \leq 2C_0 ^ 2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A - A_0\|_{W_{A_0}^{1,d}(X) } + \|\nabla_{A_0}(A - A_0)\|_{L^d(X)}.\end{aligned}$$ Third, for the case $ d < p<\infty$, $ $ \begin{aligned } \|\nabla_{A_0}(u^{-1}(A - A_0)u)\|_{L^p(X) } & \leq 2\|\nabla_{A_0}u_0\|_{L^\infty(X)}\|A - A_0\|_{L^p(X) } + \|\nabla_{A_0}(A - A_0)\|_{L^p(X) } \\ & \leq 2C_0\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A - A_0\|_{L^p(X) } + \|\nabla_{A_0}(A - A_0)\|_{L^p(X) } \\ & \leq 2C_0\|u_0\|_{W_{A_0}^{2,p}(X)}\|A - A_0\|_{L^p(X) } + \|\nabla_{A_0}(A - A_0)\|_{L^p(X)}.\end{aligned}$$ Hence, the combination of the preceding three cases gives $ $ \begin{gathered } \|\nabla_{A_0}(u^{-1}(A - A_0)u)\|_{L^p(X) } \leq 2C_0 ^ 2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A - A_0\|_{W_{A_0}^{1,p}(X) } + \|\nabla_{A_0}(A - A_0)\|_{L^p(X) }, \\ \quad\text{for } p<\infty \text { and } d/2 \leq phosphorus \leq q.\end{gathered}$$ and

C_0^2\|\nahla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + \|\nabla_{A_0}(A-A_0)\|_{U^d(X)} \\ &\leq 2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{C_{Q_0}^{1,d}(X)} + \|\naula_{A_0}(A-A_0)\|_{L^s(X)}.\end{aliened}$$ Third, for the case $d<p<\inhty$, $$\vegin{qligned} \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} &\led 2\|\nabla_{A_0}u_0\|_{L^\pnfty(X)}\|A-A_0\|_{L^p(X)} + \|\nebla_{A_0}(A-A_0)\|_{L^p(X)} \\ &\leq 2C_0\|\negla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{L^p(S)} + \|\nabne_{A_0}(A-A_0)\|_{L^p(X)} \\ &\leq 2C_0\|u_0\|_{W_{A_0}^{2,k}(X)}\|A-A_0\|_{L^p(X)} + \|\nabld_{A_0}(A-A_0)\|_{L^p(X)}.\end{aligtea}$$ Kence, the combination of the precedigg threr fases gives $$\bedin{gsehersd} \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} \leq 2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,p}(X)} + \|\nabma_{A_0}(A-A_0)\|_{L^p(Q)}, \\ \quad\text{for } p<\ingty \text{ and } d/2 \leq p \leq e.\end{hathered}$$ and

C_0^2\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^d(X)} \\ &\leq 2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + for case $d<p<\infty$, \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} &\leq 2\|\nabla_{A_0}u_0\|_{L^\infty(X)}\|A-A_0\|_{L^p(X)} + \\ &\leq 2C_0\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{L^p(X)} \|\nabla_{A_0}(A-A_0)\|_{L^p(X)}.\end{aligned}$$ Hence, the of the preceding three cases gives \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} \leq 2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,p}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^p(X)}, \\ \quad\text{for } p<\infty \text{ and } d/2 p \leq q.\end{gathered}$$ and

C_0^2\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + \|\nabla_{A_0}(A-a_0)\|_{L^d(X)} \\ &\leq 2C_0^2\|u_0\|_{W_{a_0}^{2,p}(X)}\|A-A_0\|_{w_{A_0}^{1,d}(x)} + \|\naBlA_{A_0}(A-A_0)\|_{l^d(X)}.\eNd{aligned}$$ Third, FOr thE case $d<p<\infty$, $$\begin{alignEd} \|\nabLa_{a_0}(U^{-1}(A-A_0)u)\|_{l^P(X)} &\Leq 2\|\naBla_{A_0}u_0\|_{L^\iNFtY(x)}\|a-A_0\|_{L^P(X)} + \|\NaBla_{a_0}(A-a_0)\|_{l^p(x)} \\ &\leq 2C_0\|\NabLa_{A_0}u_0\|_{W_{A_0}^{1,d}(x)}\|A-A_0\|_{L^p(X)} + \|\nablA_{A_0}(A-a_0)\|_{L^P(X)} \\ &\leq 2C_0\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-a_0\|_{l^p(x)} + \|\nabla_{A_0}(A-A_0)\|_{L^P(X)}.\eNd{aligned}$$ HenCe, tHe combInAtiON of thE prEcediNg threE Cases gIves $$\begin{GaTHered} \|\nABla_{A_0}(u^{-1}(A-A_0)U)\|_{l^P(X)} \Leq 2C_0^2\|U_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,p}(X)} + \|\nabla_{A_0}(a-a_0)\|_{L^P(x)}, \\ \quad\text{for } p<\iNfty \teXt{ ANd } D/2 \LEq p \Leq Q.\end{gatherEd}$$ And

C_0^2\|\nabla_{A_0}u_0\|_{ W_{A_0}^{1 ,d}(X )}\ |A- A_ 0\|_ {W_{ A_0}^{1,d}(X)} + \| \nabla_{A_0}(A-A_0)\|_ {L^d( X) } \\& \l eq 2C _0^2\|u _ 0\ | _ {W_ {A _0 }^{ 2, p }( X)}\| A-A _0\|_{W _{A_0}^{1, d}( X) } + \|\nabla _ {A _0}(A-A_0) \|_ {L^d(X)}.\en d{a ligned }$ $ T h ird,for thecase $ d <p<\in fty$, $$\ be g in{ali g ned} \| \ n ab la_{ A_0}(u^{-1}(A-A_0 ) u) \ |_{L^p(X)} &\l eq 2\| \n a bl a _ {A_ 0}u _0\|_{L^\i nf ty(X) } \|A-A_0 \ |_ { L ^ p(X ) } + \|\nabla_ {A_0}(A-A_0 ) \|_ {L^p(X )} \\ &\leq2C_0\ |\ n abl a_{A_0}u_0\ |_{W _{A_0}^{1 ,d}(X) } \|A-A_0 \ |_{L^p( X)} +\|\ nab la_{ A _0 }( A-A _0 ) \|_ { L^ p(X ) } \ \ &\leq2C _0 \|u_0 \|_{ W _ { A _0}^ {2, p}(X )}\|A -A_0\|_{L^p(X )}+ \| \ nab la_{A _0}(A -A_0 )\ |_{L^ p(X)}. \end{ al igned}$$ Hence, the combinat ion o f t he prec e ding t hre e c ases gi ves $$\ b egi n{ g a t he red} \|\nabla_{A_0 }( u ^ {- 1}(A-A_0 )u)\|_ { L^ p( X )} \leq2C _0^ 2\|u _ 0 \|_{W _{A_ 0 }^ {2,p}(X) }\|A-A _ 0\ |_ {W_{A_0 }^ {1,p}( X) } + \| \nabl a _{A_ 0}(A-A _0)\|_{L ^p(X) } , \\ \quad\tex t {for } p<\inf t y\ t ex t { an d } d/2 \leq p \le q q.\ end{ g at her e d}$$and

C_0^2\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^d(X)} \\ &\leq_2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,d}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^d(X)}.\end{aligned}$$_Third, for the case_$d<p<\infty$, $$\begin{aligned} \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} &\leq 2\|\nabla_{A_0}u_0\|_{L^\infty(X)}\|A-A_0\|_{L^p(X)} +_\|\nabla_{A_0}(A-A_0)\|_{L^p(X)} \\ &\leq_2C_0\|\nabla_{A_0}u_0\|_{W_{A_0}^{1,d}(X)}\|A-A_0\|_{L^p(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^p(X)} \\ &\leq_2C_0\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{L^p(X)} +_\|\nabla_{A_0}(A-A_0)\|_{L^p(X)}.\end{aligned}$$ Hence, the_combination of the_preceding three cases gives_$$\begin{gathered} \|\nabla_{A_0}(u^{-1}(A-A_0)u)\|_{L^p(X)} \leq 2C_0^2\|u_0\|_{W_{A_0}^{2,p}(X)}\|A-A_0\|_{W_{A_0}^{1,p}(X)} + \|\nabla_{A_0}(A-A_0)\|_{L^p(X)}, \\ \quad\text{for }_p<\infty_\text{ and } d/2 \leq p \leq q.\end{gathered}$$ and

+1 dimensions and show that the massive vector field in the Proca theory is canonically spin 1. Consider the following Proca Lagrangian \[Apl\] [L]{}\_P= -F\_F\^+ m\^2 [A]{}\_\^2. After the canonical quantization of the theory by introducing Dirac bracket, we obtain the following commutation relations: \[Acr\] && \[A\_0([**x**]{}), \_0([**y**]{})\] =0,\ && \[A\_i([**x**]{}), \^j([**y**]{})\] =i\_[ij]{}\^2([**x**]{} - [**y**]{}), where conjugate momenta $\pi^j \equiv \frac{\partial {\cal L}_P} {\partial {\dot A}_j}=F^{j0}$. Then the energy-momentum tensor $T_{\mu\nu}$ for the Proca theory (\[Apl\]) is given as \[Apem\] T\_ =-F\_[F\_]{}\^ +m\^2A\_A\_-g\_[L]{}\_P. With the energy-momentum tensor, the Poincaré generators can be expressed as following forms, \[App\] & P\_i=d\^2 x T\_[0i]{}=d\^2 x \^k(x)\_i A\_k(x),\ & H=d\^2 x T\_[00]{}=d\^2 x {\^i(x)\^i(x) +F\^[ij]{}(x)F\_[ij]{}(x)+(\_i\^i(x))\^2+ A\^i(x)A\^i(x)},\ &L=d\^2 x \^[ij]{} x\_i T\_[0j]{}= d\^2 x \^[ij]{} x\_i\^k(x)\_[j]{}A\_k(x) -d\^2 x \^[ij]{}\^i(x)A\_j(x),\ &M\_[i0]{}=d\^2 x x\^i {\^j(x)\^j(x) +F\^[jk]{}(x)F\_[jk]{}(x)+(\_j\^j(x))\^2+ A\^j(x)A\^j(x)}-tP\_i, using the equation of motion, $A_0=-\frac{1}{m^2}\partial_i F^{i0} =-\frac{1}{m^2}\partial_i \pi^i$. After

+1 dimensions and show that the massive vector field in the Proca theory is canonically spin 1. think the take after Proca Lagrangian \[Apl\ ] [ L]{}\_P= -F\_F\^+ m\^2 [ A]{}\_\^2. After the canonical quantization of the theory by introducing Dirac bracket, we prevail the following commutation relation: \[Acr\ ] & & \[A\_0([**x * * ] { }), \_0([**y**]{})\ ] = 0,\ & & \[A\_i([**x * * ] { }), \^j([**y**]{})\ ] = i\_[ij]{}\^2([**x * * ] { } - [ * * y * * ] { }), where conjugate solution momenta $ \pi^j \equiv \frac{\partial { \cal L}_P } { \partial { \dot A}_j}=F^{j0}$. Then the energy - momentum tensor $ T_{\mu\nu}$ for the Proca theory (\[Apl\ ]) is given as \[Apem\ ] T\ _ = -F\_[F\_]{}\^ + m\^2A\_A\_-g\_[L]{}\_P. With the energy - momentum tensor, the Poincaré generators can be express as following forms, \[App\ ] & P\_i = d\^2 x T\_[0i]{}=d\^2 x \^k(x)\_i A\_k(x),\ & H = d\^2 x T\_[00]{}=d\^2 x { \^i(x)\^i(x) + F\^[ij]{}(x)F\_[ij]{}(x)+(\_i\^i(x))\^2 + A\^i(x)A\^i(x)},\ & L = d\^2 x \^[ij ] { } x\_i T\_[0j]{}= d\^2 x \^[ij ] { } x\_i\^k(x)\_[j]{}A\_k(x) -d\^2 x \^[ij]{}\^i(x)A\_j(x),\ & M\_[i0]{}=d\^2 x x\^i { \^j(x)\^j(x) + F\^[jk]{}(x)F\_[jk]{}(x)+(\_j\^j(x))\^2 + A\^j(x)A\^j(x)}-tP\_i, use the equality of motion, $ A_0=-\frac{1}{m^2}\partial_i F^{i0 } = -\frac{1}{m^2}\partial_i \pi^i$. After

+1 dilensions and show that tme massive vectot dield mn the Lroca thdory is canonically spin 1. Condieer tye following Proca Lagfangian \[Ail\] [L]{}\_P= -F\_F\^+ n\^2 [A]{}\_\^2. Efter the canonirzl quanbnzatikk of chx theory by inttoducing Dirdc bracket, we mbgann the following commutation relatiogs: \[Acr\] && \[A\_0([**d**]{}), \_0([**y**]{})\] =0,\ && \[A\_i([**x**]{}), \^j([**y**]{})\] =i\_[it]{}\^2([**x**]{} - [**j**]{}), rhers conjugate momenta $\pi^j \equiv \frac{\lartial {\cal L}_P} {\partiak {\dot A}_j}=F^{j0}$. Then the energy-lomejtum tensor $T_{\mu\nu}$ vor the Proxa trwory (\[Apl\]) is eiven as \[Akek\] T\_ =-F\_[F\_]{}\^ +m\^2A\_A\_-f\_[L]{}\_P. With the energy-momentum tenror, tke Poincaré teberwjors can be xxpresfed as following forks, \[App\] & P\_i=d\^2 x T\_[0i]{}=d\^2 x \^k(w)\_i A\_k(e),\ & H=d\^2 x T\_[00]{}=d\^2 x {\^i(x)\^i(x) +F\^[ij]{}(x)F\_[ij]{}(x)+(\_m\^i(x))\^2+ A\^i(x)A\^i(x)},\ &L=d\^2 x \^[ij]{} x\_i T\_[0j]{}= d\^2 x \^[ib]{} r\_i\^k(x)\_[j]{}A\_k(x) -d\^2 x \^[ij]{}\^i(x)A\_j(x),\ &M\_[u0]{}=d\^2 x x\^i {\^j(x)\^j(f) +F\^[jy]{}(z)F\_[jy]{}(x)+(\_j\^n(x))\^2+ A\^n(x)A\^j(x)}-tO\_i, nsing the esuation of notion, $A_0=-\frac{1}{m^2}\partiak_i D^{i0} =-\frac{1}{m^2}\partiam_i \pi^i$. Aster

+1 dimensions and show that the massive in Proca theory canonically spin 1. \[Apl\] -F\_F\^+ m\^2 [A]{}\_\^2. the canonical quantization the theory by introducing Dirac bracket, obtain the following commutation relations: \[Acr\] && \[A\_0([**x**]{}), \_0([**y**]{})\] =0,\ && \[A\_i([**x**]{}), \^j([**y**]{})\] - [**y**]{}), where conjugate momenta $\pi^j \equiv \frac{\partial {\cal L}_P} {\partial {\dot A}_j}=F^{j0}$. the tensor for Proca theory (\[Apl\]) is given as \[Apem\] T\_ =-F\_[F\_]{}\^ +m\^2A\_A\_-g\_[L]{}\_P. With the energy-momentum tensor, the Poincaré can be expressed as following forms, \[App\] & x T\_[0i]{}=d\^2 x \^k(x)\_i & H=d\^2 x T\_[00]{}=d\^2 x +F\^[ij]{}(x)F\_[ij]{}(x)+(\_i\^i(x))\^2+ &L=d\^2 x x\_i d\^2 \^[ij]{} x\_i\^k(x)\_[j]{}A\_k(x) -d\^2 \^[ij]{}\^i(x)A\_j(x),\ &M\_[i0]{}=d\^2 x x\^i {\^j(x)\^j(x) +F\^[jk]{}(x)F\_[jk]{}(x)+(\_j\^j(x))\^2+ A\^j(x)A\^j(x)}-tP\_i, using the equation of motion, $A_0=-\frac{1}{m^2}\partial_i F^{i0} =-\frac{1}{m^2}\partial_i \pi^i$. After

+1 dimensions and show that the mAssive vectOr fieLd iN thE PRoca TheoRy is canonicallY Spin 1. consider the following PrOca LaGrANgiaN \[apL\] [L]{}\_P= -F\_F\^+ M\^2 [A]{}\_\^2. After THe CANonIcAl QuaNtIZaTion oF thE theory By introducIng diRac bracket, we OBtAin the follOwiNg commutatioN reLationS: \[ACr\] && \[A\_0([**X**]{}), \_0([**Y**]{})\] =0,\ && \[A\_i([**x**]{}), \^j([**Y**]{})\] =i\_[iJ]{}\^2([**x**]{} - [**y**]{}), whEre conJUgate mOmenta $\pi^j \EqUIv \frac{\PArtial {\cAL l}_P} {\PartIal {\dot A}_j}=F^{j0}$. Then the ENeRGy-momentum tensOr $T_{\mu\nU}$ fOR tHE proCa tHeory (\[Apl\]) is GiVen as \[aPem\] T\_ =-F\_[F\_]{}\^ +m\^2a\_a\_-g\_[l]{}\_p. wIth THe energy-momenTum tensor, thE poiNcaré gEnEraTOrs can Be expReSSed As following FormS, \[App\] & P\_i=d\^2 x T\_[0I]{}=d\^2 x \^k(x)\_i a\_K(x),\ & H=d\^2 x T\_[00]{}=d\^2 X {\^I(x)\^i(x) +F\^[ij]{}(X)F\_[ij]{}(x)+(\_i\^I(x))\^2+ A\^I(x)A\^I(x)},\ &L=d\^2 X \^[Ij]{} X\_i t\_[0j]{}= d\^2 X \^[iJ]{} X\_i\^k(X)\_[J]{}A\_K(x) -d\^2 X \^[Ij]{}\^i(X)A\_j(x),\ &M\_[i0]{}=d\^2 x X\^i {\^J(x)\^J(x) +F\^[jk]{}(X)F\_[jk]{}(X)+(\_J\^J(X))\^2+ a\^j(x)A\^J(x)}-tp\_i, usIng thE equation of moTioN, $A_0=-\frAC{1}{m^2}\pArtiaL_i F^{i0} =-\fRac{1}{m^2}\PaRtial_I \pi^i$. AfTer

+1 dimensions and show tha t the mass ive v ect orfi eldin t he Proca theor y iscanonically spin 1. C onsid er thef ol lowin g Proca La g r ang ia n\[A pl \ ][L]{} \_P = -F\_F \^+ m\^2 [ A]{ }\ _\^2. Aftert he canonical qu antization o f t he the or y b y intr odu cingDiracb racket , we obta in the fo l lowingc o mm utat ion relations: \[ A cr \ ] && \[A\_0([* *x**]{ }) , \ _ 0 ([* *y* *]{})\] =0 ,\ && \ [ A\_i([* * x* * ] { }), \^j([**y**]{} )\] =i\_[ij ] {}\ ^2([** x* *]{ } - [** y**]{ }) , wh ere conjuga te m omenta $\ pi^j \ e quiv \f r ac{\par tial { \ca l L }_P} {\ pa rti al {\d o tA}_ j }=F ^{j0}$.Th en theener g y - m omen tum ten sor $ T_{\mu\nu}$ f ortheP roc a the ory ( \[Ap l\ ]) is given as \ [A pem\] T\_ =-F\_ [F\_ ]{}\^ +m\ ^2A \_ A\_ -g \_[L] { }\_P.Wit h t he ener gy-mome n tum t e n s or , the Poincaré gen er a t or s can be expre s se da s follow in g f orms , \[App \] & P\ _i=d\^2x T\_[ 0 i] {} =d\^2 x \ ^k(x)\ _i A\ _k( x),\& H=d \^2 xT\_[00]{ }=d\^ 2 x {\^i(x)\^i( x ) +F\^[ij]{}( x )F \ _ [i j ]{}( x)+ (\_i\^i(x)) \^2+ A\^i (x)A \ ^i (x) } ,\ &L =d\^2 x \^ [ ij]{} x\_i T\_[0j]{ }= d\^2x \^[ ij]{} x\_i\^k (x)\_[j]{} A \ _ k(x) -d\ ^2 x \^ [ ij]{}\^i(x)A\_ j(x), \ &M\_[i0] { }=d\^2 x x\^i {\^j(x) \^j(x) +F \ ^ [jk]{}(x )F\ _[j k]{ }(x ) + (\ _j\^j(x))\^2+ A \^j( x) A\^j(x) }-t P\_i, u sin g t heequ at ion of mo tion, $A _0 =- \f ra c{1 }{m^2 } \partial _i F^ {i 0}=-\fr a c{1}{m ^2}\p arti al _i \pi ^i$. Af t er

+1 dimensions_and show_that the massive vector_field in_the_Proca theory_is_canonically spin 1. Consider_the following Proca_Lagrangian \[Apl\] [L]{}\_P= -F\_F\^+_m\^2 [A]{}\_\^2. After_the_canonical quantization of the theory by introducing Dirac bracket, we obtain the following commutation_relations:_\[Acr\] &&_\[A\_0([**x**]{}),_\_0([**y**]{})\]_=0,\ && \[A\_i([**x**]{}), \^j([**y**]{})\] =i\_[ij]{}\^2([**x**]{} -_[**y**]{}), where conjugate momenta $\pi^j_\equiv \frac{\partial_{\cal L}_P} {\partial {\dot A}_j}=F^{j0}$. Then the energy-momentum tensor_$T_{\mu\nu}$_for the Proca_theory (\[Apl\]) is given as \[Apem\] T\_ =-F\_[F\_]{}\^ +m\^2A\_A\_-g\_[L]{}\_P._With the energy-momentum tensor, the Poincaré_generators can be_expressed_as_following forms, \[App\] &_P\_i=d\^2 x T\_[0i]{}=d\^2 x \^k(x)\_i A\_k(x),\ &_H=d\^2 x T\_[00]{}=d\^2 x {\^i(x)\^i(x) +F\^[ij]{}(x)F\_[ij]{}(x)+(\_i\^i(x))\^2+_A\^i(x)A\^i(x)},\ &L=d\^2 x \^[ij]{} x\_i T\_[0j]{}= d\^2 x_\^[ij]{} x\_i\^k(x)\_[j]{}A\_k(x) -d\^2 x \^[ij]{}\^i(x)A\_j(x),\ &M\_[i0]{}=d\^2 x_x\^i {\^j(x)\^j(x) +F\^[jk]{}(x)F\_[jk]{}(x)+(\_j\^j(x))\^2+ A\^j(x)A\^j(x)}-tP\_i, using_the equation_of motion, $A_0=-\frac{1}{m^2}\partial_i F^{i0} =-\frac{1}{m^2}\partial_i \pi^i$._After

f$ to denote its isotropic part. Lemma \[pd1\] and Proposition \[conve\] supply a function $g$ so that $$g(s,t)=\sum_{k,l=0}^{\infty}\hat{f}_{k,l}P_{k}^{m}(t)P_{l}^{M}(s), \quad t,s \in [-1,1],$$ with uniform convergence in $[-1,1]^2$. In particular, $g$ is continuous in $[-1,1]^2$. On the other hand, the same uniform convergence and the orthogonality relation mentioned at the beginning of the section imply that $$\hat{f}_{k,l} -\hat{g}_{k,l}=0, \quad k,l \in \mathbb{Z}_+.$$ Consequently, $f=g$. Positive definiteness on $S^\infty \times S^M$ ============================================== In this section, we extend Theorem \[mainPD\] to the cases in which either $m=\infty$ or $M=\infty$. Clearly, it suffices to consider the cases $m=\infty$, $M<\infty$ and $m=M=\infty$ only. Every sphere $S^{m}$ can be isometrically embedded in $S^{\infty}$. In particular, a positive definite kernel on $S^\infty \times S^{M}$ is positive definite on $S^{m}\times S^{M}$, for $m=1,2,\ldots$. Likewise, if $f$ is the isotropic part of a positive definite kernel on $S^\infty \times S^{M}$, then it is the isotropic part of a positive definite kernel on $S^{m}\times S^{M}$, for $m=1,2, \ldots$. In addition, if $f$ is continuous, then for every $m\geq 1$, we have a representation for $f$ in the form $$f(t,s)=\sum_{k,l=0}^{\infty}\hat{f}_{k,l}^{m,M} P_{k}^{m}(t) P_{l}^{M}(s), \quad t,s \in [-1,1],$$ in which $$\hat{f}_{k,l}^{m,M} =\frac{1}{\tau_{k}^{m} \tau_{l}^{M}} \int_{[-1,1

f$ to denote its isotropic part.   Lemma \[pd1\ ] and Proposition \[conve\ ] supply a function $ g$ so that $ $ g(s, t)=\sum_{k, l=0}^{\infty}\hat{f}_{k, l}P_{k}^{m}(t)P_{l}^{M}(s), \quad metric ton, s \in [ -1,1],$$ with consistent convergence in $ [ -1,1]^2$.   In particular, $ g$ is continuous in $ [ -1,1]^2$.   On the other handwriting, the same uniform overlap and the orthogonality relation mentioned at the beginning of the section entail that $ $ \hat{f}_{k, l } -\hat{g}_{k, l}=0, \quad k, l \in \mathbb{Z}_+.$$ Consequently, $ f = g$. Positive determinateness on $ S^\infty \times S^M$ = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = In this section, we extend Theorem \[mainPD\ ] to the cases in which either $ m=\infty$ or $ M=\infty$.   Clearly, it suffice to consider the cases $ m=\infty$, $ M<\infty$ and $ thousand = M=\infty$ only. Every sphere $ S^{m}$ can be isometrically implant in $ S^{\infty}$.   In particular, a positive definite kernel on $ S^\infty \times S^{M}$ is positive definite on $ S^{m}\times S^{M}$, for $ m=1,2,\ldots$.   alike, if $ f$ is the isotropic part of a positive definite kernel on $ S^\infty \times S^{M}$, then it is the isotropic part of a convinced definite kernel on $ S^{m}\times S^{M}$, for $ m=1,2, \ldots$.   In addition, if $ f$ is continuous, then for every $ m\geq 1 $, we have a representation for $ f$ in the form $ $ f(t, s)=\sum_{k, l=0}^{\infty}\hat{f}_{k, l}^{m, M } P_{k}^{m}(t) P_{l}^{M}(s), \quad t, s \in [ -1,1],$$ in which $ $ \hat{f}_{k, l}^{m, M } = \frac{1}{\tau_{k}^{m } \tau_{l}^{M } } \int_{[-1,1

f$ tl denote its isotropic pxrt. Lemma \[pd1\] and Proposmtion \[cknve\] supoly a function $g$ so that $$g(s,t)=\dun_{k,l=0}^{\indty}\hat{f}_{k,l}P_{k}^{m}(t)P_{l}^{M}(s), \quad t,s \in [-1,1],$$ wpth unifoem cibvergence mh $[-1,1]^2$. In pavciculzv, $g$ iv continuous in $[-1,1]^2$. On the otver hand, the sdmd bniform convergence and the orthogonwlity rrlwtion mentioneq at ehe gvglnning of the section imply that $$\hat{f}_{k,l} -\hat{g}_{k,l}=0, \quad l,l \in \mathbb{Z}_+.$$ Consequently, $f=g$. Plsitive definiteneds on $S^\infti \tiiws S^M$ ============================================== In this section, wt zxtend Theotem \[mainPD\] to the cases in which eithzr $m=\infty$ ot $J=\ijxty$. Clearly, mt sufsices to consider tha cases $m=\infty$, $M<\infty$ anv $m=M=\unfty$ only. Every spherx $S^{m}$ can be isometrisally embadbed in $S^{\infty}$. In partixuoar, a poshtivd dewinjtx ksrnel ln $A^\infty \timss S^{M}$ is powitive definite on $X^{m}\epkes S^{M}$, for $m=1,2,\mdots$. Lykqwise, if $f$ is the isotropic part of a pmsifive definite kernel on $S^\infty \times S^{M}$, then it is thq isotropic part of a positive definite kernel on $S^{m}\tijds W^{M}$, wir $m=1,2, \ldots$. In addition, if $f$ is continuous, then fjd tvegy $m\geq 1$, we have c representation flr $s$ in the form $$f(t,s)=\sum_{k,l=0}^{\jnfty}\hat{f}_{k,l}^{m,M} P_{k}^{m}(t) P_{l}^{M}(s), \qoad t,s \in [-1,1],$$ in wrich $$\hat{f}_{k,l}^{m,M} =\frac{1}{\tau_{k}^{m} \tau_{l}^{M}} \unt_{[-1,1

f$ to denote its isotropic part. Lemma Proposition supply a $g$ so that with convergence in $[-1,1]^2$. particular, $g$ is in $[-1,1]^2$. On the other hand, same uniform convergence and the orthogonality relation mentioned at the beginning of the imply that $$\hat{f}_{k,l} -\hat{g}_{k,l}=0, \quad k,l \in \mathbb{Z}_+.$$ Consequently, $f=g$. Positive definiteness on \times ============================================== this we extend Theorem \[mainPD\] to the cases in which either $m=\infty$ or $M=\infty$. Clearly, it suffices consider the cases $m=\infty$, $M<\infty$ and $m=M=\infty$ only. sphere $S^{m}$ can be embedded in $S^{\infty}$. In particular, positive kernel on \times is definite on $S^{m}\times for $m=1,2,\ldots$. Likewise, if $f$ is the isotropic part of a positive definite kernel on $S^\infty \times then it isotropic part a definite on $S^{m}\times S^{M}$, \ldots$. In addition, if $f$ is every $m\geq 1$, we have a representation for in the $$f(t,s)=\sum_{k,l=0}^{\infty}\hat{f}_{k,l}^{m,M} P_{k}^{m}(t) P_{l}^{M}(s), \quad t,s \in in which $$\hat{f}_{k,l}^{m,M} =\frac{1}{\tau_{k}^{m} \tau_{l}^{M}} \int_{[-1,1

f$ to denote its isotropic part. lemma \[pd1\] and propoSitIon \[CoNve\] sUpplY a function $g$ so tHAt $$g(s,T)=\sum_{k,l=0}^{\infty}\hat{f}_{k,l}P_{k}^{m}(t)P_{L}^{M}(s), \quAd T,S \in [-1,1],$$ wITh UnifoRm conveRGeNCE in $[-1,1]^2$. in PaRtiCuLAr, $G$ is coNtiNuous in $[-1,1]^2$. on the other HanD, tHe same uniforM CoNvergence aNd tHe orthogonalIty RelatiOn MenTIoned At tHe begInning OF the seCtion implY tHAt $$\hat{f}_{K,L} -\hat{g}_{k,l}=0, \QUAd K,l \in \Mathbb{Z}_+.$$ ConsequentLY, $f=G$. positive definiTeness On $s^\InFTY \tiMes s^M$ ============================================== In this seCtIon, we EXtend ThEOrEM \[MAinpd\] to the cases in Which either $M=\InfTy$ or $M=\iNfTy$. CLEarly, iT suffIcES to Consider the CaseS $m=\infty$, $M<\iNfty$ anD $M=M=\infty$ ONly. EverY spherE $S^{m}$ Can Be isOMeTrIcaLlY EmbEDdEd iN $s^{\inFty}$. In parTiCuLar, a pOsitIVE DEfinIte KernEl on $S^\Infty \times S^{M}$ iS poSitiVE deFinitE on $S^{m}\TimeS S^{m}$, for $m=1,2,\Ldots$. LIkewiSe, If $f$ is the isotropIc paRt of a posiTivE dEfiNiTe kerNEl on $S^\iNftY \tiMes S^{M}$, thEn it is tHE isOtROPIc Part of a positive defInITE kErnel on $S^{M}\times s^{m}$, fOr $M=1,2, \Ldots$. In aDdItiOn, if $F$ IS contInuoUS, tHen for evEry $m\geQ 1$, We HaVe a reprEsEntatiOn For $F$ in The foRM $$f(t,s)=\Sum_{k,l=0}^{\iNfty}\hat{f}_{K,l}^{m,M} P_{K}^{M}(t) P_{l}^{M}(s), \quad t,s \in [-1,1],$$ IN which $$\hat{f}_{k,l}^{m,m} =\FrAC{1}{\TaU_{K}^{m} \taU_{l}^{M}} \Int_{[-1,1

f$ to denote its isotropic part. Lem ma \[ pd1 \]an d Pr opos ition \[conve\ ] sup ply a function $g$ sothat$$ g (s,t ) =\ sum_{ k,l=0}^ { \i n f ty} \h at {f} _{ k ,l }P_{k }^{ m}(t)P_ {l}^{M}(s) , \ qu ad t,s \in [ - 1, 1],$$ with un iform conver gen ce in$[ -1, 1 ]^2$.  In part icular , $g$ i s continu ou s in $[ - 1,1]^2$ . On the other hand, thes am e uniform conve rgence a n dt h e o rth ogonalityre latio n mentio n ed a t th e beginning of the sectio n im ply th at $$ \ hat{f} _{k,l }- \ha t{g}_{k,l}= 0, \ quad k,l\in \m a thbb{Z} _ +.$$ Co nseque ntl y,$f=g $ . P osi ti v e d e fi nit e nes s on $S^ \i nf ty \t imes S ^ M $ == === ==== ===== ============= === ==== = === ===== === In t hi s sec tion,we ex te nd Theorem \[ma inPD \] to the ca se s i nwhich either $m =\i nfty$ o r $M=\i n fty $. C l ea rly, it suffices t oc o ns ider the cases $m =\ i nfty$, $ M< \in fty$ a nd $m =M=\ i nf ty$ only . Eve r ysp here $S ^{ m}$ ca nbeiso metri c ally embed ded in $ S^{\i n fty}$. In part i cular, a posi t iv e de f init e k ernel on $S ^\in f ty \ time s S ^{M } $ isposit iv e d e finite on $S^{m}\ti me s S^{M }$, f or $m=1,2,\ld ots$. Like w i s e, if $f $ is th e isotropic par t ofa positive definite kern el on $S ^\infty \ t i mes S^{M }$, th eniti s t he isotropicp a rt o fa posit ive defini teker nel on $ S^{m}\tim es S^{M} $, f or $ m=1 ,2, \ l dots$. I nadd it ion , if$ f$ isconti nuou s, t h enfor eve r y$ m \geq 1 $, wehav ea rep rese n tat ion for $f$ in t hef orm$$ f( t,s)=\s um_{k,l=0}^{\ in fty}\hat{f }_ {k, l}^{m, M } P_{k}^{ m}(t) P_{l}^{M}(s), \qu a d t,s \ in[-1,1 ],$$ in which $$ \hat{f }_{ k ,l}^{m ,M} =\ frac{ 1} {\t a u _{k}^ { m }\ta u_ {l}^{M}} \ i nt_ {[-1, 1

f$ to_denote its_isotropic part. Lemma \[pd1\] and_Proposition \[conve\]_supply_a function_$g$_so that $$g(s,t)=\sum_{k,l=0}^{\infty}\hat{f}_{k,l}P_{k}^{m}(t)P_{l}^{M}(s),_\quad t,s \in_[-1,1],$$ with uniform convergence_in $[-1,1]^2$. In particular,_$g$_is continuous in $[-1,1]^2$. On the other hand, the same uniform convergence and the orthogonality_relation_mentioned at_the_beginning_of the section imply that_$$\hat{f}_{k,l} -\hat{g}_{k,l}=0, \quad k,l \in_\mathbb{Z}_+.$$ Consequently,_$f=g$. Positive definiteness on $S^\infty \times S^M$ ============================================== In this section,_we_extend Theorem \[mainPD\]_to the cases in which either $m=\infty$ or $M=\infty$. Clearly,_it suffices to consider the cases_$m=\infty$, $M<\infty$ and_$m=M=\infty$_only. Every_sphere $S^{m}$ can be_isometrically embedded in $S^{\infty}$. In particular, a_positive definite kernel on $S^\infty \times_S^{M}$ is positive definite on $S^{m}\times S^{M}$,_for $m=1,2,\ldots$. Likewise, if $f$ is the_isotropic part of a positive_definite kernel_on $S^\infty \times S^{M}$, then_it is the_isotropic part_of a positive_definite kernel on $S^{m}\times S^{M}$, for_$m=1,2, \ldots$. In addition,_if $f$ is continuous, then for_every_$m\geq 1$, we_have_a_representation for_$f$ in the_form_$$f(t,s)=\sum_{k,l=0}^{\infty}\hat{f}_{k,l}^{m,M} P_{k}^{m}(t)_P_{l}^{M}(s),_\quad t,s \in [-1,1],$$ in which_$$\hat{f}_{k,l}^{m,M}_=\frac{1}{\tau_{k}^{m} \tau_{l}^{M}} \int_{[-1,1

aux, E., Bouvier, J., Stauffer, J. R., Cuillandre, J.-C.  2003, A&A, 400, 891 Muench, A. A., Lada, E. A., Lada, C. J., & Alves, J. 2002, ApJ, 573, 366 Nakajima, T., Oppenheimer, B. R., Kulkarni, S. R., Golimowski, D. A., Matthews, K., Durrance, S. T. 1995, Nature, 378, 463 Pinfield, D. J., Jameson, R. F., & Hodgkin, S. T. 1998, MNRAS, 299, 955 Pinfield, D. J., Hodgkin, S. T., Jameson, R. F., Cossburn, M. R., Hambly, N. C., & Devereux, N. 2000,, 313, 347 Preibisch, T., Stanke, T., Zinnecker, H. 2003, A&A, 409,147 Raboud, D. & Mermilliod, J.-C. 1998,, 333, 897 Rebolo, R., Zapatero-Osorio, M. R., Martín, E. L. 1995, Nature, 377, 129 Reipurth, B.  2000, AJ, 120, 3177 Reipurth, B. & Clarke, C. 2001, AJ, 122, 432 Stauffer, J. R., Schultz G., & Kirkpatrick J. D. 1998, ApJ, 499, L199 Sterzik, M. F. & Durisen, R. H. 2003, A&A, 400, 1031 van Leeuwen, F. 1983, PhD thesis, University of Leiden Zapatero Osorio, M. R., Rebolo, R., Mart[í]{}n, E. L., Hodgkin, S. T., Cossburn, M. R., Magazz[ù]{}, A., Steele, I. A., & Jameson, R. F. 1999,, 134, 537 --- abstract: | We

aux, E., Bouvier, J., Stauffer, J.   R., Cuillandre, J.-C.   2003, A&A, 400, 891 Muench, A.   A., Lada, E.   A., Lada, C.   J., & Alves, J.   2002, ApJ, 573, 366 Nakajima, T., Oppenheimer, B.   R., Kulkarni, S.   R., Golimowski, D.   A., Matthews, K., Durrance, S.   T.   1995, Nature, 378, 463 Pinfield, D.   J., Jameson, R.   F., & Hodgkin, S.   T.   1998, MNRAS, 299, 955 Pinfield, D.   J., Hodgkin, S.   T., Jameson, R.   F., Cossburn, M.   R., Hambly, N.   C., & Devereux, N.   2000, , 313, 347 Preibisch, T., Stanke, T., Zinnecker, H.   2003, A&A, 409,147 Raboud, D.   & Mermilliod, J.-C.   1998, , 333, 897 Rebolo, R., Zapatero - Osorio, M.   R., Martín, E.   L.   1995, Nature, 377, 129 Reipurth, B.    2000, AJ, 120, 3177 Reipurth, B.   & Clarke, C.   2001, AJ, 122, 432 Stauffer, J.   R., Schultz G., & Kirkpatrick J.   D.   1998, ApJ, 499, L199 Sterzik, M.   F.   & Durisen, R.   H.   2003, A&A, 400, 1031 van Leeuwen, F. 1983, PhD thesis, University of Leiden Zapatero Osorio, M.   R., Rebolo, R., Mart[í]{}n, E.   L., Hodgkin, S.   T., Cossburn, M.   R., Magazz[ù ] { }, A., Steele, I.   A., & Jameson, R.   F.   1999, , 134, 537 --- abstract: | We

aux, E., Bouvier, J., Stauffer, J. R., Cuillandre, J.-C.  2003, A&A, 400, 891 Mnench, A. Z., Lada, E. X., Lada, C. J., & Alves, J. 2002, ApJ, 573, 366 Nakejimq, T., Okienheimer, B. R., Kulkarni, S. R., Golimlwski, D. A., Matuhews, K., Durrance, S. T. 1995, Naturc, 378, 463 Pjkfielb, V. J., Jameson, R. F., & Modgkin, S. T. 1998, KNRAS, 299, 955 Pinfiend, D. L., Hodgkin, S. T., Jameson, R. F., Cossburn, M. R., Rambly, M. C., & Devereux, N. 2000,, 313, 347 Prepbysch, N., Wtanke, T., Zinnecker, H. 2003, A&A, 409,147 Rabohd, D. & Megmilliod, J.-C. 1998,, 333, 897 Rebplo, R., Zapatero-Osorio, M. R., Magtín, F. L. 1995, Nature, 377, 129 Reipurhh, B.  2000, AJ, 120, 3177 Rgjpuwrh, B. & Clarke, Z. 2001, AJ, 122, 432 Stalyfer, J. R., Scholtz G., & Kirkpatrick J. D. 1998, ApJ, 499, L199 Stdrzik, M. F. & Duriseb, E. H. 2003, W&D, 400, 1031 van Leenwen, F. 1983, PhD thesis, Univershty of Keiden Zapaterp Ovoruo, M. R., Rebolo, R., Mart[í]{}n, X. L., Hodgkin, S. T., Cossbutn, M. R., Magasz[ù]{}, A., Steele, I. A., & Jamwsin, R. F. 1999,, 134, 537 --- atstrxxt: | Wx

aux, E., Bouvier, J., Stauffer, J. R., 2003, 400, 891 A. A., Lada, & J. 2002, ApJ, 366 Nakajima, T., B. R., Kulkarni, S. R., Golimowski, A., Matthews, K., Durrance, S. T. 1995, Nature, 378, 463 Pinfield, D. J., R. F., & Hodgkin, S. T. 1998, MNRAS, 299, 955 Pinfield, D. J., S. Jameson, F., M. R., Hambly, N. C., & Devereux, N. 2000,, 313, 347 Preibisch, T., Stanke, T., Zinnecker, 2003, A&A, 409,147 Raboud, D. & Mermilliod, J.-C. 333, 897 Rebolo, R., M. R., Martín, E. L. Nature, 129 Reipurth, 2000, 120, Reipurth, B. & C. 2001, AJ, 122, 432 Stauffer, J. R., Schultz G., & Kirkpatrick J. D. 1998, ApJ, 499, Sterzik, M. Durisen, R. 2003, 400, van Leeuwen, F. thesis, University of Leiden Zapatero Osorio, R., Mart[í]{}n, E. L., Hodgkin, S. T., Cossburn, R., Magazz[ù]{}, Steele, I. A., & Jameson, R. 1999,, 134, 537 --- abstract: | We

aux, E., Bouvier, J., Stauffer, J. R., CuiLlandre, J.-C.  2003, A&a, 400, 891 MuenCh, A. a., LaDa, e. A., LaDa, C. J., & alves, J. 2002, ApJ, 573, 366 NakajIMa, T., OPpenheimer, B. R., Kulkarni, S. R., golimOwSKi, D. A., mAtThews, k., DurranCE, S. t. 1995, nAtuRe, 378, 463 piNfiElD, d. J., jamesOn, R. f., & HodgkiN, S. T. 1998, MNRAS, 299, 955 PiNfiElD, D. J., Hodgkin, S. T., jAmEson, R. F., CossBurN, M. R., Hambly, N. C., & DEveReux, N. 2000,, 313, 347 PReIbiSCh, T., StAnkE, T., ZinNecker, h. 2003, a&A, 409,147 RaboUd, D. & MermilLiOD, J.-C. 1998,, 333, 897 RebOLo, R., ZapaTERo-osorIo, M. R., Martín, E. L. 1995, NaturE, 377, 129 reIPurth, B.  2000, AJ, 120, 3177 ReipurTh, B. & ClaRkE, c. 2001, Aj, 122, 432 sTauFfeR, J. R., Schultz g., & KIrkpaTRick J. D. 1998, APj, 499, L199 sTERziK, m. F. & Durisen, R. H. 2003, A&A, 400, 1031 Van Leeuwen, F. 1983, pHD tHesis, UNiVerSIty of LEiden zaPAteRo Osorio, M. R., REbolO, R., Mart[í]{}n, E. l., HodgkIN, S. T., CossBUrn, M. R., MaGazz[ù]{}, A., steEle, i. A., & JaMEsOn, r. F. 1999,, 134, 537 --- aBsTRacT: | we

aux, E., Bouvier, J., Stau ffer, J. R ., Cu ill and re , J. -C. 2003, A&A, 40 0 , 89 1 Muench, A. A., Lada, E. A ., Lada , C . J., & Alve s ,J .  20 02 ,ApJ ,5 73 , 366 Na kajima, T., Oppen hei me r, B. R., Ku l ka rni, S. R. , G olimowski, D . A ., Mat th ews , K.,Dur rance , S. T .  1995, Nature,37 8 , 463P infield , D.  J., Jameson, R. F.,& H o dgkin, S. T. 1 998, M NR A S, 2 99, 95 5 Pinfield ,D. J. , Hodgki n ,S . T., Jameson, R. F ., Cossburn , M.  R., H am bly , N. C. , & D ev e reu x, N. 2000, , 31 3, 347 Pr eibisc h , T., S t anke, T ., Zin nec ker , H. 20 03 , A &A , 40 9 ,1 47R abo ud, D. & M er milli od,J . - C . 19 98, , 33 3, 89 7 Rebolo, R., Za pate r o-O sorio , M.R.,Ma rtín, E. L.  1995 ,Nature, 377, 12 9 Re ipurth, B . 20 00, A J, 12 0 , 3177 Re ipu rth, B.  & Clar k e,C. 2 0 01 , AJ, 122, 432 Sta uf f e r, J. R.,Schult z G ., & Kirkpa tr ick J.D .  1998 , Ap J ,499, L19 9 Ster z ik ,M. F. & D urisen ,R.H.2003, A&A, 400,1031 van Leeu w en, F. 1983, P h D thesis, Uni v er s i ty of L eid en Zapatero Oso r io,M. R . ,Reb o lo, R ., Ma rt [ í] { }n, E. L., Hodgkin, S . T.,Cossb urn, M. R., M agazz[ù]{} , A ., Steel e, I .  A . , & Jameson, R . F.1999,, 134 , 537 -- - abs tract: | We

aux, E.,_Bouvier, J.,_Stauffer, J. R., Cuillandre, J.-C. _2003, A&A,_400,_891 Muench,_A. A.,_Lada, E. A., Lada,_C. J., & Alves,_J. 2002, ApJ, 573, 366_Nakajima, T., Oppenheimer,_B. R.,_Kulkarni, S. R., Golimowski, D. A., Matthews, K., Durrance, S. T. 1995, Nature, 378, 463 Pinfield, D. J., Jameson,_R. F.,_& Hodgkin,_S. T. 1998,_MNRAS,_299, 955 Pinfield, D. J., Hodgkin,_S. T., Jameson, R. F., Cossburn, M. R.,_Hambly, N. C.,_& Devereux, N. 2000,, 313, 347 Preibisch, T., Stanke,_T.,_Zinnecker, H. 2003, A&A,_409,147 Raboud, D. & Mermilliod, J.-C. 1998,, 333, 897 Rebolo, R.,_Zapatero-Osorio, M. R., Martín, E. L. 1995, Nature, 377,_129 Reipurth, B.  2000,_AJ,_120,_3177 Reipurth, B. & Clarke,_C. 2001, AJ, 122, 432 Stauffer, J. R.,_Schultz G., & Kirkpatrick J. D. 1998, ApJ,_499, L199 Sterzik, M. F. & Durisen, R. H. 2003, A&A,_400, 1031 van Leeuwen, F. 1983,_PhD thesis, University of Leiden_Zapatero Osorio,_M. R., Rebolo, R., Mart[í]{}n, E. L.,_Hodgkin, S. T., Cossburn,_M. R., Magazz[ù]{},_A., Steele, I. A.,_& Jameson, R. F. 1999,, 134, 537 --- abstract:_| _ We

d'})$ consisting of all $d$-dimensional subspaces of $\bbc^{d'}.$ Indeed, for any $I=\{i_1,\cdots, i_d\}\subset \{1, \cdots, d'\}$ and $J=\{1,\cdots, d\}$, let $M^g_{d'\times d}(I)$ be the subset of $M_{d'\times d}(d, I,J)$ consisting of the matrices $A$ satisfying that $\bigtriangleup_{(I;J)}(A)$ are identity matrices. Then there is a finite stratification $$\mathrm{Gr}_{d}(\bbc^{d'})=\bigsqcup_{I}M^g_{d'\times d}(I).$$ For any $A\in M_{d'\times d}(d, I, J)$, we substitute the identity matrix for the submatrix $\bigtriangleup_{(I;J)}(A)$ and then $A$ corresponds to a unqiue matrix $A'\in M^g_{d'\times d}(I)$. For every pair of multi-indices $I=\{i_1,\cdots, i_r\}$ and $J=\{j_1,\cdots, j_r\},$ we will define the following morphism of varieties: $$\Upsilon_{(r, I, J)}^{1}: M_{d'\times d}(r, I, J)\rightarrow M_{d\times (d-r)}(d-r),$$ $$\hspace{-0.8cm}\Upsilon_{(r, I, J)}^{2}: M_{d'\times d}(r, I, J)\rightarrow \mathrm{Gr}_{d-r}(\bbc^{d}),$$ $$\hspace{0.3cm}\Omega_{(r, I, J)}^{1}: M_{d'\times d}(r, I, J)\rightarrow M_{(d'-r)\times d'}(d'-r),$$ and $$\hspace{-1.2cm}\Omega_{(r, I, J)}^2: M_{d'\times d}(r, I, J)\rightarrow \mathrm{Gr}_{r}(\bbc^{d'}).$$ Let $P_{ij}(k)$ be the elementary matrix of size $k\times k$ transposing the $i$-th row and the $j$-th row. Set $P_{I}(d')=P_{r,i_r}(d')\cdots P_{1, i_1}(d')$ and $P_{

d'})$ consisting of all $ d$-dimensional subspaces of $ \bbc^{d'}.$ Indeed, for any $ I=\{i_1,\cdots, i_d\}\subset \{1, \cdots, d'\}$ and $ J=\{1,\cdots, d\}$, let $ M^g_{d'\times d}(I)$ be the subset of $ M_{d'\times d}(d, I, J)$ consisting of the matrix $ A$ meet that $ \bigtriangleup_{(I;J)}(A)$ are identity matrices. Then there is a finite stratification $ $ \mathrm{Gr}_{d}(\bbc^{d'})=\bigsqcup_{I}M^g_{d'\times d}(I).$$ For any $ A\in M_{d'\times d}(d, I, J)$, we substitute the identity matrix for the submatrix $ \bigtriangleup_{(I;J)}(A)$ and then $ A$ corresponds to a unqiue matrix $ A'\in M^g_{d'\times d}(I)$. For every pair of multi - indices $ I=\{i_1,\cdots, i_r\}$ and $ J=\{j_1,\cdots, j_r\},$ we will specify the following morphism of varieties: $ $ \Upsilon_{(r, I, J)}^{1 }: M_{d'\times d}(r, I, J)\rightarrow M_{d\times (d - r)}(d - r),$$ $ $ \hspace{-0.8cm}\Upsilon_{(r, I, J)}^{2 }: M_{d'\times d}(r, I, J)\rightarrow \mathrm{Gr}_{d - r}(\bbc^{d}),$$ $ $ \hspace{0.3cm}\Omega_{(r, I, J)}^{1 }: M_{d'\times d}(r, I, J)\rightarrow M_{(d'-r)\times d'}(d'-r),$$ and $ $ \hspace{-1.2cm}\Omega_{(r, I, J)}^2: M_{d'\times d}(r, I, J)\rightarrow \mathrm{Gr}_{r}(\bbc^{d'}).$$ permit $ P_{ij}(k)$ be the elementary matrix of size $ k\times k$ transposing the $ i$-th row and the $ j$-th quarrel. Set $ P_{I}(d')=P_{r, i_r}(d')\cdots P_{1, i_1}(d')$ and $ P _ {

d'})$ clnsisting of all $d$-dimenslonal subspaces of $\bbc^{d'}.$ Indees, for anh $I=\{i_1,\cdots, i_d\}\subset \{1, \cdots, d'\}$ end $H=\{1,\cdotw, d\}$, let $M^g_{d'\times d}(I)$ be the subsvt of $M_{d'\tumes e}(d, I,J)$ consmating on the latrncxs $A$ satisfying that $\bigtsiangleup_{(I;J)}(A)$ ase ibentity matrices. Then there is a finyte strstlfication $$\mathtm{Gr}_{d}(\nfc^{d'})=\bjgsqcup_{I}M^g_{d'\times d}(I).$$ For any $A\in M_{d'\fimes d}(v, I, J)$, we substiyute the identity matrix flr tje submatrix $\bigtrlangleup_{(I;J)}(A)$ and rhen $A$ correrponds to a unqiue mattix $A'\in M^g_{d'\times d}(I)$. For every paif of kulti-indicgs $I=\{l_1,\wdots, i_r\}$ anv $J=\{j_1,\cdjts, j_r\},$ we will defina the fpllowing morphlsm oh vaeieties: $$\Upsilon_{(r, I, J)}^{1}: K_{d'\times d}(r, I, J)\rightwrrow M_{d\thmzs (d-r)}(d-r),$$ $$\hspace{-0.8cm}\Upsilin_{(e, I, J)}^{2}: K_{d'\tikes a}(e, I, J)\rjgitadrow \mahhrj{Gr}_{d-r}(\bbc^{d}),$$ $$\gspace{0.3cm}\Ometa_{(r, I, J)}^{1}: M_{d'\times d}(r, I, J)\wpbhtarrow M_{(d'-r)\fimes q'}(d'-w),$$ and $$\hspace{-1.2cm}\Omega_{(r, I, J)}^2: M_{d'\times d}(r, I, J)\rpghtzrrow \mathrm{Gr}_{r}(\bbc^{d'}).$$ Let $P_{ij}(k)$ be the elementaty matrix jf size $k\times k$ transposing the $i$-th row and the $b$-th rkd. Stt $P_{I}(d')=P_{f,u_r}(f')\cdots P_{1, i_1}(d')$ and $P_{

d'})$ consisting of all $d$-dimensional subspaces of for $I=\{i_1,\cdots, i_d\}\subset \cdots, d'\}$ and be subset of $M_{d'\times I,J)$ consisting of matrices $A$ satisfying that $\bigtriangleup_{(I;J)}(A)$ are matrices. Then there is a finite stratification $$\mathrm{Gr}_{d}(\bbc^{d'})=\bigsqcup_{I}M^g_{d'\times d}(I).$$ For any $A\in M_{d'\times I, J)$, we substitute the identity matrix for the submatrix $\bigtriangleup_{(I;J)}(A)$ and then corresponds a matrix M^g_{d'\times d}(I)$. For every pair of multi-indices $I=\{i_1,\cdots, i_r\}$ and $J=\{j_1,\cdots, j_r\},$ we will define the morphism of varieties: $$\Upsilon_{(r, I, J)}^{1}: M_{d'\times d}(r, J)\rightarrow M_{d\times (d-r)}(d-r),$$ $$\hspace{-0.8cm}\Upsilon_{(r, J)}^{2}: M_{d'\times d}(r, I, J)\rightarrow $$\hspace{0.3cm}\Omega_{(r, J)}^{1}: M_{d'\times I, M_{(d'-r)\times and $$\hspace{-1.2cm}\Omega_{(r, I, M_{d'\times d}(r, I, J)\rightarrow \mathrm{Gr}_{r}(\bbc^{d'}).$$ Let $P_{ij}(k)$ be the elementary matrix of size $k\times k$ transposing the row and row. Set P_{1, and

d'})$ consisting of all $d$-dimensioNal subspacEs of $\bBc^{d'}.$ indEeD, for Any $I=\{I_1,\cdots, i_d\}\subset \{1, \CDots, D'\}$ and $J=\{1,\cdots, d\}$, let $M^g_{d'\times D}(I)$ be tHe SUbseT Of $m_{d'\timEs d}(d, I,J)$ cONsISTinG oF tHe mAtRIcEs $A$ saTisFying thAt $\bigtrianGleUp_{(i;J)}(A)$ are identiTY mAtrices. TheN thEre is a finite StrAtificAtIon $$\MAthrm{gr}_{d}(\Bbc^{d'})=\bIgsqcuP_{i}M^g_{d'\tiMes d}(I).$$ For aNy $a\In M_{d'\tiMEs d}(d, I, J)$, wE SUbStitUte the identity matRIx FOr the submatrix $\BigtriAnGLeUP_{(i;J)}(A)$ And Then $A$ correSpOnds tO A unqiue MAtRIX $a'\in m^G_{d'\times d}(I)$. For eVery pair of mULti-IndiceS $I=\{I_1,\cdOTs, i_r\}$ anD $J=\{j_1,\cdOtS, J_r\},$ wE will define The fOllowing mOrphisM Of varieTIes: $$\UpsiLon_{(r, I, J)}^{1}: m_{d'\tImeS d}(r, I, j)\RiGhTarRoW m_{d\tIMeS (d-r)}(D-R),$$ $$\hsPace{-0.8cm}\UpSiLoN_{(r, I, J)}^{2}: M_{D'\timES D}(R, i, J)\riGhtArroW \mathRm{Gr}_{d-r}(\bbc^{d}),$$ $$\hspAce{0.3Cm}\OmEGa_{(r, i, J)}^{1}: M_{d'\tImes d}(R, I, J)\rIgHtarrOw M_{(d'-r)\tImes d'}(D'-r),$$ And $$\hspace{-1.2cm}\OmegA_{(r, I, J)}^2: m_{d'\times d}(r, i, J)\rIgHtaRrOw \matHRm{Gr}_{r}(\bBc^{d'}).$$ let $p_{ij}(k)$ be tHe elemeNTarY mATRIx Of size $k\times k$ transPoSINg The $i$-th roW and thE $J$-tH rOW. Set $P_{I}(d')=P_{R,i_R}(d')\cDots p_{1, I_1}(D')$ and $P_{

d'})$ consisting of all $d $-dimensio nal s ubs pac es of$\bb c^{d'}.$ Indee d , fo r any $I=\{i_1,\cdots, i_d\ }\ s ubse t \ {1, \ cdots,d '\ } $ an d$J =\{ 1, \ cd ots,d\} $, let$M^g_{d'\t ime sd}(I)$ be th e s ubset of $ M_{ d'\times d}( d,I,J)$co nsi s tingofthe m atrice s $A$ s atisfying t h at $\b i gtriang l e up _{(I ;J)}(A)$ are iden t it y matrices. The n ther ei sa fin ite stratific at ion $ $ \mathrm { Gr } _ { d}( \ bbc^{d'})=\bi gsqcup_{I}M ^ g_{ d'\tim es d} ( I).$$For a ny $A\ in M_{d'\ti mesd}(d, I,J)$, w e substi t ute the ident ity ma trix fo rthe s u bma t ri x $ \ big triangle up _{ (I;J) }(A) $ a n d th en$A$corre sponds to a u nqi ue m a tri x $A' \in M ^g_{ d' \time s d}(I )$. Fo r every pair of mul ti-indice s $ I= \{i _1 ,\cdo t s, i_r \}$ an d $J=\{ j_1,\cd o ts, j _ r \ }, $ we will define t he f ol lowing m orphis m o fv arieties :$$\ Upsi l o n_{(r , I, J) }^{1}: M _{d'\t i me sd}(r, I ,J)\rig ht arr owM_{d\ t imes (d-r) }(d-r),$ $ $$\ h space{-0.8cm}\ U psilon_{(r, I , J ) } ^{ 2 }: M _{d '\times d}( r, I , J)\ righ t ar row \math rm{Gr }_ { d- r }(\bbc^{d}),$$ $$\h sp ace{0. 3cm}\ Omega_{(r, I, J)}^{1}:M _ { d'\times d}( r ,I , J)\rightarro w M_{ (d'-r)\tim e s d'}(d' -r),$ $ and $$ \hspace{- 1 . 2cm}\Ome ga_ {(r , I , J ) } ^2 : M_{d'\times d }(r, I , J)\ri ght arrow \ mat hrm {Gr }_{ r} (\bbc^{d' }).$$ Le t$P _{ ij }(k )$ be the elem en tar ymat rix o f size$k\ti mesk$ t r ans posingt he $ i$-t hro w an d t he $j$- th r o w.Set $P_ {I}(d')=P _{r , i_r} (d ') \cdotsP_{1, i_1}(d' )$ and $P_{

d'})$ consisting_of all_$d$-dimensional subspaces of $\bbc^{d'}.$_Indeed, for_any_$I=\{i_1,\cdots, i_d\}\subset_\{1,_\cdots, d'\}$ and_$J=\{1,\cdots, d\}$, let_$M^g_{d'\times d}(I)$ be the_subset of $M_{d'\times_d}(d,_I,J)$ consisting of the matrices $A$ satisfying that $\bigtriangleup_{(I;J)}(A)$ are identity matrices. Then there_is_a finite_stratification_$$\mathrm{Gr}_{d}(\bbc^{d'})=\bigsqcup_{I}M^g_{d'\times_d}(I).$$ For any $A\in M_{d'\times_d}(d, I, J)$, we substitute_the identity_matrix for the submatrix $\bigtriangleup_{(I;J)}(A)$ and then $A$_corresponds_to a unqiue_matrix $A'\in M^g_{d'\times d}(I)$. For every pair of multi-indices $I=\{i_1,\cdots,_i_r\}$ and $J=\{j_1,\cdots, j_r\},$ we will_define the following_morphism_of_varieties: $$\Upsilon_{(r, I, J)}^{1}:_M_{d'\times d}(r, I, J)\rightarrow M_{d\times (d-r)}(d-r),$$ $$\hspace{-0.8cm}\Upsilon_{(r,_I, J)}^{2}: M_{d'\times d}(r, I, J)\rightarrow \mathrm{Gr}_{d-r}(\bbc^{d}),$$ $$\hspace{0.3cm}\Omega_{(r,_I, J)}^{1}: M_{d'\times d}(r, I, J)\rightarrow M_{(d'-r)\times d'}(d'-r),$$_and $$\hspace{-1.2cm}\Omega_{(r, I, J)}^2: M_{d'\times d}(r, I,_J)\rightarrow \mathrm{Gr}_{r}(\bbc^{d'}).$$ Let $P_{ij}(k)$ be_the elementary_matrix of size $k\times k$_transposing the $i$-th_row and_the $j$-th row._Set $P_{I}(d')=P_{r,i_r}(d')\cdots P_{1, i_1}(d')$ and $P_{

. D. Husemoller, *Fibre Bundles*, 3rd Ed., Graduate Texts in Mathematics, **20**, Springer, New York, NY, 1994. T. Kato, *Perturbation Theory for Linear Operators*, Classics in Mathematics, Springer-Verlag, Berlin, 1995. A. V. Knyazev and M. E. Argentati, “Majorization for changes in angles between subspaces, Ritz values, and graph Laplacian spectra,” *SIAM J. Matrix Anal. Appl*, **29** (2006), no. 1, pp. 15–32. G. Lerman and T. Zhang, “Robust recovery of multiple subspaces by geometric $l_p$ minimization,” *Ann. Statist.*, **39** (2011), no. 5, pp. 2686–2715. H. Li and A. Li, “Utilizing improved Bayesian algorithm to identify blog comment spam,” *Proc. IEEE Symp. Robot. Appl.* (ISRA), **1** (2012), pp. 423–426. J. Liesen and Z. Strakoš, *Krylov Subspace Methods*, Oxford University Press, Oxford, 2013. L.-H. Lim, K. S.-W. Wong, and K. Ye, “Statistical estimation and the grassmannian of affine subspaces,” *prepirint*, (2016), <http://www.stat.uchicago.edu/~lekheng/work/affine.pdf>. D. J. Love, R. W. Heath, Jr., and T. Strohmer, “Grassmannian beamforming for multiple-input multiple-output wireless systems,” *IEEE Trans. Inform. Theory*, **49** (2003), no. 10, pp. 2735–2747. D. Luo and H. Huang, “Video motion segmentation using new adaptive manifold denoising model,” pp. 65–72, *Proc. IEEE Conf. Computer Vis. Pattern Recognit.* (CVPR), Columbus, OH, 2014. Y. Ma, A

. D.   Husemoller, * Fibre Bundles *, 3rd Ed. , Graduate Texts in Mathematics, * * 20 * *, Springer, New York, NY, 1994. T.   Kato, * Perturbation Theory for Linear Operators *, Classics in Mathematics, Springer - Verlag, Berlin, 1995. A.   V.   Knyazev and M.   E.   Argentati, “ Majorization for changes in angles between subspace, Ritz value, and graph Laplacian spectra, ” * SIAM J.   Matrix Anal.   Appl *, * * 29 * * (2006), no.   1, pp.   15–32. G.   Lerman and T.   Zhang, “ full-bodied recovery of multiple subspace by geometric $ l_p$ minimization, ” * Ann.   Statist. *, * * 39 * * (2011), no.   5, pp.   2686–2715. H.   Li and A.   Li, “ Utilizing improved Bayesian algorithm to identify web log comment spam, ” * Proc.   IEEE Symp.   Robot.   Appl. * (ISRA), * * 1 * * (2012), pp.   423–426. J.   Liesen and Z.   Strakoš, * Krylov Subspace Methods *, Oxford University Press, Oxford, 2013. L.-H.   Lim, K.   S.-W.   Wong, and K.   Ye, “ Statistical estimation and the grassmannian of affine subspaces, ” * prepirint *, (2016), < http://www.stat.uchicago.edu/~lekheng/work/affine.pdf >. D.   J.   Love, R.   W.   Heath, Jr., and T.   Strohmer, “ Grassmannian beamforming for multiple - input signal multiple - output wireless systems, ” * IEEE Trans.   Inform.   Theory *, * * 49 * * (2003), no.   10, pp.   2735–2747. D.   Luo and H.   Huang, “ Video apparent motion segmentation using newfangled adaptive manifold denoising model, ” pp.   65–72, * Proc.   IEEE Conf.   Computer Vis.   Pattern Recognit. * (CVPR), Columbus, OH, 2014. Y.   Ma, A

. D. Hudemoller, *Fibre Bundles*, 3ra Ed., Graduate Tgxrs in Kathemztics, **20**, Soringer, New York, NY, 1994. T. Kato, *Pecturvatiob Theory for Linear Opdrators*, Cpassics un Methematics, Sprinjsr-Verlan, Bermln, 1995. A. V. Niyazev and M. E. Arnentati, “Majmrization for whxnyes in angles between subspaces, Ritz values, ajd graph Laplasian fpecfga,” *SIAM J. Matrix Anal. Appl*, **29** (2006), no. 1, pl. 15–32. G. Lermai and T. Zhang, “Ronust recovery of multiple dubsoaces by geometric $l_p$ minimizqtiog,” *Ann. Statist.*, **39** (2011), no. 5, pp. 2686–2715. H. Li and A. Li, “Utimizing improved Bayesian algorighm tp identify boog womment span,” *Prob. IEEE Symp. Robot. Appl.* (IVRA), **1** (2012), pl. 423–426. J. Liesen and Z. Xtrekoš, *Jrylov Subspace Methovs*, Oxford University Press, Oxxoxd, 2013. L.-H. Lim, K. S.-W. Wong, and J. Yw, “Stajistiwal dwtioatjoi ahd the gressmannian kf affine sybspaces,” *prepirint*, (2016), <henl://www.stat.uchidago.edt/~lqkheng/work/affine.pdf>. D. J. Love, R. W. Heath, Jr., ang T. Atrohmer, “Grassmannian bwamforming for multippe-input mtltiple-output wireless systems,” *IEEE Trans. Inform. Thaory*, **49** (2003), no. 10, pi. 2735–2747. D. Luu ajd H. Huang, “Video motion segmentation using new zdspnive manifold denjising modek,” op. 65–72, *Kroc. IEEE Conf. Cumputex Vjs. Pattern Recognit.* (CVPR), Cjlumbys, OH, 2014. Y. Ma, A

. D. Husemoller, *Fibre Bundles*, 3rd Ed., in **20**, Springer, York, NY, 1994. Linear Classics in Mathematics, Berlin, 1995. A. Knyazev and M. E. Argentati, “Majorization changes in angles between subspaces, Ritz values, and graph Laplacian spectra,” *SIAM J. Anal. Appl*, **29** (2006), no. 1, pp. 15–32. G. Lerman and T. Zhang, recovery multiple by $l_p$ minimization,” *Ann. Statist.*, **39** (2011), no. 5, pp. 2686–2715. H. Li and A. Li, “Utilizing Bayesian algorithm to identify blog comment spam,” *Proc. Symp. Robot. Appl.* (ISRA), (2012), pp. 423–426. J. Liesen Z. *Krylov Subspace Oxford Press, 2013. L.-H. Lim, S.-W. Wong, and K. Ye, “Statistical estimation and the grassmannian of affine subspaces,” *prepirint*, (2016), <http://www.stat.uchicago.edu/~lekheng/work/affine.pdf>. D. Love, R. Jr., and Strohmer, beamforming multiple-input multiple-output wireless Trans. Inform. Theory*, **49** (2003), no. D. Luo and H. Huang, “Video motion segmentation new adaptive denoising model,” pp. 65–72, *Proc. IEEE Computer Vis. Pattern Recognit.* (CVPR), Columbus, OH, 2014. Ma, A

. D. Husemoller, *Fibre Bundles*, 3rd ed., Graduate texts In MAthEmAticS, **20**, SprInger, New York, NY, 1994. t. kato, *perturbation Theory for LInear opERatoRS*, CLassiCs in MatHEmATIcs, spRiNgeR-VERlAg, BerLin, 1995. a. V. KnyazEv and M. E. ArgEntAtI, “MajorizatioN FoR changes in AngLes between suBspAces, RiTz ValUEs, and GraPh LapLacian SPectra,” *sIAM J. MatrIx aNal. AppL*, **29** (2006), No. 1, pp. 15–32. G. LeRMAn And T. zhang, “Robust recoveRY oF Multiple subspaCes by gEoMEtRIC $l_p$ MinImization,” *ANn. statiST.*, **39** (2011), no. 5, pp. 2686–2715. H. LI AnD a. lI, “UtILizing improveD Bayesian alGOriThm to iDeNtiFY blog cOmmenT sPAm,” *PRoc. IEEE Symp. roboT. Appl.* (ISRA), **1** (2012), Pp. 423–426. J. LieSEn and Z. STRakoš, *KrYlov SuBspAce methODs*, oxForD UNIveRSiTy PREss, oxford, 2013. L.-H. liM, K. s.-W. WonG, and k. yE, “sTatiStiCal eStimaTion and the graSsmAnniAN of AffinE subsPaceS,” *pRepirInt*, (2016), <httP://www.sTaT.uchicago.edu/~lekHeng/Work/affinE.pdF>. D. j. LoVe, r. W. HeaTH, Jr., and t. StRohMer, “GrasSmanniaN BeaMfORMInG for multiple-input mUlTIPlE-output wIrelesS SyStEMs,” *IEEE TrAnS. InForm. tHEory*, **49** (2003), nO. 10, pp. 2735–2747. D. lUo And H. HuanG, “Video MOtIoN segmenTaTion usInG neW adAptivE ManiFold deNoising mOdel,” pP. 65–72, *proc. IEEE Conf. CoMPuter Vis. PatteRN RECOgNIt.* (CVpR), COlumbus, OH, 2014. Y. MA, A

. D. Husemoller, *Fibre B undles*, 3 rd Ed .,Gra du ateText s in Mathemati c s, * *20**, Springer, New Y ork,NY , 199 4 . T. K ato, *P e rt u r bat io nThe or y f or Li nea r Opera tors*, Cla ssi cs in Mathemat i cs , Springer -Ve rlag, Berlin , 1 995. A.  V. Knyaz evand M . E. A r gentat i, “Major iz a tion f o r chang e s i n an gles between subs p ac e s, Ritz values , andgr a ph L apl aci an spectra ,” *SIA M J. Mat r ix A n al. Appl*, **29** (2006), no .  1, pp. 1 5– 32. G. Le rmanan d T.  Zhang, “Ro bust recovery of mu l tiple s u bspaces by ge ome tri c $l _ p$ m ini mi z ati o n, ” * A nn.  Statist .* ,**39* * (2 0 1 1 ) , no . 5 , pp . 268 6–2715. H. L i a nd A .  Li , “Ut ilizi ng i mp roved Bayes ian a lg orithm to ident ifyblog comm ent s pam ,” *Pro c . IEEE Sy mp.  Robot.  Appl.* (IS RA ) , ** 1** (2012), pp. 42 3– 4 2 6. J. Lie sen an d Z .S trakoš,*K ryl ov S u b space Met h od s*, Oxfo rd Uni v er si ty Pres s, Oxfor d, 20 13. L.- H . Li m, K.S.-W. Wo ng, a n d K. Ye, “Stat i stical estima t io n an d the gr assmannianof a f fine sub s pa ces , ” *pr epiri nt * ,( 2016), <http://www. st at.uch icago .edu/~lekheng /work/affi n e . pdf>. D . J. Lo v e, R. W. Heath , Jr. , and T. S t rohmer,“Gras smannian beamform i n g for mu lti ple -in put m ul tiple-outputw i rele ss system s,” *IEEETra ns.  In for m.  Theory*, **49**(2 00 3) ,no.  10,p p. 2735– 27 47. D.Luo a n d H. H uang, “Vi de om oti on segm e nt a t ionus in g ne w a da ptive man i fol d denoi sing mode l,” pp.65 –7 2, *Pro c. IEEE Conf.  C omputer Vi s.  Pa tternR e cognit.* (CVPR), Columbus, OH,2 014. Y . M a, A

. D. Husemoller, *Fibre_Bundles*, 3rd_Ed., Graduate Texts in_Mathematics, **20**,_Springer,_New York,_NY,_1994. T. Kato, *Perturbation Theory_for Linear Operators*,_Classics in Mathematics, Springer-Verlag,_Berlin, 1995. A. V. Knyazev and_M. E. Argentati,_“Majorization for changes in angles between subspaces, Ritz values, and graph Laplacian spectra,” *SIAM_J. Matrix_Anal. Appl*, **29**_(2006),_no. 1,_pp. 15–32. G. Lerman and T. Zhang, “Robust recovery_of multiple subspaces by geometric_$l_p$ minimization,”_*Ann. Statist.*, **39** (2011), no. 5, pp. 2686–2715. H. Li and A. Li, “Utilizing_improved_Bayesian algorithm to_identify blog comment spam,” *Proc. IEEE Symp. Robot. Appl.* (ISRA), **1** (2012),_pp. 423–426. J. Liesen and Z. Strakoš, *Krylov Subspace Methods*,_Oxford University Press,_Oxford,_2013. L.-H. Lim,_K. S.-W. Wong, and K. Ye, “Statistical_estimation and the grassmannian of affine_subspaces,” *prepirint*, (2016), <http://www.stat.uchicago.edu/~lekheng/work/affine.pdf>. D. J. Love, R. W. Heath, Jr.,_and T. Strohmer, “Grassmannian beamforming for multiple-input multiple-output_wireless systems,” *IEEE Trans. Inform. Theory*, **49** (2003),_no. 10, pp. 2735–2747. D. Luo and H. Huang, “Video_motion segmentation_using new adaptive manifold denoising_model,” pp. 65–72, *Proc. IEEE_Conf. Computer Vis. Pattern_Recognit.* (CVPR), Columbus,_OH, 2014. Y. Ma, A

). A. K. Ekert, C. M. Alves, D. K. L. Oi, M. Horodecki, P. Horodecki, and L. C. Kwek, Phys. Rev. Lett. [**88**]{}, 217901 (2002). S. Hill and W. K. Wootters, Phys. Rev. Lett. [**78**]{}, 5022 (1997); W. K. Wootters, [*ibid.*]{} [**80**]{}, 2245 (1998). C. H. Bennett, D. P. DiVincenzo, J. A. Smolin, and W. K. Wootters, Phys. Rev. A [**54**]{}, 3824 (1996). P. Horodecki, Phys. Rev. Lett. [**90**]{}, 167901 (2003). H. A. Carteret, Phys. Rev. Lett. [**94**]{}, 040502 (2005). H. A. Carteret, quant-ph/0309212. T. A. Brun, Quant. Inf. Comp. [**4**]{}, 401 (2004). M. S. Leifer, N. Linden, and A. Winter, Phys. Rev. A [**69**]{}, 052304 (2004). M. Grassl, M. Roetteler, and T. Beth, Phys. Rev. A [**58**]{}, 1833-1839 (1998). P. Horodecki, Phys. Rev. A [**67**]{}, 060101(R) (2003). S. Lloyd, A. J. Landahl, and J.-J. E. Slotine, Phys. Rev. A [**69**]{}, 012305 (2004). By $\mathbbm{1}_{\mathcal{H}}$ we denote an identity operator on $\mathcal{H}$. We use standard Pauli matrices, i.e., $\sigma_{x}={|0\rangle}{\langle1|}+{|1\rangle}{\langle0|},\sigma_{y}= -i{|0\rangle}{\langle1|}+i{|1\rangle}{\langle0|}, \sigma_{z}={{|0\rangle}{\langle0|}}-{{|1\rangle}{\langle1|}}$. P. Horodecki, Acta Phys. Pol. A [**101**]{}, 399 (2002). O. Rudolph, quant-

). A. K. Ekert, C. M. Alves, D. K. L. Oi, M. Horodecki, P. Horodecki, and L. C. Kwek, Phys. Rev. Lett. [ * * 88 * * ] { }, 217901 (2002). S. Hill and W. K. Wootters, Phys. Rev. Lett. [ * * 78 * * ] { }, 5022 (1997); W. K. Wootters, [ * ibid. * ] { } [ * * 80 * * ] { }, 2245 (1998). C. H. Bennett, D. P. DiVincenzo, J. A. Smolin, and W. K. Wootters, Phys. Rev. A [ * * 54 * * ] { }, 3824 (1996). P. Horodecki, Phys. Rev. Lett. [ * * 90 * * ] { }, 167901 (2003). H. A. Carteret, Phys. Rev. Lett. [ * * 94 * * ] { }, 040502 (2005). H. A. Carteret, quant - ph/0309212. T. A. Brun, Quant. Inf. Comp. [ * * 4 * * ] { }, 401 (2004). M. S. Leifer, N. Linden, and A. Winter, Phys. Rev. A [ * * 69 * * ] { }, 052304 (2004). M. Grassl, M. Roetteler, and T. Beth, Phys. Rev. A [ * * 58 * * ] { }, 1833 - 1839 (1998). P. Horodecki, Phys. Rev. A [ * * 67 * * ] { }, 060101(R) (2003). S. Lloyd, A. J. Landahl, and J.-J. E. Slotine, Phys. Rev. A [ * * 69 * * ] { }, 012305 (2004). By $ \mathbbm{1}_{\mathcal{H}}$ we denote an identity operator on $ \mathcal{H}$. We use standard Pauli matrices, i.e., $ \sigma_{x}={|0\rangle}{\langle1|}+{|1\rangle}{\langle0|},\sigma_{y}= -i{|0\rangle}{\langle1|}+i{|1\rangle}{\langle0| }, \sigma_{z}={{|0\rangle}{\langle0|}}-{{|1\rangle}{\langle1|}}$. P. Horodecki, Acta Phys. Pol. A [ * * 101 * * ] { }, 399 (2002). O. Rudolph, quant-

). A. K. Ekert, C. M. Alves, D. K. L. Ol, M. Horodecki, P. Kirodecni, and L. C. Kwey, Phys. Rev. Lett. [**88**]{}, 217901 (2002). S. Hill and W. K. Woitters, Phys. Rev. Lett. [**78**]{}, 5022 (1997); W. K. Woonters, [*ibie.*]{} [**80**]{}, 2245 (1998). R. H. Bennett, D. P. VjVincendj, J. Z. Smonmn, and W. K. Wootjers, Phys. Ree. A [**54**]{}, 3824 (1996). P. Horodewkk, 'hys. Rev. Lett. [**90**]{}, 167901 (2003). H. A. Carteret, Phys. Red. Lett. [**94**]{}, 040502 (2005). J. A. Carteret, qoant-pn/0309212. E. A. Gguk, Quant. Inf. Comp. [**4**]{}, 401 (2004). M. S. Leifer, N. Minden, end A. Winter, Phus. Rev. A [**69**]{}, 052304 (2004). M. Grassl, M. Roetheleg, and T. Beth, Phys. Gev. A [**58**]{}, 1833-1839 (1998). P. Hirodqxki, Phys. Rev. A [**67**]{}, 060101(R) (2003). S. Lloyd, A. J. Landzhl, and J.-J. E. Slotine, Phys. Rev. A [**69**]{}, 012305 (2004). By $\mathbbm{1}_{\majkxal{J}}$ we denote en idegtity operator on $\madhcal{H}$. Wr use standard Panli natrices, i.e., $\sigma_{x}={|0\ranjle}{\langle1|}+{|1\rangle}{\langlg0|},\sigma_{y}= -i{|0\ratgme}{\langle1|}+i{|1\rangle}{\labgoe0|}, \sicma_{z}={{|0\sanguw}{\lavglt0|}}-{{|1\raigls}{\langlf1|}}$. P. Iorodecki, Adta Phys. Poo. A [**101**]{}, 399 (2002). O. Rudolph, quamt-

). A. K. Ekert, C. M. Alves, L. M. Horodecki, Horodecki, and L. [**88**]{}, (2002). S. Hill W. K. Wootters, Rev. Lett. [**78**]{}, 5022 (1997); W. Wootters, [*ibid.*]{} [**80**]{}, 2245 (1998). C. H. Bennett, D. P. DiVincenzo, J. A. and W. K. Wootters, Phys. Rev. A [**54**]{}, 3824 (1996). P. Horodecki, Phys. Lett. 167901 H. Carteret, Phys. Rev. Lett. [**94**]{}, 040502 (2005). H. A. Carteret, quant-ph/0309212. T. A. Brun, Quant. Inf. [**4**]{}, 401 (2004). M. S. Leifer, N. Linden, A. Winter, Phys. Rev. [**69**]{}, 052304 (2004). M. Grassl, Roetteler, T. Beth, Rev. [**58**]{}, (1998). P. Horodecki, Rev. A [**67**]{}, 060101(R) (2003). S. Lloyd, A. J. Landahl, and J.-J. E. Slotine, Phys. Rev. A 012305 (2004). we denote identity on We use standard i.e., $\sigma_{x}={|0\rangle}{\langle1|}+{|1\rangle}{\langle0|},\sigma_{y}= -i{|0\rangle}{\langle1|}+i{|1\rangle}{\langle0|}, \sigma_{z}={{|0\rangle}{\langle0|}}-{{|1\rangle}{\langle1|}}$. P. Horodecki, A [**101**]{}, 399 (2002). O. Rudolph, quant-

). A. K. Ekert, C. M. Alves, D. K. L. Oi, M. HorodEcki, P. HorodEcki, aNd L. c. KwEk, phys. rev. LEtt. [**88**]{}, 217901 (2002). S. Hill and W. K. WOOtteRs, Phys. Rev. Lett. [**78**]{}, 5022 (1997); W. K. WootterS, [*ibid.*]{} [**80**]{}, 2245 (1998). c. H. bEnneTT, D. p. DiViNcenzo, J. a. smOLIn, aNd w. K. wooTtERs, phys. REv. A [**54**]{}, 3824 (1996). p. HorodeCki, Phys. Rev. letT. [**90**]{}, 167901 (2003). H. a. Carteret, PhyS. reV. Lett. [**94**]{}, 040502 (2005). H. A. CarTerEt, quant-ph/0309212. T. A. BRun, quant. INf. comP. [**4**]{}, 401 (2004). m. S. LeiFer, n. LindEn, and A. wInter, PHys. Rev. A [**69**]{}, 052304 (2004). M. GRaSSl, M. RoeTTeler, anD t. beTh, PhYs. Rev. A [**58**]{}, 1833-1839 (1998). P. Horodecki, PHYs. rEv. A [**67**]{}, 060101(R) (2003). S. Lloyd, A. J. LaNdahl, aNd j.-j. E. sLOtiNe, PHys. Rev. A [**69**]{}, 012305 (2004). By $\mAtHbbm{1}_{\mAThcal{H}}$ wE DeNOTE an IDentity operatOr on $\mathcal{h}$. we uSe stanDaRd PAUli matRices, I.e., $\SIgmA_{x}={|0\rangle}{\lanGle1|}+{|1\rAngle}{\langLe0|},\sigmA_{Y}= -i{|0\ranglE}{\Langle1|}+i{|1\Rangle}{\LanGle0|}, \SigmA_{Z}={{|0\rAnGle}{\LaNGle0|}}-{{|1\RAnGle}{\LAngLe1|}}$. P. HorodEcKi, acta PHys. POL. a [**101**]{}, 399 (2002). o. rudoLph, QuanT-

). A. K. Ekert, C. M. Alv es, D. K.L. Oi , M . H or odec ki,P. Horodecki,a nd L . C. Kwek, Phys. Rev.Lett. [ * *88* * ]{ }, 21 7901 (2 0 02 ) . S .Hi llan d W . K.Woo tters,Phys. Rev. Le tt . [**78**]{} , 5 022 (1997) ; W . K. Wootter s,[*ibid .* ]{} [**80 **] {}, 2 245 (1 9 98). C. H. Ben ne t t, D.P . DiVin c e nz o, J . A. Smolin, andW .K . Wootters, Ph ys. Re v. A[ * *54 **] {}, 3824 ( 19 96).P. Horo d ec k i , Ph y s. Rev. Lett. [**90**]{} , 16 7901 ( 20 03) . H. A . Car te r et, Phys. Rev. Let t. [**94* *]{},0 40502 ( 2 005). H. A.Car ter et,q ua nt -ph /0 3 092 1 2. T . A. Brun, Q ua nt . Inf . Co m p . [**4 **] {},401 ( 2004). M. S. Le ifer , N. Lind en, a nd A .Winte r, Phy s. Re v. A [**69**]{},0523 04 (2004) . M. Gr as sl, M . Roett ele r,and T.Beth, P h ys. R e v . A [**58**]{}, 1833- 18 3 9 ( 1998). P. Hor o de ck i , Phys.Re v.A [* * 6 7**]{ }, 0 6 01 01(R) (2 003).S. L loyd, A .J. Lan da hl, an d J.- J . E. Sloti ne, Phys . Rev . A [**69**]{}, 012305 (2004) . B y $ \ math bbm {1}_{\mathc al{H } }$ w e de n ot e a n iden tityop e ra t or on $\mathcal{H}$ . We us e sta ndard Pauli m atrices, i . e . , $\sigm a_{x } ={ | 0\rangle}{\lan gle1| }+{|1\rang l e}{\lang le0|} ,\sigma_ {y}= -i{| 0 \ rangle}{ \la ngl e1| }+i { | 1\ rangle}{\lang l e 0|}, \ sigma_{ z}= {{|0\ra ngl e}{ \la ngl e0 |}}-{{|1\ rangle}{ \l an gl e1 |}} $. P . Horodec ki , A ct a P hys.P ol. A[**10 1**] {} ,3 99(2002). O . Rudo lp h, qua nt-

). A. K._Ekert, C._M. Alves, D. K._L. Oi,_M._Horodecki, P._Horodecki,_and L. C._Kwek, Phys. Rev._Lett. [**88**]{}, 217901 (2002). S._Hill and W._K._Wootters, Phys. Rev. Lett. [**78**]{}, 5022 (1997); W. K. Wootters, [*ibid.*]{} [**80**]{}, 2245 (1998). C._H._Bennett, D._P._DiVincenzo,_J. A. Smolin, and W._K. Wootters, Phys. Rev. A_[**54**]{}, 3824_(1996). P. Horodecki, Phys. Rev. Lett. [**90**]{}, 167901 (2003). H._A._Carteret, Phys. Rev._Lett. [**94**]{}, 040502 (2005). H. A. Carteret, quant-ph/0309212. T. A. Brun,_Quant. Inf. Comp. [**4**]{}, 401 (2004). M._S. Leifer, N._Linden,_and_A. Winter, Phys. Rev._A [**69**]{}, 052304 (2004). M. Grassl, M._Roetteler, and T. Beth, Phys. Rev._A [**58**]{}, 1833-1839 (1998). P. Horodecki, Phys. Rev._A [**67**]{}, 060101(R) (2003). S. Lloyd, A._J. Landahl, and J.-J. E._Slotine, Phys._Rev. A [**69**]{}, 012305 (2004). By_$\mathbbm{1}_{\mathcal{H}}$ we denote_an identity_operator on $\mathcal{H}$. We_use standard Pauli matrices, i.e., $\sigma_{x}={|0\rangle}{\langle1|}+{|1\rangle}{\langle0|},\sigma_{y}= -i{|0\rangle}{\langle1|}+i{|1\rangle}{\langle0|},_\sigma_{z}={{|0\rangle}{\langle0|}}-{{|1\rangle}{\langle1|}}$. P. Horodecki, Acta_Phys. Pol. A [**101**]{}, 399 (2002). O._Rudolph,_quant-

a commutative ring and $M$ is an $R$-module, then a prime ideal $p$ of $R$ is associated to $M$, if there is an $x\in M$ such that $p={{\mathop{\mathrm{ann\,}}}}x$. We shall make use of the following fact: if $R$ is a Noetherian ring and $M$ is a finite $R$-module, then there are finitely many primes associated to $M$, and furthermore, each zerodivisor on $M$ is contained in one of them, see [@eisen Theorem 3.1]. Thus, in our setting, ${{\mathop{\mathfrak{m}}}}$ is a subset of the finite union of the associated primes. Now, the prime avoidance lemma [@eisen Theorem 3.3] states that ${{\mathop{\mathfrak{m}}}}$ is contained in one of the associated primes, and hence, itself is an associated prime. Therefore, ${{\mathop{\mathfrak{m}}}}x=0$ for some nonzero $x\in M$ and we put $L=xR$. Conversely, suppose $L\subset M$ is a nonzero submodule such that ${{\mathop{\mathfrak{m}}}}L=0$. Then, every $r\in {{\mathop{\mathfrak{m}}}}\backslash\{0\}$ is a zerodivisor. Since $M\not=0$ is a finite $R$-module, Nakayama’s lemma [@eisen Corollary 4.8] implies that $M\not={{\mathop{\mathfrak{m}}}}M$. So, ${\mathop{\mathrm{depth}}\nolimits _{{{\mathop{\mathfrak{m}}}}}} M=0$. For the proof of Theorem \[thm\_coh\_stalk\] we shall need a number of Lemmas that are algebraic in nature. Recall the notion of localization at a prime. Suppose $R$ is a ring, $p\subset R$ a prime ideal, and $M$ an $R$-module. Consider the multiplicatively closed set $S=R\backslash p$, then the localization of $M$ at $p$ is $$M_p=M\times S/\sim\,,$$ where $(v,s)\sim(w,t)$ means $q(vt-ws)=0$ for some $q\in S$. Elements of $M_p$ are written as fractions $v/s$. The usual rules for operating with fractions turn $R_

a commutative ring and $ M$ is an $ R$-module, then a prime ideal $ p$ of $ R$ is associated to $ M$, if there is an $ x\in M$ such that $ p={{\mathop{\mathrm{ann\,}}}}x$. We shall seduce habit of the following fact: if $ R$ is a Noetherian ring and $ M$ is a finite $ R$-module, then there are finitely many primes consort to $ M$, and furthermore, each zerodivisor on $ M$ is contained in one of them, see [ @eisen Theorem 3.1 ]. therefore, in our setting, $ { { \mathop{\mathfrak{m}}}}$ is a subset of the finite union of the associated primes. Now, the premier avoidance lemma [ @eisen Theorem 3.3 ] states that $ { { \mathop{\mathfrak{m}}}}$ is contained in one of the associated prime, and hence, itself is an associated prime. consequently, $ { { \mathop{\mathfrak{m}}}}x=0 $ for some nonzero $ x\in M$ and we put $ L = xR$. Conversely, suppose $ L\subset M$ is a nonzero submodule such that $ { { \mathop{\mathfrak{m}}}}L=0$. Then, every $ r\in { { \mathop{\mathfrak{m}}}}\backslash\{0\}$ is a zerodivisor. Since $ M\not=0 $ is a finite $ R$-module, Nakayama ’s lemma [ @eisen Corollary 4.8 ] implies that $ M\not={{\mathop{\mathfrak{m}}}}M$. So, $ { \mathop{\mathrm{depth}}\nolimits _ { { { \mathop{\mathfrak{m } } } } } } M=0$. For the proof of Theorem \[thm\_coh\_stalk\ ] we shall need a number of Lemmas that are algebraic in nature. remember the notion of localization at a prime. presuppose $ R$ is a ring, $ p\subset R$ a prime ideal, and $ M$ an $ R$-module. Consider the multiplicatively closed put $ S = R\backslash p$, then the localization of $ M$ at $ p$ is $ $ M_p = M\times S/\sim\,,$$ where $ (v, s)\sim(w, t)$ means $ q(vt - ws)=0 $ for some $ q\in S$. Elements of $ M_p$ are write as fractions $ v / s$. The usual rules for operating with fractions call on $ R _

a fommutative ring and $M$ ir an $R$-module, thgn a prike idezl $p$ of $F$ is associated to $M$, if therx is an $x\un M$ such that $p={{\mathop{\oathrm{ann\,}}}}q$. We shalo majw use of tis following fzgt: if $C$ is a Noetherisn ring ang $M$ is a finita $F$-mldule, then there are finitely many [rimes ssdociated to $M$, wnd gtrthsgmire, each zerodivisor on $M$ is dontaintd in one of them, xee [@eisen Theorem 3.1]. Thus, in lur detting, ${{\mathop{\mathvrak{m}}}}$ is a wubsqr of the finkte union of the assocjated primes. Now, the prime avoiaance lemma [@eisgu Thfmrem 3.3] statew than ${{\mathop{\mathfvsk{m}}}}$ is contaimed in one of bhe avsoxiated primes, and henre, itself is an assosiated prhmz. Therefore, ${{\mathop{\matyfeak{m}}}}x=0$ for somd novzedo $x\jn M$ ajd xe put $L=xR$. Cknversely, syppose $L\subset M$ is a bonzero submosule stcr that ${{\mathop{\mathfrak{m}}}}L=0$. Then, every $r\in {{\katgop{\mathfrak{m}}}}\backslash\{0\}$ iw a zerodivisor. Since $M\not=0$ is w finite $R$-module, Nakayama’s lemma [@eisen Corollary 4.8] implmer tkqt $M\nur={{\mwthop{\mathfrak{m}}}}M$. So, ${\mathop{\mathrm{depth}}\nolimits _{{{\iztnoi{\mathfrak{m}}}}}} M=0$. For tme proof of Theorek \[hhk\_soh\_stalk\] we sfall nzsd a number of Lemmad that wre aogebraic yn nsture. Recall the notion of oocalization qt a prime. Suppose $R$ is a riny, $p\subxet R$ a prime ideal, and $M$ an $R$-mosule. Considfr the mumgiplicatively clusec vet $S=R\backslash p$, then the localizavion pf $M$ at $p$ ix $$M_p=M\tymes S/\sim\,,$$ ahere $(v,s)\sim(w,t)$ means $q(vt-wd)=0$ for smme $q\in S$. Flements of $M_p$ are written as fcections $v/s$. Thg uvuan rules yor opcrating with frwctions turn $R_

a commutative ring and $M$ is an a ideal $p$ $R$ is associated an M$ such that We shall make of the following fact: if $R$ a Noetherian ring and $M$ is a finite $R$-module, then there are finitely primes associated to $M$, and furthermore, each zerodivisor on $M$ is contained in of see Theorem Thus, in our setting, ${{\mathop{\mathfrak{m}}}}$ is a subset of the finite union of the associated primes. the prime avoidance lemma [@eisen Theorem 3.3] states ${{\mathop{\mathfrak{m}}}}$ is contained in of the associated primes, and itself an associated Therefore, for nonzero $x\in M$ we put $L=xR$. Conversely, suppose $L\subset M$ is a nonzero submodule such that ${{\mathop{\mathfrak{m}}}}L=0$. Then, every $r\in is a $M\not=0$ is finite Nakayama’s [@eisen Corollary 4.8] $M\not={{\mathop{\mathfrak{m}}}}M$. So, ${\mathop{\mathrm{depth}}\nolimits _{{{\mathop{\mathfrak{m}}}}}} M=0$. For Theorem \[thm\_coh\_stalk\] we shall need a number of that are in nature. Recall the notion of at a prime. Suppose $R$ is a ring, R$ a prime ideal, and $M$ an $R$-module. Consider the multiplicatively closed set $S=R\backslash p$, localization of $M$ at is $$M_p=M\times S/\sim\,,$$ $(v,s)\sim(w,t)$ $q(vt-ws)=0$ some S$. Elements $M_p$ are written as fractions $v/s$. The usual rules for operating fractions turn $R_

a commutative ring and $M$ is an $R$-Module, then A primE idEal $P$ oF $R$ is AssoCiated to $M$, if theRE is aN $x\in M$ such that $p={{\mathop{\maThrm{aNn\,}}}}X$. we shALl Make uSe of the FOlLOWinG fAcT: if $r$ iS A NOetheRiaN ring anD $M$ is a finitE $R$-mOdUle, then there ARe Finitely maNy pRimes associaTed To $M$, and FuRthERmore, EacH zeroDivisoR On $M$ is cOntained iN oNE of theM, See [@eiseN tHeOrem 3.1]. thus, in our setting, ${{\mAThOP{\mathfrak{m}}}}$ is a sUbset oF tHE fINIte UniOn of the assOcIated PRimes. NoW, ThE PRIme AVoidance lemma [@Eisen TheoreM 3.3] StaTes thaT ${{\mAthOP{\mathfRak{m}}}}$ iS cONtaIned in one of The aSsociated Primes, ANd hence, ITself is An assoCiaTed PrimE. thErEfoRe, ${{\MAthOP{\mAthFRak{M}}}}x=0$ for somE nOnZero $x\In M$ aND WE Put $L=XR$. COnveRsely, Suppose $L\subseT M$ iS a noNZerO submOdule Such ThAt ${{\matHop{\matHfrak{M}}}}L=0$. then, every $r\in {{\matHop{\mAthfrak{m}}}}\bAckSlAsh\{0\}$ Is A zeroDIvisor. sinCe $M\Not=0$ is a fInite $R$-mODulE, NAKAYaMa’s lemma [@eisen CorolLaRY 4.8] ImPlies thaT $M\not={{\mAThOp{\MAthfrak{m}}}}m$. SO, ${\maThop{\MAThrm{dEpth}}\NOlImits _{{{\matHop{\matHFrAk{M}}}}}} M=0$. For thE pRoof of thEorEm \[tHm\_coh\_STalk\] We shalL need a nuMber oF lemmas that are aLGebraic in natuRE. RECAlL The nOtiOn of localizAtioN At a pRime. sUpPosE $r$ is a rIng, $p\sUbSEt r$ A prime ideal, and $M$ an $R$-mOdUle. ConSider The multiplicaTively closED SEt $S=R\backSlasH P$, tHEn the localizatIon of $m$ at $p$ is $$M_p=M\tIMes S/\sim\,,$$ wHere $(v,S)\sim(w,t)$ meAns $q(vt-ws)=0$ fOR Some $q\in S$. eleMenTs oF $M_p$ ARE wRitten as fractIONs $v/s$. thE usual rUleS for opeRatIng WitH frAcTions turn $r_

a commutative ring and $M $ is an $R $-mod ule , t he n aprim e ideal $p$ of $R$is associated to $M$,if th er e isa n$x\in M$ suc h t h a t $ p= {{ \ma th o p{ \math rm{ ann\,}} }}x$. We s hal lmake use oft he following fa ct: if $R$ i s a Noeth er ian ringand $M$is a f i nite $ R$-module ,t hen th e re aref i ni tely many primes asso c ia t ed to $M$, and furth er m or e , ea chzerodiviso ron $M $ is con t ai n e d in one of them,see [@eisen The orem 3 .1 ].Thus,in ou rs ett ing, ${{\ma thop {\mathfra k{m}}} } $ is as ubset o f thefin ite uni o nof th ea sso c ia ted pri mes. Now ,th e pri me a v o i d ance le mma[@eis en Theorem 3. 3]stat e s t hat $ {{\ma thop {\ mathf rak{m} }}}$is contained in o ne o f the ass oci at edpr imes, and he nce , i tself i s an as s oci at e d pr ime. Therefore, ${ {\ m a th op{\math frak{m } }} }x = 0$ for s om e n onze r o $x\i n M$ an d we put $L=xR $ . C onverse ly , supp os e $ L\s ubset M$ i s a no nzero su bmodu l e such that ${ { \mathop{\math f ra k { m} } }}L= 0$. Then, ever y $r \ in { {\ma t ho p{\ m athfr ak{m} }} } \b a ckslash\{0\}$ is aze rodivi sor.Since $M\not= 0$ is a fi n i t e $R$-mo dule , N a kayama’s lemma [@ei sen Coroll a ry 4.8]impli es that$M\not={{ \ m athop{\m ath fra k{m }}} } M $. So, ${\matho p { \mat hr m{depth }}\ nolimit s _ {{{ \ma tho p{ \mathfrak {m}}}}}} M =0 $. For thep roof ofTh eor em \[ thm\_ c oh\_st alk\] wesh al l ne ed a nu m be r of L em ma s th atar e alg ebra i c i n natur e. Recall th e not io nof loca lization at a p rime. Supp os e $ R$ isa ring, $p \subset R$ a prime idea l , and $ M$an $R $-mo dule. Con sid er the mu l tiplic ativel y clo se d s e t $S=R \ b ac ksl as h p$, then t helocal iz atio n of $M $ at $p$ is $$M_p= M \ti mes S/\sim\,, $$wher e $( v,s ) \s i m(w ,t ) $ m e a ns $q(vt-ws)=0$ for some$q \ in S$. Eleme n tsof $M_p$are wri ttena s fract ions $v/s $. The us ua l ru l e s f or operati ng withfractions turn$ R_

a_commutative ring_and $M$ is an_$R$-module, then_a_prime ideal_$p$_of $R$ is_associated to $M$,_if there is an_$x\in M$ such_that_$p={{\mathop{\mathrm{ann\,}}}}x$. We shall make use of the following fact: if $R$ is a Noetherian_ring_and $M$_is_a_finite $R$-module, then there are_finitely many primes associated to_$M$, and_furthermore, each zerodivisor on $M$ is contained in_one_of them, see_[@eisen Theorem 3.1]. Thus, in our setting, ${{\mathop{\mathfrak{m}}}}$ is a_subset of the finite union of_the associated primes._Now,_the_prime avoidance lemma [@eisen_Theorem 3.3] states that ${{\mathop{\mathfrak{m}}}}$ is_contained in one of the associated_primes, and hence, itself is an associated_prime. Therefore, ${{\mathop{\mathfrak{m}}}}x=0$ for some nonzero_$x\in M$ and we put_$L=xR$. Conversely, suppose_$L\subset M$ is a nonzero_submodule such that_${{\mathop{\mathfrak{m}}}}L=0$. Then,_every $r\in {{\mathop{\mathfrak{m}}}}\backslash\{0\}$_is a zerodivisor. Since $M\not=0$ is_a finite $R$-module,_Nakayama’s lemma [@eisen Corollary 4.8] implies_that_$M\not={{\mathop{\mathfrak{m}}}}M$. So, ${\mathop{\mathrm{depth}}\nolimits__{{{\mathop{\mathfrak{m}}}}}}_M=0$. For_the proof_of Theorem \[thm\_coh\_stalk\]_we_shall need_a_number of Lemmas that are algebraic_in_nature. Recall the notion of localization at_a prime. Suppose $R$_is_a ring, $p\subset R$_a prime ideal, and $M$_an $R$-module. Consider the multiplicatively closed_set $S=R\backslash_p$, then_the localization of $M$ at $p$ is $$M_p=M\times S/\sim\,,$$ where $(v,s)\sim(w,t)$_means $q(vt-ws)=0$ for some $q\in S$._Elements of $M_p$ are_written as_fractions_$v/s$. The usual_rules_for operating_with fractions turn $R_

\,0}(\phi,\theta,\psi)=P_l(\cos\theta)$, where $\phi,\theta,\psi$ are the Euler angles on $SU(2)$ (see [@ViKl:repspecfunc1 Chapter 6]). From the orthogonality relation of the Legendre polynomials, the fact that their only common value is at $P_l(1)=1$, and considering that $\theta=0$ denotes a pole of $SU(2)$, we find that the delta function at the identity of $SU(2)$ restricted to $S^2$ can be represented as $$\delta_0(\phi,\theta)=\sum_l (2l+1)\, P_l(\cos\theta) =\sum_l (2l+1)\, t^{(l)}_{0\,0}(\phi,\theta). \label{eq:delta}$$ Recall that a coquasitriangular structure $\Ru:H\tens H\to\k$ on a quantum group $H$ determines a braiding between right comodules $V$ and $W$ via $$\psi(v\tens w)=w\i1\tens v\i1 \Ru(v\i2\tens w\i2)$$ for $v\in V$ and $w\in W$. (We use here Sweedler’s coproduct notation for the coaction.) For calculations we need the functionals $u$ and $v$ defined with $\Ru$ as (see e.g.[@Maj:qgroups]) $$u(a)\defeq\Ru(a\i2\tens \antip a\i1), \quad v(a)\defeq\Ru(a\i1\tens \antip a\i2) \label{eq:defcqtruv}$$ for $a\in H$. For $H=\SUq2$ in our basis they are $$u(t^{(l)}_{m\,n}) =\delta_{m,n}\, q^{-2l(l+1)+2m},\quad v(t^{(l)}_{m\,n}) =\delta_{m,n}\, q^{-2l(l+1)-2m}. \label{eq:sl2uv}$$ We also note that property (\[eq:Rprop\]) is satisfied, i.e., $$\Ru\left(t

\,0}(\phi,\theta,\psi)=P_l(\cos\theta)$, where $ \phi,\theta,\psi$ are the Euler angles on $ SU(2)$ (see [ @ViKl: repspecfunc1 Chapter   6 ]). From the orthogonality relation of the Legendre polynomials, the fact that their lone coarse value is at $ P_l(1)=1 $, and considering that $ \theta=0 $ denotes a pole of $ SU(2)$, we determine that the delta function at the identity of $ SU(2)$ restricted to $ S^2 $ can be represented as $ $ \delta_0(\phi,\theta)=\sum_l (2l+1)\, P_l(\cos\theta) = \sum_l (2l+1)\, t^{(l)}_{0\,0}(\phi,\theta). \label{eq: delta}$$ remember that a coquasitriangular structure $ \Ru: H\tens H\to\k$ on a quantum group $ H$ determines a braid between proper comodules $ V$ and $ W$ via $ $ \psi(v\tens w)=w\i1\tens v\i1 \Ru(v\i2\tens w\i2)$$ for $ v\in V$ and $ w\in W$. (We use here Sweedler ’s coproduct note for the coaction .) For calculations we need the functionals $ u$ and $ v$ defined with $ \Ru$ as (determine e.g.[@Maj: qgroups ]) $ $ u(a)\defeq\Ru(a\i2\tens \antip a\i1), \quad v(a)\defeq\Ru(a\i1\tens \antip a\i2) \label{eq: defcqtruv}$$ for $ a\in H$. For $ H=\SUq2 $ in our basis they are $ $ u(t^{(l)}_{m\,n }) = \delta_{m, n}\, q^{-2l(l+1)+2m},\quad v(t^{(l)}_{m\,n }) = \delta_{m, n}\, q^{-2l(l+1)-2 m }. \label{eq: sl2uv}$$ We also note that property (\[eq: Rprop\ ]) is quenched, i.e., $ $ \Ru\left(t

\,0}(\phi,\hheta,\psi)=P_l(\cos\theta)$, where $\phi,\theta,\psi$ arg rhe Euner anfles on $RU(2)$ (see [@ViKl:repspecfunc1 Chaptxr 6]). Feom tye orthogonality relatkon of thv Legendrw pootnomials, tis fact bkat tgcir oulb common value ls at $P_l(1)=1$, ang considering dhxt $\theta=0$ denotes a pole of $SU(2)$, we find that tne delta functiog at ehe jdentity of $SU(2)$ restricted to $S^2$ can be repgesented as $$\delta_0(\lhi,\theta)=\sum_l (2l+1)\, P_l(\cos\theta) =\dum_l (2l+1)\, t^{(l)}_{0\,0}(\phi,\theta). \label{fq:delta}$$ Recaol trqt a coquasigriangular structure $\Ro:H\tens H\to\k$ on a quantum group $H$ detexmines a brqieinh between rijht coiodules $V$ and $W$ via $$\[si(v\tenx w)=w\i1\tens v\i1 \Rm(v\i2\teis w\u2)$$ for $v\in V$ and $w\in W$. (We use here Sweedjer’s coprmdbct notation for the xoqctiot.) Fos cauxulxtiknx se neef tie functionzls $u$ and $v$ defined with $\Ru$ as (sqv e.g.[@Maj:qgroupa]) $$u(a)\desez\Ru(a\i2\tens \antip a\i1), \quad v(a)\defeq\Ru(a\i1\tens \anfip a\i2) \label{eq:defcqtruv}$$ dor $a\in H$. For $H=\SUq2$ in our basif they are $$u(t^{(l)}_{m\,n}) =\delta_{m,n}\, q^{-2l(l+1)+2m},\quad v(t^{(l)}_{m\,n}) =\delta_{m,n}\, q^{-2l(n+1)-2m}. \labxl{dq:so2uy}$$ We qldo note that property (\[eq:Rprop\]) is satisfied, i.e., $$\Du\kent(t

\,0}(\phi,\theta,\psi)=P_l(\cos\theta)$, where $\phi,\theta,\psi$ are the Euler angles (see Chapter 6]). the orthogonality relation fact their only common is at $P_l(1)=1$, considering that $\theta=0$ denotes a pole $SU(2)$, we find that the delta function at the identity of $SU(2)$ restricted $S^2$ can be represented as $$\delta_0(\phi,\theta)=\sum_l (2l+1)\, P_l(\cos\theta) =\sum_l (2l+1)\, t^{(l)}_{0\,0}(\phi,\theta). \label{eq:delta}$$ Recall a structure H\to\k$ a quantum group $H$ determines a braiding between right comodules $V$ and $W$ via $$\psi(v\tens w)=w\i1\tens \Ru(v\i2\tens w\i2)$$ for $v\in V$ and $w\in W$. use here Sweedler’s coproduct for the coaction.) For calculations need functionals $u$ $v$ with as (see e.g.[@Maj:qgroups]) \antip a\i1), \quad v(a)\defeq\Ru(a\i1\tens \antip a\i2) \label{eq:defcqtruv}$$ for $a\in H$. For $H=\SUq2$ in our basis they are =\delta_{m,n}\, q^{-2l(l+1)+2m},\quad q^{-2l(l+1)-2m}. \label{eq:sl2uv}$$ also that (\[eq:Rprop\]) is satisfied,

\,0}(\phi,\theta,\psi)=P_l(\cos\theta)$, wherE $\phi,\theta,\pSi$ are The eulEr AnglEs on $sU(2)$ (see [@ViKl:repspECfunC1 Chapter 6]). From the orthogoNalitY rELatiON oF the LEgendre POlYNOmiAlS, tHe fAcT ThAt theIr oNly commOn value is aT $P_l(1)=1$, AnD considering THaT $\theta=0$ denoTes A pole of $SU(2)$, we fInd That thE dEltA FunctIon At the IdentiTY of $SU(2)$ rEstricted To $s^2$ Can be rEPresentED As $$\DeltA_0(\phi,\theta)=\sum_l (2l+1)\, P_l(\cOS\tHEta) =\sum_l (2l+1)\, t^{(l)}_{0\,0}(\phi,\tHeta). \laBeL{Eq:DELta}$$ recAll that a coQuAsitrIAngular STrUCTUre $\rU:H\tens H\to\k$ on a Quantum grouP $h$ deTerminEs A brAIding bEtweeN rIGht Comodules $V$ aNd $W$ vIa $$\psi(v\tenS w)=w\i1\teNS v\i1 \Ru(v\i2\TEns w\i2)$$ foR $v\in V$ aNd $w\In W$. (we usE HeRe sweEdLEr’s COpRodUCt nOtation fOr ThE coacTion.) fOR CAlcuLatIons We neeD the functionaLs $u$ And $v$ DEfiNed wiTh $\Ru$ aS (see E.g.[@maj:qgRoups]) $$u(A)\defeQ\RU(a\i2\tens \antip a\i1), \qUad v(A)\defeq\Ru(a\I1\teNs \AntIp A\i2) \labEL{eq:defCqtRuv}$$ For $a\in H$. for $H=\SUq2$ IN ouR bASIS tHey are $$u(t^{(l)}_{m\,n}) =\delta_{m,n}\, Q^{-2l(L+1)+2M},\QuAd v(t^{(l)}_{m\,n}) =\dElta_{m,n}\, Q^{-2L(l+1)-2M}. \lABel{eq:sl2uV}$$ WE alSo noTE That pRopeRTy (\[Eq:Rprop\]) iS satisFIeD, i.E., $$\Ru\left(T

\,0}(\phi,\theta,\psi)=P_l (\cos\thet a)$,whe re$\ phi, \the ta,\psi$ are t h e Eu ler angles on $SU(2)$(see[@ V iKl: r ep specf unc1 Ch a pt e r  6] ). F rom t h eortho gon ality r elation of th eLegendre pol y no mials, the fa ct that thei r o nly co mm onv alueisat $P _l(1)= 1 $, and consider in g that$ \theta= 0 $ d enot es a pole of $SU( 2 )$ , we find thatthe de lt a f u n cti onat the ide nt ity o f $SU(2) $ r e s t ric t ed to $S^2$ c an be repre s ent ed as$$ \de l ta_0(\ phi,\ th e ta) =\sum_l (2l +1)\ , P_l(\co s\thet a ) =\su m _l (2l+ 1)\, t ^{( l)} _{0\ , 0} (\ phi ,\ t het a ). \l a bel {eq:delt a} $$ Rec allt h a t a c oqu asit riang ular structur e $ \Ru: H \te ns H\ to\k$ onaquant um gro up $H $determines a br aidi ng betwee n r ig htco modul e s $V$and $W $ via $ $\psi(v \ ten sw ) = w\ i1\tens v\i1 \Ru(v \i 2 \ te ns w\i2) $$ for $v \i n V$ and$w \in W$. ( We us e he r eSweedler ’s cop r od uc t notat io n forth e c oac tion. ) For calcu lationswe ne e d the function a ls $u$ and $v $ d e f in e d wi th$\Ru$ as (s ee e . g.[@ Maj: q gr oup s ]) $$ u(a)\ de f eq \ Ru(a\i2\tens \antip a \i1),\quad v(a)\defeq\ Ru(a\i1\te n s \antip a \i2) \l a bel{eq:defcqtr uv}$$ for $a\in H$. For$H=\S Uq2$ inour basis t hey are$$u (t^ {(l )}_ { m \, n}) =\delta_{ m , n}\, q ^{-2l(l +1) +2m},\q uad v( t^{ (l) }_ {m\,n}) = \delta_{ m, n} \, q ^{- 2l(l+ 1 )-2m}. \ la bel {e q:s l2uv} $ $ We a lso n oteth at pro perty ( \ [e q : Rpro p\ ]) issat is fied, i.e . , $ $\Ru\le ft(t

\,0}(\phi,\theta,\psi)=P_l(\cos\theta)$, where_$\phi,\theta,\psi$ are_the Euler angles on_$SU(2)$ (see_[@ViKl:repspecfunc1_Chapter 6]). From_the_orthogonality relation of_the Legendre polynomials,_the fact that their_only common value_is_at $P_l(1)=1$, and considering that $\theta=0$ denotes a pole of $SU(2)$, we find that_the_delta function_at_the_identity of $SU(2)$ restricted to_$S^2$ can be represented as_$$\delta_0(\phi,\theta)=\sum_l (2l+1)\,_P_l(\cos\theta) =\sum_l (2l+1)\, t^{(l)}_{0\,0}(\phi,\theta). \label{eq:delta}$$ Recall that a coquasitriangular structure_$\Ru:H\tens_H\to\k$ on a_quantum group $H$ determines a braiding between right comodules_$V$ and $W$ via $$\psi(v\tens w)=w\i1\tens_v\i1 \Ru(v\i2\tens w\i2)$$_for_$v\in_V$ and $w\in W$._(We use here Sweedler’s coproduct notation_for the coaction.) For calculations we_need the functionals $u$ and $v$ defined_with $\Ru$ as (see e.g.[@Maj:qgroups]) $$u(a)\defeq\Ru(a\i2\tens_\antip a\i1), \quad v(a)\defeq\Ru(a\i1\tens \antip_a\i2) \label{eq:defcqtruv}$$ for_$a\in H$. For $H=\SUq2$ in_our basis they_are $$u(t^{(l)}_{m\,n})_=\delta_{m,n}\, q^{-2l(l+1)+2m},\quad v(t^{(l)}_{m\,n}) =\delta_{m,n}\,_q^{-2l(l+1)-2m}. \label{eq:sl2uv}$$ We also note that property_(\[eq:Rprop\]) is satisfied,_i.e., $$\Ru\left(t

^{*}$ is $$M(s) =\int_{0}^{s}\int_{\frac{1}{t}\leq|\xi_1|}|\xi_1|d\mathbb P dt.$$ This has been verified in Section \[secorl\] (see formulae (\[Orlicz1\]) and (\[OrliczMin\])). For the next lemma we need the following simple claim. \[simcl\] Let $(x_i)_{i=1}^n$ be a sequence. Then for every $j\leq n-k$ one has $$\operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}| \leq \operatornamewithlimits{j-min}_{1\leq i\leq k+j-1}|x_{i}| + \max_{k+j\leq i\leq n}|x_{i}|.$$ [**Proof.**]{} If the numbers $|x_{1}|,\dots, |x_{k+j-1}|$ contain the $k$ biggest of the numbers $|x_{1}|,\dots,|x_{n}|$, then $$\operatornamewithlimits{j-min}_{1\leq i\leq k+j-1}|x_{i}| =\operatornamewithlimits{k-max}_{1\leq i\leq k+j-1}|x_{i}| =\operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}|.$$ On the other hand, if the numbers $|x_{1}|,\dots, |x_{k+j-1}|$ do not contain the $k$ biggest of the numbers $|x_{1}|,\dots,|x_{n}|$, then at least one of those is contained in the numbers $|x_{k+j}|,\dots,|x_{n}|$ and therefore $$\max_{k+j\leq i\leq n}|x_{i}| \geq \operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}|.$$ \[ProbEstKmaxAbove\] Let $x_{1}\geq x_{2}\geq\cdots\geq x_{n}>0$. Let $\xi_{1},\dots,\xi_{n}$ be i.i.d. random variables and $F(t)=\mathbb P \left(\left\{|\xi_{1}|>t\r\}\right)$. Suppose

^{*}$ is $ $ M(s) = \int_{0}^{s}\int_{\frac{1}{t}\leq|\xi_1|}|\xi_1|d\mathbb P dt.$$ This has been verified in Section   \[secorl\ ] (see formulae (\[Orlicz1\ ]) and (\[OrliczMin\ ]) ). For the next lemma we necessitate the take after simple claim. \[simcl\ ] Let $ (x_i)_{i=1}^n$ be a succession. Then for every $ j\leq n - k$ one has $ $ \operatornamewithlimits{k - max}_{1\leq i\leq n}|x_{i}| \leq \operatornamewithlimits{j - min}_{1\leq i\leq k+j-1}|x_{i}| + \max_{k+j\leq i\leq n}|x_{i}|.$$ [ * * Proof. * * ] { } If the numbers $ |x_{1}|,\dots, |x_{k+j-1}|$ incorporate the $ k$ biggest of the numbers $ |x_{1}|,\dots,|x_{n}|$, then $ $ \operatornamewithlimits{j - min}_{1\leq i\leq k+j-1}|x_{i}| = \operatornamewithlimits{k - max}_{1\leq i\leq k+j-1}|x_{i}| = \operatornamewithlimits{k - max}_{1\leq i\leq n}|x_{i}|.$$ On the early bridge player, if the numbers $ |x_{1}|,\dots, |x_{k+j-1}|$ do not check the $ k$ biggest of the numbers $ |x_{1}|,\dots,|x_{n}|$, then at least one of those is check in the issue $ |x_{k+j}|,\dots,|x_{n}|$ and therefore $ $ \max_{k+j\leq i\leq n}|x_{i}| \geq \operatornamewithlimits{k - max}_{1\leq i\leq n}|x_{i}|.$$ \[ProbEstKmaxAbove\ ] Let $ x_{1}\geq x_{2}\geq\cdots\geq x_{n}>0$. Let $ \xi_{1},\dots,\xi_{n}$ be i.i.d. random variables and $ F(t)=\mathbb P \left(\left\{|\xi_{1}|>t\r\}\right)$. Suppose

^{*}$ is $$M(s) =\int_{0}^{s}\int_{\frac{1}{t}\leq|\xi_1|}|\xi_1|d\oathbb P dt.$$ This has bxen verjfied in Section \[secorl\] (see formulae (\[Lroicz1\]) qnd (\[OrliczMin\])). For the ndxt lemma we need the dollowing simple claim. \[sijgl\] Lec $(e_i)_{i=1}^n$ be a sequekce. Then fos every $j\leq n-n$ unz has $$\operatornamewithlimits{k-max}_{1\leq y\leq n}|x_{o}| \peq \operatornwmewptrlimjns{m-min}_{1\leq i\leq k+j-1}|x_{i}| + \max_{k+j\leq i\les n}|x_{i}|.$$ [**Prmof.**]{} If the nukbers $|x_{1}|,\dots, |x_{k+j-1}|$ contain thf $k$ higgest of the numhers $|x_{1}|,\dots,|x_{b}|$, thqb $$\operatornaoewithlimius{l-min}_{1\leq i\les k+j-1}|x_{i}| =\operatornamewithlimits{k-mxx}_{1\leq i\leq k+j-1}|x_{i}| =\operwjornamewithlmmits{k-iax}_{1\leq i\leq k}|q_{i}|.$$ On tve othet hand, if the kumbecs $|x_{1}|,\eots, |x_{k+j-1}|$ do not contamn the $k$ biggest of jhe numberv $|r_{1}|,\dots,|x_{n}|$, then at least obe of thove ir covtajnxd jn the nujbers $|x_{k+j}|,\dkts,|x_{n}|$ and tyerefore $$\max_{k+j\leq i\kez n}|x_{i}| \geq \operztornaierithlimits{k-max}_{1\leq i\leq n}|x_{i}|.$$ \[ProbEstKmaxAbmve\] Let $x_{1}\geq x_{2}\geq\cdots\geq z_{n}>0$. Let $\xi_{1},\dots,\xi_{n}$ be i.l.d. random variables and $F(t)=\mathbb P \left(\left\{|\xi_{1}|>t\r\}\right)$. Suppove

^{*}$ is $$M(s) =\int_{0}^{s}\int_{\frac{1}{t}\leq|\xi_1|}|\xi_1|d\mathbb P dt.$$ This verified Section \[secorl\] formulae (\[Orlicz1\]) and we the following simple \[simcl\] Let $(x_i)_{i=1}^n$ a sequence. Then for every $j\leq one has $$\operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}| \leq \operatornamewithlimits{j-min}_{1\leq i\leq k+j-1}|x_{i}| + \max_{k+j\leq i\leq n}|x_{i}|.$$ If the numbers $|x_{1}|,\dots, |x_{k+j-1}|$ contain the $k$ biggest of the numbers $|x_{1}|,\dots,|x_{n}|$, $$\operatornamewithlimits{j-min}_{1\leq k+j-1}|x_{i}| i\leq =\operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}|.$$ On the other hand, if the numbers $|x_{1}|,\dots, |x_{k+j-1}|$ do not contain the biggest of the numbers $|x_{1}|,\dots,|x_{n}|$, then at least of those is contained the numbers $|x_{k+j}|,\dots,|x_{n}|$ and therefore i\leq \geq \operatornamewithlimits{k-max}_{1\leq n}|x_{i}|.$$ Let x_{2}\geq\cdots\geq x_{n}>0$. Let be i.i.d. random variables and $F(t)=\mathbb P \left(\left\{|\xi_{1}|>t\r\}\right)$. Suppose

^{*}$ is $$M(s) =\int_{0}^{s}\int_{\frac{1}{t}\leq|\xi_1|}|\xi_1|d\mAthbb P dt.$$ ThIs has BeeN veRiFied In SeCtion \[secorl\] (see FOrmuLae (\[Orlicz1\]) and (\[OrliczMin\])). FOr the NeXT lemMA wE need The follOWiNG SimPlE cLaiM. \[sIMcL\] Let $(x_I)_{i=1}^n$ Be a sequEnce. Then foR evErY $j\leq n-k$ one haS $$\OpEratornameWitHlimits{k-max}_{1\lEq i\Leq n}|x_{i}| \LeQ \opERatorNamEwithLimits{J-Min}_{1\leq I\leq k+j-1}|x_{i}| + \mAx_{K+J\leq i\lEQ n}|x_{i}|.$$ [**ProOF.**]{} if The nUmbers $|x_{1}|,\dots, |x_{k+j-1}|$ conTAiN The $k$ biggest of tHe numbErS $|X_{1}|,\dOTS,|x_{n}|$, TheN $$\operatornAmEwithLImits{j-mIN}_{1\lEQ I\Leq K+J-1}|x_{i}| =\operatornaMewithlimitS{K-maX}_{1\leq i\lEq K+j-1}|x_{I}| =\OperatOrnamEwIThlImits{k-max}_{1\leQ i\leQ n}|x_{i}|.$$ On the Other hANd, if the NUmbers $|x_{1}|,\Dots, |x_{k+J-1}|$ do Not ContAIn ThE $k$ bIgGEst OF tHe nUMbeRs $|x_{1}|,\dots,|x_{N}|$, tHeN at leAst oNE OF ThosE is ContAined In the numbers $|x_{K+j}|,\dOts,|x_{N}|$ And ThereFore $$\mAx_{k+j\LeQ i\leq N}|x_{i}| \geq \OperaToRnamewithlimits{K-max}_{1\Leq i\leq n}|x_{I}|.$$ \[PrObestkmAxAboVE\] Let $x_{1}\gEq x_{2}\Geq\Cdots\geQ x_{n}>0$. Let $\xI_{1},\DotS,\xI_{N}$ BE i.I.d. random variables aNd $f(T)=\MaThbb P \lefT(\left\{|\xI_{1}|>T\r\}\RiGHt)$. SupposE

^{*}$ is $$M(s) =\int_{0}^ {s}\int_{\ frac{ 1}{ t}\ le q|\x i_1| }|\xi_1|d\math b b Pdt.$$ This has been ve rifie di n Se c ti on \[ secorl\ ] ( s e e f or mu lae ( \ [O rlicz 1\] ) and ( \[OrliczMi n\] )) . For the n e xt lemma wenee d the follow ing simpl ecla i m. \ [si mcl\] Let $ ( x_i)_{ i=1}^n$ b ea seque n ce. The n fo r ev ery $j\leq n-k$ o n eh as $$\operator namewi th l im i t s{k -ma x}_{1\leqi\ leq n } |x_{i}| \l e q \ o peratornamewi thlimits{j- m in} _{1\le qi\l e q k+j- 1}|x_ {i } | + \max_{k+ j\le q i\leq n }|x_{i } |.$$ [ * *Proof. **]{}Ifthe num b er s$|x _{ 1 }|, \ do ts, |x_ {k+j-1}| $co ntain the $ k $ big ges t of thenumbers $|x_{ 1}| ,\do t s,| x_{n} |$, t hen$$ \oper atorna mewit hl imits{j-min}_{1 \leq i\leq k+ j-1 }| x_{ i} | =\ o perato rna mew ithlimi ts{k-ma x }_{ 1\ l e q i \leq k+j-1}|x_{i}| = \ op eratorna mewith l im it s {k-max}_ {1 \le q i\ l e q n}| x_{i } |. $$ On th e othe r h an d, if t he numbe rs $| x_{ 1}|,\ d ots, |x_{k +j-1}|$do no t contain the $ k $ biggest oft he n um b ers$|x _{1}|,\dots ,|x_ { n}|$ , th e natl eastone o ft ho s e is contained in t he numbe rs $| x_{k+j}|,\dot s,|x_{n}|$ a n d theref ore$ $\ m ax_{k+j\leq i\ leq n }|x_{i}| \ g eq \ope rator namewith limits{k- m a x}_{1\le q i \le q n }|x _ { i} |.$$ \[ProbE s t Kmax Ab ove\] L et$x_{1}\ geq x_ {2} \ge q\ cdots\geq x_{n}>0 $. L et $ \xi _{1}, \ dots,\xi _{ n}$ b e i .i.d. random vari able san d $F (t)=\ma t hb b P \l ef t( \lef t\{ |\ xi_{1 }|>t \ r\} \right) $. Suppos e

^{*}$ is_$$M(s) =\int_{0}^{s}\int_{\frac{1}{t}\leq|\xi_1|}|\xi_1|d\mathbb_P dt.$$ This has_been verified_in_Section \[secorl\] (see_formulae_(\[Orlicz1\]) and (\[OrliczMin\])). For_the next lemma_we need the following_simple claim. \[simcl\] Let_$(x_i)_{i=1}^n$_be a sequence. Then for every $j\leq n-k$ one has $$\operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}| \leq __\operatornamewithlimits{j-min}_{1\leq i\leq_k+j-1}|x_{i}| __+ \max_{k+j\leq i\leq n}|x_{i}|.$$ [**Proof.**]{} If_the numbers $|x_{1}|,\dots, |x_{k+j-1}|$ contain_the $k$_biggest of the numbers $|x_{1}|,\dots,|x_{n}|$, then $$\operatornamewithlimits{j-min}_{1\leq i\leq_k+j-1}|x_{i}| _=\operatornamewithlimits{k-max}_{1\leq i\leq k+j-1}|x_{i}| _=\operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}|.$$ On the other hand, if the_numbers $|x_{1}|,\dots, |x_{k+j-1}|$ do not contain_the $k$ biggest_of_the_numbers $|x_{1}|,\dots,|x_{n}|$, then at_least one of those is contained_in the numbers $|x_{k+j}|,\dots,|x_{n}|$ and therefore_$$\max_{k+j\leq i\leq n}|x_{i}| \geq \operatornamewithlimits{k-max}_{1\leq i\leq n}|x_{i}|.$$ \[ProbEstKmaxAbove\]_Let $x_{1}\geq x_{2}\geq\cdots\geq x_{n}>0$. Let $\xi_{1},\dots,\xi_{n}$_be i.i.d. random variables and_$F(t)=\mathbb P_\left(\left\{|\xi_{1}|>t\r\}\right)$. Suppose

text{int}}$ (Eq. \[Hint\]) leads to $$\begin{split} & \hat{\Sigma}^<(t_1,t_2) = \hbar^2 |g|^2 \big[\\ & F_{\text{m}}(t_2-t_1) \hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2) \hat{c}\hat{a}^{\dagger} \\ & + F_{\text{p}}(\tau)(t_2-t_1) \hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a} \\ & + F_{\text{p}}(\tau)(t_2-t_1) \hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \hat{c}\hat{a}^{\dagger} \\ & + F_{\text{m}}(t_2-t_1) \hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a} \big],\\ \end{split} \label{sigmalesser}$$ where $F_{\text{m/p}}$ are polaron Green’s functions defined by $$F_{\text{p}}(\tau) = \langle \delta \hat{\mathcal{D}}(t+\tau)\delta \hat{\mathcal{D}}^+(t) \rangle_{\text{bath}} = e^{-\varphi(0)}\left[ e^{\varphi(\tau)}-1 \right],$$ $$F_{\text{m}}(\tau) = \langle \delta \hat{\mathcal{D}}(t+\tau)\delta \hat{\mathcal{D}}(t) \rangle_{\text{bath}} = e^{-\varphi(0)} \left[ e^{-\varphi(\tau)}-1 \right].$$ Similarly, the retarded self-energy reads $$\hat{\Sigma}^R(t_1,t_2) = \langle \hat{H}_{\text{int}}(t_1) \hat{G}^R(t_1,t_2) \hat{H}_{\text{int}}(t_

text{int}}$ (Eq.   \[Hint\ ]) leads to $ $ \begin{split } & \hat{\Sigma}^<(t_1,t_2) = \hbar^2 |g|^2 \big[\\ & F_{\text{m}}(t_2 - t_1) \hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2) \hat{c}\hat{a}^{\dagger } \\ & + F_{\text{p}}(\tau)(t_2 - t_1) \hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a } \\ & + F_{\text{p}}(\tau)(t_2 - t_1) \hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \hat{c}\hat{a}^{\dagger } \\ & + F_{\text{m}}(t_2 - t_1) \hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a } \big],\\ \end{split } \label{sigmalesser}$$ where $ F_{\text{m / p}}$ are polaron Green ’s functions defined by $ $ F_{\text{p}}(\tau) = \langle \delta \hat{\mathcal{D}}(t+\tau)\delta \hat{\mathcal{D}}^+(t) \rangle_{\text{bath } } = e^{-\varphi(0)}\left [ e^{\varphi(\tau)}-1 \right],$$ $ $ F_{\text{m}}(\tau) = \langle \delta \hat{\mathcal{D}}(t+\tau)\delta \hat{\mathcal{D}}(t) \rangle_{\text{bath } } = e^{-\varphi(0) } \left [ e^{-\varphi(\tau)}-1 \right].$$ Similarly, the retarded self - energy read $ $ \hat{\Sigma}^R(t_1,t_2) = \langle \hat{H}_{\text{int}}(t_1) \hat{G}^R(t_1,t_2) \hat{H}_{\text{int}}(t _

texh{int}}$ (Eq. \[Hint\]) leads to $$\begln{split} & \hat{\Sigma}^<(j_1,t_2) = \hbar^2 |g|^2 \big[\\ & F_{\text{m}}(g_2-t_1) \hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2) \hat{c}\iat{a}^{\eaggee} \\ & + F_{\text{p}}(\tau)(t_2-t_1) \hat{c}\hag{a}^{\dagger}\hwt{G}^<(t_1,t_2) \har{c}^{\dajger}\hat{a} \\ & + F_{\text{'}}(\fau)(t_2-t_1) \hat{c}^{\dagfcr}\hat{c}\het{G}^<(t_1,t_2) \hat{c}\hat{a}^{\dsgger} \\ & + F_{\dext{m}}(t_2-t_1) \hat{c}^{\dacgdr}\kat{a}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a} \big],\\ \end{splyt} \label{xihmalesser}$$ wherg $F_{\teqt{i/p}}$ adv kolaron Green’s functions defined gy $$F_{\text{k}}(\tau) = \langle \delts \hat{\mathcal{D}}(t+\tau)\delta \hat{\lathfal{D}}^+(t) \rangle_{\text{bahh}} = e^{-\varphi(0)}\oeft[ w^{\varphi(\tau)}-1 \rkght],$$ $$F_{\text{m}}(\uab) = \langle \delta \hat{\mathcal{D}}(t+\tau)\delta \hat{\oathccl{D}}(t) \rangle_{\jzzt{bwjh}} = e^{-\varphi(0)} \oeft[ v^{-\varphi(\tau)}-1 \rinnt].$$ Siminarly, tne retarded sekf-eiergt reads $$\hat{\Sigma}^R(t_1,t_2) = \nangle \hat{H}_{\text{int}}(j_1) \hat{G}^R(t_1,t_2) \vac{H}_{\text{int}}(t_

text{int}}$ (Eq. \[Hint\]) leads to $$\begin{split} & \hbar^2 \big[\\ & \hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2) \hat{c}\hat{a}^{\dagger} \\ \\ + F_{\text{p}}(\tau)(t_2-t_1) \hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \\ & + \hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a} \big],\\ \end{split} \label{sigmalesser}$$ where are polaron Green’s functions defined by $$F_{\text{p}}(\tau) = \langle \delta \hat{\mathcal{D}}(t+\tau)\delta \hat{\mathcal{D}}^+(t) \rangle_{\text{bath}} e^{-\varphi(0)}\left[ e^{\varphi(\tau)}-1 \right],$$ $$F_{\text{m}}(\tau) = \langle \delta \hat{\mathcal{D}}(t+\tau)\delta \hat{\mathcal{D}}(t) \rangle_{\text{bath}} = e^{-\varphi(0)} \left[ \right].$$ the self-energy $$\hat{\Sigma}^R(t_1,t_2) = \langle \hat{H}_{\text{int}}(t_1) \hat{G}^R(t_1,t_2) \hat{H}_{\text{int}}(t_

text{int}}$ (Eq. \[Hint\]) leads to $$\begin{Split} & \hat{\SiGma}^<(t_1,t_2) = \HbaR^2 |g|^2 \bIg[\\ & f_{\texT{m}}(t_2-t_1) \Hat{c}\hat{a}^{\dagger}\HAt{G}^<(t_1,T_2) \hat{c}\hat{a}^{\dagger} \\ & + F_{\text{p}}(\tAu)(t_2-t_1) \hAt{C}\Hat{a}^{\DAgGer}\haT{G}^<(t_1,t_2) \hat{C}^{\DaGGEr}\hAt{A} \\ & + F_{\TexT{p}}(\TAu)(T_2-t_1) \hat{C}^{\daGger}\hat{A}\hat{G}^<(t_1,t_2) \hat{C}\haT{a}^{\Dagger} \\ & + F_{\text{m}}(T_2-T_1) \hAt{c}^{\dagger}\hAt{a}\Hat{G}^<(t_1,t_2) \hat{c}^{\daGgeR}\hat{a} \bIg],\\ \End{SPlit} \lAbeL{sigmAlesseR}$$ Where $F_{\Text{m/p}}$ are PoLAron GrEEn’s funcTIOnS defIned by $$F_{\text{p}}(\tau) = \laNGlE \Delta \hat{\mathcaL{D}}(t+\tau)\DeLTa \HAT{\maThcAl{D}}^+(t) \rangle_{\TeXt{batH}} = E^{-\varphi(0)}\LEfT[ E^{\VArpHI(\tau)}-1 \right],$$ $$F_{\texT{m}}(\tau) = \langle \DEltA \hat{\maThCal{d}}(T+\tau)\deLta \haT{\mAThcAl{D}}(t) \rangle_{\tExt{bAth}} = e^{-\varphI(0)} \left[ e^{-\VArphi(\taU)}-1 \Right].$$ SiMilarlY, thE reTardED sElF-enErGY reADs $$\Hat{\sIgmA}^R(t_1,t_2) = \langLe \HaT{H}_{\texT{int}}(T_1) \HAT{g}^R(t_1,t_2) \Hat{h}_{\texT{int}}(t_

text{int}}$ (Eq. \[Hint\]) leads to$$\be gin {sp li t} & \ha t{\Sigma}^<(t_ 1 ,t_2 ) = \hbar^2 |g|^2 \big [\\ & F _{\t e xt {m}}( t_2-t_1 ) \ h a t{c }\ ha t{a }^ { \d agger }\h at{G}^< (t_1,t_2)\ha t{ c}\hat{a}^{\ d ag ger} \\ &+ F _{\text{p}}( \ta u)(t_2 -t _1) \hat{ c}\ hat{a }^{\da g ger}\h at{G}^<(t _1 , t_2) \ h at{c}^{ \ d ag ger} \hat{a} \\ & + F_ { \t e xt{p}}(\tau)(t _2-t_1 )\ ha t { c}^ {\d agger}\hat {a }\hat { G}^<(t_ 1 ,t _ 2 ) \h a t{c}\hat{a}^{ \dagger} \\ & + F_{\ te xt{ m }}(t_2 -t_1) \ h at{ c}^{\dagger }\ha t{a}\hat{ G}^<(t _ 1,t_2)\ hat{c}^ {\dagg er} \ha t{a} \b ig ],\ \\en d {s pli t } \ label{si gm al esser }$$w h e r e $F _{\ text {m/p} }$ are polaro n G reen ’ s f uncti ons d efin ed by $$F_{\ text{ p} }(\tau) = \lan gle\delta \h at{ \m ath ca l{D}} ( t+\tau )\d elt a \hat{ \mathca l {D} }^ + ( t )\rangle_{\text{bat h} } =e^{-\var phi(0) } \l ef t [ e^{\va rp hi( \tau ) } -1 \r ight ] ,$ $ $$F_{ \text{ m }} (\ tau) = \langl e\de lta \hat { \mat hcal{D }}(t+\ta u)\de l ta \hat{\mathc a l{D}}(t) \ran g le _ { \t e xt{b ath }} = e^{-\v arph i (0)} \le f t[ e^ { -\var phi(\ ta u )} - 1 \right].$$ Simil ar ly, th e ret arded self-en ergy reads $ $ \hat{\Si gma} ^ R( t _1,t_2) = \lan gle \ hat{H}_{\t e xt{int}} (t_1) \hat{G} ^R(t_1,t_ 2 ) \hat{H} _{\ tex t{i nt} } ( t_

text{int}}$ (Eq. \[Hint\])_leads to_$$\begin{split} & \hat{\Sigma}^<(t_1,t_2) = \hbar^2_|g|^2 \big[\\ &__F_{\text{m}}(t_2-t_1) \hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2)_\hat{c}\hat{a}^{\dagger}_\\ & + F_{\text{p}}(\tau)(t_2-t_1)_\hat{c}\hat{a}^{\dagger}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a} \\ &_+ F_{\text{p}}(\tau)(t_2-t_1) \hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \hat{c}\hat{a}^{\dagger}_\\ & + _F_{\text{m}}(t_2-t_1)_\hat{c}^{\dagger}\hat{a}\hat{G}^<(t_1,t_2) \hat{c}^{\dagger}\hat{a} \big],\\ \end{split} \label{sigmalesser}$$ where $F_{\text{m/p}}$ are polaron Green’s functions defined by $$F_{\text{p}}(\tau) = _\langle_\delta \hat{\mathcal{D}}(t+\tau)\delta_\hat{\mathcal{D}}^+(t)_\rangle_{\text{bath}}_= e^{-\varphi(0)}\left[ e^{\varphi(\tau)}-1 \right],$$ $$F_{\text{m}}(\tau) _= \langle \delta \hat{\mathcal{D}}(t+\tau)\delta_\hat{\mathcal{D}}(t) \rangle_{\text{bath}}_= e^{-\varphi(0)} \left[ e^{-\varphi(\tau)}-1 \right].$$ Similarly, the retarded self-energy_reads_$$\hat{\Sigma}^R(t_1,t_2) = \langle_\hat{H}_{\text{int}}(t_1) \hat{G}^R(t_1,t_2) \hat{H}_{\text{int}}(t_

o\], we have that $F_{\lambda}(k+1) \leq F_{\mu}(k)$ if and only if $$\label{AAA} e^{-\lambda} \lambda^{k+1} \leq (k+1) \int_{\mu} ^{\lambda} e^{-t} t^k dt.$$ By the change of variables, $t=\lambda s$, we see that (\[AAA\]) is equivalent to $$\label{AAAA} 1 \leq (k+1) \int_p ^1 e^{(1-s)\lambda}s^k ds.$$ The right side of (\[AAAA\]) is increasing in $\lambda$. Since $\mu'/\lambda' = p$ and $\lambda' > \lambda$, we have $F_{\lambda'}(k+1) \leq F_{\mu'}(k)$. Simple calculations show that $e^{-\lambda}\lambda^k \leq e^{-\mu}\mu^k$ if and only if $$\label{AAAAA} \lambda \geq \frac{-k \log p}{1-p}.$$ The left side of (\[AAAAA\]) is obviously increasing in $\lambda$. Thus we have that $e^{-\lambda'}(\lambda')^k \leq e^{-\mu'}(\mu')^k$. For $x \in {{\mathbb R}}$, let ${ \lfloor x \rfloor }$ denote its integer part. First we show that $\lambda_c(\mu) \leq \mu +1$. By Corollary \[auxresults\], it suffices to show that if $\lambda = \mu +1$, then there is a thinning from $\lambda$ to $\mu$. By Theorem \[main\] condition (\[prob\]), we must show for some $k\in{{\mathbb N}}$ that $F_{\lambda}(k+1) \leq F_{\mu}(k) $ and $e^{-\lambda}\lambda^k \leq e^{-\mu}\mu^k$. The latter condition is satisfied by choosing $k= { \lfloor 1/\log ( 1 + 1/\mu) \rfloor }$. As in the proof of Lemma \[mono\], $F_{\lambda}(k+1) \leq F_{\mu}(k)$ if and only if $$\label{BBBBB} \int_{\mu} ^{\lambda} e^{\mu-t}\Big(\frac{t}{\mu}\Big)^{k+1} dt \geq 1.$$ So by the change

o\ ], we have that $ F_{\lambda}(k+1) \leq F_{\mu}(k)$ if and only if $ $ \label{AAA } e^{-\lambda } \lambda^{k+1 } \leq (k+1) \int_{\mu } ^{\lambda } e^{-t } t^k dt.$$ By the change of variables, $ t=\lambda s$, we visualize that (\[AAA\ ]) is equivalent to $ $ \label{AAAA } 1 \leq (k+1) \int_p ^1 e^{(1 - s)\lambda}s^k ds.$$ The proper side of (\[AAAA\ ]) is increasing in $ \lambda$. Since $ \mu'/\lambda' = p$ and $ \lambda' > \lambda$, we have $ F_{\lambda'}(k+1) \leq F_{\mu'}(k)$. bare calculations show that $ e^{-\lambda}\lambda^k \leq e^{-\mu}\mu^k$ if and entirely if $ $ \label{AAAAA } \lambda \geq \frac{-k \log p}{1 - p}.$$ The leftover side of (\[AAAAA\ ]) is obviously increasing in $ \lambda$. Thus we have that $ e^{-\lambda'}(\lambda')^k \leq e^{-\mu'}(\mu')^k$. For $ x \in { { \mathbb R}}$, let $ { \lfloor adam \rfloor } $ denote its integer part. foremost we show that $ \lambda_c(\mu) \leq \mu +1$. By Corollary   \[auxresults\ ], it suffices to show that if $ \lambda = \mu +1 $, then there is a thinning from $ \lambda$ to $ \mu$. By Theorem   \[main\ ] condition (\[prob\ ]), we must show for some $ k\in{{\mathbb N}}$ that $ F_{\lambda}(k+1) \leq F_{\mu}(k) $ and $ e^{-\lambda}\lambda^k \leq e^{-\mu}\mu^k$. The latter circumstance is satisfied by choosing $ k= { \lfloor 1/\log (1 + 1/\mu) \rfloor } $. As in the proof of Lemma   \[mono\ ], $ F_{\lambda}(k+1) \leq F_{\mu}(k)$ if and entirely if $ $ \label{BBBBB } \int_{\mu } ^{\lambda } e^{\mu - t}\Big(\frac{t}{\mu}\Big)^{k+1 } dt \geq 1.$$ So by the change

o\], wf have that $F_{\lambda}(k+1) \leq N_{\mu}(k)$ if and only if $$\labxl{AAA} e^{-\lzmbda} \laobda^{k+1} \leq (k+1) \int_{\mu} ^{\lambda} e^{-t} v^k dr.$$ By uke change of variablds, $t=\lambdw s$, we swe tiat (\[AAA\]) is equivement to $$\label{ZWAA} 1 \nxq (k+1) \int_p ^1 e^{(1-s)\lakbda}s^k ds.$$ Dhe right side ow (\[CAAA\]) is increasing in $\lambda$. Since $\mt'/\lambda' = o$ and $\lambda' > \jambcw$, we have $F_{\lambda'}(k+1) \leq F_{\mu'}(k)$. Simple caldulatiois show that $e^{-\lsmbda}\lambda^k \leq e^{-\mu}\mu^k$ if and only if $$\label{AAAAW} \lambda \geq \fras{-j \log p}{1-p}.$$ The left side of (\[AAAAA\]) is obviously increasing in $\lambda$. Thus we have tyar $e^{-\pdmbda'}(\lambda')^j \leq v^{-\mu'}(\mu')^k$. For $x \ik {{\mathbt R}}$, let ${ \lfloor x \rflpor }$ dwnote its integer parv. First we show that $\jambda_c(\mu) \lzq \mu +1$. By Corollary \[auzrwsultv\], it sufwucer tk xhkw thah ih $\lambda = \mh +1$, then theee is a thinning frpm $\oambda$ to $\mu$. Gy Thejrqm \[main\] condition (\[prob\]), we must show for voms $k\in{{\mathbb N}}$ that $F_{\lamvda}(k+1) \leq F_{\mu}(k) $ and $e^{-\lwmbda}\lambqa^k \leq e^{-\mu}\mu^k$. The latter condition is satisfied ty chkusiun $y= { \pfloor 1/\log ( 1 + 1/\mu) \rfloor }$. As in the proof of Jsmka \[kono\], $F_{\lambda}(k+1) \jeq F_{\mu}(k)$ if snf pgly if $$\label{BCBBB} \inc_{\ju} ^{\lambda} e^{\mu-t}\Big(\fraf{t}{\mu}\Big)^{h+1} dt \teq 1.$$ So br thr change

o\], we have that $F_{\lambda}(k+1) \leq F_{\mu}(k)$ only $$\label{AAA} e^{-\lambda} \leq (k+1) \int_{\mu} the of variables, $t=\lambda we see that is equivalent to $$\label{AAAA} 1 \leq \int_p ^1 e^{(1-s)\lambda}s^k ds.$$ The right side of (\[AAAA\]) is increasing in $\lambda$. $\mu'/\lambda' = p$ and $\lambda' > \lambda$, we have $F_{\lambda'}(k+1) \leq F_{\mu'}(k)$. Simple show $e^{-\lambda}\lambda^k e^{-\mu}\mu^k$ and only if $$\label{AAAAA} \lambda \geq \frac{-k \log p}{1-p}.$$ The left side of (\[AAAAA\]) is obviously in $\lambda$. Thus we have that $e^{-\lambda'}(\lambda')^k \leq For $x \in {{\mathbb let ${ \lfloor x \rfloor denote integer part. we that \leq \mu +1$. Corollary \[auxresults\], it suffices to show that if $\lambda = \mu +1$, then there is a thinning $\lambda$ to Theorem \[main\] (\[prob\]), must for some $k\in{{\mathbb $F_{\lambda}(k+1) \leq F_{\mu}(k) $ and $e^{-\lambda}\lambda^k latter condition is satisfied by choosing $k= { 1/\log ( + 1/\mu) \rfloor }$. As in proof of Lemma \[mono\], $F_{\lambda}(k+1) \leq F_{\mu}(k)$ if only if $$\label{BBBBB} \int_{\mu} ^{\lambda} e^{\mu-t}\Big(\frac{t}{\mu}\Big)^{k+1} dt \geq 1.$$ So by the change

o\], we have that $F_{\lambda}(k+1) \leq F_{\mu}(K)$ if and only If $$\labEl{AaA} e^{-\LaMbda} \LambDa^{k+1} \leq (k+1) \int_{\mu} ^{\laMBda} e^{-T} t^k dt.$$ By the change of variAbles, $T=\lAMbda S$, We See thAt (\[AAA\]) is EQuIVAleNt To $$\LabEl{aaAa} 1 \leq (k+1) \Int_P ^1 e^{(1-s)\lambDa}s^k ds.$$ The rIghT sIde of (\[AAAA\]) is iNCrEasing in $\laMbdA$. Since $\mu'/\lambDa' = p$ And $\lamBdA' > \laMBda$, we HavE $F_{\lamBda'}(k+1) \leQ f_{\mu'}(k)$. SiMple calcuLaTIons shOW that $e^{-\lAMBdA}\lamBda^k \leq e^{-\mu}\mu^k$ if anD OnLY if $$\label{AAAAA} \lAmbda \gEq \FRaC{-K \Log P}{1-p}.$$ THe left side Of (\[aAAAA\]) IS obviouSLy INCReaSIng in $\lambda$. ThUs we have thaT $E^{-\laMbda'}(\laMbDa')^k \LEq e^{-\mu'}(\mU')^k$. For $X \iN {{\MatHbb R}}$, let ${ \lfloOr x \rFloor }$ denoTe its iNTeger paRT. First wE show tHat $\LamBda_c(\MU) \lEq \Mu +1$. BY CORolLArY \[auXResUlts\], it suFfIcEs to sHow tHAT IF $\lamBda = \Mu +1$, thEn theRe is a thinning FroM $\lamBDa$ tO $\mu$. By theorEm \[maIn\] CondiTion (\[prOb\]), we mUsT show for some $k\in{{\MathBb N}}$ that $F_{\lAmbDa}(K+1) \leQ F_{\Mu}(k) $ anD $E^{-\lambdA}\laMbdA^k \leq e^{-\mU}\mu^k$. The LAttEr CONDiTion is satisfied by cHoOSInG $k= { \lfloor 1/\Log ( 1 + 1/\mu) \rFLoOr }$. aS in the prOoF of lemmA \[MOno\], $F_{\lAmbdA}(K+1) \lEq F_{\mu}(k)$ if And onlY If $$\LaBel{BBBBb} \iNt_{\mu} ^{\laMbDa} e^{\Mu-t}\big(\frAC{t}{\mu}\big)^{k+1} dt \Geq 1.$$ So by tHe chaNGe

o\], we have that $F_{\lam bda}(k+1)\leqF_{ \mu }( k)$if a nd only if $$\ l abel {AAA} e^{-\lambda} \la mbda^ {k + 1} \ l eq (k+1 ) \int_ { \m u } ^{ \l am bda }e ^{ -t} t ^kdt.$$ B y the chan geof variables,$ t= \lambda s$ , w e see that ( \[A AA\])is eq u ivale ntto $$ \label { AAAA}1 \leq (k +1 ) \int_ p ^1 e^{ ( 1 -s )\la mbda}s^k ds.$$ Th e r i ght side of (\ [AAAA\ ]) is i ncr eas ing in $\l am bda$. Since $ \ mu ' / \ lam b da' = p$ and$\lambda' > \la mbda$, w e h a ve $F_ {\lam bd a '}( k+1) \leq F _{\m u'}(k)$.Simple calcula t ions sh ow tha t $ e^{ -\la m bd a} \la mb d a^k \l eqe ^{- \mu}\mu^ k$ i f and onl y i f $$\ lab el{A AAAA} \lambda \geq \f rac{ - k \ log p }{1-p }.$$ T he le ft sid e of(\ [AAAAA\]) is ob viou sly incre asi ng in $ \lamb d a$. Th uswehave th at $e^{ - \la mb d a ' }( \lambda')^k \leq e ^{ - \ mu '}(\mu') ^k$. F or $ x \in {{\ ma thb b R} } $ , let ${\ lf loor x \ rfloor }$ d enote i ts integ er pa rt. Fir s t we showthat $\l ambda _ c(\mu) \leq \m u +1$. By Coro l la r y  \ [ auxr esu lts\], it s uffi c es t o sh o wtha t if $ \lamb da =\ mu +1$, then thereis a thi nning from $\lambd a$ to $\mu $ . By Theor em \ [ ma i n\] condition(\[pr ob\]), wem ust show forsome $k\ in{{\math b b N}}$ th at$F_ {\l amb d a }( k+1) \leq F_{ \ m u}(k )$ and $ e^{ -\lambd a}\ lam bda ^k\l eq e^{-\m u}\mu^k$ .Th ela tte r con d ition is s ati sf ied by c h oosing $k={ \l fl oo r 1 /\log ( 1+ 1/\m u) \rfl oor } $. As int heproof o f Lemma \ [mo n o\], $ F_ {\lambd a}(k+1) \leqF_ {\mu}(k)$if an d only i f $$\lab el{BBBBB} \int_{\mu} ^{ \ lambda} e^ {\mu- t}\B ig(\frac{ t}{ \mu}\B ig) ^ {k+1}dt \ge q 1.$ $Sob y thec h an ge

o\], we_have that_$F_{\lambda}(k+1) \leq F_{\mu}(k)$ if and_only if_$$\label{AAA} e^{-\lambda}_\lambda^{k+1} \leq_(k+1)_\int_{\mu} ^{\lambda} e^{-t}_t^k dt.$$ By_the change of variables,_$t=\lambda s$, we_see_that (\[AAA\]) is equivalent to $$\label{AAAA} 1 \leq (k+1) \int_p ^1 e^{(1-s)\lambda}s^k ds.$$ The right_side_of (\[AAAA\])_is_increasing_in $\lambda$. Since $\mu'/\lambda' =_p$ and $\lambda' > \lambda$,_we have_$F_{\lambda'}(k+1) \leq F_{\mu'}(k)$. Simple calculations show that $e^{-\lambda}\lambda^k_\leq_e^{-\mu}\mu^k$ if and_only if $$\label{AAAAA} \lambda \geq \frac{-k \log p}{1-p}.$$ The left_side of (\[AAAAA\]) is obviously increasing_in $\lambda$. Thus_we_have_that $e^{-\lambda'}(\lambda')^k \leq e^{-\mu'}(\mu')^k$. For $x_\in {{\mathbb R}}$, let ${ \lfloor_x \rfloor }$ denote its integer_part. First we show that $\lambda_c(\mu) \leq \mu_+1$. By Corollary \[auxresults\], it suffices to_show that if $\lambda =_\mu +1$,_then there is a thinning_from $\lambda$ to_$\mu$. By_Theorem \[main\] condition (\[prob\]),_we must show for some $k\in{{\mathbb_N}}$ that $F_{\lambda}(k+1)_\leq F_{\mu}(k) $ and $e^{-\lambda}\lambda^k \leq_e^{-\mu}\mu^k$._The latter condition_is_satisfied_by choosing_$k= { \lfloor__1/\log (_1_+ 1/\mu) \rfloor }$. As in_the_proof of Lemma \[mono\], $F_{\lambda}(k+1) \leq F_{\mu}(k)$ if and_only if $$\label{BBBBB} \int_{\mu} ^{\lambda}_e^{\mu-t}\Big(\frac{t}{\mu}\Big)^{k+1}_dt \geq 1.$$ So_by the change

\left( \sum_i \hat{s_i} \right).$$ This will follow if we can prove that, for positive real numbers $t_1, \dots, t_k$, $$\label{eqn:to_prove_about_phi_flipped} t_1\dots t_k \Phi(t^{-1}_1, \dots, t^{-1}_k) {\leqslant}\frac{\sigma_k}{k}(t_1 + \dots + t_k).$$ To prove (\[eqn:to\_prove\_about\_phi\_flipped\]), first observe that equality holds in the case that all of the $t_i$ are equal. Indeed, when $t_i = 1$ the relation follows from the definition of $\sigma$, and for other constant values of $t_i$ the equality follows by homogeneity. Set $\Theta(t_1, \dots, t_k) = t_1 \dots t_k \Phi(t^{-1}_1, \dots, t^{-1}_k)$. To prove that $\Theta(t_1, \dots, t_k)$ achieves its maximum value (with $t_1 + \dots + t_k$ fixed) when all of the $t_i$ are equal, observe that it will suffice to prove the following claim. \[clm:make\_si\_equal\] If $t_1 + t_2$ is fixed (as well as each of $t_3, \dots, t_k$), then $\Theta(t_1, \dots, t_k)$ achieves its maximum when $t_1 = t_2$. To see that this claim is sufficient, observe that we may repeatedly replace the largest and smallest of the $t_i$ with their average. In doing so, $\max t_i - \min t_i$ will tend to 0, and we can use the continuity of $\Theta$ to obtain the result. To prove Claim \[clm:make\_si\_equal\], recall the expression for $\Theta(t_1, \dots, t_k)$: $$\begin{aligned} \Theta(t_1, \dots, t_k) &= t_1t_2 (t_3\dots t_k) (I_{t^{-1}_1} \ast I_{t^{-1

\left (\sum_i \hat{s_i } \right).$$ This will follow if we can prove that, for positive real number $ t_1, \dots, t_k$, $ $ \label{eqn: to_prove_about_phi_flipped } t_1\dots t_k \Phi(t^{-1}_1, \dots, t^{-1}_k) { \leqslant}\frac{\sigma_k}{k}(t_1 + \dots + t_k).$$ To rise (\[eqn: to\_prove\_about\_phi\_flipped\ ]), first observe that equality holds in the lawsuit that all of the $ t_i$ are equal. Indeed, when $ t_i = 1 $ the relation surveil from the definition of $ \sigma$, and for other constant value of $ t_i$ the equality follows by homogeneity. Set $ \Theta(t_1, \dots, t_k) = t_1 \dots t_k \Phi(t^{-1}_1, \dots, t^{-1}_k)$. To raise that $ \Theta(t_1, \dots, t_k)$ achieves its maximum value (with $ t_1 + \dots + t_k$ specify) when all of the $ t_i$ are equal, observe that it will suffice to prove the following claim. \[clm: make\_si\_equal\ ] If $ t_1 + t_2 $ is fixed (equally well as each of $ t_3, \dots, t_k$), then $ \Theta(t_1, \dots, t_k)$ achieves its maximum when $ t_1 = t_2$. To see that this call is sufficient, observe that we may repeatedly replace the big and smallest of the $ t_i$ with their average. In make so, $ \max t_i - \min t_i$ will tend to 0, and we can use the continuity of $ \Theta$ to obtain the result. To prove Claim \[clm: make\_si\_equal\ ], echo the expression for $ \Theta(t_1, \dots, t_k)$: $ $ \begin{aligned } \Theta(t_1, \dots, t_k) & = t_1t_2 (t_3\dots t_k) (I_{t^{-1}_1 } \ast I_{t^{-1

\levt( \sum_i \hat{s_i} \right).$$ This will follow if we can prove that, fof positive real numbers $t_1, \dovs, t_j$, $$\labtj{eqn:to_prove_about_phk_flipped} t_1\fots t_k \Phi(t^{-1}_1, \eots, t^{-1}_k) {\leqslant}\frac{\sigmz_n}{k}(t_1 + \vots + t_k).$$ To provg (\[eqn:to\_prove\_dbout\_phi\_flippeg\]), wixst observe that equality holds in tre case tjat all of the $t_i$ swe eslao. Indeed, when $t_i = 1$ the relatikn follmws from the cefinition of $\sigma$, and fog otjer constant valued of $t_i$ the equwoity follows by homogeneity. Set $\Theja(t_1, \dots, t_k) = t_1 \dots t_k \Phi(t^{-1}_1, \dots, t^{-1}_k)$. Tp prove thqt $\Thfja(t_1, \dots, t_k)$ echievvs its maximum value (fith $t_1 + \dots + t_k$ fixec) wien qll of the $t_i$ are equel, observe that it wyll suffiwe to prove the folooqing wlaik. \[clm:oqke\_ri\_esuel\] Jf $t_1 + h_2$ ia fixed (as well as eaxh of $t_3, \dots, t_k$), them $\Eyeta(t_1, \dots, t_k)$ achiedef its maximum when $t_1 = t_2$. To see that this clzim is sufficient, obserce that we may repeatgdly replase the largest and smallest of the $t_i$ with their dveraje. In djkbg so, $\max t_i - \min t_i$ will tend to 0, and we can ufs uhe continuity of $\Theta$ to ontwim the result. To prove Clzim \[clm:make\_si\_equal\], recall the wxpressiog fot $\Theta(t_1, \dots, t_k)$: $$\begin{alignwd} \Theta(t_1, \dotf, t_k) &= t_1t_2 (t_3\dots t_k) (I_{c^{-1}_1} \ast I_{t^{-1

\left( \sum_i \hat{s_i} \right).$$ This will follow can that, for real numbers $t_1, \Phi(t^{-1}_1, t^{-1}_k) {\leqslant}\frac{\sigma_k}{k}(t_1 + + t_k).$$ To (\[eqn:to\_prove\_about\_phi\_flipped\]), first observe that equality holds the case that all of the $t_i$ are equal. Indeed, when $t_i = the relation follows from the definition of $\sigma$, and for other constant values $t_i$ equality by Set $\Theta(t_1, \dots, t_k) = t_1 \dots t_k \Phi(t^{-1}_1, \dots, t^{-1}_k)$. To prove that $\Theta(t_1, \dots, achieves its maximum value (with $t_1 + \dots t_k$ fixed) when all the $t_i$ are equal, observe it suffice to the claim. If $t_1 + is fixed (as well as each of $t_3, \dots, t_k$), then $\Theta(t_1, \dots, t_k)$ achieves its maximum $t_1 = see that claim sufficient, that we may the largest and smallest of the average. In doing so, $\max t_i - \min will tend 0, and we can use the of $\Theta$ to obtain the result. To prove \[clm:make\_si\_equal\], recall the expression for $\Theta(t_1, \dots, t_k)$: $$\begin{aligned} \Theta(t_1, \dots, t_k) &= t_1t_2 (t_3\dots \ast I_{t^{-1

\left( \sum_i \hat{s_i} \right).$$ This wilL follow if wE can pRovE thAt, For pOsitIve real numbers $T_1, \Dots, T_k$, $$\label{eqn:to_prove_about_Phi_flIpPEd} t_1\dOTs T_k \Phi(T^{-1}_1, \dots, t^{-1}_k) {\LEqSLAnt}\FrAc{\SigMa_K}{K}(t_1 + \Dots + t_K).$$ To Prove (\[eqN:to\_prove\_abOut\_PhI\_flipped\]), firsT ObServe that eQuaLity holds in tHe cAse thaT aLl oF The $t_i$ Are Equal. indeed, WHen $t_i = 1$ tHe relatioN fOLlows fROm the deFINiTion Of $\sigma$, and for otheR CoNStant values of $t_I$ the eqUaLItY FOllOws By homogeneItY. Set $\THEta(t_1, \dotS, T_k) = T_1 \DOTs t_K \phi(t^{-1}_1, \dots, t^{-1}_k)$. To pRove that $\TheTA(t_1, \dOts, t_k)$ aChIevES its maXimum VaLUe (wIth $t_1 + \dots + t_k$ fIxed) When all of The $t_i$ aRE equal, oBServe thAt it wiLl sUffIce tO PrOvE thE fOLloWInG clAIm. \[cLm:make\_si\_EqUaL\] If $t_1 + t_2$ Is fiXED (AS welL as Each Of $t_3, \doTs, t_k$), then $\Theta(T_1, \doTs, t_k)$ AChiEves iTs maxImum WhEn $t_1 = t_2$. TO see thAt thiS cLaim is sufficienT, obsErve that wE maY rEpeAtEdly rEPlace tHe lArgEst and sMallest OF thE $t_I$ WITh Their average. In doinG sO, $\MAx T_i - \min t_i$ wIll tenD To 0, AnD We can use ThE coNtinUITy of $\THeta$ TO oBtain the Result. tO pRoVe Claim \[ClM:make\_sI\_eQuaL\], reCall tHE expRessioN for $\ThetA(t_1, \dotS, T_k)$: $$\begin{aligned} \tHeta(t_1, \dots, t_k) &= t_1t_2 (T_3\DoTS T_k) (i_{T^{-1}_1} \ast i_{t^{-1

\left( \sum_i \hat{s_i} \ right).$$Thiswil l f ol lowif w e can prove th a t, f or positive real numbe rs $t _1 , \do t s, t_k$ , $$\la b el { e qn: to _p rov e_ a bo ut_ph i_f lipped} t_1\dotst_k \ Phi(t^{-1}_1 , \ dots, t^{- 1}_ k) {\leqslan t}\ frac{\ si gma _ k}{k} (t_ 1 + \ dots + t_k).$ $ To pro ve (\[eqn : to\_pro v e \_ abou t\_phi\_flipped\] ) ,f irst observe t hat eq ua l it y hol dsin the cas ethata ll of t h e$ t _ i$a re equal. Ind eed, when $ t _i= 1$ t he re l ationfollo ws fro m the defin itio n of $\si gma$,a nd foro ther co nstant va lue s of $t _i $ t he equ a li tyf oll ows by h om og eneit y. S e t $\Th eta (t_1 , \do ts, t_k) = t_ 1 \ dots t_k \Phi (t^{- 1}_1 ,\dots , t^{- 1}_k) $. To prove that$\Th eta(t_1,\do ts , t _k )$ ac h ievesits ma ximum v alue (w i th$t _ 1 +\dots + t_k$ fixed )w h en all ofthe $t _ i$ a r e equal, o bse rvet h at it wil l s uffice t o prov e t he follow in g clai m. \ [cl m:mak e \_si \_equa l\] If $ t_1 + t_2$ is fixed( as well as ea c ho f $ t _3,\do ts, t_k$),then $\Th eta( t _1 , \ d ots,t_k)$ a c hi e ves its maximum whe n$t_1 = t_2$ . To see tha t this cla i m is suffi cien t ,o bserve that we mayrepeatedly replacethe l argest a nd smalle s t of the$t_ i$wit h t h e ir average. Ind o ingso , $\max t_ i - \mi n t _i$ wi llte nd to 0,and we c an u se t heconti n uity of$\ The ta $ t o obt a in the resu lt. T op rov e Claim \[ c l m:ma ke \_ si\_ equ al \], r ecal l th e expre ssion for $\ T heta (t _1 , \dots , t_k)$: $$\b eg in{aligned }\Th eta(t_ 1 , \dots,t_k) &= t_1t_2 (t_3\dot s t_k) ( I_{ t^{-1 }_1} \ast I_{ t^{ -1

\left(_\sum_i \hat{s_i}_\right).$$ This will follow_if we_can_prove that,_for_positive real numbers_$t_1, \dots, t_k$,_$$\label{eqn:to_prove_about_phi_flipped} t_1\dots t_k \Phi(t^{-1}_1, \dots,_t^{-1}_k) {\leqslant}\frac{\sigma_k}{k}(t_1 +_\dots_+ t_k).$$ To prove (\[eqn:to\_prove\_about\_phi\_flipped\]), first observe that equality holds in the case that all_of_the $t_i$_are_equal._Indeed, when $t_i = 1$_the relation follows from the_definition of_$\sigma$, and for other constant values of $t_i$_the_equality follows by_homogeneity. Set $\Theta(t_1, \dots, t_k) = t_1 \dots t_k \Phi(t^{-1}_1,_\dots, t^{-1}_k)$. To prove that $\Theta(t_1,_\dots, t_k)$ achieves_its_maximum_value (with $t_1 +_\dots + t_k$ fixed) when all_of the $t_i$ are equal, observe_that it will suffice to prove the_following claim. \[clm:make\_si\_equal\] If $t_1 + t_2$_is fixed (as well as_each of_$t_3, \dots, t_k$), then $\Theta(t_1,_\dots, t_k)$ achieves_its maximum_when $t_1 =_t_2$. To see that this claim is_sufficient, observe that_we may repeatedly replace the largest_and_smallest of the_$t_i$_with_their average._In doing so,_$\max_t_i -_\min_t_i$ will tend to 0, and_we_can use the continuity of $\Theta$ to_obtain the result. To prove_Claim_\[clm:make\_si\_equal\], recall the expression_for $\Theta(t_1, \dots, t_k)$: $$\begin{aligned} \Theta(t_1,_\dots, t_k) &= t_1t_2 (t_3\dots t_k)_(I_{t^{-1}_1} \ast_I_{t^{-1

vertices. Pairwise beta model ------------------- The tilted Dirichlet model has been successfully used for applications [e.g. @coles1991], although it suffers from a lack of interpretability of the parameters. @cooley2010 proposed a similar model but with easily interpretable parameters. The definition of their model is based on a geometric approach. Specifically, they considered the symmetric pairwise beta function $$\begin{aligned} h^*(w_i,w_j) = \frac{\Gamma( 2 \beta_{i,j} )}{\Gamma^2( \beta_{i,j} )} \bigg( \frac{w_i}{w_i + w_j} \bigg)^{\beta_{i,j}-1} \bigg( \frac{w_j}{w_i + w_j} \bigg)^{\beta_{i,j}-1},i,j\in I,\end{aligned}$$ where $w_i$ and $w_j$ are two elements of ${{\boldsymbol w}}$ and $\beta_{i,j}>0$. This function has its center at the point $(1/d,\ldots,1/d)$ and it verifies the first moment conditions. Then, the angular pairwise beta density is defined by summing together all the $d(d-1)/2$ possible pairs of variables, namely $$\begin{aligned} h ( {{\boldsymbol w}};{{\boldsymbol \theta}}) = \frac{2 ( d-3 )!\Gamma ( \alpha d +1 )}{d (d-1)\Gamma ( 2\alpha +1 )\Gamma \{ \alpha (d-2) \}} \sum_{i,j\in I,i<j} h(w_i,w_j), \quad {{\boldsymbol w}}\in \sW,\end{aligned}$$ where $$h(w_i,w_j)=(w_i+w_j)^{2\alpha-1}\{1-(w_i+w_j)\}^{\alpha(d-2)-d+2}\,h^*( w_i,w_j)$$ and ${{\boldsymbol \theta}}=(\alpha,\{\beta_{i,j}\}_{i,j\in I})$ with $\alpha>0$. Each parameter $\beta_{i,j}$ controls the level of dependence between the $i^{th}$ and the $j^{th}$ components and the dependence increases for increasing values

vertices. Pairwise beta model ------------------- The tilted Dirichlet model has been successfully used for application [ for example @coles1991 ], although it suffers from a lack of interpretability of the parameters. @cooley2010 aim a like model but with easily explainable parameters. The definition of their model is based on a geometric access. Specifically, they considered the symmetric pairwise beta routine $ $ \begin{aligned } h^*(w_i, w_j) = \frac{\Gamma (2 \beta_{i, j }) } { \Gamma^2 (\beta_{i, j }) } \bigg (\frac{w_i}{w_i + w_j } \bigg)^{\beta_{i, j}-1 } \bigg (\frac{w_j}{w_i + w_j } \bigg)^{\beta_{i, j}-1},i, j\in I,\end{aligned}$$ where $ w_i$ and $ w_j$ are two elements of $ { { \boldsymbol w}}$ and $ \beta_{i, j}>0$. This routine experience its center at the point $ (1 / d,\ldots,1 / d)$ and it verifies the inaugural moment conditions. Then, the angular pairwise beta density is define by summing together all the $ d(d-1)/2 $ possible pairs of variable, namely $ $ \begin{aligned } h ({ { \boldsymbol w}};{{\boldsymbol \theta } }) = \frac{2 (d-3)! \Gamma (\alpha d +1) } { d (d-1)\Gamma (2\alpha +1) \Gamma \ { \alpha (d-2) \ } } \sum_{i, j\in I, i < j } h(w_i, w_j), \quad { { \boldsymbol w}}\in \sW,\end{aligned}$$ where $ $ h(w_i, w_j)=(w_i+w_j)^{2\alpha-1}\{1-(w_i+w_j)\}^{\alpha(d-2)-d+2}\,h^ * (w_i, w_j)$$ and $ { { \boldsymbol \theta}}=(\alpha,\{\beta_{i, j}\}_{i, j\in I})$ with $ \alpha>0$. Each parameter $ \beta_{i, j}$ controls the level of dependence between the $ i^{th}$ and the $ j^{th}$ component and the dependence increases for increasing values

vegtices. Pairwise beta modeu ------------------- The tilted Dirnxhlet kodel gas been successfully used for appliratiins [e.t. @coles1991], although it suwfers frol a lack of mnterpretability of the icramefcrs. @cmiley2010 proposed s similar kodel but with exsnly interpretable parameters. The defynition ov their model ys bsfed kn a geometric approach. Specificalmy, they considered tne symmetric pairwise beta funftion $$\begin{aligned} j^*(w_i,w_j) = \frac{\Tammw( 2 \beta_{i,j} )}{\Gamoa^2( \beta_{i,j} )} \bigg( \frac{w_i}{w_j + w_j} \bigg)^{\beta_{i,j}-1} \bigg( \frac{w_j}{w_i + d_j} \biyg)^{\beta_{i,j}-1},i,j\ib U,\enf{dligned}$$ whece $w_i$ wnd $w_j$ are two elemetts of ${{\noldsymbol w}}$ akd $\beva_{i,j}>0$. This function has itv center at the poynt $(1/d,\ldotv,1/d)$ and it verifies rhw firvt mmmeng covdiuiois. Fhen, tje engular paidwise beta eensity is defined ny wumming togetger alj ehe $d(d-1)/2$ possible pairs of variables, nameny $$\gegin{aligned} h ( {{\boldsymbil w}};{{\boldsymbol \theta}}) = \frac{2 ( d-3 )!\Damma ( \alpha d +1 )}{d (d-1)\Gamma ( 2\alpha +1 )\Gamma \{ \alpha (d-2) \}} \sgm_{i,j\ii K,i<j} h(r_k,q_j), \quad {{\boldsymbol w}}\in \sW,\end{aligned}$$ where $$h(w_i,w_j)=(r_j+w_k)^{2\akpha-1}\{1-(w_i+w_j)\}^{\alpha(d-2)-b+2}\,h^*( w_i,w_j)$$ and ${{\boldxylbpj \theta}}=(\alpha,\{\bgta_{i,j}\}_{i,j\nh J})$ with $\alpha>0$. Each oarametgr $\betq_{i,j}$ contrjls yhe level of dependence betqeen the $i^{th}$ qnd the $j^{th}$ componznts and the depgndencr increases for increasnng vamues

vertices. Pairwise beta model ------------------- The tilted has successfully used applications [e.g. @coles1991], lack interpretability of the @cooley2010 proposed a model but with easily interpretable parameters. definition of their model is based on a geometric approach. Specifically, they considered symmetric pairwise beta function $$\begin{aligned} h^*(w_i,w_j) = \frac{\Gamma( 2 \beta_{i,j} )}{\Gamma^2( \beta_{i,j} )} \frac{w_i}{w_i w_j} \bigg( + w_j} \bigg)^{\beta_{i,j}-1},i,j\in I,\end{aligned}$$ where $w_i$ and $w_j$ are two elements of ${{\boldsymbol w}}$ and $\beta_{i,j}>0$. function has its center at the point $(1/d,\ldots,1/d)$ it verifies the first conditions. Then, the angular pairwise density defined by together the possible pairs of namely $$\begin{aligned} h ( {{\boldsymbol w}};{{\boldsymbol \theta}}) = \frac{2 ( d-3 )!\Gamma ( \alpha d +1 )}{d ( 2\alpha \{ \alpha \}} I,i<j} \quad {{\boldsymbol w}}\in $$h(w_i,w_j)=(w_i+w_j)^{2\alpha-1}\{1-(w_i+w_j)\}^{\alpha(d-2)-d+2}\,h^*( w_i,w_j)$$ and ${{\boldsymbol \theta}}=(\alpha,\{\beta_{i,j}\}_{i,j\in I})$ parameter $\beta_{i,j}$ controls the level of dependence between $i^{th}$ and $j^{th}$ components and the dependence increases increasing values

vertices. Pairwise beta model ------------------- the tilted DIrichLet ModEl Has bEen sUccessfully useD For aPplications [e.g. @coles1991], althOugh iT sUFferS FrOm a laCk of intERpRETabIlItY of ThE PaRametErs. @Cooley2010 pRoposed a siMilAr Model but with EAsIly interprEtaBle parameterS. ThE definItIon OF theiR moDel is Based oN A geomeTric approAcH. specifICally, thEY CoNsidEred the symmetric pAIrWIse beta functioN $$\begin{AlIGnED} H^*(w_i,W_j) = \fRac{\Gamma( 2 \beTa_{I,j} )}{\GamMA^2( \beta_{i,j} )} \BIgG( \FRAc{w_I}{W_i + w_j} \bigg)^{\beta_{i,J}-1} \bigg( \frac{w_j}{W_I + w_j} \Bigg)^{\beTa_{I,j}-1},i,J\In I,\end{AlignEd}$$ WHerE $w_i$ and $w_j$ are Two eLements of ${{\BoldsyMBol w}}$ and $\BEta_{i,j}>0$. ThIs funcTioN haS its CEnTeR at ThE PoiNT $(1/d,\LdoTS,1/d)$ aNd it veriFiEs The fiRst mOMENT conDitIons. then, tHe angular pairWisE betA DenSity iS defiNed bY sUmminG togetHer alL tHe $d(d-1)/2$ possible paiRs of Variables, NamElY $$\beGiN{aligNEd} h ( {{\bolDsyMboL w}};{{\boldsYmbol \thETa}}) = \fRaC{2 ( D-3 )!\gAmMa ( \alpha d +1 )}{d (d-1)\Gamma ( 2\alpHa +1 )\gAMmA \{ \alpha (d-2) \}} \sUm_{i,j\in i,I<j} H(w_I,W_j), \quad {{\boLdSymBol w}}\IN \SW,\end{AligNEd}$$ Where $$h(w_i,W_j)=(w_i+w_j)^{2\ALpHa-1}\{1-(W_i+w_j)\}^{\alpHa(D-2)-d+2}\,h^*( w_i,w_J)$$ aNd ${{\bOldSymboL \ThetA}}=(\alpha,\{\Beta_{i,j}\}_{i,j\In I})$ wiTH $\alpha>0$. Each paraMEter $\beta_{i,j}$ conTRoLS ThE LeveL of Dependence bEtweEN the $I^{th}$ aND tHe $j^{TH}$ compOnentS aND tHE dependence increaseS fOr incrEasinG values

vertices. Pairwise betamodel ---- ----- --- --- -- -- Thetilted Dirichl e t mo del has been successfu lly u se d for ap plica tions [ e .g . @co le s1 991 ], al thoug h i t suffe rs from alac kof interpret a bi lity of th e p arameters. @ coo ley201 0pro p oseda s imila r mode l but w ith easil yi nterpr e table p a r am eter s. The definition of their model is based o n a g eom etr ic approac h. Spec i fically , t h e y co n sidered the s ymmetric pa i rwi se bet afun c tion $ $\beg in { ali gned} h^*(w _i,w _j) = \fr ac{\Ga m ma( 2 \ b eta_{i, j} )}{ \Ga mma ^2(\ be ta _{i ,j } )} \b igg ( \f rac{w_i} {w _i + w_ j} \ b i g g )^{\ bet a_{i ,j}-1 } \bigg( \fra c{w _j}{ w _i+ w_j } \bi gg)^ {\ beta_ {i,j}- 1},i, j\ in I,\end{align ed}$ $ where $ w_i $and $ w_j$a re two el eme nts of${{\bol d sym bo l w }} $ and $\beta_{i,j} >0 $ . T his func tion h a sit s centerat th e po i n t $(1 /d,\ l do ts,1/d)$ and i t v er ifies t he first m ome ntcondi t ions . Then , the an gular pairwise betad ensity is def i ne d by summ ing together a ll t h e $d (d-1 ) /2 $ p o ssibl e pai rs of variables, namely $ $\ begin{ align ed} h ( {{\bo ldsymbol w } } ; {{\bolds ymbo l \ t heta}}) = \fra c{2 ( d-3 )!\Ga m ma ( \al pha d +1 )}{d (d-1)\Ga m m a ( 2\al pha +1 )\ Gam m a \ { \alpha (d-2 ) \}}\s um_{i,j \in I,i<j} h (w_ i,w _j) ,\quad {{\ boldsymb ol w }} \i n \ sW,\e n d{aligne d} $$wh ere $$h( w _i,w_j )=(w_ i+w_ j) ^{ 2 \al pha-1}\ { 1- ( w _i+w _j )\ }^{\ alp ha (d-2) -d+2 } \,h ^*( w_i ,w_j)$$ a nd$ {{\b ol ds ymbol \ theta}}=(\alp ha ,\{\beta_{ i, j}\ }_{i,j \ i n I})$ w ith $\alpha>0$. Each pa r ameter$\b eta_{ i,j} $ control s t he lev elo f depe ndence betw ee n t h e $i^{ t h }$ an dthe $j^{th } $ co mpone nt s an d the d ependence increase s fo r increasingval ues

vertices. Pairwise_beta model ------------------- The_tilted Dirichlet model has_been successfully_used_for applications_[e.g._@coles1991], although it_suffers from a_lack of interpretability of_the parameters. @cooley2010_proposed_a similar model but with easily interpretable parameters. The definition of their model is_based_on a_geometric_approach._Specifically, they considered the symmetric_pairwise beta function $$\begin{aligned} h^*(w_i,w_j) =_\frac{\Gamma( 2_\beta_{i,j} )}{\Gamma^2( \beta_{i,j} )} \bigg( \frac{w_i}{w_i + w_j} \bigg)^{\beta_{i,j}-1} \bigg(_\frac{w_j}{w_i_+ w_j} \bigg)^{\beta_{i,j}-1},i,j\in_I,\end{aligned}$$ where $w_i$ and $w_j$ are two elements of_${{\boldsymbol w}}$ and $\beta_{i,j}>0$. This function_has its center_at_the_point $(1/d,\ldots,1/d)$ and it_verifies the first moment conditions. Then,_the angular pairwise beta density is_defined by summing together all the $d(d-1)/2$_possible pairs of variables, namely $$\begin{aligned} h_( {{\boldsymbol w}};{{\boldsymbol \theta}}) =_\frac{2 (_d-3 )!\Gamma ( \alpha d_+1 )}{d (d-1)\Gamma_( 2\alpha_+1 )\Gamma \{_\alpha (d-2) \}} \sum_{i,j\in I,i<j} h(w_i,w_j), \quad_{{\boldsymbol w}}\in \sW,\end{aligned}$$_where $$h(w_i,w_j)=(w_i+w_j)^{2\alpha-1}\{1-(w_i+w_j)\}^{\alpha(d-2)-d+2}\,h^*( w_i,w_j)$$ and ${{\boldsymbol \theta}}=(\alpha,\{\beta_{i,j}\}_{i,j\in_I})$_with $\alpha>0$. Each_parameter_$\beta_{i,j}$_controls the_level of dependence_between_the $i^{th}$_and_the $j^{th}$ components and the dependence_increases_for increasing values

9 0.0316 9 0.0100 24 : Ratio of the wind X-ray luminosities as a function of wind momentum ratio, $\eta$, for colliding wind systems with equal, spatially invariant wind speeds, equal abundances, and near to the adiabatic limit.[]{data-label="tab:ad_lim"} $L_{1}/L_{2}$ in-between these limits {#sec:inbetween} ------------------------------------- Estimates of the luminosity ratio for systems where the shocked winds are in-between the limiting radiative and adiabatic cases must be done on a case-by-case basis. Somewhere in this region of parameter space, the wind which dominates the X-ray emission must switch over from the one with the faster pre-shock speed (radiative limit) to the one with the slower pre-shock speed (adiabatic limit). In contrast, situations where one wind is clearly radiative and the other is closer to being adiabatic will have their X-ray emission dominated by the former (in $\gamma^{2}$ Velorum, where $\chi_{1} \ll 1$ and $\chi_{2} > 1$, the X-ray emission is dominated by the shocked WR wind). Summary ======= In this letter we have re-examined the issue of which wind is the dominant X-ray emitter in colliding wind binaries, following the discovery of some confusion in the existing literature. Our work supports the earlier conclusions in Luo ([@LMM1990]) and Myasnikov & Zhekov ([@MZ1993]), though is sometimes in disagreement with the relevant analytical equations in Usov ([@U1992]). For systems near the radiative limit (typically short period binaries), we find that the primary influence on $L_{1}/L_{2}$ is the ratio of the wind speeds, $v_{1}/v_{2}$, since it controls the energy flux ratio. For $v_{1} = v_{2}$, $L_{1} \approx L_{2}$ irrespective of the wind momentum ratio, $\eta$. When $v_{1} \neq v_{2}$, the faster wind normally dominates the luminosity, although it is unlikely to do so by more than a factor of 5. The equations in Usov ([@U1992]) predict values for $L_{1}/L_{2}$ which can be in error by up to a factor of 4. For systems near the adiabatic limit (long-

9 0.0316 9 0.0100 24 : Ratio of the wind X - ray luminosities as a affair of fart momentum ratio, $ \eta$, for colliding wind system with equal, spatially invariant tip speeds, equal abundance, and near to the adiabatic limit.[]{data - label="tab: ad_lim " } $ L_{1}/L_{2}$ in - between these limits { # sec: inbetween } ------------------------------------- Estimates of the luminosity proportion for systems where the shocked winds are in - between the limiting radiative and adiabatic case must be done on a subject - by - case basis. Somewhere in this area of parameter space, the wind which dominate the X - ray discharge must switch over from the one with the quicker pre - shock speed (radiative terminus ad quem) to the one with the slower pre - shock speed (adiabatic limit). In contrast, situation where one wind is clearly radiative and the other is closer to being adiabatic will have their ten - ray emission dominated by the former (in $ \gamma^{2}$   Velorum, where $ \chi_{1 } \ll 1 $ and $ \chi_{2 } > 1 $, the X - ray emission is dominated by the shocked WR wind). Summary = = = = = = = In this letter we have re - analyze the issue of which wind is the prevailing ten - re emitter in colliding wind binaries, following the discovery of some confusion in the existing literature. Our work supports the early conclusions in Luo ([ @LMM1990 ]) and Myasnikov & Zhekov ([ @MZ1993 ]), though is sometimes in disagreement with the relevant analytical equations in Usov ([ @U1992 ]). For systems near the radiative terminus ad quem (typically short period binaries), we find that the primary influence on $ L_{1}/L_{2}$ is the ratio of the wind amphetamine, $ v_{1}/v_{2}$, since it controls the energy flux ratio. For $ v_{1 } = v_{2}$, $ L_{1 } \approx L_{2}$ irrespective of the fart momentum ratio, $ \eta$. When $ v_{1 } \neq v_{2}$, the faster wind normally dominate the luminosity, although it is unlikely to do so by more than a factor of 5. The equality in Usov ([ @U1992 ]) predict values for $ L_{1}/L_{2}$ which can be in mistake by up to a factor of 4. For organization near the adiabatic limit (long-

9 0.0316 9 0.0100 24 : Rxtio of the winb X-ray numinoaities ar a function of wind momentul eatio, $\eta$, for colliding wina systems with eqyal, wpatially iibariant wind aieeds, xqual abundancex, and near to the adiabadiz pimit.[]{data-label="tab:ad_lim"} $L_{1}/L_{2}$ in-between ehese lomlts {#sec:inbetwegn} ------------------------------------- Estpmwtes of the luminosity ratio for systejs whert the shocked windx are in-between the limitijg rwdiative and adiabwtic cases nust ve done on a case-by-cast yasis. Somewgere in this region of parametef spaee, the wind wyicj dominates vhe X-rwy emission must swidch ovet from the one wivh tye faster pre-shock spxed (radiative limit) jo the one wnth the slower pre-shoxk speeg (adhabaguc uimjt). Ih contgasv, situationa where one wind is clearly raciwnove and the kther ys closer to being adiabatic will have thtir X-day emission dominated vy the former (in $\gammw^{2}$ Velorum, rhere $\chi_{1} \ll 1$ and $\chi_{2} > 1$, the X-ray emission is domitated cy uhc ryofked WR wind). Summary ======= In this letter we have re-evzmonvd the issue of wmich wind is the dpmlnsgt X-ray emittgr in colljding wind binaried, folloring rhe discodery of some confusion in the ezisting litegatuee. Our work supporcs the earlizr convlusipns in Luo ([@LMM1990]) and Myasuikov & Zhekov ([@MZ1993]), hhough is rometimes in disxgrvemett with ufe relevant analyeical equetionx in Usuv ([@U1992]). Gor syftems near the vddiative limit (typlcalli shord period blnaries), we find that the primarb influence om $N_{1}/L_{2}$ ps the racio of the wind speqds, $v_{1}/v_{2}$, since ij controlf the energy fllx ratio. Hor $v_{1} = v_{2}$, $L_{1} \a[prox L_{2}$ irres[gctive of the wind moientym rqtio, $\etx$. When $v_{1} \neq v_{2}$, yhe fasteg cind normqlly dominates the luojnosity, althougk ut is unlikely yo ao fo bb morq than a factmr ow 5. Gne eqjations in Msox ([@U1992]) lredict values for $L_{1}/N_{2}$ whjch can be in errot ny up to q factor of 4. For systeks near the adiabauic likit (long-

9 0.0316 9 0.0100 24 : Ratio wind luminosities as function of wind wind with equal, spatially wind speeds, equal and near to the adiabatic limit.[]{data-label="tab:ad_lim"} in-between these limits {#sec:inbetween} ------------------------------------- Estimates of the luminosity ratio for systems where shocked winds are in-between the limiting radiative and adiabatic cases must be done a basis. in region of parameter space, the wind which dominates the X-ray emission must switch over from the with the faster pre-shock speed (radiative limit) to one with the slower speed (adiabatic limit). In contrast, where wind is radiative the is closer to adiabatic will have their X-ray emission dominated by the former (in $\gamma^{2}$ Velorum, where $\chi_{1} \ll 1$ $\chi_{2} > X-ray emission dominated the WR wind). Summary this letter we have re-examined the wind is the dominant X-ray emitter in colliding binaries, following discovery of some confusion in the literature. Our work supports the earlier conclusions in ([@LMM1990]) and Myasnikov & Zhekov ([@MZ1993]), though is sometimes in disagreement with the relevant analytical Usov ([@U1992]). For systems the radiative limit short binaries), find the primary on $L_{1}/L_{2}$ is the ratio of the wind speeds, $v_{1}/v_{2}$, since controls the energy flux ratio. For $v_{1} = v_{2}$, $L_{1} irrespective the wind momentum $\eta$. When $v_{1} \neq the wind normally dominates the it unlikely by than factor of 5. The in Usov ([@U1992]) predict values $L_{1}/L_{2}$ which can be a factor of 4. For systems near the limit (long-

9 0.0316 9 0.0100 24 : Ratio of the wind X-ray luminosIties as a fuNctioN of WinD mOmenTum rAtio, $\eta$, for collIDing Wind systems with equal, spAtialLy INvarIAnT wind Speeds, eQUaL ABunDaNcEs, aNd NEaR to thE adIabatic Limit.[]{data-lAbeL="tAb:ad_lim"} $L_{1}/L_{2}$ in-bETwEen these liMitS {#sec:inbetweeN} ------------------------------------- EsTimateS oF thE LuminOsiTy ratIo for sYStems wHere the shOcKEd windS Are in-beTWEeN the Limiting radiative ANd ADiabatic cases mUst be dOnE On A CAse-By-cAse basis. SoMeWhere IN this reGIoN OF ParAMeter space, the Wind which doMInaTes the x-rAy eMIssion Must sWiTCh oVer from the oNe wiTh the fastEr pre-sHOck speeD (RadiatiVe limiT) to The One wITh ThE slOwER prE-ShOck SPeeD (adiabatIc LiMit). In ContRAST, SituAtiOns wHere oNe wind is clearLy rAdiaTIve And thE otheR is cLoSer to Being aDiabaTiC will have their X-Ray eMission doMinAtEd bY tHe forMEr (in $\gaMma^{2}$ velOrum, wheRe $\chi_{1} \ll 1$ ANd $\cHi_{2} > 1$, THE x-rAy emission is dominaTeD BY tHe shockeD WR winD). suMmARy ======= In this LeTteR we hAVE re-exAminED tHe issue oF which WInD iS the domInAnt X-raY eMitTer In colLIdinG wind bInaries, fOllowINg the discovery OF some confusioN In THE eXIstiNg lIterature. OuR worK SuppOrts THe EarLIer coNclusIoNS iN luo ([@LMM1990]) and Myasnikov & ZHeKov ([@MZ1993]), tHough Is sometimes in DisagreemeNT WIth the reLevaNT aNAlytical equatiOns in usov ([@U1992]). For sySTems near The raDiative lImit (typicALLy short pEriOd bInaRieS), WE fInd that the priMARy inFlUence on $l_{1}/L_{2}$ iS the ratIo oF thE wiNd sPeEds, $v_{1}/v_{2}$, sincE it contrOlS tHe EnErgY flux RAtio. For $v_{1} = V_{2}$, $L_{1} \AppRoX L_{2}$ iRrespECtive oF the wInd mOmEnTUm rAtio, $\eta$. wHeN $V_{1} \Neq v_{2}$, ThE fAsteR wiNd NormaLly dOMinAtes the LuminositY, alTHougH iT iS unlikeLy to do so by morE tHan a factor Of 5. the EquatiONS in Usov ([@U1992]) Predict values for $L_{1}/L_{2}$ which CAn be in eRroR by up To a fActor of 4. FoR syStems nEar THe adiaBatic lImit (lOnG-

9 0.0316 9 0 .0100 24 :Ra tioof t he wind X-rayl umin osities as a functionof wi nd mome n tu m rat io, $\e t a$ , for c ol lid in g w ind s yst ems wit h equal, s pat ia lly invarian t w ind speeds , e qual abundan ces , andne art o the ad iabat ic lim i t.[]{d ata-label =" t ab:ad_ l im"} $ L _ {1 }/L_ {2}$ in-between t h es e limits {#sec: inbetw ee n }- - --- --- ---------- -- ----- - ------- - -- - Est i mates of theluminosityr ati o forsy ste m s wher e the s h ock ed winds ar e in -betweenthe li m iting r a diative and a dia bat ic c a se smus tb e d o ne on a c ase-by-c as ebasis . So m e w h ereinthis regi on of paramet erspac e , t he wi nd wh ichdo minat es the X-ra yemission must s witc h over fr omth e o ne with the fa ste r p re-shoc k speed (ra di a t i ve limit) to the one w i t hthe slow er pre - sh oc k speed ( ad iab atic l imit) . In co ntrast,situat i on swhere o ne windis cl ear ly ra d iati ve and the oth er is closer to bein g adiabatic wi l lh a ve thei r X -ray emissi on d o mina tedb ythe forme r (in $ \ ga m ma^{2}$ Velorum, wh er e $\ch i_{1} \ll 1$ and $ \chi_{2} > 1 $ , the X- raye mi s sion is domina ted b y the shoc k ed WR wi nd). Summary =======I n this l ett erwehav e re -examined the i ssue o f which wi nd is t hedom ina ntX- ray emitt er in co ll id in gwin d bin a ries, fo ll owi ng th e dis c overyof so me c on fu s ion in the ex i s ting l it erat ure .Our w orks upp orts th e earlier co n clus io ns in Luo ([@LMM1990]) a nd Myasnik ov &Zhekov ( [@MZ1993 ]), though is sometimes in disa gre ement wit h the rel eva nt ana lyt i cal eq uation s inUs ov( [ @U199 2 ] ). F or systems n e a r t he ra di ativ e limit (typically shortp eri od binaries), we fin d th att he pri ma r y i n f luence on $L_{1 }/L_{2}$ i st he ratio oft hewi nd spee ds, $v_ {1}/v _ {2}$, s ince it c ontrols t he ene r g y f lux ratio. For $v_ {1} = v_{ 2 }$, $ L _{ 1} \a ppr ox L_{ 2} $ i rresp ective ofthe w ind mo me ntum r atio, $ \eta$. W hen $v_{1} \neq v_{2}$, the f aster wi nd normal lyd omi nates the lum inosity, a lth oug h itisu nlike ly t o d o s o by m oret han a fac t or of 5 .The equatio n s inUsov([@ U 1992]) pre dict values for $ L _{1}/L_{2}$ wh ichc a n b e i n err or by up to a fa cto ro f 4. For s ystems near the adi ab a tic l imit ( long-

9 _ 0.0316_ _ __ 9 __ 0.0100 _ _ 24 _ : Ratio_of_the wind X-ray luminosities as a function of wind momentum ratio, $\eta$, for colliding_wind_systems with_equal,_spatially_invariant wind speeds, equal abundances,_and near to the adiabatic_limit.[]{data-label="tab:ad_lim"} $L_{1}/L_{2}$ in-between_these limits {#sec:inbetween} ------------------------------------- Estimates of the luminosity ratio for_systems_where the shocked_winds are in-between the limiting radiative and adiabatic cases_must be done on a case-by-case_basis. Somewhere in_this_region_of parameter space, the_wind which dominates the X-ray emission_must switch over from the one_with the faster pre-shock speed (radiative limit)_to the one with the slower_pre-shock speed (adiabatic limit). In_contrast, situations_where one wind is clearly_radiative and the_other is_closer to being_adiabatic will have their X-ray emission_dominated by the_former (in $\gamma^{2}$ Velorum, where $\chi_{1} \ll_1$_and $\chi_{2} >_1$,_the_X-ray emission_is dominated by_the_shocked WR_wind). Summary ======= In_this letter we have re-examined the_issue_of which wind is the dominant X-ray_emitter in colliding wind_binaries,_following the discovery of_some confusion in the existing_literature. Our work supports the earlier_conclusions in_Luo ([@LMM1990])_and Myasnikov & Zhekov ([@MZ1993]), though is sometimes in disagreement with_the relevant analytical equations in Usov_([@U1992]). For systems near the_radiative limit_(typically_short period binaries),_we_find that_the primary influence on $L_{1}/L_{2}$ is the_ratio of_the wind speeds, $v_{1}/v_{2}$, since it_controls the energy flux_ratio._For $v_{1} = v_{2}$, $L_{1} \approx_L_{2}$ irrespective of the wind momentum_ratio, $\eta$. When $v_{1} \neq_v_{2}$,_the_faster wind normally dominates the_luminosity, although it is unlikely to_do so by_more than a factor of 5. The_equations_in Usov ([@U1992]) predict values for_$L_{1}/L_{2}$_which can be in error by_up_to_a factor of 4. For systems_near the adiabatic limit (long-

j,n})$ realizes the $A$-invariant type $p_1 \otimes \cdots \otimes p_n$ over $A'q_0 \subseteq {\operatorname{dcl}}(q_0,a_{1,1},\ldots,a_{j-1,n}A)$. Applying Lemma \[shuffle-chaos\], we find a small model $M \supseteq A'$ and a non-constant global $M$-invariant type $r$ such that $$(a_{j,1},\ldots,a_{j,n},q_0) \models p_1 \otimes \cdots \otimes p_n \otimes r|M.$$ The type $r$ extends ${\operatorname{tp}}(q_0/M)$ and therefore lives in the $M$-definable set $Q$. By Lemma \[coheir-magic-1\], it follows that $${\operatorname{tp}}(a_{j,1},\ldots,a_{j,n},q_0/M)$$ is broad, as a type in $X_1 \times \cdots \times X_n \times Q$. This contradicts Lemma \[expander\]: the $X_i, P, Q$ are all $M$-definable, the type of $(a_{j,1},\ldots,a_{j,n},q_0)$ over $M$ is broad, the tuple $(a_{j,1},\ldots,a_{j,n})$ lives on $P$, and for $$s = a_{j,1} + \cdots + a_{j,n} + q_0 = s_j + q_0,$$ we have $$q_0 \in {\operatorname{acl}}(A',s_j + q_0) \subseteq {\operatorname{acl}}(sM). \qedhere$$ \[stacking\] Let $Y$ be a critical set and $Q_1, \ldots, Q_n$ be quasi-minimal. Then for every $m$ there exist $\{q_{i,j}\}_{i \in [n], j \in [m]}$ such that - For fixed $i \in [n]$, the sequence $q_{i,1},\ldots,q_{i,m}$ consists of $m$ distinct elements of $Q_i$. - The intersection $$\bigcap_{\eta : [n]

j, n})$ realizes the $ A$ -invariant type $ p_1 \otimes \cdots \otimes p_n$ over $ A'q_0 \subseteq { \operatorname{dcl}}(q_0,a_{1,1},\ldots, a_{j-1,n}A)$. Applying Lemma   \[shuffle - chaos\ ], we find a humble exemplar $ M \supseteq A'$ and a non - constant global $ M$-invariant character $ r$ such that $ $ (a_{j,1},\ldots, a_{j, n},q_0) \models p_1 \otimes \cdots \otimes p_n \otimes r|M.$$ The type $ r$ extends $ { \operatorname{tp}}(q_0 / M)$ and therefore be in the $ M$-definable set $ Q$. By Lemma   \[coheir - magic-1\ ], it follows that $ $ { \operatorname{tp}}(a_{j,1},\ldots, a_{j, n},q_0 / M)$$ is wide, as a type in $ X_1 \times \cdots \times X_n \times Q$. This contradicts Lemma   \[expander\ ]: the $ X_i, phosphorus, Q$ are all $ M$-definable, the type of $ (a_{j,1},\ldots, a_{j, n},q_0)$ over $ M$ is broad, the tuple $ (a_{j,1},\ldots, a_{j, n})$ lives on $ P$, and for $ $ s = a_{j,1 } + \cdots + a_{j, n } + q_0 = s_j + q_0,$$ we accept $ $ q_0 \in { \operatorname{acl}}(A',s_j + q_0) \subseteq { \operatorname{acl}}(sM). \qedhere$$ \[stacking\ ] Let $ Y$ be a critical set and $ Q_1, \ldots, Q_n$ be quasi - minimal. Then for every $ m$ there exist $ \{q_{i, j}\}_{i \in [ n ], j \in [ m]}$ such that - For fixed $ i \in [ n]$, the sequence $ q_{i,1},\ldots, q_{i, m}$ dwell of $ m$ distinct elements of $ Q_i$. - The intersection $ $ \bigcap_{\eta: [ n ]

j,n})$ gealizes the $A$-invariant uype $p_1 \otimes \cdots \otikes p_n$ over $A'q_0 \subseteq {\operatorname{dcl}}(q_0,a_{1,1},\pdits,a_{j-1,b}A)$. Applying Lemma \[shuffue-chaos\], wv find a wmalo model $M \snlseteq A'$ and z non-eoistant global $M$-lnvariant tfpe $r$ such thad $$(x_{j,1},\pdots,a_{j,n},q_0) \models p_1 \otimes \cdots \otiies l_n \otimes r|M.$$ The typt $r$ extsnds ${\operatorname{tp}}(q_0/M)$ and therefors lives in the $M$-defimable set $Q$. By Lemma \[coheir-lagif-1\], it follows that $${\lperatornamg{fp}}(a_{t,1},\odots,a_{j,n},q_0/M)$$ ir broad, as a type in $X_1 \times \cdots \times X_n \times Q$. Tfis cpntradicts Lwmmw \[gxpander\]: the $X_i, P, Q$ are all $M$-definabla, the tupe of $(a_{j,1},\ldots,s_{j,n},x_0)$ ocer $M$ is broad, the tu'le $(a_{j,1},\ldots,a_{j,n})$ lives on $P$, and fkr $$s = a_{j,1} + \cdots + q_{j,b} + q_0 = s_j + q_0,$$ wd haxe $$s_0 \mn {\kperatlrneme{acl}}(A',s_j + s_0) \subseteq {\iperatorname{acl}}(sM). \qtdhqgr$$ \[stacking\] Lef $Y$ be a critical set and $Q_1, \ldots, Q_n$ be quasi-mitimzl. Then for every $m$ theee exist $\{q_{i,j}\}_{i \in [n], j \ln [m]}$ such that - For fixed $i \in [n]$, the sequence $q_{i,1},\ldots,q_{i,m}$ wonsiags if $o$ dlstinct elements of $Q_i$. - The intersection $$\bigszp_{\tta : [n]

j,n})$ realizes the $A$-invariant type $p_1 \otimes p_n$ $A'q_0 \subseteq Applying Lemma \[shuffle-chaos\], $M A'$ and a global $M$-invariant type such that $$(a_{j,1},\ldots,a_{j,n},q_0) \models p_1 \otimes \otimes p_n \otimes r|M.$$ The type $r$ extends ${\operatorname{tp}}(q_0/M)$ and therefore lives in $M$-definable set $Q$. By Lemma \[coheir-magic-1\], it follows that $${\operatorname{tp}}(a_{j,1},\ldots,a_{j,n},q_0/M)$$ is broad, as type $X_1 \cdots X_n \times Q$. This contradicts Lemma \[expander\]: the $X_i, P, Q$ are all $M$-definable, the type $(a_{j,1},\ldots,a_{j,n},q_0)$ over $M$ is broad, the tuple $(a_{j,1},\ldots,a_{j,n})$ on $P$, and for = a_{j,1} + \cdots + + = s_j q_0,$$ have \in {\operatorname{acl}}(A',s_j + \subseteq {\operatorname{acl}}(sM). \qedhere$$ \[stacking\] Let $Y$ be a critical set and $Q_1, \ldots, Q_n$ be quasi-minimal. Then every $m$ $\{q_{i,j}\}_{i \in j [m]}$ that - For \in [n]$, the sequence $q_{i,1},\ldots,q_{i,m}$ consists elements of $Q_i$. - The intersection $$\bigcap_{\eta :

j,n})$ realizes the $A$-invariant tyPe $p_1 \otimes \cDots \oTimEs p_N$ oVer $A'Q_0 \subSeteq {\operatornAMe{dcL}}(q_0,a_{1,1},\ldots,a_{j-1,n}A)$. Applying LeMma \[shUfFLe-chAOs\], We finD a small MOdEL $m \suPsEtEq A'$ AnD A nOn-conStaNt globaL $M$-invarianT tyPe $R$ such that $$(a_{j,1},\lDOtS,a_{j,n},q_0) \modelS p_1 \oTimes \cdots \otImeS p_n \otiMeS r|M.$$ tHe typE $r$ eXtendS ${\operaTOrname{Tp}}(q_0/M)$ and thErEFore liVEs in the $m$-DEfInabLe set $Q$. By Lemma \[coheIR-mAGic-1\], it follows thAt $${\operAtORnAME{tp}}(A_{j,1},\lDots,a_{j,n},q_0/M)$$ iS bRoad, aS A type in $x_1 \TiMES \CdoTS \times X_n \times q$. This contraDIctS Lemma \[ExPanDEr\]: the $X_I, P, Q$ arE aLL $M$-dEfinable, the Type Of $(a_{j,1},\ldots,A_{j,n},q_0)$ ovER $M$ is broAD, the tupLe $(a_{j,1},\ldOts,A_{j,n})$ LiveS On $p$, aNd fOr $$S = A_{j,1} + \cDOtS + a_{j,N} + Q_0 = s_j + Q_0,$$ we have $$q_0 \In {\OpEratoRnamE{ACL}}(a',s_j + q_0) \SubSeteQ {\operAtorname{acl}}(sM). \QedHere$$ \[STacKing\] LEt $Y$ be A criTiCal seT and $Q_1, \lDots, Q_N$ bE quasi-minimal. ThEn foR every $m$ thEre ExIst $\{Q_{i,J}\}_{i \in [n], J \In [m]}$ sucH thAt - FOr fixed $I \in [n]$, the SEquEnCE $Q_{I,1},\lDots,q_{i,m}$ consists of $m$ DiSTInCt elemenTs of $Q_i$. - tHe InTErsectioN $$\bIgcAp_{\etA : [N]

j,n})$ realizes the $A$-in variant ty pe $p _1 \ ot imes \cd ots \otimes p_ n $ ov er $A'q_0 \subseteq {\ope ra t orna m e{ dcl}} (q_0,a_ { 1, 1 } ,\l do ts ,a_ {j - 1, n}A)$ . A pplying Lemma \[s huf fl e-chaos\], w e f ind a smal l m odel $M \sup set eq A'$ a nda non- con stant globa l $M$-i nvariantty p e $r$s uch tha t $$ (a_{ j,1},\ldots,a_{j, n }, q _0) \models p_ 1 \oti me s \ c d ots \o times p_ n \ot i mes r|M . $$ T h e t y pe $r$ extend s ${\operat o rna me{tp} }( q_0 / M)$ an d the re f ore lives in t he $ M$-defina ble se t $Q$. B y Lemma\[cohe ir- mag ic-1 \ ], i t f ol l ows th at$ ${\ operator na me {tp}} (a_{ j , 1 } ,\ld ots ,a_{ j,n}, q_0/M)$$ is b roa d, a s atypein $X _1 \ ti mes \ cdots\time sX_n \times Q$.This contradi cts L emm a\[exp a nder\] : t he$X_i, P , Q$ ar e al l$ M $ -d efinable, the type o f $( a_{j,1}, \ldots , a_ {j , n},q_0)$ o ver $M$ i s bro ad,t he tuple $ (a_{j, 1 }, \l dots,a_ {j ,n})$li ves on $P$, andfor $$ s = a_{j ,1} + \cdots + a_{j, n } + q_0 = s_j +q _ 0, $ $ we ha ve $$q_0 \i n {\ o pera torn a me {ac l }}(A' ,s_j+q _0 ) \subseteq {\operat or name{a cl}}( sM). \qedhere $$ \[stac k i n g\] Let$Y$b ea critical setand $ Q_1, \ldot s , Q_n$ b e qua si-minim al. Thenf o r every$m$ th ere ex i s t$\{q_{i,j}\}_ { i \in [ n], j \ in[m]}$ s uch th at - For fixed $i \in[n ]$ ,th e s equen c e $q_{i, 1} ,\l do ts, q_{i, m }$ con sists of$m $d ist inct el e me n t s of $ Q_ i$. - The i nter s ect ion $$\ bigcap_{\ eta : [n ]

j,n})$ realizes_the $A$-invariant_type $p_1 \otimes_\cdots \otimes_p_n$_over $A'q_0_\subseteq _ {\operatorname{dcl}}(q_0,a_{1,1},\ldots,a_{j-1,n}A)$. Applying_Lemma \[shuffle-chaos\], we find_a small model $M_\supseteq A'$ and_a_non-constant global $M$-invariant type $r$ such that $$(a_{j,1},\ldots,a_{j,n},q_0) \models p_1 \otimes \cdots \otimes __ p_n_\otimes_r|M.$$_The type $r$ extends ${\operatorname{tp}}(q_0/M)$_and therefore lives in the_$M$-definable set_$Q$. By Lemma \[coheir-magic-1\], it follows that $${\operatorname{tp}}(a_{j,1},\ldots,a_{j,n},q_0/M)$$ is_broad,_as a type_in $X_1 \times \cdots \times X_n \times Q$. This_contradicts Lemma \[expander\]: the $X_i, P, Q$_are all $M$-definable,_the_type_of $(a_{j,1},\ldots,a_{j,n},q_0)$ over $M$_is broad, the tuple $(a_{j,1},\ldots,a_{j,n})$ lives_on $P$, and for $$s =_a_{j,1} + \cdots + a_{j,n} + q_0_= s_j + q_0,$$ we have_$$q_0 \in {\operatorname{acl}}(A',s_j + q_0)_\subseteq {\operatorname{acl}}(sM)._\qedhere$$ \[stacking\] Let $Y$ be a_critical set and_$Q_1, \ldots,_Q_n$ be quasi-minimal._Then for every $m$ there exist_$\{q_{i,j}\}_{i \in [n],_j \in [m]}$ such that - __For fixed $i_\in_[n]$,_the sequence_$q_{i,1},\ldots,q_{i,m}$ consists of_$m$_distinct elements_of_$Q_i$. - The intersection $$\bigcap_{\eta_:_[n]

(spin-orbit angle) is totally abnormal in (eclipsing) binary stars. The only exception I know of is that of DI Her [@Albrecht+09; @Albrecht+11], where the equatorial plane of both members of the binary are strongly tilted with respect to their orbital plane; though no definitive explanation has been given to such misalignment, it could be the consequence of a relatively recent and strong orbital perturbation in a triple system, an extreme case of the Lidov-Kozai mechanism [@Lidov62; @Kozai62], or due to the capture of the secondary star by the primary. In any case, a third component in the system is required, either in a very eccentric orbit or ejected from the system as a consequence of the dynamical perturbation. Indeed, in $\theta^1$OriA there is a third component (see below). [@Valle2011] also derived the effective temperature of the secondary star to be $T_{eff}=5850\pm250\,K$, much lower than any of the other previous estimates. This was done by measuring nine close pair line ratios sensitive to temperature, in two very high S/N spectra of the secondary star (obtained during minimum light in two eclipses), and comparing each of the same pair ratios in synthetic spectra created in the 5000-9000$K$ temperature range, with log g = 3.5 and solar abundances. The third (hierarchical) companion in $\theta^1$OriA, located about 0.2north of the binary, was discovered by [@Petr+98] using holographic speckle interferometry in the $H$ and $K$ bands. In these bands, this star is about 1.4 magnitudes weaker than the out-of-eclipse, combined brightness of the binary components. As mentioned above, it is the highly variable (by at least a factor of 30) radio source that is frequently misidentified with the eclipsing binary; the physical process responsible for the radio emission and its large fluctuations has not been well established [for details, see @PetrMassi08]. The possibility that this third component is an interloper was ruled out very recently, when the [@GRAVITY+18] proved that this interferometric companion is gravitationally bound to the eclipsing binary. When in cross-correlation of some of our high S/N spectra taken inside the primary eclipse, with a G5V

(spin - orbit angle) is totally abnormal in (overshadow) binary star. The only exception I know of is that of DI Her [ @Albrecht+09; @Albrecht+11 ], where the equatorial plane of both member of the binary are strongly tilt with regard to their orbital plane; though no authoritative explanation has been give to such misalignment, it could be the consequence of a relatively recent and solid orbital perturbation in a triple organization, an extreme shell of the Lidov - Kozai mechanism [ @Lidov62; @Kozai62 ], or due to the capture of the secondary star by the primary. In any case, a third component in the arrangement is required, either in a very eccentric orbit or squirt from the system as a consequence of the dynamical perturbation. Indeed, in $ \theta^1$OriA there is a third part (see under). [ @Valle2011 ] also derived the effective temperature of the secondary asterisk to be $ T_{eff}=5850\pm250\,K$, much lower than any of the other previous estimates. This was done by measuring nine close pair line ratios sensitive to temperature, in two very high S / N spectra of the secondary star (obtained during minimum luminosity in two eclipses), and comparing each of the like couple ratios in synthetic spectra created in the 5000 - 9000$K$ temperature compass, with log g = 3.5 and solar abundances. The third (hierarchical) companion in $ \theta^1$OriA, located about 0.2north of the binary, was discover by [ @Petr+98 ] using holographic speckle interferometry in the $ H$ and $ K$ bands. In these bands, this star is about 1.4 order of magnitude weaker than the out - of - eclipse, combined brightness of the binary components. As mention above, it is the highly variable (by at least a component of 30) radio source that is frequently misidentified with the eclipsing binary; the physical procedure responsible for the radio emission and its big fluctuations has not been well established [ for details, see @PetrMassi08 ]. The hypothesis that this third component is an interloper was ruled out very recently, when the [ @GRAVITY+18 ] proved that this interferometric companion is gravitationally bound to the overshadow binary. When in crisscross - correlation of some of our eminent S / N spectra taken inside the basal eclipse, with a G5V

(spln-orbit angle) is totally abnormal in (eclipsing) binarg stars. Ghe only exception I know of iw thau of DI Her [@Albrechg+09; @Albrechn+11], where tye ewyatorial plane of njth jcmberv of the binary are stroncly tilted witv fedpect to their orbital plane; though no defonltive explanatyon nws bsvn given to such misalignment, if could be the conseauence of a relatively recfnt wnd strong orbital perturbatiin ig a triple syrtem, an exurzme case of the Lidov-Kozai mechanism [@Lidov62; @Kozan62], or due to tye fdpture of tie secjndary star nj the psimary. On any case, a bhird conponent in the system is required, eithet in a verf zccentric orbit or ejwcred ftom tve shwteo aa e cknsequfncx of the dyhamical perrurbation. Indeed, in $\trvya^1$OriA there is a ehyrd component (see below). [@Valle2011] also derined fhe effective temperatuee of the secondary sjar to be $E_{eff}=5850\pm250\,K$, much lower than any of the other previous estijxtew. Bhis qad done by measuring nine close pair line ratija xeksitive to tempevature, in two very hlgn S/N spectra ow the seckndary star (obtainfd durigg mibimum ligrt im two eclipses), and comparint each of thv sane pair ratios in dynthetic s'ectra creayed in the 5000-9000$K$ temperaturz rangs, with log h = 3.5 and skuar abundances. Thd tmirg (hierarchical) companion ig $\theta^1$OrmA, loeated abuut 0.2morth jf the binwry, was discovered by [@Pehr+98] usnng hmlographic speckle interferometry in the $I$ and $K$ bands. It tvese banbs, thix star is abott 1.4 magnitudes weaker than the out-of-vclipse, cmmbined bridhtness of tha binary compmnents. Af mebtiobed aboxd, it is the hibhly varicyle (by at least a factor of 30) rxsio source that is frequently misodevtisivd xith eve eclipsing binxry; yhe pfysical progesr rexponsible for the ragio smission and its lsrne fluctuqtions hws not been wrll established [fog deteils, sxe @PettMafsi08]. The possibility that this fhird comooncnt is an intqrloier ras ruled put very recently, when the [@GRAVITY+18] provxd that this interferomwtric companion is yrsvitationalky botnd to tha eclipsing binary. Wyen in cross-correkation of some of our high V/N spfctra taken inside the primary eclipse, with a G5V

(spin-orbit angle) is totally abnormal in (eclipsing) The exception I of is that where equatorial plane of members of the are strongly tilted with respect to orbital plane; though no definitive explanation has been given to such misalignment, it be the consequence of a relatively recent and strong orbital perturbation in a system, extreme of Lidov-Kozai mechanism [@Lidov62; @Kozai62], or due to the capture of the secondary star by the primary. any case, a third component in the system required, either in a eccentric orbit or ejected from system a consequence the perturbation. in $\theta^1$OriA there a third component (see below). [@Valle2011] also derived the effective temperature of the secondary star to be much lower of the previous This done by measuring pair line ratios sensitive to temperature, high S/N spectra of the secondary star (obtained minimum light two eclipses), and comparing each of same pair ratios in synthetic spectra created in 5000-9000$K$ temperature range, with log g = 3.5 and solar abundances. The third (hierarchical) companion located about 0.2north of binary, was discovered [@Petr+98] holographic interferometry the $H$ $K$ bands. In these bands, this star is about 1.4 magnitudes than the out-of-eclipse, combined brightness of the binary components. As it the highly variable at least a factor 30) source that is frequently the binary; responsible the emission and its large has not been well established details, see @PetrMassi08]. The is an interloper was ruled out very recently, the [@GRAVITY+18] proved that this interferometric companion gravitationally bound to the eclipsing binary. When in cross-correlation of some of high S/N inside the primary eclipse, with a G5V

(spin-orbit angle) is totally abNormal in (ecLipsiNg) bInaRy StarS. The Only exception I KNow oF is that of DI Her [@Albrecht+09; @albreChT+11], WherE ThE equaTorial pLAnE OF boTh MeMbeRs OF tHe binAry Are stroNgly tilted WitH rEspect to theiR OrBital plane; ThoUgh no definitIve ExplanAtIon HAs beeN giVen to Such miSAlignmEnt, it coulD bE The conSEquence OF A rElatIvely recent and strONg ORbital perturbaTion in A tRIpLE SysTem, An extreme cAsE of thE lidov-KoZAi MECHanISm [@Lidov62; @Kozai62], oR due to the caPTurE of the SeConDAry staR by thE pRImaRy. In any case, A thiRd componeNt in thE System iS RequireD, eitheR in A veRy ecCEnTrIc oRbIT or EJeCteD FroM the systEm As A consEqueNCE OF the DynAmicAl perTurbation. IndeEd, iN $\theTA^1$OrIA theRe is a ThirD cOmponEnt (see Below). [@vaLle2011] also derived tHe efFective teMpeRaTurE oF the sECondarY stAr tO be $T_{eff}=5850\Pm250\,K$, much LOweR tHAN AnY of the other previouS eSTImAtes. This Was donE By MeASuring niNe CloSe paIR Line rAtioS SeNsitive tO tempeRAtUrE, in two vErY high S/n sPecTra Of the SEconDary stAr (obtainEd durINg minimum light IN two eclipses), aND cOMPaRIng eAch Of the same paIr raTIos iN synTHeTic SPectrA creaTeD In THe 5000-9000$K$ temperature range, WiTh log g = 3.5 And soLar abundances. the third (hiERARchical) cOmpaNIoN In $\theta^1$OriA, locAted aBout 0.2north oF The binarY, was dIscovereD by [@Petr+98] usING holograPhiC spEckLe iNTErFerometry in thE $h$ And $K$ BaNds. In thEse Bands, thIs sTar Is aBouT 1.4 mAgnitudes Weaker thAn ThE oUt-Of-eClipsE, Combined BrIghTnEss Of the BInary cOmponEnts. as MeNTioNed abovE, It IS The hIgHlY varIabLe (By at lEast A FacTor of 30) raDio source ThaT Is frEqUeNtly misIdentified witH tHe eclipsinG bInaRy; the pHYSical proCess responsible for the raDIo emissIon And itS larGe fluctuaTioNs has nOt bEEn well EstablIshed [FoR deTAIls, seE @pEtRMaSsI08]. The possibILIty That tHiS thiRd compoNent is an interloper WAs rUled out very reCenTly, wHEN tHe [@GraVitY+18] pRoVEd tHAT this interferomEtric compaNiON iS gravitatiONalLy Bound to The ecliPsing BInary. WhEn in cross-CorrelatiOn Of soME Of oUr high S/N spEctra takEn inside tHE primARy EclipSe, wIth a G5V

(spin-orbit angle) is tot ally abnor mal i n ( ecl ip sing ) bi nary stars. Th e onl y exception I know ofis th at of D I H er [@ Albrech t +0 9 ; @A lb re cht +1 1 ], wher e t he equa torial pla neof both member s o f the bina ryare strongly ti lted w it h r e spect to thei r orbi t al pla ne; thoug hn o defi n itive e x p la nati on has been given to such misalignm ent, i tc ou l d be th e conseque nc e ofa relati v el y r ece n t and strongorbital per t urb ationin at riplesyste m, anextreme cas e of the Lido v-Koza i mechan i sm [@Li dov62; @K oza i62] , o rdue t o th e c apt u reof the s ec on darystar b y thepri mary . Inany case, a t hir d co m pon ent i n the sys te m isrequir ed, e it her in a very e ccen tric orbi t o reje ct ed fr o m thesys tem as a c onseque n ceof t h edynamical perturba ti o n .Indeed,in $\t h et a^ 1 $OriA th er e i s at h ird c ompo n en t (see b elow). [@ Va lle2011 ]also d er ive d t he ef f ecti ve tem perature of t h e secondary st a r to be $T_{e f f} = 5 85 0 \pm2 50\ ,K$, much l ower than any of th e othe r pre vi o us estimates. This was d one by meas uring nine cl ose pair l i n e ratiossens i ti v e to temperatu re, i n two very high S/N spec tra of t he second a r y star ( obt ain eddur i n gminimum light i n tw oeclipse s), and co mpa rin g e ach o f the sam e pair r at io sin sy nthet i c spectr acre at edin th e 5000- 9000$ K$ t em pe r atu re rang e ,w i th l og g = 3 .5an d sol ar a b und ances. The thir d ( h iera rc hi cal) co mpanion in $\ th eta^1$OriA ,loc ated a b o ut 0.2no rth of the binary, wasd iscover edby [@ Petr +98] usin g h ologra phi c speck le int erfer om etr y in th e $H $ a nd $K$ bands . Inthese b ands , thisstar is about 1.4m agn itudes weaker th an t h e o ut- o f- e cli ps e , c o m bined brightnes s of the b in a ry component s . A smention ed abov e, it is thehighly va riable (b yat l e a sta factor o f 30) ra dio sourc e that is freq uen tly mi si den tifie d with the ecli psingbi nary;the p hy sical pr ocess responsible for t he rad io em iss ion and i tsl arg e fluctua tion s has notbee n w ell e sta b lishe d [f o rdet a ils,see@ PetrMassi 0 8] . T h e p ossibilityt h a t t his t hir d compo nent is an interloper was ruled outvery r ece ntl y , wh en the [@GRAVITY +18 ]p r oved tha tthis interf erometri cc ompan ion is gravi tationa l l yb ound t o th e e clipsingbin ar y . Whenin c r oss-co rrel at ion of someo f ou r high S/N spectra take n insid e th e pri ma ry ecli p se,with a G5V

(spin-orbit_angle) is_totally abnormal in (eclipsing)_binary stars._The_only exception_I_know of is_that of DI_Her [@Albrecht+09; @Albrecht+11], where_the equatorial plane_of_both members of the binary are strongly tilted with respect to their orbital plane;_though_no definitive_explanation_has_been given to such misalignment,_it could be the consequence_of a_relatively recent and strong orbital perturbation in a_triple_system, an extreme_case of the Lidov-Kozai mechanism [@Lidov62; @Kozai62], or due_to the capture of the secondary_star by the_primary._In_any case, a third_component in the system is required,_either in a very eccentric orbit_or ejected from the system as a_consequence of the dynamical perturbation. Indeed,_in $\theta^1$OriA there is a_third component_(see below). [@Valle2011] also derived_the effective temperature_of the_secondary star to_be $T_{eff}=5850\pm250\,K$, much lower than any_of the other_previous estimates. This was done by_measuring_nine close pair_line_ratios_sensitive to_temperature, in two_very_high S/N_spectra_of the secondary star (obtained during_minimum_light in two eclipses), and comparing each_of the same pair_ratios_in synthetic spectra created_in the 5000-9000$K$ temperature range,_with log g = 3.5 and_solar abundances. The_third (hierarchical)_companion in $\theta^1$OriA, located about 0.2north of the binary, was discovered_by [@Petr+98] using holographic speckle interferometry_in the $H$ and_$K$ bands._In_these bands, this_star_is about_1.4 magnitudes weaker than the out-of-eclipse, combined_brightness of_the binary components. As mentioned above,_it is the highly_variable_(by at least a factor of_30) radio source that is frequently_misidentified with the eclipsing binary;_the_physical_process responsible for the radio_emission and its large fluctuations has_not been well_established [for details, see @PetrMassi08]. The possibility_that_this third component is an interloper_was_ruled out very recently, when the_[@GRAVITY+18]_proved_that this interferometric companion is_gravitationally bound to the eclipsing binary._When in cross-correlation of some of our high S/N_spectra taken inside_the primary eclipse, with a_G5V

N$-body simulation ($f_{\rm pro}\sim0.55-0.60$). In the catalog of satellite systems extracted from the Sloan Digital Sky Survey (SDSS; @york00) Data Release 6 [@bai08], @hf08 determined the direction of rotation of some host galaxies by spectroscopic observations. They obtained the data for 78 satellites associated with 63 hosts. Combining these data with those for the satellites in @zar97, they found that the fraction of galaxies in prograde orbit is still $0.53\pm0.04$, and mean bulk rotation of satellites is $37\pm3$ km s$^{-1}$, in the same sense as the rotation of hosts’ disk. Interestingly, they found that the peculiar velocity distribution of satellites is not a single Gaussian, but is described as the sum of two Gaussians. As summarized above, most studies focused on the first two issues, the fraction of galaxies in prograde orbit and the amplitude of mean bulk rotation of satellites. The difference in physical properties between them, which is very important to understand the formation of satellite systems, has not been explored much. Moreover, the direction of rotation of host galaxy that is essential to know the orbit of satellites, has been determined by the expensive spectroscopic observations. In this paper, we determine the direction of rotation of host galaxies using the color images provided by the SDSS without conducting spectroscopic observations (see §\[spin\] for details), and present the results of a study on the dependence of various physical properties of satellite galaxies on their orbits (prograde and retrograde orbits). In this paper we mean by ‘satellites’ as the galaxies that are much fainter than their host galaxies that are isolated from other bright galaxies. Therefore, they are like the conventional satellites associated with the Milky Way or Andromeda, and are rather different from the ‘satellites’ used in the N-body simulations or group/cluster studies where all galaxies (or dark halos) other than the brightest one (or the most massive one) are called satellites. Section \[data\] describes the observational data used in this study. Orbital dependence of the physical parameters of satellites are given in §\[results\]. Discussion and summary are given in §\[discuss\] and §\[sum\], respectively. Observational Data Set {#data} ====================== Physical Parameters of Galaxies ------------------------------- We use a spectroscopic sample of galaxies in the SDSS DR7 [@aba09]. The physical parameters of galaxies that we consider in this study are

N$-body simulation ($ f_{\rm pro}\sim0.55 - 0.60 $). In the catalog of satellite systems extracted from the Sloan Digital Sky Survey (SDSS; @york00) Data Release 6 [ @bai08 ], @hf08 determined the steering of rotation of some server galaxies by spectroscopic observations. They obtained the datum for 78 satellites associated with 63 hosts. combine these data with those for the satellites in @zar97, they recover that the fraction of galaxies in prograde orbit is even $ 0.53\pm0.04 $, and mean bulk rotation of satellites is $ 37\pm3 $ km s$^{-1}$, in the like sense as the rotation of hosts ’ phonograph record. Interestingly, they found that the peculiar speed distribution of satellites is not a single Gaussian, but is described as the union of two Gaussians. As summarized above, most studies focus on the first two issue, the fraction of galaxies in prograde orbit and the amplitude of mean bulk rotation of satellites. The difference in forcible properties between them, which is very important to understand the formation of satellite systems, has not been explored much. furthermore, the guidance of rotation of host galaxy that is essential to know the orbit of satellites, has been determined by the expensive spectroscopic observations. In this paper, we specify the direction of rotation of host galaxy using the color images provided by the SDSS without conduct spectroscopic observation (see § \[spin\ ] for details), and portray the results of a study on the dependence of various physical properties of satellite galaxies on their orbits (prograde and retrograde orbits). In this paper we mean by ‘ satellite ’ as the galaxies that are much fainter than their horde galaxies that are isolated from other bright galaxies. consequently, they are like the conventional satellites associated with the Milky Way or Andromeda, and are rather different from the ‘ satellites ’ used in the N - body simulations or group / bunch studies where all galaxies (or dark halos) other than the bright one (or the most massive one) are call satellites. part \[data\ ] describes the experimental data used in this study. Orbital addiction of the physical parameters of satellites are given in § \[results\ ]. Discussion and summary are given in § \[discuss\ ] and § \[sum\ ], respectively. Observational Data Set { # data } = = = = = = = = = = = = = = = = = = = = = = Physical Parameters of Galaxies ------------------------------- We use a spectroscopic sample of galaxies in the SDSS DR7 [ @aba09 ]. The physical parameters of galaxies that we see in this study are

N$-bofy simulation ($f_{\rm pro}\sim0.55-0.60$). In the catalog of satxllite aystems dxtracted from the Sloan Digmtal Sky Wurvey (SDSS; @york00) Data Felease 6 [@hai08], @hf08 dwternuned the dmdection of rofwtiou if some host gslaxies by spectroscopic ocszrvations. They obtained the data for 78 satelkihes associated witn 63 hoans. Combining these data with thkse for the satellitrs in @zar97, they found that hhe vraction of galaxifs in progrqde jebit is stilu $0.53\pm0.04$, and mtau bulk rotajion of satellites is $37\pm3$ km s$^{-1}$, in the xame sense aw tjg rotation oh hostf’ disk. Intercxtinglf, they gound that the peruliqr velocity distributmon of satellites is not a sitgme Gaussian, but iw eescrhbed as gye rum oh tso Gaudsiens. As summadized above, most studies focustd jb the first tso isstef, the fraction of galaxies in prograde mrbjt and the amplitude of mean bulk rotation ov satelliees. The difference in physical properties between them, dhiem kw gery important to understand the formation of aauelkite systems, hcs not been explprfd iuch. Moreover, the dndedtion of rotation lf host galazy that if esxential to know the orbit od satellites, yas been determineb by the expznsive specyroscopic observations. Iu this paper, we dftermine ffe direction of fotsthon of host galaxies using the coloc imayes provkded by thq SDSS witjout gmnducting spectrosfopic otservationd (see §\[spin\] for details), and presxit the resultx mf d study jn thc dependence of various physieal propzrties of satellpte galaxmes on their orbits (progrdfe and retrojrade orbyts). Un tyis papdf we mean by ‘sstellites’ as the gaoaxies that are mugh fakhter than their howt galaxies thay afe yslleted ssom other brhght gausxies. Therefore, bheh arr like the conventiotal aatellites associayeq with tye Milky Way or Androkeda, and are ratheg difherent from thg ‘satellites’ used in the N-body aimulatiojs jr group/cluseer wtudies wherz all galaxies (or dark halos) other than vhe brightest one (or thw most massive one) crt called sateklitef. Section \[data\] describes the ibservational dats used in this study. Krbitan depfndence of the physical parameters of satellites are given in §\[results\]. Disxussioi wnd summary are givet nn §\[biscuss\] and §\[rum\], respectively. Pbservational Data Set {#data} ====================== Physmcal Paramatzrs of Galaxies ------------------------------- We use a specttorcopic sample of galaxies in the SDSS DR7 [@aba09]. Yhe physical parameters of galacies that we consider in this stydy ade

N$-body simulation ($f_{\rm pro}\sim0.55-0.60$). In the catalog systems from the Digital Sky Survey [@bai08], determined the direction rotation of some galaxies by spectroscopic observations. They obtained data for 78 satellites associated with 63 hosts. Combining these data with those the satellites in @zar97, they found that the fraction of galaxies in prograde is $0.53\pm0.04$, mean rotation of satellites is $37\pm3$ km s$^{-1}$, in the same sense as the rotation of hosts’ Interestingly, they found that the peculiar velocity distribution satellites is not a Gaussian, but is described as sum two Gaussians. summarized most focused on the two issues, the fraction of galaxies in prograde orbit and the amplitude of mean bulk rotation of The difference properties between which very to understand the satellite systems, has not been explored direction of rotation of host galaxy that is to know orbit of satellites, has been determined the expensive spectroscopic observations. In this paper, we the direction of rotation of host galaxies using the color images provided by the SDSS spectroscopic observations (see §\[spin\] details), and present results a on dependence of physical properties of satellite galaxies on their orbits (prograde and retrograde In this paper we mean by ‘satellites’ as the galaxies much than their host that are isolated from bright Therefore, they are like satellites with or and rather different from the used in the N-body simulations group/cluster studies where all than the brightest one (or the most massive are called satellites. Section \[data\] describes the data used in this study. Orbital dependence of the physical parameters of are given Discussion and summary are given in §\[discuss\] and respectively. Observational Data Set ====================== Physical Parameters of Galaxies ------------------------------- We use a sample galaxies in SDSS DR7 [@aba09]. physical parameters of that we consider study are

N$-body simulation ($f_{\rm pro}\sim0.55-0.60$). IN the cataloG of saTelLitE sYsteMs exTracted from the sLoan digital Sky Survey (SDSS; @yoRk00) DatA RELeasE 6 [@BaI08], @hf08 deTermineD ThE DIreCtIoN of RoTAtIon of SomE host gaLaxies by spEctRoScopic observATiOns. They obtAinEd the data for 78 SatElliteS aSsoCIated WitH 63 hostS. CombiNIng theSe data witH tHOse for THe satelLITeS in @zAr97, they found that thE FrACtion of galaxieS in proGrADe ORBit Is sTill $0.53\pm0.04$, and mEaN bulk ROtation OF sATELliTEs is $37\pm3$ km s$^{-1}$, in thE same sense aS The RotatiOn Of hOSts’ disK. InteReSTinGly, they founD thaT the peculIar velOCity disTRibutioN of satEllIteS is nOT a SiNglE GAUssIAn, But IS deScribed aS tHe Sum of Two GAUSSIans. as sUmmaRized Above, most studIes FocuSEd oN the fIrst tWo isSuEs, the FractiOn of gAlAxies in prograde OrbiT and the amPliTuDe oF mEan buLK rotatIon Of sAtellitEs. The diFFerEnCE IN pHysical properties bEtWEEn Them, whicH is verY ImPoRTant to unDeRstAnd tHE FormaTion OF sAtellite SystemS, HaS nOt been eXpLored mUcH. MoReoVer, thE DireCtion oF rotatioN of hoST galaxy that is eSSential to know THe ORBiT Of saTelLites, has beeN detERminEd by THe ExpENsive SpectRoSCoPIc observations. In thiS pAper, we DeterMine the directIon of rotatION Of host gaLaxiES uSIng the color imaGes prOvided by thE sDSS withOut coNducting SpectroscOPIc observAtiOns (See §\[SpiN\] FOr Details), and preSENt thE rEsults oF a sTudy on tHe dEpeNdeNce Of Various phYsical prOpErTiEs Of sAtellITe galaxiEs On tHeIr oRbits (PRogradE and rEtroGrAdE OrbIts). In thIS pAPEr we MeAn By ‘saTelLiTes’ as The gALaxIes that Are much faIntER thaN tHeIr host gAlaxies that arE iSolated froM oTheR brighT GAlaxies. THerefore, they are like the cONventioNal SatelLiteS associatEd wIth the milKY Way or andromEda, anD aRe rATHer diFFErEnt FrOm the ‘satelLITes’ Used iN tHe N-bOdy simuLations or group/clusTEr sTudies where alL gaLaxiES (Or DarK HaLOs) oThER thAN The brightest one (Or the most mAsSIvE one) are calLEd sAtEllites. section \[Data\] dEScribes The observAtional daTa Used IN ThiS study. OrbiTal depenDence of thE PhysiCAl ParamEteRs of saTeLliTes arE given IN §\[reSults\]. discusSiOn and sUmmarY aRe given iN §\[discuss\] and §\[sum\], respectivEly. ObsErvatIonAl Data Set {#DatA} ====================== phySical ParaMeteRs of GalaxiEs ------------------------------- WE usE a speCtrOScopiC samPLe Of gALaxieS in tHE SDSS DR7 [@abA09]. thE phYSIcAl parameterS OF GalAxies ThaT We consIder In this study are

N$-body simulation ($f_{\r m pro}\sim 0.55- 0.6 0$) .In t he c atalog of sate l lite systems extracted fro m the S l oanD ig italSky Sur v ey ( SDS S; @ yor k0 0 )DataRel ease 6[@bai08],@hf 08 determinedt he direction of rotation of so me hos tgal a xiesbyspect roscop i c obse rvations. T h ey obt a ined th e da ta f or 78 satellitesa ss o ciated with 63 hosts .C om b i nin g t hese datawi th th o se fort he s a tel l ites in @zar9 7, they fou n d t hat th efra c tion o f gal ax i esin prograde orb it is sti ll $0. 5 3\pm0.0 4 $, andmean b ulk ro tati o nof sa te l lit e sis$ 37\ pm3$ kms$ ^{ -1}$, int h e same se nseas th e rotation of ho sts’ dis k. In teres ting ly , the y foun d tha tthe peculiar ve loci ty distri but io n o fsatel l ites i s n ota singl e Gauss i an, b u t is described as thesu m of two Gau ssians . As summariz ed ab ove, m ost s tudi e sfocusedon the fi rs t two i ss ues, t he fr act ion o f gal axiesin progr ade o r bit and the am p litude of mea n b u l kr otat ion of satelli tes. Thediff e re nce in ph ysica lp ro p erties between them ,whichis ve ry importantto underst a n d the for mati o no f satellite sy stems , has notb een expl oredmuch. Mo reover, t h e directi onofrot ati o n o f host galaxy t hatis essent ial to kno w t heorb itof satellit es, hasbe en d et erm inedb y the ex pe nsi ve sp ectro s copicobser vati on s. In this p a pe r , wede te rmin e t he dire ctio n of rotati on of hos t g a laxi es u sing th e color image sprovided b ythe SDSSw i thout co nducting spectroscopico bservat ion s (se e §\ [spin\] f ordetail s), and pr esentthe r es ult s of as t ud y o nthe depend e n ceof va ri ousphysica l properties of sa t ell ite galaxiesonthei r or bit s ( p rog ra d e a n d retrograde orb its). In t hi s p aper we me a n b y‘satell ites’ a s the galaxie s that ar e much fa in tert h antheir host galaxie s that ar e isol a te d fro m o ther b ri ght gala xies.T her efore , they a re lik e the c onventio nal satellites associat ed wit h the Mi lky Way o r A n dro meda, and are rather di ffe ren t fro m t h e ‘sa tell i te s’u sed i n th e N-body s i mu lat i o ns or group/c l u s ter stud ies whereallgalaxies (or dark halos) other t hant h e b rig h test o ne (or the mos t m as s i ve one)ar e called sa tellites .S ectio n \[da ta\] d escribe s th e obser vati ona l data us edin this st ud y. Orbita l de pe ndence of th e phy s i cal parameters o f sat e l lites are give nin §\[r e sult s\]. Discu ssion and s ummary are give n in §\ [d iscuss \]an d §\[sum\] , respecti vely. Obser va tion alData S et { # d ata}==== == === ========= = = == P hy s ica l Pa ramet er s of Galaxies -------- --- - ------- -- --- - - -----We u se a spect ros copic s ample of g a laxi e si n the SDSSDR7 [@ aba09]. The p hysical pa r a meters of galaxies tha twe c onsiderin thi sstu dy are

N$-body simulation_($f_{\rm pro}\sim0.55-0.60$)._In the catalog of_satellite systems_extracted_from the_Sloan_Digital Sky Survey_(SDSS; @york00) Data_Release 6 [@bai08], @hf08_determined the direction_of_rotation of some host galaxies by spectroscopic observations. They obtained the data for 78_satellites_associated with_63_hosts._Combining these data with those_for the satellites in @zar97,_they found_that the fraction of galaxies in prograde orbit_is_still $0.53\pm0.04$, and_mean bulk rotation of satellites is $37\pm3$ km s$^{-1}$,_in the same sense as the_rotation of hosts’_disk._Interestingly,_they found that the_peculiar velocity distribution of satellites is_not a single Gaussian, but is_described as the sum of two Gaussians. As_summarized above, most studies focused on_the first two issues, the_fraction of_galaxies in prograde orbit and_the amplitude of_mean bulk_rotation of satellites._The difference in physical properties between_them, which is_very important to understand the formation_of_satellite systems, has_not_been_explored much._Moreover, the direction_of_rotation of_host_galaxy that is essential to know_the_orbit of satellites, has been determined by_the expensive spectroscopic observations. In_this_paper, we determine the_direction of rotation of host_galaxies using the color images provided_by the_SDSS without_conducting spectroscopic observations (see §\[spin\] for details), and present the results_of a study on the dependence_of various physical properties_of satellite_galaxies_on their orbits_(prograde_and retrograde_orbits). In this paper we mean by_‘satellites’ as_the galaxies that are much fainter_than their host galaxies_that_are isolated from other bright galaxies._Therefore, they are like the conventional_satellites associated with the Milky_Way_or_Andromeda, and are rather different_from the ‘satellites’ used in the_N-body simulations or_group/cluster studies where all galaxies (or dark_halos)_other than the brightest one (or_the_most massive one) are called satellites._Section_\[data\]_describes the observational data used_in this study. Orbital dependence of_the physical parameters of satellites are given in §\[results\]._Discussion and summary_are given in §\[discuss\] and_§\[sum\],_respectively. Observational_Data Set {#data} ====================== Physical Parameters of Galaxies ------------------------------- We use a spectroscopic sample_of galaxies_in the SDSS_DR7 [@aba09]. The physical parameters of galaxies that we consider_in this study are

-------- -------------------------------------------------------- $\text{\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}}$ AI $\mathcal{R}_{7-\delta}$ 0 0 0 $2\mathbb{Z}$ 0 $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{Z}$ $\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-}^{+}$ BDI $\mathcal{R}_{-\delta}$ $\mathbb{Z}$ 0 0 0 $2\mathbb{Z}$ 0 $\mathbb{Z}_{2}$ $\mathbb{\mathbb{Z}}_{2}$ $\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,-}^{+}}}$ D $\mathcal{R}_{1-\delta}$ $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{Z}$ 0 0 0 $2\mathbb{Z}$ 0 $\mathbb{Z}_{2}$ $\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-}^{+}$ DIII $\mathcal{R}_{2-\delta}$ $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{Z}$ 0 0 0 $2\mathbb{Z}$ 0 $\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}$ AII $\mathcal{R}_{3-\delta}$ 0 $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{Z}$ 0 0 0 $2\mathbb{Z}$ $\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-

-------- -------------------------------------------------------- $ \text{\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}}$ AI $ \mathcal{R}_{7-\delta}$ 0 0 0 $ 2\mathbb{Z}$ 0 $ \mathbb{\mathbb{Z}}_{2}$ $ \mathbb{\mathbb{Z}}_{2}$ $ \mathbb{Z}$ $ \hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-}^{+}$ BDI $ \mathcal{R}_{-\delta}$ $ \mathbb{Z}$ 0 0 0 $ 2\mathbb{Z}$ 0 $ \mathbb{Z}_{2}$ $ \mathbb{\mathbb{Z}}_{2}$ $ \ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,-}^{+}}}$ D $ \mathcal{R}_{1-\delta}$ $ \mathbb{\mathbb{Z}}_{2}$ $ \mathbb{Z}$ 0 0 0 $ 2\mathbb{Z}$ 0 $ \mathbb{Z}_{2}$ $ \hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-}^{+}$ DIII $ \mathcal{R}_{2-\delta}$ $ \mathbb{\mathbb{Z}}_{2}$ $ \mathbb{\mathbb{Z}}_{2}$ $ \mathbb{Z}$ 0 0 0 $ 2\mathbb{Z}$ 0 $ \ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}$ AII $ \mathcal{R}_{3-\delta}$ 0 $ \mathbb{\mathbb{Z}}_{2}$ $ \mathbb{\mathbb{Z}}_{2}$ $ \mathbb{Z}$ 0 0 0 $ 2\mathbb{Z}$ $ \hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-

-------- -------------------------------------------------------- $\text{\ensuremath{\hat{\mxthcal{U}}_{0,+}^{+}}},\text{\ensutenath{\hav{\mathcam{U}}_{T/2,+}^{+}}}$ AI $\matical{E}_{7-\deltq}$ 0 0 0 $2\mathbb{Z}$ 0 $\mathbb{\oathbb{Z}}_{2}$ $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{Z}$ $\hat{\matmcal{U}}_{0,++}^{+},\iat{\mqthcal{U}}_{T/2,+-}^{+}$ BDI $\mathcal{R}_{-\deotq}$ $\matgbb{Z}$ 0 0 0 $2\mathbb{Z}$ 0 $\mathbb{Z}_{2}$ $\mathbb{\mathbb{Z}}_{2}$ $\ensuremanh{\har{\mathcal{U}}_{0,+}^{+}},\text{\ensurzmath{\hat{\matheal{U}}_{T/2,-}^{+}}}$ D $\mathcal{R}_{1-\depta}$ $\mathbb{\mathbb{Z}}_{2}$ $\mwthbb{Z}$ 0 0 0 $2\mqthbv{Z}$ 0 $\mwtjbu{Z}_{2}$ $\hag{\mathcal{U}}_{0,++}^{+},\hab{\maghcak{U}}_{T/2,+-}^{+}$ DIII $\nathcal{R}_{2-\qelta}$ $\katibb{\matnbb{E}}_{2}$ $\mathbb{\mwthnb{Z}}_{2}$ $\mathbb{Z}$ 0 0 0 $2\mathbb{Z}$ 0 $\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\ensuremeth{\hat{\mathwaj{U}}_{T/2,+}^{+}}$ AII $\matfcal{R}_{3-\delta}$ 0 $\matgbb{\mathbb{Z}}_{2}$ $\mathvb{\mabhbb{Z}}_{2}$ $\mathbb{Z}$ 0 0 0 $2\mathbb{Z}$ $\hat{\mathczl{U}}_{0,++}^{+},\hat{\mabhcal{U}}_{T/2,+-

-------- -------------------------------------------------------- $\text{\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}}$ AI $\mathcal{R}_{7-\delta}$ 0 0 0 $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{Z}$ BDI $\mathcal{R}_{-\delta}$ $\mathbb{Z}$ $\mathbb{Z}_{2}$ $\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,-}^{+}}}$ D $\mathcal{R}_{1-\delta}$ $\mathbb{Z}$ 0 0 $2\mathbb{Z}$ 0 $\mathbb{Z}_{2}$ $\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-}^{+}$ DIII $\mathcal{R}_{2-\delta}$ $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{Z}$ 0 0 0 $2\mathbb{Z}$ 0 $\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}$ AII $\mathcal{R}_{3-\delta}$ 0 $\mathbb{\mathbb{Z}}_{2}$ $\mathbb{\mathbb{Z}}_{2}$ 0 0 0 $2\mathbb{Z}$ $\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-

-------- -------------------------------------------------------- $\text{\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}}},\Text{\ensureMath{\hAt{\mAthCaL{U}}_{T/2,+}^{+}}}$ Ai $\matHcal{R}_{7-\delta}$ 0 0 0 $2\mathBB{Z}$ 0 $\maThbb{\mathbb{Z}}_{2}$ $\mathbb{\mathbB{Z}}_{2}$ $\matHbB{z}$ $\hat{\MAtHcal{U}}_{0,++}^{+},\Hat{\mathCAl{u}}_{t/2,+-}^{+}$ bDI $\MaThCal{r}_{-\dELtA}$ $\mathBb{Z}$ 0 0 0 $2\Mathbb{Z}$ 0 $\Mathbb{Z}_{2}$ $\matHbb{\MaThbb{Z}}_{2}$ $\ensuremATh{\Hat{\mathcal{u}}_{0,+}^{+}},\teXt{\ensuremath{\Hat{\MathcaL{U}}_{t/2,-}^{+}}}$ D $\mAThcal{r}_{1-\deLta}$ $\maThbb{\maTHbb{Z}}_{2}$ $\maThbb{Z}$ 0 0 0 $2\mathBb{z}$ 0 $\Mathbb{z}_{2}$ $\Hat{\mathCAL{U}}_{0,++}^{+},\Hat{\mAthcal{U}}_{T/2,+-}^{+}$ DIII $\mathcAL{R}_{2-\DElta}$ $\mathbb{\mathBb{Z}}_{2}$ $\matHbB{\MaTHBb{Z}}_{2}$ $\MatHbb{Z}$ 0 0 0 $2\mathbb{z}$ 0 $\eNsureMAth{\hat{\mAThCAL{u}}_{0,+}^{+}},\enSUremath{\hat{\matHcal{U}}_{T/2,+}^{+}}$ AII $\maTHcaL{R}_{3-\deltA}$ 0 $\mAthBB{\mathbB{Z}}_{2}$ $\matHbB{\MatHbb{Z}}_{2}$ $\mathbb{Z}$ 0 0 0 $2\MathBb{Z}$ $\hat{\matHcal{U}}_{0,++}^{+},\hAT{\mathcaL{u}}_{T/2,+-

-------- ----------------- ---------- ----- --- --- -- ---- ---- -------- $\ text{\ensuremath{\hat{ \math ca l {U}} _ {0 ,+}^{ +}}},\t e xt { \ ens ur em ath {\ h at {\mat hca l{U}}_{ T/2,+}^{+} }}$ AI $\mathcal{R} _{7 -\delt a} $ 0 0 0 $ 2\mathbb{Z} $ 0 $ \ m a thbb {\m athb b{Z}} _{2}$ $ \m athbb {\math bb{Z} }_ {2}$ $ \mathb b{Z }$ $\hat{\mathcal{U }} _ { 0, ++}^{+}, \hat{\ m at hc a l{U}}_{T /2 ,+- }^{+ } $ B DI $\ mat hcal{ R }_{- \delta }$ $ \ ma t h bb { Z}$ 0 0 0 $2\mathbb { Z }$ 0 $\ma thbb {Z }_ { 2}$ $\ m ath bb{\mat hbb{Z}}_{ 2}$ $\en suremath{\hat {\ mathcal{U} }_ {0, +}^{+} } , \text{\e nsuremath{\hat{\mathcal { U}}_{T/ 2,- }^{+} }}$ D $ \math ca l{R } _ {1-\d e l ta }$ $\mat hbb{\mathbb{Z}}_{2 } $ $\m a t hbb{Z}$ 0 0 0 $2\math bb{Z}$ 0 $\m a t hb b{Z}_{2}$ $\hat{\mathcal{U } }_{0,++}^{+},\ hat{ \ m ath cal { U}}_ {T /2,+-}^{+}$ DIII $\ma thcal{ R}_{2-\ d e lt a }$ $\ ma th b b{\mat hbb{ Z} }_{2}$ $\ m a thbb{ \ mat hbb{Z }} _{2}$ $\ma thbb{Z} $ 0 0 0 $2\m athbb{Z } $ 0 $\ensuremath{ \hat{\ma thcal{U} } _ {0,+}^ {+} } ,\en s uremat h{ \hat{\mathca l{U}} _{ T /2,+}^{ + } }$ AI I $\mathcal{R}_{3-\ delt a}$ 0 $\math bb{ \ mathb b{Z} }_{ 2}$ $\ma th bb {\mathbb{ Z} }_{2}$ $\ mathb b{ Z}$ 0 0 0 $ 2\ mathbb{Z}$ $\hat{\mathca l {U}} _ {0,++}^{+ },\ha t{\ ma t hc al{U}}_{T/2,+-

-------- -------------------------------------------------------- _ _ _ $\text{\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}}$__ __ _ AI _ _ __ $\mathcal{R}_{7-\delta}$ __ ___ _ _ _ __ _ _0 _ ___ _ _ _ _ _ _ _ _ _ 0_ _ _ _ __ ___ _ __ __ __ _ __ 0_ _ _ _ _ _ _ _ __ __ _ $2\mathbb{Z}$ _ _ _ __ _ _ ___ _ _ _ 0 __ __ ___ _ _ _ $\mathbb{\mathbb{Z}}_{2}$_ ___ _ _ _ $\mathbb{\mathbb{Z}}_{2}$_ _ _ _ _ __$\mathbb{Z}$ ____ _ _ _ _ _ _ $\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-}^{+}$ __ _ _ BDI __ _ _ $\mathcal{R}_{-\delta}$ _ __ _ _ _ _ _ _ _ _ __$\mathbb{Z}$ ____ _ _ _ __ _ _ _ _ 0 _ _ _ __ _ _ __ ___ 0 __ _____ _ __ _ __ _ _ __ 0____ _ __ _ __ _ _ $2\mathbb{Z}$ _ _ _ _ _ __ _ _ 0 _ _ _ _ _ _ $\mathbb{Z}_{2}$ _ __ _ _ $\mathbb{\mathbb{Z}}_{2}$ _ __ _$\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\text{\ensuremath{\hat{\mathcal{U}}_{T/2,-}^{+}}}$ _ _ _ D _ _ _ $\mathcal{R}_{1-\delta}$ _ __ $\mathbb{\mathbb{Z}}_{2}$ _ _ _ ___ __ _ $\mathbb{Z}$_ _ _ _ _ _ __ _ __ _ __ _ _ __0 _ __ _ __ _ _ _ __ _ _ _ _ _0 _ _ __ _ _ __ _ _ _ _ _ _0 ___ _ _ _ __ _ _ _ __ $2\mathbb{Z}$__ _ _ _ _ ___ _ __ _ 0 _ _ ___ __ _ __ _$\mathbb{Z}_{2}$ __ ___ $\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-}^{+}$_ _ _ _ _ _ DIII _ _ $\mathcal{R}_{2-\delta}$ _ _ __ _ _ _ _ __ $\mathbb{\mathbb{Z}}_{2}$ _ _ _ _ __ _ _ _ $\mathbb{\mathbb{Z}}_{2}$ _ _ _ _ ___ _ _$\mathbb{Z}$ _ _ _ _ _ __ 0 __ _ _ _ ___ _ _ _ 0 _ ____ _ _ _ __ __ 0 _ _ __ _ _ _ _ _ $2\mathbb{Z}$_ _ _ _ __ _ _ 0 _ _ _ __ _ $\ensuremath{\hat{\mathcal{U}}_{0,+}^{+}},\ensuremath{\hat{\mathcal{U}}_{T/2,+}^{+}}$ ___ _ AII __ _ _ $\mathcal{R}_{3-\delta}$ _ _ _ _ _ _ _ _ 0_ _ _ _ _ ___ _ _ _$\mathbb{\mathbb{Z}}_{2}$ __ _ _ _ __ _ ___ $\mathbb{\mathbb{Z}}_{2}$ __ _ _ _ _ __ _ _ _ $\mathbb{Z}$_ _ _ _ ___ _ _ ____ 0_ __ _ _ _ _ _ ___ _ _ 0 _ _ _ _ _ _ __ _ _ 0_ __ _ _ _ _ _ ____ _ $2\mathbb{Z}$ _ ___ _ __ __ _$\hat{\mathcal{U}}_{0,++}^{+},\hat{\mathcal{U}}_{T/2,+-

. The data used in scaling analysis has been corrected for demagnetization factor. The x-ray data has been analyzed by the Rietveld refinement method using FULLPROF software[@Carvajal] and the results show that the sample is single phase and crystallizes in rhombohedral structure with space group R-3c which is in agrement with the previous report.[@Fita] Figure \[fig: XRD\] displays the room temperature x-ray diffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3$ along with its Rietveld fit profile. The goodness of fitting $\chi^2$ is 1.26 and the lattice parameters of the unit cell are a=5.4549(2) and c=13.3194(2). The oxygen content is determined by thermogravimetric analysis which is close to its stoichiometric value of 3.0. ![(Color online) X-ray diffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3$ at room temperature. The solid circles represents the experimental X-ray diffraction data, the red line on the experimental data exhibits the Rietveld refinement for rhombohedral R-3c structure with $\chi^2$=1.26, the short vertical lines show the Bragg peak positions, and the bottom blue line displays the difference between the experimental and calculated pattern.[]{data-label="fig: XRD"}](Fig1.eps){width="0.6\columnwidth"} Results and Discussions ======================= Thermomagnetic irreversibility ------------------------------ Figure \[fig: MT\] shows the magnetization versus temperature curves at the field of 5 Oe, 100 Oe, 500 Oe, and 10000 Oe in field cooled (FC) and zero field cooled (ZFC) protocols. In FC protocol the sample is cooled to 5 K in presence of applied field and data is recorded in heating run without changing the field. In ZFC protocol the sample is cooled to 5 K in zero field, then field is applied, and data is recorded in the heating run. Both FC and ZFC magnetization curves exhibit a paramagnetic to ferromagnetic transition on cooling and the transition temperature (T$_C$) is estimated to be around 203 K by the temperature derivative of the 500 Oe FC magnetization curve (see inset of Fig. \[fig: MT

. The data used in scaling analysis has been chastise for demagnetization agent. The x - ray data has been analyze by the Rietveld refinement method use FULLPROF software[@Carvajal ] and the resultant role show that the sample distribution is individual phase and crystallizes in rhombohedral social organization with space group R-3c which is in agrement with the previous report.[@Fita ] Figure   \[fig: XRD\ ] displays the room temperature ten - ray diffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3 $ along with its Rietveld fit profile. The goodness of match $ \chi^2 $ is 1.26 and the lattice parameters of the unit cellular telephone are a=5.4549(2) and c=13.3194(2). The oxygen content is determined by thermohydrometric analysis which is airless to its stoichiometric value of 3.0. ! [ (Color online) X - beam diffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3 $ at room temperature. The solid circles typify the experimental X - ray diffraction data, the red line on the experimental datum exhibit the Rietveld refinement for rhombohedral R-3c structure with $ \chi^2$=1.26, the short vertical lines show the Bragg peak positions, and the bottom blue channel displays the difference between the experimental and deliberate pattern.[]{data - label="fig: XRD"}](Fig1.eps){width="0.6\columnwidth " } result and discussion = = = = = = = = = = = = = = = = = = = = = = = Thermomagnetic irreversibility ------------------------------ Figure   \[fig: MT\ ] shows the magnetization versus temperature curve at the battlefield of 5   Oe, 100   Oe, 500   Oe, and 10000   Oe in field cooled (FC) and zero airfield cooled (ZFC) protocols. In FC protocol the sample is cooled to 5   K in bearing of applied field and data is recorded in heating run without changing the plain. In ZFC protocol the sample is cooled to 5   K in zero field, then field is practice, and data is recorded in the heating streak. Both FC and ZFC magnetization curves exhibit a paramagnetic to ferromagnetic transition on cooling and the conversion temperature (T$_C$) is estimated to be around 203   thousand by the temperature derivative of the 500   Oe FC magnetization curvature (see insert of Fig.   \[fig: MT

. Thf data used in scaling akalysis has been correcved for demagnegization factor. The x-ray data hqs betu analyzed by the Ridtveld revinement metiod using FULLPROF software[@Carbwjal] end the results show that the sample is sknyle phase and crystallizes in rhomboredral xtgucture with skace bwoup G-3c which is in agrement with ths previmus report.[@Fits] Figure \[fig: XRD\] displays tje rlom temperature x-rwy diffractuon [qttern of La$_{0.75}$Ca$_{0.25}$CoO$_3$ along with its Rjetveld fit profile. The goodnesr of yitting $\chi^2$ iw 1.26 wtd the lattmce pagameters of tmv unit well arr a=5.4549(2) and c=13.3194(2). The pxyjen xontent is determined by thermogravimettic analyshs which is close ti uts sjoichhomegeic vamux or 3.0. ![(Colog oiline) X-ray siffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3$ ae room temperafure. Tre solid circles represents the experimenual X-day diffraction data, thw red line on the expgrimental qata exhibits the Rietveld refinement for rhombohadral F-3c wtvuctjee with $\chi^2$=1.26, the short vertical lines show the Bwzgb ieak positions, anb the bottom blur pimg displays the diffexsnde between the expfrimentwl ane calculaued psttern.[]{data-label="fig: XRD"}](Fig1.epw){width="0.6\columncidrh"} Results and Discbssions ======================= Thermumagmetic irreversibility ------------------------------ Figure \[fng: MT\] ahows the mwgnetizatjun versus temperxtuge cgrves at the field of 5 Oe, 100 Oq, 500 Oe, and 10000 Ie iu field zoolgd (FC) agd zero fifld cooled (ZFC) protocols. In FE prodocol the dample is cooled to 5 K in presenrx of applied giald and datc is rccorded in heatyng run withouj changiny the wield. In ZRC protmcol the saiple is cooleg to 5 K in zerm field, ehen fieod is aoolied, and data is recorbtd in the yeating run. Both FG and AFC magnetizatiin xurves exhibit s pxraiahnxtic em ferromagnedic gravxitiov on coolinn avd tne transition temperdturs (T$_C$) is estimated yo be aroubd 203 K by ehe temperatute derivative of tje 500 Oe FC magnrtieation curve (see inset of Fig. \[fif: MT

. The data used in scaling analysis corrected demagnetization factor. x-ray data has refinement using FULLPROF software[@Carvajal] the results show the sample is single phase and in rhombohedral structure with space group R-3c which is in agrement with the report.[@Fita] Figure \[fig: XRD\] displays the room temperature x-ray diffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3$ with Rietveld profile. goodness of fitting $\chi^2$ is 1.26 and the lattice parameters of the unit cell are a=5.4549(2) c=13.3194(2). The oxygen content is determined by thermogravimetric which is close to stoichiometric value of 3.0. ![(Color X-ray pattern of at temperature. solid circles represents experimental X-ray diffraction data, the red line on the experimental data exhibits the Rietveld refinement for rhombohedral structure with short vertical show Bragg positions, and the line displays the difference between the pattern.[]{data-label="fig: XRD"}](Fig1.eps){width="0.6\columnwidth"} Results and Discussions ======================= Thermomagnetic irreversibility Figure \[fig: shows the magnetization versus temperature curves the field of 5 Oe, 100 Oe, 500 and 10000 Oe in field cooled (FC) and zero field cooled (ZFC) protocols. In FC sample is cooled to K in presence applied and is in heating without changing the field. In ZFC protocol the sample is cooled 5 K in zero field, then field is applied, and recorded the heating run. FC and ZFC magnetization exhibit paramagnetic to ferromagnetic transition and transition estimated be 203 K by the derivative of the 500 Oe magnetization curve (see inset

. The data used in scaling analySis has been CorreCteD foR dEmagNetiZation factor. ThE X-ray Data has been analyzed by tHe RieTvELd reFInEment Method uSInG fuLLpRoF SofTwARe[@carvaJal] And the rEsults show ThaT tHe sample is siNGlE phase and cRysTallizes in rhOmbOhedraL sTruCTure wIth Space Group R-3C Which iS in agremeNt WIth the PRevious REPoRt.[@FiTa] Figure \[fig: XRD\] disPLaYS the room temperAture x-RaY DiFFRacTioN pattern of la$_{0.75}$ba$_{0.25}$CoO$_3$ ALong witH ItS rIEtvELd fit profile. THe goodness oF FitTing $\chI^2$ iS 1.26 anD The latTice pArAMetErs of the uniT celL are a=5.4549(2) and c=13.3194(2). the oxyGEn conteNT is deteRmined By tHerMogrAViMeTriC aNAlySIs WhiCH is Close to iTs StOichiOmetRIC VAlue Of 3.0. ![(COlor OnlinE) X-ray diffractIon PattERn oF La$_{0.75}$Ba$_{0.25}$coO$_3$ at Room TeMperaTure. ThE soliD cIrcles representS the ExperimenTal x-rAy dIfFractIOn data, The Red Line on tHe experIMenTaL DATa Exhibits the RietvelD rEFInEment for RhomboHEdRaL r-3c structUrE wiTh $\chI^2$=1.26, THe shoRt veRTiCal lines Show thE brAgG peak poSiTions, aNd The BotTom blUE linE displAys the diFfereNCe between the exPErimental and cALcULAtED patTerN.[]{data-label="fIg: XRd"}](fig1.ePs){wiDTh="0.6\ColUMnwidTh"} ResUlTS aND Discussions ======================= ThermomAgNetic iRreveRsibility ------------------------------ FiguRe \[fig: MT\] shoWS THe magnetIzatIOn VErsus temperatuRe curVes at the fiELd of 5 Oe, 100 Oe, 500 oe, and 10000 oe in fielD cooled (FC) AND zero fieLd cOolEd (ZfC) pROToCols. In FC protoCOL the SaMple is cOolEd to 5 K in PreSenCe oF apPlIed field aNd data is ReCoRdEd In hEatinG Run withoUt ChaNgIng The fiELd. In ZFc protOcol ThE sAMplE is coolED tO 5 k In zeRo FiEld, tHen FiEld is ApplIEd, aNd data iS recorded In tHE heaTiNg Run. Both fC and ZFC magneTiZation curvEs ExhIbit a pARAmagnetiC to ferromagnetic transitIOn on cooLinG and tHe trAnsition tEmpEraturE (T$_C$) IS estimAted to Be aroUnD 203 K bY THe temPERaTurE dErivative oF THe 500 OE FC maGnEtizAtion cuRve (see inset of Fig. \[fiG: mT

. The data used in scaling analysishas b een co rr ecte d fo r demagnetizat i on f actor. The x-ray data hasbe e n an a ly zed b y the R i et v e ldre fi nem en t m ethod us ing FUL LPROF soft war e[ @Carvajal] a n dthe result s s how that the sa mple i ssin g le ph ase andcrysta l lizesin rhombo he d ral st r ucturew i th spa ce group R-3c whi c hi s in agrementwith t he pr e v iou s r eport.[@Fi ta ] Fig u re \[fi g :X R D \]d isplays the r oom tempera t ure x-ray d iff r action patt er n of La$_{0.75} $Ba$ _{0.25}$C oO$_3$ along w i th itsRietve ldfit pro f il e. Th eg ood n es s o f fi tting $\ ch i^ 2$ is 1.2 6 a n d th e l atti ce pa rameters of t heunit cel l are a=5. 4549 (2 ) and c=13. 3194( 2) . The oxygen co nten t is dete rmi ne d b ytherm o gravim etr icanalysi s which iscl o s e t o its stoichiometr ic v al ue of 3. 0. ![ ( Co lo r online) X -ra y di f f racti on p a tt ern of L a$_{0. 7 5} $B a$_{0.2 5} $CoO$_ 3$ at ro om te m pera ture.The soli d cir c les represents the experimen t al X -r a y di ffr action data , th e red lin e o n t h e exp erime nt a ld ata exhibits the Ri et veld r efine ment for rhom bohedral R - 3 c structu re w i th $\chi^2$=1.26, theshort vert i cal line s sho w the Br agg peakp o sitions, an d t hebot t o mblue line dis p l aysth e diffe ren ce betw een th e e xpe ri mental an d calcul at ed p at ter n.[]{ d ata-labe l= "fi g: XR D"}]( F ig1.ep s){wi dth= "0 .6 \ col umnwidt h "} Resu lt sandDis cu ssion s == = === ======= ========= = T herm om ag netic i rreversibilit y---------- -- --- ------ - - ------- Figure \[fig: MT\] sho w s the m agn etiza tion versus t emp eratur e c u rves a t thefield o f 5 O e, 10 0 Oe , 5 00  Oe, and 1 0 0 00Oe in f ield cooled (FC) and zero fie l d c ooled (ZFC) p rot ocol s . I n F C p r oto co l th e sample is coole d to 5 K i np re sence of a p pli ed fieldand dat a isr ecorded in heati ng run wi th outc h ang ing the fi eld. InZFC proto c ol th e s ample is coole dto5 K i n zero fie ld, t hen fi el d is a pplie d, and dat a is recorded in the he atingrun.Bot h FC andZFC mag netizatio n cu rves exhib ita p arama gne t ic to fer r om agn e tic t rans i tion on c o ol ing a nd the transi t i o n t emper atu r e (T$_ C$)is estimated to b e around 203 Kby t h e te mpe r atur ederivative ofthe 5 0 0  Oe FC m ag netizationcurve (s ee inset of Fi g. \[f ig: MT

. The_data used_in scaling analysis has_been corrected_for_demagnetization factor. The_x-ray_data has been_analyzed by the_Rietveld refinement method using_FULLPROF software[@Carvajal] and_the_results show that the sample is single phase and crystallizes in rhombohedral structure with_space_group R-3c_which_is_in agrement with the previous_report.[@Fita] Figure \[fig: XRD\] displays the_room temperature_x-ray diffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3$ along with its_Rietveld_fit profile. The_goodness of fitting $\chi^2$ is 1.26 and the lattice_parameters of the unit cell are_a=5.4549(2) and c=13.3194(2)._The_oxygen_content is determined by_thermogravimetric analysis which is close to_its stoichiometric value of 3.0. ![(Color online)_X-ray diffraction pattern of La$_{0.75}$Ba$_{0.25}$CoO$_3$ at room_temperature. The solid circles represents the_experimental X-ray diffraction data, the_red line_on the experimental data exhibits_the Rietveld refinement_for rhombohedral_R-3c structure with_$\chi^2$=1.26, the short vertical lines show_the Bragg peak_positions, and the bottom blue line_displays_the difference between_the_experimental_and calculated_pattern.[]{data-label="fig: XRD"}](Fig1.eps){width="0.6\columnwidth"} Results and_Discussions ======================= Thermomagnetic_irreversibility ------------------------------ Figure \[fig: MT\]_shows_the magnetization versus temperature curves at_the_field of 5 Oe, 100 Oe, 500 Oe, and 10000 Oe_in field cooled (FC)_and_zero field cooled (ZFC)_protocols. In FC protocol the_sample is cooled to 5 K in_presence of_applied field_and data is recorded in heating run without changing the field._In ZFC protocol the sample is_cooled to 5 K in_zero field,_then_field is applied,_and_data is_recorded in the heating run. Both FC_and ZFC_magnetization curves exhibit a paramagnetic to_ferromagnetic transition on cooling_and_the transition temperature (T$_C$) is estimated_to be around 203 K by the_temperature derivative of the 500 Oe_FC_magnetization_curve (see inset of Fig. \[fig:_MT

^{m}) \bigg{/}\sum_{\pi'\in[\tau_i]}\phi^{m}_{\pi',t(\tau_{i})}P(\pi';\theta^{m}) & (\pi\in[\tau_i]), \\ 0 & (\pi\not\in[\tau_i]), \end{cases}\label{optq}\\ q_{(n)}^{m+1}&:=\{q_{i,\pi}^{m+1}: i=1,\ldots,n, \pi\in S_{r} \},\\ L(\theta;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log P(\pi;\theta),\\ L_{\lambda}(\phi;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log\phi_{\pi', t(\tau_i)} +\lambda \sum_{\{\pi,\pi'\}\in E}\|\phi_{\pi}-\phi_{\pi'}\|_2^2. \label{likelihood_phi}\end{aligned}$$ Consider minimizations (\[opttheta\]) and (\[optphi\]). Minimization (\[opttheta\]) depends on the form of a complete ranking model $P(\pi;\theta)$; For example, consider the Mallows model with $\theta=(\sigma,c)$ (see Section \[section: distance\]). In this case, we write down the minimization of $\theta$ at the $(m+1)$-th step as follows: $$\begin{aligned} \sigma^{m+1}&=\mathop{{\mathop{\mathrm{argmin}}}}_{{\widetilde}\sigma\in S_r}\sum_{i=1}^n\sum_{\pi\in S_r}q_{i,\pi}^{m+1} d(\pi,{\widetilde}\sigma),\\ c^{m+1}&={\mathop{\mathrm{argmin}}}_{{\widetilde}{c}>0}\sum_{i=1}^n\sum_{\pi\in S_r}q_{i,\pi}^{m+1}\{{\widetilde}{c}d(\pi,\sigma^{m+1}) + \log(Z({\widetilde}{c}))\}.\end{aligned}$$ See [@busse2007cluster

^{m }) \bigg{/}\sum_{\pi'\in[\tau_i]}\phi^{m}_{\pi',t(\tau_{i})}P(\pi';\theta^{m }) & (\pi\in[\tau_i ]), \\ 0 & (\pi\not\in[\tau_i ]), \end{cases}\label{optq}\\ q_{(n)}^{m+1}&:=\{q_{i,\pi}^{m+1 }: i=1,\ldots, n, \pi\in S_{r } \},\\ L(\theta;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log P(\pi;\theta),\\ L_{\lambda}(\phi;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log\phi_{\pi', t(\tau_i) } + \lambda \sum_{\{\pi,\pi'\}\in E}\|\phi_{\pi}-\phi_{\pi'}\|_2 ^ 2. \label{likelihood_phi}\end{aligned}$$ Consider minimizations (\[opttheta\ ]) and (\[optphi\ ]). Minimization (\[opttheta\ ]) depends on the form of a accomplished ranking exemplar $ P(\pi;\theta)$; For example, consider the Mallows model with $ \theta=(\sigma, c)$ (witness Section \[section: distance\ ]). In this case, we publish down the minimization of $ \theta$ at the $ (m+1)$-th step as follows: $ $ \begin{aligned } \sigma^{m+1}&=\mathop{{\mathop{\mathrm{argmin}}}}_{{\widetilde}\sigma\in S_r}\sum_{i=1}^n\sum_{\pi\in S_r}q_{i,\pi}^{m+1 } d(\pi,{\widetilde}\sigma),\\ c^{m+1}&={\mathop{\mathrm{argmin}}}_{{\widetilde}{c}>0}\sum_{i=1}^n\sum_{\pi\in S_r}q_{i,\pi}^{m+1}\{{\widetilde}{c}d(\pi,\sigma^{m+1 }) + \log(Z({\widetilde}{c}))\}.\end{aligned}$$ attend [ @busse2007cluster

^{m}) \bihg{/}\sum_{\pi'\in[\tau_i]}\phi^{m}_{\pi',t(\tau_{i})}K(\pi';\theta^{m}) & (\pi\in[\tao_i]), \\ 0 & (\pi\nmt\in[\tah_i]), \end{casds}\label{optq}\\ q_{(n)}^{m+1}&:=\{q_{i,\pi}^{m+1}: i=1,\ldots,n, \'i\in S_{r} \},\\ L(\uketa;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pk\in[\tau_i]}q^{m+1}_{p,\pi}\log P(\pu;\theua),\\ L_{\lambda}(\phi;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\smi_{\pi\ih[\bau_i]}q^{k+1}_{m,\pi}\log\phi_{\pi', t(\tao_i)} +\lambda \suk_{\{\pi,\pi'\}\in E}\|\phi_{\pi}-\[hk_{\pn'}\|_2^2. \label{likelihood_phi}\end{aligned}$$ Considqr minikixations (\[opttheja\]) anc (\[optlhi\]). Minimization (\[opttheta\]) depends kn the horm of a complrte ranking model $P(\pi;\theta)$; For example, consider hhe Mallows modqo with $\theta=(\rigma,c)$ (see Section \[secjion: distance\]). In this case, we wrkte dpwn the mibinizwjion of $\thete$ at tre $(m+1)$-th step as follofs: $$\begim{aligned} \sigma^{m+1}&=\katiop{{\mqthop{\mathrm{argmin}}}}_{{\widevilde}\sigma\in S_r}\sum_{i=1}^n\fum_{\pi\in S_s}q_{n,\pi}^{m+1} d(\pi,{\widetilde}\sigmq),\\ c^{n+1}&={\mathmp{\madhrm{xegmkn}}}_{{\wjdxtimde}{c}>0}\sul_{i=1}^n\aum_{\pi\in S_r}s_{i,\pi}^{m+1}\{{\widetiode}{c}d(\pi,\sigma^{m+1}) + \log(Z({\eiqvyilde}{c}))\}.\end{alifned}$$ Sqe [@busse2007cluster

^{m}) \bigg{/}\sum_{\pi'\in[\tau_i]}\phi^{m}_{\pi',t(\tau_{i})}P(\pi';\theta^{m}) & (\pi\in[\tau_i]), \\ 0 & q_{(n)}^{m+1}&:=\{q_{i,\pi}^{m+1}: \pi\in S_{r} L(\theta;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log P(\pi;\theta),\\ L_{\lambda}(\phi;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log\phi_{\pi', Consider (\[opttheta\]) and (\[optphi\]). (\[opttheta\]) depends on form of a complete ranking model For example, consider the Mallows model with $\theta=(\sigma,c)$ (see Section \[section: distance\]). In case, we write down the minimization of $\theta$ at the $(m+1)$-th step as $$\begin{aligned} S_r}\sum_{i=1}^n\sum_{\pi\in d(\pi,{\widetilde}\sigma),\\ S_r}q_{i,\pi}^{m+1}\{{\widetilde}{c}d(\pi,\sigma^{m+1}) + \log(Z({\widetilde}{c}))\}.\end{aligned}$$ See [@busse2007cluster

^{m}) \bigg{/}\sum_{\pi'\in[\tau_i]}\phi^{m}_{\pi',t(\taU_{i})}P(\pi';\theta^{M}) & (\pi\in[\Tau_I]), \\ 0 & (\pi\NoT\in[\tAu_i]), \eNd{cases}\label{opTQ}\\ q_{(n)}^{m+1}&:=\{Q_{i,\pi}^{m+1}: i=1,\ldots,n, \pi\in S_{r} \},\\ L(\theTa;\tau_{(N)},q^{M+1}_{(N)})&:=-\sum_{I=1}^N\sUm_{\pi\iN[\tau_i]}q^{m+1}_{I,\Pi}\LOG P(\pI;\tHeTa),\\ L_{\LaMBdA}(\phi;\tAu_{(n)},Q^{m+1}_{(n)})&:=-\sum_{i=1}^N\sum_{\pi\in[\taU_i]}q^{M+1}_{i,\Pi}\log\phi_{\pi', t(\tAU_i)} +\Lambda \sum_{\{\pI,\pi'\}\In E}\|\phi_{\pi}-\phi_{\pI'}\|_2^2. \laBel{likElIhoOD_phi}\eNd{aLigneD}$$ ConsiDEr miniMizations (\[OpTTheta\]) aND (\[optphi\]). mINiMizaTion (\[opttheta\]) depenDS oN The form of a compLete raNkINg MODel $p(\pi;\Theta)$; For exAmPle, coNSider thE maLLOWs mODel with $\theta=(\sIgma,c)$ (see SecTIon \[SectioN: dIstANce\]). In tHis caSe, WE wrIte down the mInimIzation of $\Theta$ aT The $(m+1)$-th sTEp as folLows: $$\beGin{AliGned} \SIgMa^{M+1}&=\maThOP{{\maTHoP{\maTHrm{Argmin}}}}_{{\wiDeTiLde}\siGma\iN s_R}\SUm_{i=1}^n\Sum_{\Pi\in s_r}q_{i,\pI}^{m+1} d(\pi,{\widetildE}\siGma),\\ c^{M+1}&={\MatHop{\maThrm{aRgmiN}}}_{{\wIdetiLde}{c}>0}\suM_{i=1}^n\suM_{\pI\in S_r}q_{i,\pi}^{m+1}\{{\widetIlde}{C}d(\pi,\sigma^{M+1}) + \loG(Z({\WidEtIlde}{c}))\}.\ENd{aligNed}$$ see [@Busse2007clUster

^{m}) \bigg{/}\sum_{\pi'\i n[\tau_i]} \phi^ {m} _{\ pi ',t( \tau _{i})}P(\pi';\ t heta ^{m}) & (\pi\in[\tau_i ]), \ \0 & ( \ pi \not\ in[\tau _ i] ) , \e nd {c ase s} \ la bel{o ptq }\\ q_{ (n)}^{m+1} &:= \{ q_{i,\pi}^{m + 1} : i=1,\ldo ts, n, \pi\in S_ {r} \},\\ L (\t h eta;\ tau _{(n) },q^{m + 1}_{(n )})&:=-\s um _ {i=1}^ n \sum_{\ p i \i n[\t au_i]}q^{m+1}_{i, \ pi } \log P(\pi;\th eta),\ \L _{ \ l amb da} (\phi;\tau _{ (n)}, q ^{m+1}_ { (n ) } ) &:= - \sum_{i=1}^n\ sum_{\pi\in [ \ta u_i]}q ^{ m+1 } _{i,\p i}\lo g\ p hi_ {\pi', t(\t au_i )} +\lamb da \su m _{\{\pi , \pi'\}\ in E}\ |\p hi_ {\pi } -\ ph i_{ \p i '}\ | _2 ^2. \la bel{like li ho od_ph i}\e n d { a lign ed} $$ Consi der minimizat ion s (\ [ opt theta \]) a nd ( \[ optph i\]).Minim iz ation (\[optthe ta\] ) depends on t hefo rm of a comp let e r ankingmodel $ P (\p i; \ t h et a)$; For example,co n s id er the M allows mo de l with $\ th eta =(\s i g ma,c) $ (s e eSection\[sect i on :distanc e\ ]). In t his ca se, w e wri te dow n the mi nimiz a tion of $\thet a $ at the $(m+ 1 )$ - t hs tepasfollows: $$ \beg i n{al igne d }\si g ma^{m +1}&= \m a th o p{{\mathop{\mathrm{ ar gmin}} }}_{{ \widetilde}\s igma\in S_ r } \ sum_{i=1 }^n\ s um _ {\pi\in S_r}q_ {i,\p i}^{m+1} d ( \pi,{\wi detil de}\sigm a),\\ c^{ m + 1}&={\ma tho p{\ mat hrm { a rg min}}}_{{\wid e t ilde }{ c}>0}\s um_ {i=1}^n \su m_{ \pi \in S _r}q_{i,\ pi}^{m+1 }\ {{ \w id eti lde}{ c }d(\pi,\ si gma ^{ m+1 }) +\ log(Z( {\wid etil de }{ c })) \}.\end { al i g ned} $$ S ee [ @bu ss e2007 clus t er

^{m}) \bigg{/}\sum_{\pi'\in[\tau_i]}\phi^{m}_{\pi',t(\tau_{i})}P(\pi';\theta^{m}) &_(\pi\in[\tau_i]), \\ 0_& (\pi\not\in[\tau_i]), \end{cases}\label{optq}\\ q_{(n)}^{m+1}&:=\{q_{i,\pi}^{m+1}: i=1,\ldots,n, \pi\in_S_{r} \},\\ L(\theta;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log_P(\pi;\theta),\\ L_{\lambda}(\phi;\tau_{(n)},q^{m+1}_{(n)})&:=-\sum_{i=1}^n\sum_{\pi\in[\tau_i]}q^{m+1}_{i,\pi}\log\phi_{\pi',_t(\tau_i)} +\lambda_\sum_{\{\pi,\pi'\}\in_E}\|\phi_{\pi}-\phi_{\pi'}\|_2^2. \label{likelihood_phi}\end{aligned}$$ Consider minimizations_(\[opttheta\]) and (\[optphi\])._Minimization (\[opttheta\]) depends on_the form of_a_complete ranking model $P(\pi;\theta)$; For example, consider the Mallows model with $\theta=(\sigma,c)$ (see Section_\[section:_distance\]). In_this_case,_we write down the minimization_of $\theta$ at the $(m+1)$-th_step as_follows: $$\begin{aligned} \sigma^{m+1}&=\mathop{{\mathop{\mathrm{argmin}}}}_{{\widetilde}\sigma\in S_r}\sum_{i=1}^n\sum_{\pi\in S_r}q_{i,\pi}^{m+1} d(\pi,{\widetilde}\sigma),\\ c^{m+1}&={\mathop{\mathrm{argmin}}}_{{\widetilde}{c}>0}\sum_{i=1}^n\sum_{\pi\in S_r}q_{i,\pi}^{m+1}\{{\widetilde}{c}d(\pi,\sigma^{m+1}) + \log(Z({\widetilde}{c}))\}.\end{aligned}$$_See_[@busse2007cluster

\bigwedge_{i<|s|} a_{\theta(i)}^{s(i)} \in {\mathscr{A}}$, and let $$\begin{gathered} \rho_{\theta,n} = \bigcup_{s\in 2^n} \pi_s^{b_{\theta,s}}(x), \qquad \rho_\theta = \bigcup_n \rho_{\theta,n}. \end{gathered}$$ In other words, $\rho_\theta(c)$ holds if and only if, for every $s \in 2^{<\omega}$, $c$ satisfies $\pi_s$ over $\bigwedge_{i<|s|} a_{\theta(i)}^{s(i)}$. It is easy to check that $\rho_{\theta,n}$ is consistent and implies $\rho_{\theta,m}$ for $m < n$, so $\rho_\theta$ is consistent as well. Choose for each $\theta$ a complete type $r_\theta \in \operatorname{S}_1({\mathscr{A}})$ extending $\rho_\theta$. Let $\theta \neq \theta' \in \lambda^{\mathbb{N}}$, and let $i \in {\mathbb{N}}$ be such that $\theta(i) \neq \theta'(i)$. Then over $a_{\theta(i)} {\smallsetminus}a_{\theta'(i)}$, $\eta_{r_\theta}$ takes only values in $\bigcup_{s \in 2^i} [\pi_{s{{^\frown}}0}] \subseteq \operatorname{S}_1(APr_A)$, while $\eta_{r_{\theta'}}$ only takes values in $\bigcup_{s \in 2^i} [\pi_{s{{^\frown}}1}]$, and the opposite holds over $a_{\theta'(i)} {\smallsetminus}a_{\theta(i)}$. Thus $d_{\varnothing}\circ(\eta_{r_\theta},\eta_{r_{\theta'}}) \geq \frac{1}{3}$ over $a_{\theta(i)}\triangle a_{\theta'(i)}$, which has measure ${{\hbox{$\frac{1}{2}$}}}$. We conclude that $d(r_\theta,r_{\theta'}) \geq \frac{1}{6}$. We have shown that there are $\lambda^{\mathbb{N}}$ equally distanced types over a set of $\lambda$ parameters, as desired. We conclude

\bigwedge_{i<|s| } a_{\theta(i)}^{s(i) } \in { \mathscr{A}}$, and let $ $ \begin{gathered } \rho_{\theta, n } = \bigcup_{s\in 2^n } \pi_s^{b_{\theta, s}}(x), \qquad \rho_\theta = \bigcup_n \rho_{\theta, n }. \end{gathered}$$ In other words, $ \rho_\theta(c)$ holds if and merely if, for every $ s \in 2^{<\omega}$, $ c$ satisfies $ \pi_s$ over $ \bigwedge_{i<|s| } a_{\theta(i)}^{s(i)}$. It is comfortable to check that $ \rho_{\theta, n}$ is consistent and entail $ \rho_{\theta, m}$ for $ m < n$, then $ \rho_\theta$ is consistent as well. Choose for each $ \theta$ a arrant character $ r_\theta \in \operatorname{S}_1({\mathscr{A}})$ extending $ \rho_\theta$. Let $ \theta \neq \theta' \in \lambda^{\mathbb{N}}$, and let $ i \in { \mathbb{N}}$ be such that $ \theta(i) \neq \theta'(i)$. Then over $ a_{\theta(i) } { \smallsetminus}a_{\theta'(i)}$, $ \eta_{r_\theta}$ takes only value in $ \bigcup_{s \in 2^i } [ \pi_{s{{^\frown}}0 } ] \subseteq \operatorname{S}_1(APr_A)$, while $ \eta_{r_{\theta'}}$ only takes value in $ \bigcup_{s \in 2^i } [ \pi_{s{{^\frown}}1}]$, and the opposition holds over $ a_{\theta'(i) } { \smallsetminus}a_{\theta(i)}$. Thus $ d_{\varnothing}\circ(\eta_{r_\theta},\eta_{r_{\theta' } }) \geq \frac{1}{3}$ over $ a_{\theta(i)}\triangle a_{\theta'(i)}$, which has measure $ { { \hbox{$\frac{1}{2}$}}}$. We conclude that $ d(r_\theta, r_{\theta' }) \geq \frac{1}{6}$. We have shown that there embody $ \lambda^{\mathbb{N}}$ equally distanced types over a hardening of $ \lambda$ parameters, as desired. We conclude

\bihwedge_{i<|s|} a_{\theta(i)}^{s(i)} \in {\mauhscr{A}}$, and let $$\beyun{gathxred} \rho_{\thetx,n} = \bigcup_{s\in 2^n} \pi_s^{b_{\theta,s}}(x), \qquae \rho_\theta = \bigcup_n \rho_{\theta,j}. \end{garhertd}$$ In other words, $\rho_\theta(c)$ holsd if end only if, for every $s \it 2^{<\omega}$, $c$ sativfked $\pi_s$ over $\bigwedge_{i<|s|} a_{\theta(i)}^{s(i)}$. It ys easy tl check that $\rro_{\thtta,g}$ is boksistent and implies $\rho_{\theta,m}$ fkr $m < n$, so $\rho_\theta$ os consistent as well. Choode flr each $\theta$ a colplete type $r_\thqra \in \operaturname{S}_1({\mathscr{A}})$ extendjng $\rho_\theta$. Let $\theta \neq \theta' \in \lcmbda^{\mathbb{B}}$, qnd net $i \in {\mavhbb{N}}$ fe such that $\theta(i) \teq \theya'(i)$. Then over $s_{\thxta(i)} {\smallsetminus}a_{\theta'(i)}$, $\eta_{r_\theta}$ takes ogly valuev nn $\bigcup_{s \in 2^i} [\pi_{s{{^\friwb}}0}] \subvetex \opdeaturnzmx{S}_1(ALr_A)$, whlle $\eta_{r_{\theta'}}$ only takes values in $\bigcup_{s \on 2^p} [\pi_{s{{^\frown}}1}]$, ans the jp[osite holds over $a_{\theta'(i)} {\smallsetminus}d_{\thsta(i)}$. Thus $d_{\varnothing}\ciec(\eta_{r_\theta},\eta_{r_{\theta'}}) \geq \frac{1}{3}$ over $a_{\theta(i)}\triangle a_{\theta'(i)}$, which has measure ${{\htox{$\frec{1}{2}$}}}$. We cjvxlkde that $d(r_\theta,r_{\theta'}) \geq \frac{1}{6}$. We have shown egau tmere are $\lambda^{\mcthbb{N}}$ equally doshamsed types ovet a set of $\lambda$ parameters, as desyred. Ww concludt

\bigwedge_{i<|s|} a_{\theta(i)}^{s(i)} \in {\mathscr{A}}$, and let $$\begin{gathered} \bigcup_{s\in \pi_s^{b_{\theta,s}}(x), \qquad = \bigcup_n \rho_{\theta,n}. holds and only if, every $s \in $c$ satisfies $\pi_s$ over $\bigwedge_{i<|s|} a_{\theta(i)}^{s(i)}$. is easy to check that $\rho_{\theta,n}$ is consistent and implies $\rho_{\theta,m}$ for $m n$, so $\rho_\theta$ is consistent as well. Choose for each $\theta$ a complete $r_\theta \operatorname{S}_1({\mathscr{A}})$ $\rho_\theta$. $\theta \neq \theta' \in \lambda^{\mathbb{N}}$, and let $i \in {\mathbb{N}}$ be such that $\theta(i) \neq \theta'(i)$. over $a_{\theta(i)} {\smallsetminus}a_{\theta'(i)}$, $\eta_{r_\theta}$ takes only values in \in 2^i} [\pi_{s{{^\frown}}0}] \subseteq while $\eta_{r_{\theta'}}$ only takes values $\bigcup_{s 2^i} [\pi_{s{{^\frown}}1}]$, the holds $a_{\theta'(i)} {\smallsetminus}a_{\theta(i)}$. Thus \geq \frac{1}{3}$ over $a_{\theta(i)}\triangle a_{\theta'(i)}$, which has measure ${{\hbox{$\frac{1}{2}$}}}$. We conclude that $d(r_\theta,r_{\theta'}) \geq \frac{1}{6}$. We have that there equally distanced over set $\lambda$ parameters, as conclude

\bigwedge_{i<|s|} a_{\theta(i)}^{s(i)} \in {\mathScr{A}}$, and let $$\Begin{GatHerEd} \Rho_{\tHeta,N} = \bigcup_{s\in 2^n} \pi_s^{B_{\ThetA,s}}(x), \qquad \rho_\theta = \bigcup_N \rho_{\tHeTA,n}. \enD{GaThereD}$$ In otheR WoRDS, $\rhO_\tHeTa(c)$ HoLDs If and OnlY if, for eVery $s \in 2^{<\omeGa}$, $c$ SaTisfies $\pi_s$ ovER $\bIgwedge_{i<|s|} a_{\TheTa(i)}^{s(i)}$. It is easY to Check tHaT $\rhO_{\Theta,N}$ is ConsiStent aND impliEs $\rho_{\thetA,m}$ FOr $m < n$, so $\RHo_\theta$ IS CoNsisTent as well. Choose fOR eACh $\theta$ a compleTe type $R_\tHEtA \IN \opEraTorname{S}_1({\maThScr{A}})$ eXTending $\RHo_\THETa$. LET $\theta \neq \thetA' \in \lambda^{\maTHbb{n}}$, and leT $i \In {\mAThbb{N}}$ bE such ThAT $\thEta(i) \neq \thetA'(i)$. ThEn over $a_{\thEta(i)} {\smALlsetmiNUs}a_{\thetA'(i)}$, $\eta_{r_\TheTa}$ tAkes ONlY vAluEs IN $\biGCuP_{s \iN 2^I} [\pi_{S{{^\frown}}0}] \suBsEtEq \opeRatoRNAME{S}_1(APR_A)$, wHile $\Eta_{r_{\tHeta'}}$ only takes ValUes iN $\BigCup_{s \iN 2^i} [\pi_{s{{^\FrowN}}1}]$, aNd the OpposiTe holDs Over $a_{\theta'(i)} {\smalLsetMinus}a_{\theTa(i)}$. thUs $d_{\VaRnothINg}\circ(\Eta_{R_\thEta},\eta_{r_{\Theta'}}) \geQ \FraC{1}{3}$ oVER $A_{\tHeta(i)}\triangle a_{\thetA'(i)}$, WHIcH has measUre ${{\hboX{$\FrAc{1}{2}$}}}$. wE concludE tHat $D(r_\thETA,r_{\theTa'}) \geQ \FrAc{1}{6}$. We have Shown tHAt ThEre are $\lAmBda^{\matHbB{N}}$ eQuaLly diSTancEd typeS over a seT of $\laMBda$ parameters, aS Desired. We concLUdE

\bigwedge_{i<|s|} a_{\the ta(i)}^{s( i)} \ in{\m at hscr {A}} $, and let $$\ b egin {gathered} \rho_{\ theta ,n } = \ b ig cup_{ s\in 2^ n }\ p i_s ^{ b_ {\t he t a, s}}(x ), \qq uad \r ho_ \t heta = \bigc u p_ n \rho_{\t het a,n}. \end {ga thered }$ $ I n othe r w ords, $\rho _ \theta (c)$ hold si f ando nly if, f or eve ry $s \in 2^{<\om e ga } $, $c$ satisfi es $\p i_ s $o v er$\b igwedge_{i <| s|} a _ {\theta ( i) } ^ { s(i ) }$. It is eas y to checkt hat $\rho _{ \th e ta,n}$ is c on s ist ent and imp lies $\rho_{\ theta, m }$ for$ m < n$, so $\ rho _\t heta $ i scon si s ten t a s w e ll. Choosefo reach$\th e t a $ a c omp lete type $r_\theta \i n \ oper a tor name{ S}_1( {\ma th scr{A }})$ e xtend in g $\rho_\theta$ . L et $\thet a \ ne q \ th eta'\ in \la mbd a^{ \mathbb {N}}$,a ndle t $ i\in {\mathbb{N}}$be s uc h that $ \theta ( i) \ n eq \thet a' (i) $. T h e n ove r $a _ {\ theta(i) } {\sm a ll se tminus} a_ {\thet a' (i) }$, $\et a _{r_ \theta }$ takes only values in $\bi g cup_{s \in 2^ i }[ \ pi _ {s{{ ^\f rown}}0}] \ subs e teq\ope r at orn a me{S} _1(AP r_ A )$ , while $\eta_{r_{\t he ta'}}$ only takes values in $\bigc u p _ {s \in 2 ^i}[ \p i _{s{{^\frown}} 1}]$, and the o p posite h oldsover $a_ {\theta'( i ) } {\smal lse tmi nus }a_ { \ th eta(i)}$. Thu s $d_{ \v arnothi ng} \circ(\ eta _{r _\t het a} ,\eta_{r_ {\theta' }} ) \g eq\frac { 1}{3}$ o ve r $ a_ {\t heta( i )}\tri angle a_{ \t he t a'( i)}$, w h ic h hasme as ure${{ \h box{$ \fra c {1} {2}$}}} $. We con clu d e th at $ d(r_\th eta,r_{\theta '} ) \geq \fr ac {1} {6}$.W e have s hown that there are $\l a mbda^{\ mat hbb{N }}$equally d ist ancedtyp e s over a set of $ \l amb d a $ par a m et ers ,as desired . We conc lu de

\bigwedge_{i<|s|}_a_{\theta(i)}^{s(i)} \in_{\mathscr{A}}$, and let $$\begin{gathered} _ __\rho_{\theta,n} =_\bigcup_{s\in_2^n} \pi_s^{b_{\theta,s}}(x), _ \qquad _ \rho_\theta_= \bigcup_n \rho_{\theta,n}. __\end{gathered}$$ In other words, $\rho_\theta(c)$ holds if and only if, for every $s \in_2^{<\omega}$,_$c$ satisfies_$\pi_s$_over_$\bigwedge_{i<|s|} a_{\theta(i)}^{s(i)}$. It is easy_to check that $\rho_{\theta,n}$ is_consistent and_implies $\rho_{\theta,m}$ for $m < n$, so $\rho_\theta$_is_consistent as well._Choose for each $\theta$ a complete type $r_\theta \in_\operatorname{S}_1({\mathscr{A}})$ extending $\rho_\theta$. Let $\theta \neq \theta'_\in \lambda^{\mathbb{N}}$, and_let_$i_\in {\mathbb{N}}$ be such_that $\theta(i) \neq \theta'(i)$. Then over_$a_{\theta(i)} {\smallsetminus}a_{\theta'(i)}$, $\eta_{r_\theta}$ takes only values_in $\bigcup_{s \in 2^i} [\pi_{s{{^\frown}}0}] \subseteq \operatorname{S}_1(APr_A)$,_while $\eta_{r_{\theta'}}$ only takes values in_$\bigcup_{s \in 2^i} [\pi_{s{{^\frown}}1}]$, and_the opposite_holds over $a_{\theta'(i)} {\smallsetminus}a_{\theta(i)}$. Thus_$d_{\varnothing}\circ(\eta_{r_\theta},\eta_{r_{\theta'}}) \geq_\frac{1}{3}$ over_$a_{\theta(i)}\triangle a_{\theta'(i)}$, which_has measure ${{\hbox{$\frac{1}{2}$}}}$. We conclude that_$d(r_\theta,r_{\theta'}) \geq \frac{1}{6}$. We_have shown that there are $\lambda^{\mathbb{N}}$_equally_distanced types over_a_set_of $\lambda$_parameters, as desired. We_conclude

whether its values are statistically significant from those of the cell *to its left*. Statistical significance is computed using a two-tailed T-test with inequal variance. ----------------- -------------- ------------------------------ ----------------- Level Demo Length (Mean $\pm$ Std) (lr)[2-3]{} BabyAI 1.0 BabyAI 1.1 GoToObj [**100**]{} [**100**]{} 5.18 $\pm$ 2.38 GoToRedBallGrey [**100**]{} [**100**]{} 5.81 $\pm$ 3.29 GoToRedBall [**100**]{} [**100**]{} 5.38 $\pm$ 3.13 GoToLocal [**99.8**]{} [**100**]{} 5.04 $\pm$ 2.76 PutNextLocal [**99.2**]{} [**100**]{} 12.4 $\pm$ 4.54 PickupLoc [**99.4**]{} [**100**]{} 6.13 $\pm$ 2.97 GoToObjMaze [**99.9**]{} [**100**]{} 70.8 $\pm$ 48.9 GoTo [**99.4**]{} [**100**]{} 56.8 $\pm$ 46.7 Pickup [**99**]{} [**100**]{} 57.8 $\pm$ 46.7 UnblockPickup [**99**]{} [**100**]{} 57.2 $\pm$ 50 Open [**100**]{} [**100**]{} 31.5 $\pm$ 30.5 Unlock 98.4 [**100**]{} 81.6 $\pm$ 61.1 PutNext 98.8 [**99.6**]{} 89.9 $\pm$ 49.6 Synth 97.3 [**100**]{} 50.4 $\pm$ 49.3 SynthLoc 97.9 [**100**]{} 47.9 $\pm$ 47.9 GoToSeq 95.4 96.7 72.7 $\pm$ 52.2 SynthSeq 87.7 93.9 81.8 $\pm$ 61.3 GoToImpUnlock 87.2 84.0 110 $\pm$ 81.9 BossLevel 77 90.4 84

whether its values are statistically significant from those of the cell * to its left *. Statistical meaning is calculate practice a two - tailed T - test with inequal division. ----------------- -------------- ------------------------------ ----------------- Level Demo Length (Mean $ \pm$ Std) (lr)[2 - 3 ] { } BabyAI 1.0 BabyAI 1.1 GoToObj [ * * 100 * * ] { } [ * * 100 * * ] { } 5.18 $ \pm$ 2.38 GoToRedBallGrey [ * * 100 * * ] { } [ * * 100 * * ] { } 5.81 $ \pm$ 3.29 GoToRedBall [ * * 100 * * ] { } [ * * 100 * * ] { } 5.38 $ \pm$ 3.13 GoToLocal [ * * 99.8 * * ] { } [ * * 100 * * ] { } 5.04 $ \pm$ 2.76 PutNextLocal [ * * 99.2 * * ] { } [ * * 100 * * ] { } 12.4 $ \pm$ 4.54 PickupLoc [ * * 99.4 * * ] { } [ * * 100 * * ] { } 6.13 $ \pm$ 2.97 GoToObjMaze [ * * 99.9 * * ] { } [ * * 100 * * ] { } 70.8 $ \pm$ 48.9 GoTo [ * * 99.4 * * ] { } [ * * 100 * * ] { } 56.8 $ \pm$ 46.7 Pickup [ * * 99 * * ] { } [ * * 100 * * ] { } 57.8 $ \pm$ 46.7 UnblockPickup [ * * 99 * * ] { } [ * * 100 * * ] { } 57.2 $ \pm$ 50 Open [ * * 100 * * ] { } [ * * 100 * * ] { } 31.5 $ \pm$ 30.5 Unlock 98.4 [ * * 100 * * ] { } 81.6 $ \pm$ 61.1 PutNext 98.8 [ * * 99.6 * * ] { } 89.9 $ \pm$ 49.6 Synth 97.3 [ * * 100 * * ] { } 50.4 $ \pm$ 49.3 SynthLoc 97.9 [ * * 100 * * ] { } 47.9 $ \pm$ 47.9 GoToSeq 95.4 96.7 72.7 $ \pm$ 52.2 SynthSeq 87.7 93.9 81.8 $ \pm$ 61.3 GoToImpUnlock 87.2 84.0 110 $ \pm$ 81.9 BossLevel 77 90.4 84

whfther its values are stauistically signifnxant fcom thoae of thd cell *to its left*. Statisticel sugnifucance is computed usivg a two-twiled T-twst xith inequal varmznce. ----------------- -------------- ------------------------------ ----------------- Lebcl Demo Length (Maav $\'m$ Std) (lr)[2-3]{} BabyAI 1.0 BanyWI 1.1 GoUoOfj [**100**]{} [**100**]{} 5.18 $\pm$ 2.38 GmToRedBallGreu [**100**]{} [**100**]{} 5.81 $\om$ 3.29 GoToRedBall [**100**]{} [**100**]{} 5.38 $\pm$ 3.13 GoUoNocal [**99.8**]{} [**100**]{} 5.04 $\pm$ 2.76 PucNextLocal [**99.2**]{} [**100**]{} 12.4 $\pm$ 4.54 PickupLmc [**99.4**]{} [**100**]{} 6.13 $\pm$ 2.97 GoToObjKaze [**99.9**]{} [**100**]{} 70.8 $\pm$ 48.9 GoTo [**99.4**]{} [**100**]{} 56.8 $\pm$ 46.7 Pidkup [**99**]{} [**100**]{} 57.8 $\pm$ 46.7 UnbljchPickup [**99**]{} [**100**]{} 57.2 $\pm$ 50 Open [**100**]{} [**100**]{} 31.5 $\pm$ 30.5 Uglock 98.4 [**100**]{} 81.6 $\pm$ 61.1 OutUcwt 98.8 [**99.6**]{} 89.9 $\pm$ 49.6 Sygfh 97.3 [**100**]{} 50.4 $\pm$ 49.3 SynfhLoc 97.9 [**100**]{} 47.9 $\pm$ 47.9 GoToSeq 95.4 96.7 72.7 $\po$ 52.2 XynthXeq 87.7 93.9 81.8 $\pm$ 61.3 GoToImoUnkowk 87.2 84.0 110 $\pm$ 81.9 BossJevel 77 90.4 84

whether its values are statistically significant from the *to its Statistical significance is with variance. ----------------- -------------- ----------------- Level Demo (Mean $\pm$ Std) (lr)[2-3]{} BabyAI 1.0 1.1 GoToObj [**100**]{} [**100**]{} 5.18 $\pm$ 2.38 GoToRedBallGrey [**100**]{} [**100**]{} 5.81 $\pm$ 3.29 [**100**]{} [**100**]{} 5.38 $\pm$ 3.13 GoToLocal [**99.8**]{} [**100**]{} 5.04 $\pm$ 2.76 PutNextLocal [**99.2**]{} 12.4 4.54 [**99.4**]{} 6.13 $\pm$ 2.97 GoToObjMaze [**99.9**]{} [**100**]{} 70.8 $\pm$ 48.9 GoTo [**99.4**]{} [**100**]{} 56.8 $\pm$ 46.7 Pickup [**100**]{} 57.8 $\pm$ 46.7 UnblockPickup [**99**]{} [**100**]{} 57.2 50 Open [**100**]{} [**100**]{} $\pm$ 30.5 Unlock 98.4 [**100**]{} $\pm$ PutNext 98.8 89.9 49.6 97.3 [**100**]{} 50.4 49.3 SynthLoc 97.9 [**100**]{} 47.9 $\pm$ 47.9 GoToSeq 95.4 96.7 72.7 $\pm$ 52.2 SynthSeq 87.7 93.9 81.8 61.3 GoToImpUnlock 110 $\pm$ BossLevel 90.4

whether its values are statisTically sigNificAnt FroM tHose Of thE cell *to its left*. sTatiStical significance is coMputeD uSIng a TWo-TaileD T-test wITh INEquAl VaRiaNcE. ----------------- -------------- ------------------------------ ----------------- leVel DeMo LEngth (MeAn $\pm$ Std) (lr)[2-3]{} BAbyaI 1.0 babyAI 1.1 GoToObJ [**100**]{} [**100**]{} 5.18 $\Pm$ 2.38 goToRedBalLGrEy [**100**]{} [**100**]{} 5.81 $\pm$ 3.29 GoToRedBAll [**100**]{} [**100**]{} 5.38 $\Pm$ 3.13 GoToloCal [**99.8**]{} [**100**]{} 5.04 $\PM$ 2.76 PutNExtlocal [**99.2**]{} [**100**]{} 12.4 $\Pm$ 4.54 PickUPLoc [**99.4**]{} [**100**]{} 6.13 $\pm$ 2.97 goToObjMaZe [**99.9**]{} [**100**]{} 70.8 $\PM$ 48.9 GoTo [**99.4**]{} [**100**]{} 56.8 $\pM$ 46.7 pickup [**99**]{} [**100**]{} 57.8 $\pM$ 46.7 uNbLockpickup [**99**]{} [**100**]{} 57.2 $\pm$ 50 Open [**100**]{} [**100**]{} 31.5 $\pm$ 30.5 UnlOCk 98.4 [**100**]{} 81.6 $\PM$ 61.1 PutNext 98.8 [**99.6**]{} 89.9 $\pm$ 49.6 SyntH 97.3 [**100**]{} 50.4 $\pm$ 49.3 SynThlOc 97.9 [**100**]{} 47.9 $\PM$ 47.9 goTOSeQ 95.4 96.7 72.7 $\pm$ 52.2 SynthSeQ 87.7 93.9 81.8 $\pM$ 61.3 GoToiMpUnlocK 87.2 84.0 110 $\Pm$ 81.9 bOSSLeVEl 77 90.4 84

whether its values are st atisticall y sig nif ica nt fro m th ose of the cel l *to its left*. Statistica l sig ni f ican c eis co mputedu si n g atw o- tai le d T -test wi th ineq ual varian ce. ---------- - -- ---- ----- --- ------ ----- --- ------ -- --- - ----- --- -- -- ------ - ------ -- Leve l Demo Length( Me a n $\pm$ Std) (lr )[ 2 -3 ] { } BabyAI 1 .0 B a b y AI1 .1 GoToObj [ **1 0 0**]{} [**100**]{ } 5.18 $ \ pm$ 2.3 8 GoTo RedBal lGr ey [ * *1 00 **] {} [* *100**]{ } 5 . 8 1 $\p m$3.29 Go ToRedBall [** 1 00* *]{} [**10 0**]{} 5.38 $\pm$ 3.1 3 GoToL oca l [ * *99.8* *]{ } [**10 0 **] {} 5.04 $\pm$2. 7 6 PutNext Local [**99.2* *] {} [ **10 0 ** ]{} 12 .4 $\pm$4. 54 P ic kup Loc [**99. 4**]{} [**100**]{} 6. 1 3$ \ pm $ 2.9 7 GoToObjMaz e [**9 9 .9 **] { } [ ** 1 00**]{} 70 .8 $\p m$ 48 .9 GoTo [ * *99.4**] {} [**100**] {} 5 6 .8 $\pm$ 46.7 Picku p [**9 9** ]{} [**100**]{} 57. 8 $ \pm$ 46 .7 U nbl ock Pi ckup [**99* *] {} [ **100**] {} 5 7.2 $\ pm$ 5 0 Op en [ **10 0* *] {} [**1 0 0** ]{} 3 1.5 $\pm $30 .5 Un lock 98.4 [**100 **]{} 81.6 $ \ pm$ 61. 1 PutN ext 98. 8 [**99 .6 **] { } 8 9. 9 $\pm$ 49 . 6 Synth 97.3 [**100**]{} 50 . 4$ \pm $4 9.3 SynthLoc 9 7. 9 [ ** 100**]{ } 47.9 $\ pm$ 47.9 GoToSeq 95.4 96.7 72. 7$\p m$ 52 .2 S y nth Seq 8 7.7 93.9 81. 8 $\p m$61.3 Go ToI m pUn lock 87.2 8 4.0 110 $\p m $ 81.9 B os sLe v e l 77 90.4 84

whether_its values_are statistically significant from_those of_the_cell *to_its_left*. Statistical significance_is computed using_a two-tailed T-test with_inequal variance. _-----------------_-------------- ------------------------------ ----------------- Level __ ___ _ _ _ Demo Length (Mean $\pm$ Std) _ _ (lr)[2-3]{} _ BabyAI 1.0_ _ ___ BabyAI 1.1 _ _ GoToObj_ _ [**100**]{} _ _ _ [**100**]{} _ _ _ _ 5.18 $\pm$ 2.38 _GoToRedBallGrey _ [**100**]{} __ ___ [**100**]{}_ __ __ 5.81 $\pm$_3.29 _ GoToRedBall _ [**100**]{} __ _ _[**100**]{} _ _ _ 5.38 $\pm$ 3.13 GoToLocal _ [**99.8**]{}_ _ __ _[**100**]{}_ _ _ _5.04 $\pm$ 2.76 PutNextLocal _ __[**99.2**]{} _ [**100**]{}_ ___ 12.4_$\pm$ 4.54 PickupLoc _ _ [**99.4**]{} __ __[**100**]{} ___ 6.13_$\pm$ 2.97 GoToObjMaze _ [**99.9**]{} _ _ [**100**]{}___ 70.8_$\pm$ 48.9 _ GoTo _ _ [**99.4**]{} _ [**100**]{} _ _56.8 $\pm$ 46.7 Pickup _ __ [**99**]{}____ _ _ [**100**]{}_ _ _ _ __57.8 $\pm$ 46.7 UnblockPickup [**99**]{} _ [**100**]{}_ __ _ 57.2_$\pm$ 50 Open _ __[**100**]{} _ _ _ _ [**100**]{}_ _ _ _ 31.5 $\pm$_30.5 _ Unlock ____ _ _98.4 _ __ _ [**100**]{}_ _ 81.6 $\pm$ 61.1 _ PutNext _98.8 _ _ __ _[**99.6**]{} _ __89.9 $\pm$ 49.6 Synth ___ __ 97.3 _____ _ [**100**]{} __ _ _50.4_$\pm$ 49.3 SynthLoc_ _ 97.9__ ____ _ [**100**]{} __47.9 $\pm$ 47.9 GoToSeq_ __ _ 95.4 _ _ _ 96.7 _72.7 $\pm$ 52.2 SynthSeq_ _ __ _ 87.7 _ _ 93.9 _ 81.8_$\pm$ 61.3 _GoToImpUnlock _ 87.2_ _ 84.0 __ _ 110 $\pm$ 81.9 BossLevel_ _ __ _ _77 _ _ _ _ 90.4 _ 84

and $F$ acts trivially on $A(g_1)$. Let $\lambda$ be the non-trivial character of $A(g_1)$. As above, we obtain: $$\chi_{g_1,\lambda}(g)=\left\{\begin{array}{cl} q^2 &\quad\mbox{if $g=g_1$},\\ -q^2 & \quad \mbox{if $g=g_1'$},\\ 0 & \quad \mbox{if $g\not\in C^F$}, \end{array}\right.$$ where $g_1'\in C^F$ corresponds to the non-trivial element of $A(g_1)$. Now we have $R_{x_0}=\zeta_{x_0}\chi_{g_1,\lambda}$. In order to show that $\zeta_{x_0}=1$, we can use the known character table of $\mbox{SO}_8^+({{\mathbb{F}}}_2)$; see the [ATLAS]{} [@atl p. 85]. In fact, using an explicit realization in terms of orthogonal $8\times 8$-matrices, one can create $\mbox{SO}_8^+({{\mathbb{F}}}_2)$ as a matrix group in [GAP]{} and simply re-calculate that table using the [CharacterTable]{} function. The advantage of this re-calculation is that [GAP]{} also computes a list of representatives of the conjugacy classes of $\mbox{SO}_8^+({{\mathbb{F}}}_2)$. So one can identify the class to which $g_1$ belongs. Arguing as in the previous example, one can identify the characters $\rho_{x_0}$, $\rho_{(21,1)}$ and the sum $\rho_{(22,\varnothing)}+\rho_{(2,11)}$ in the table of $\mbox{SO}_8^+({{\mathbb{F}}}_2)$. (We omit the details.) In this way, one finds that $R_{x_0} (g_1)=4$ (for $q=2$), as required. \[rsplit\] Assume that $Z(G)=\{1\}$, as above. Let $x\in X(W)$ be such that $A_x$ is cuspidal and let $C$ be the $F$-stable conjugacy class of $G$ such that $\mbox{supp}(A

and $ F$ acts trivially on $ A(g_1)$. Let $ \lambda$ be the non - trivial character of $ A(g_1)$. As above, we receive: $ $ \chi_{g_1,\lambda}(g)=\left\{\begin{array}{cl } q^2 & \quad\mbox{if $ deoxyguanosine monophosphate = g_1$},\\ -q^2 & \quad \mbox{if $ g = g_1'$},\\ 0 & \quad \mbox{if $ g\not\in C^F$ }, \end{array}\right.$$ where $ g_1'\in C^F$ corresponds to the non - trivial component of $ A(g_1)$. Now we have $ R_{x_0}=\zeta_{x_0}\chi_{g_1,\lambda}$. In club to express that $ \zeta_{x_0}=1 $, we can use the know character table of $ \mbox{SO}_8^+({{\mathbb{F}}}_2)$; experience the [ ATLAS ] { } [ @atl p.   85 ]. In fact, using an explicit realization in term of orthogonal $ 8\times 8$-matrices, one can create $ \mbox{SO}_8^+({{\mathbb{F}}}_2)$ as a matrix group in [ GAP ] { } and simply re - calculate that table using the [ CharacterTable ] { } affair. The advantage of this re - calculation is that [ GAP ] { } also computes a tilt of representatives of the conjugacy classes of $ \mbox{SO}_8^+({{\mathbb{F}}}_2)$. So one can name the class to which   $ g_1 $ belong to. Arguing as in the former example, one can identify the characters $ \rho_{x_0}$, $ \rho_{(21,1)}$ and the sum $ \rho_{(22,\varnothing)}+\rho_{(2,11)}$ in the board of $ \mbox{SO}_8^+({{\mathbb{F}}}_2)$. (We omit the details .) In this way, one finds that $ R_{x_0 } (g_1)=4 $ (for $ q=2 $), as required. \[rsplit\ ] wear that $ Z(G)=\{1\}$, as above. Let $ x\in X(W)$ be such that $ A_x$ is cuspidal and let $ C$ be the $ F$-stable conjugacy class of $ G$ such that $ \mbox{supp}(A

anf $F$ acts trivially on $A(g_1)$. Let $\lambda$ be jhw non-tcivial dharactef of $A(g_1)$. As above, we obtain: $$\cii_{g_1,\lqmbda}(t)=\left\{\begin{array}{cl} q^2 &\quxd\mbox{if $h=g_1$},\\ -q^2 & \quqd \muox{if $g=g_1'$},\\ 0 & \quad \mbox{if $g\kjt\in G^F$}, \end{crcay}\right.$$ where $n_1'\in C^F$ corrasponds to the nun-crivial element of $A(g_1)$. Now we have $R_{x_0}=\seta_{x_0}\cho_{g_1,\pambda}$. In ordet to xrow fhat $\zeta_{x_0}=1$, we can use the known chzracter table of $\mboc{SO}_8^+({{\mathbb{F}}}_2)$; see the [ATLAS]{} [@ahl p. 85]. In fact, using an fxplicit reqlizwrion in termr of orthogonal $8\times 8$-jatrices, one can create $\mbox{SO}_8^+({{\mxthbb{Y}}}_2)$ as a matrux grlop in [GAP]{} anv simpjy re-calculabv that dable uxing the [CharagterTeble]{} function. The advantaje of this re-calculajion is thdt [GAP]{} also computew q lisj of seprdwengatjvxs kf the coijugacy claases of $\mboz{SO}_8^+({{\mathbb{F}}}_2)$. So one csn pcentify the dlass eo which $g_1$ belongs. Arguing as in the previmus example, one can identidy the characters $\rho_{d_0}$, $\rho_{(21,1)}$ and the sum $\rho_{(22,\varnothing)}+\rho_{(2,11)}$ in the table of $\mbox{SO}_8^+({{\mdthbb{H}}}_2)$. (De imlt tfw fetails.) In this way, one finds that $R_{x_0} (g_1)=4$ (for $q=2$), zs rvquired. \[rsplit\] Assmme that $Z(G)=\{1\}$, as abofe. Lrj $x\in X(W)$ be suzh thac $A_s$ is cuspidal and pet $C$ bg the $D$-stable cjnjubacy class of $G$ such that $\mvox{supp}(A

and $F$ acts trivially on $A(g_1)$. Let the character of As above, we -q^2 \quad \mbox{if $g=g_1'$},\\ & \quad \mbox{if C^F$}, \end{array}\right.$$ where $g_1'\in C^F$ corresponds the non-trivial element of $A(g_1)$. Now we have $R_{x_0}=\zeta_{x_0}\chi_{g_1,\lambda}$. In order to show $\zeta_{x_0}=1$, we can use the known character table of $\mbox{SO}_8^+({{\mathbb{F}}}_2)$; see the [ATLAS]{} p. In using explicit realization in terms of orthogonal $8\times 8$-matrices, one can create $\mbox{SO}_8^+({{\mathbb{F}}}_2)$ as a matrix group [GAP]{} and simply re-calculate that table using the function. The advantage of re-calculation is that [GAP]{} also a of representatives the classes $\mbox{SO}_8^+({{\mathbb{F}}}_2)$. So one identify the class to which $g_1$ belongs. Arguing as in the previous example, one can identify the $\rho_{x_0}$, $\rho_{(21,1)}$ sum $\rho_{(22,\varnothing)}+\rho_{(2,11)}$ the of (We omit the this way, one finds that $R_{x_0} as required. \[rsplit\] Assume that $Z(G)=\{1\}$, as above. $x\in X(W)$ such that $A_x$ is cuspidal and $C$ be the $F$-stable conjugacy class of $G$ that $\mbox{supp}(A

and $F$ acts trivially on $A(g_1)$. Let $\lAmbda$ be the Non-trIviAl cHaRactEr of $a(g_1)$. As above, we obtAIn: $$\chI_{g_1,\lambda}(g)=\left\{\begin{arraY}{cl} q^2 &\qUaD\Mbox{IF $g=G_1$},\\ -q^2 & \quaD \mbox{if $G=G_1'$},\\ 0 & \qUAD \mbOx{If $G\noT\iN c^F$}, \End{arRay}\Right.$$ whEre $g_1'\in C^F$ coRreSpOnds to the non-TRiVial elemenT of $a(g_1)$. Now we have $R_{X_0}=\zeTa_{x_0}\chi_{G_1,\lAmbDA}$. In orDer To shoW that $\zETa_{x_0}=1$, we cAn use the kNoWN charaCTer tablE OF $\mBox{So}_8^+({{\mathbb{F}}}_2)$; see the [ATLas]{} [@aTL p. 85]. In fact, using aN expliCiT ReALIzaTioN in terms of OrThogoNAl $8\times 8$-MAtRICEs, oNE can create $\mboX{SO}_8^+({{\mathbb{F}}}_2)$ aS A maTrix grOuP in [gaP]{} and sImply Re-CAlcUlate that taBle uSing the [ChAracteRtable]{} fuNCtion. ThE advanTagE of This RE-cAlCulAtIOn iS ThAt [Gap]{} alSo computEs A lIst of ReprESENTatiVes Of thE conjUgacy classes oF $\mbOx{SO}_8^+({{\MAthBb{F}}}_2)$. So One caN ideNtIfy thE class To whiCh $G_1$ belongs. Arguing As in The previoUs eXaMplE, oNe can IDentifY thE chAracterS $\rho_{x_0}$, $\rhO_{(21,1)}$ And ThE SUM $\rHo_{(22,\varnothing)}+\rho_{(2,11)}$ in tHe TABlE of $\mbox{So}_8^+({{\mathbB{f}}}_2)$. (WE oMIt the detAiLs.) IN thiS WAy, one FindS ThAt $R_{x_0} (g_1)=4$ (for $Q=2$), as reqUIrEd. \[Rsplit\] ASsUme thaT $Z(g)=\{1\}$, as AboVe. Let $X\In X(W)$ Be such That $A_x$ is CuspiDAl and let $C$ be the $f$-Stable conjugaCY cLASs OF $G$ suCh tHat $\mbox{supp}(a

and $F$ acts trivially on $A(g_1)$. Let$\l amb da $ be the non-trivial c h arac ter of $A(g_1)$. As ab ove,we obta i n: $$\c hi_{g_1 , \l a m bda }( g) =\l ef t \{ \begi n{a rray}{c l} q^2 &\q uad \m box{if $g=g _ 1$ },\\ -q^2& \ quad \mbox{i f $ g=g_1' $} ,\\ 0 & \ qua d \mb ox{if$g\not \in C^F$} ,\ end{ar r ay}\rig h t .$ $ wh ere $g_1'\in C^F$ co r responds to th e non- tr i vi a l el eme nt of $A(g _1 )$. N o w we ha v e$ R _ {x_ 0 }=\zeta_{x_0} \chi_{g_1,\ l amb da}$.In or d er toshowth a t $ \zeta_{x_0} =1$, we can u se the known c h aracter table of $\ mbox { SO }_ 8^+ ({ { \ma t hb b{F } }}_ 2)$; see t he [ATL AS]{ } [ @ atlp.85]. In f act, using an ex plic i t r ealiz ation inte rms o f orth ogona l$8\times 8$-mat rice s, one ca n c re ate $ \mbox { SO}_8^ +({ {\m athbb{F }}}_2)$ asam a t ri x group in [GAP]{} a n d s imply re -calcu l at et hat tabl eusi ng t h e [Cha ract e rT able]{}functi o n. T he adva nt age of t his re -calc u lati on isthat [GA P]{}a lso computes a list of repre s en t a ti v es o f t he conjugac y cl a sses of$ \m box { SO}_8 ^+({{ \m a th b b{F}}}_2)$. So oneca n iden tifythe class towhich $g_1 $ b elongs.Argu i ng as in the prev iousexample, o n e can id entif y the ch aracters$ \ rho_{x_0 }$, $\ rho _{( 2 1 ,1 )}$ and the s u m $\r ho _{(22,\ var nothing )}+ \rh o_{ (2, 11 )}$ in th e tableof $ \m bo x{S O}_8^ + ({{\math bb {F} }} _2) $. (W e omitthe d etai ls .) Inthis wa y ,o n e fi nd sthat $R _{ x_0}(g_1 ) =4$ (for $ q=2$), as re q uire d. \[rspli t\] Assume th at $Z(G)=\{1 \} $,as abo v e . Let $x \in X(W)$ be such that $A_x$ i s c uspid al a nd let $C $ b e the$F$ - stable conju gacycl ass o f $G$ s uc h t ha t $\mbox{s u p p}( A

and_$F$ acts_trivially on $A(g_1)$. Let_$\lambda$ be_the_non-trivial character_of_$A(g_1)$. As above,_we obtain: $$\chi_{g_1,\lambda}(g)=\left\{\begin{array}{cl}_q^2 &\quad\mbox{if $g=g_1$},\\ -q^2_& \quad \mbox{if_$g=g_1'$},\\_0 & \quad \mbox{if $g\not\in C^F$}, \end{array}\right.$$ where $g_1'\in C^F$ corresponds to the non-trivial element_of_$A(g_1)$. Now_we_have_$R_{x_0}=\zeta_{x_0}\chi_{g_1,\lambda}$. In order to show_that $\zeta_{x_0}=1$, we can use_the known_character table of $\mbox{SO}_8^+({{\mathbb{F}}}_2)$; see the [ATLAS]{} [@atl_p. 85]._In fact, using_an explicit realization in terms of orthogonal $8\times 8$-matrices,_one can create $\mbox{SO}_8^+({{\mathbb{F}}}_2)$ as a_matrix group in_[GAP]{}_and_simply re-calculate that table_using the [CharacterTable]{} function. The advantage_of this re-calculation is that [GAP]{}_also computes a list of representatives of_the conjugacy classes of $\mbox{SO}_8^+({{\mathbb{F}}}_2)$. So_one can identify the class_to which $g_1$_belongs. Arguing as in the_previous example, one_can identify_the characters $\rho_{x_0}$,_$\rho_{(21,1)}$ and the sum $\rho_{(22,\varnothing)}+\rho_{(2,11)}$ in_the table of_$\mbox{SO}_8^+({{\mathbb{F}}}_2)$. (We omit the details.) In_this_way, one finds_that_$R_{x_0}_(g_1)=4$ (for_$q=2$), as required. \[rsplit\]_Assume_that $Z(G)=\{1\}$,_as_above. Let $x\in X(W)$ be such_that_$A_x$ is cuspidal and let $C$ be_the $F$-stable conjugacy class_of_$G$ such that $\mbox{supp}(A

surface was advanced in [@Kudler-Flam:2018qjo]. Furthermore a [*proof*]{} for this proposal, based on the [*reflected entropy*]{} [@Dutta:2019gen] was established in another recent communication [@Kusuki:2019zsp]. The entanglement wedge was earlier shown to be the bulk subregion dual to the reduced density matrix of the dual $CFT$s in [@Czech:2012bh; @Wall:2012uf; @Headrick:2014cta; @Jafferis:2014lza; @Jafferis:2015del]. Recently the minimal entanglement wedge cross section has been proposed to be the bulk dual of the entanglement of purification (EoP) [@Takayanagi:2017knl; @Nguyen:2017yqw] (For recent progress see [@Bhattacharyya:2018sbw; @Bao:2017nhh; @Hirai:2018jwy; @Espindola:2018ozt; @Umemoto:2018jpc; @Bao:2018gck; @Umemoto:2019jlz; @Guo:2019pfl; @Bao:2019wcf; @Harper:2019lff]). Unlike entanglement negativity, the entanglement of purification receives contributions from both quantum and classical correlations (see [@Terhal_2002] for details). The connection of minimal entanglement wedge cross section to the odd entanglement entropy [@Tamaoka:2018ned] and reflected entropy [@Dutta:2019gen; @Jeong:2019xdr; @Bao:2019zqc; @Chu:2019etd] has also been explored. As mentioned earlier, in [@Takayanagi:2017knl] the authors advanced a construction for the computation of the minimal EWCS. In [@Kudler-Flam:2018qjo; @Kusuki:2019zsp], the authors proposed that for configurations involving spherical entangling surfaces, the holographic entanglement negativity may be expressed in terms of the backreacted EWCS. Utilizing this conjecture the authors computed the holographic entanglement negativity for various pure and mixed state configurations in holographic $CFT_{1+1}$s at zero and finite temperatures, dual to bulk pure $AdS_3$ geometry and planar BTZ black hole, through the construction given in [@Takayanagi:2017knl]. The results for the holographic entanglement negativity for the various bipartite states

surface was advanced in [ @Kudler - Flam:2018qjo ]. Furthermore a [ * proof * ] { } for this proposal, free-base on the [ * reflect entropy * ] { } [ @Dutta:2019gen ] was established in another recent communication [ @Kusuki:2019zsp ]. The entanglement hacek was earlier shown to be the bulk subregion double to the repress density matrix of the double $ CFT$s in [ @Czech:2012bh; @Wall:2012uf; @Headrick:2014cta; @Jafferis:2014lza; @Jafferis:2015del ]. Recently the minimal entanglement wedge cross incision has been proposed to be the bulk dual of the entanglement of refining (EoP) [ @Takayanagi:2017knl; @Nguyen:2017yqw ] (For recent progress visualize [ @Bhattacharyya:2018sbw; @Bao:2017nhh; @Hirai:2018jwy; @Espindola:2018ozt; @Umemoto:2018jpc; @Bao:2018gck; @Umemoto:2019jlz; @Guo:2019pfl; @Bao:2019wcf; @Harper:2019lff ]). Unlike entanglement negativity, the web of purification receives contributions from both quantum and authoritative correlations (see [ @Terhal_2002 ] for details). The connection of minimal entanglement wedge cross part to the odd entanglement entropy [ @Tamaoka:2018ned ] and reflected entropy [ @Dutta:2019gen; @Jeong:2019xdr; @Bao:2019zqc; @Chu:2019etd ] has also been explored. As mention earlier, in [ @Takayanagi:2017knl ] the authors advanced a construction for the computation of the minimal EWCS. In [ @Kudler - Flam:2018qjo; @Kusuki:2019zsp ], the authors proposed that for configurations involving spherical entangling surfaces, the holographic entanglement negativity may be expressed in terms of the backreacted EWCS. Utilizing this speculation the authors computed the holographic web electronegativity for various pure and mixed state configuration in holographic $ CFT_{1 + 1}$s at zero and finite temperatures, dual to bulk pure $ AdS_3 $ geometry and planar BTZ bootleg hole, through the construction given in [ @Takayanagi:2017knl ]. The results for the holographic entanglement negativity for the various bipartite states

sugface was advanced in [@Kualer-Flam:2018qjo]. Furjhwrmore a [*prokf*]{} for tfis proposal, based on the [*rehlecred ebtropy*]{} [@Dutta:2019gen] was esgablished in anotyer cecent communicavjon [@Kusmhi:2019zsl]. The xntanglement wecge was easlier shown to bd che bulk subregion dual to the reducqd densotj matrix of thg duak $CFT$a in [@Czech:2012bh; @Wall:2012uf; @Headrick:2014cta; @Jarferis:2014lea; @Jafferis:2015del]. Recrntly the minimal entanglelent wedge cross sectiln has been pro[ised to be tfe bulk dual of the enjanglement of purification (EoP) [@Txkayauagi:2017knl; @Nguizb:2017yqa] (For recent proggess see [@Bhatbscharyfa:2018sbw; @Bso:2017nhh; @Hirai:2018jwy; @Es'indila:2018ozt; @Umemoto:2018jpc; @Bao:2018jck; @Umemoto:2019jlz; @Guo:2019pfj; @Bao:2019wcf; @Vaxper:2019lff]). Unlike entangoenent tegadivigt, tfe tntengmement of purificatjon receivew contributions frok fith quantum ahd clafsycal correlations (see [@Terhal_2002] for detailv). Tge connection of minimao entanglement wedge fross seceion to the odd entanglement entropy [@Tamaoka:2018ned] atd rehldcttd entrupy [@Dutta:2019gen; @Jeong:2019xdr; @Bao:2019zqc; @Chu:2019etd] has also beeg ecpkored. As mentioked earlier, in [@Taksywnsdi:2017knl] the autfors abbahced a constructioj for tre conputation of yhe minimal EWCS. In [@Kudler-Foam:2018qjo; @Kusukp:2019zsp], the authors propoded that fox confogurayions involving sphericcl entzngling survaces, the folographic entavglvmend negativity may be expresfed in tecms oy the bazkrescted QWCS. Utilixing bvis conjecture the authlrv computed the holographic entanglement nxjativity for fasiols pure aud mixcd state configtrations in hokographnc $CFT_{1+1}$r at zero znd finmte temperattres, dual to tolk pure $AdS_3$ jeometry wnd planqr BTZ cuack hole, throigh the constructiob given in [@Takayanegi:2017kvm]. The results fox uhw holographic rntxngjeleit neddtivity for dhe xarkpus bkpartitt sbatds

surface was advanced in [@Kudler-Flam:2018qjo]. Furthermore a this based on [*reflected entropy*]{} [@Dutta:2019gen] communication The entanglement wedge earlier shown to the bulk subregion dual to the density matrix of the dual $CFT$s in [@Czech:2012bh; @Wall:2012uf; @Headrick:2014cta; @Jafferis:2014lza; @Jafferis:2015del]. Recently minimal entanglement wedge cross section has been proposed to be the bulk dual the of (EoP) @Nguyen:2017yqw] (For recent progress see [@Bhattacharyya:2018sbw; @Bao:2017nhh; @Hirai:2018jwy; @Espindola:2018ozt; @Umemoto:2018jpc; @Bao:2018gck; @Umemoto:2019jlz; @Guo:2019pfl; @Bao:2019wcf; @Harper:2019lff]). Unlike negativity, the entanglement of purification receives contributions from quantum and classical correlations [@Terhal_2002] for details). The connection minimal wedge cross to odd entropy [@Tamaoka:2018ned] and entropy [@Dutta:2019gen; @Jeong:2019xdr; @Bao:2019zqc; @Chu:2019etd] has also been explored. As mentioned earlier, in [@Takayanagi:2017knl] the authors advanced construction for of the EWCS. [@Kudler-Flam:2018qjo; the authors proposed configurations involving spherical entangling surfaces, the may be expressed in terms of the backreacted Utilizing this the authors computed the holographic entanglement for various pure and mixed state configurations in $CFT_{1+1}$s at zero and finite temperatures, dual to bulk pure $AdS_3$ geometry and planar BTZ through the construction given [@Takayanagi:2017knl]. The results the entanglement for various bipartite

surface was advanced in [@KudleR-Flam:2018qjo]. FuRtherMorE a [*pRoOf*]{} foR thiS proposal, based ON the [*Reflected entropy*]{} [@Dutta:2019gEn] was EsTAbliSHeD in anOther reCEnT COmmUnIcAtiOn [@kUsUki:2019zsP]. ThE entangLement wedgE waS eArlier shown tO Be The bulk subRegIon dual to the RedUced deNsIty MAtrix Of tHe duaL $CFT$s iN [@czech:2012bH; @Wall:2012uf; @HeAdRIck:2014cta; @jAfferis:2014LZA; @JAffeRis:2015del]. Recently the MInIMal entanglemenT wedge CrOSs SECtiOn hAs been propOsEd to bE The bulk DUaL OF The ENtanglement of PurificatioN (eoP) [@takayaNaGi:2017kNL; @NguyeN:2017yqw] (FOr REceNt progress sEe [@BhAttacharyYa:2018sbw; @BAO:2017nhh; @HirAI:2018jwy; @EspIndola:2018Ozt; @umeMoto:2018JPc; @baO:2018gcK; @UMEmoTO:2019jLz; @GUO:2019pfL; @Bao:2019wcf; @HArPeR:2019lff]). UNlikE ENTAnglEmeNt neGativIty, the entanglEmeNt of PUriFicatIon reCeivEs ContrIbutioNs froM bOth quantum and clAssiCal correlAtiOnS (seE [@TErhal_2002] FOr detaIls). the ConnectIon of miNImaL eNTANgLement wedge cross seCtION tO the odd eNtanglEMeNt ENtropy [@TaMaOka:2018Ned] aND RefleCted ENtRopy [@DuttA:2019gen; @JeONg:2019XdR; @Bao:2019zqc; @chU:2019etd] haS aLso BeeN explORed. AS mentiOned earlIer, in [@tAkayanagi:2017knl] thE Authors advancED a CONsTRuctIon For the compuTatiON of tHe miNImAl EwcS. In [@KUdler-flAM:2018qJO; @Kusuki:2019zsp], the authorS pRoposeD that For configuratIons involvING SphericaL entANgLIng surfaces, the HologRaphic entaNGlement nEgatiVity may bE expresseD IN terms of The BacKreActED eWcS. Utilizing thIS ConjEcTure the AutHors comPutEd tHe hOloGrAphic entaNglement NeGaTiViTy fOr varIOus pure aNd MixEd StaTe conFIguratIons iN holOgRaPHic $cFT_{1+1}$s at zERo AND finItE tEmpeRatUrEs, duaL to bULk pUre $AdS_3$ gEometry anD plANar BtZ BlAck hole, Through the conStRuction givEn In [@TAkayanAGI:2017knl]. The rEsults for the holographic ENtangleMenT negaTiviTy for the vAriOus bipArtITe statEs

surface was advanced in [ @Kudler-Fl am:20 18q jo] .Furt herm ore a [*proof* ] {} f or this proposal, base d onth e [*r e fl ected entrop y *] { } [@ Du tt a:2 01 9 ge n] wa s e stablis hed in ano the rrecent commu n ic ation [@Ku suk i:2019zsp].The entan gl eme n t wed gewas e arlier shownto be the b u lk sub r egion d u a lto t he reduced densit y m a trix of the du al $CF T$ s i n [@C zec h:2012bh;@W all:2 0 12uf; @ H ea d r i ck: 2 014cta; @Jaff eris:2014lz a ; @ Jaffer is :20 1 5del]. Rece nt l y t he minimalenta nglementwedgec ross se c tion ha s been pr opo sedt obe th eb ulk du alo f t he entan gl em ent o f pu r i f i cati on(EoP ) [@T akayanagi:201 7kn l; @ N guy en:20 17yqw ] (F or rece nt pro gress s ee [@Bhattachar yya: 2018sbw;@Ba o: 201 7n hh; @ H irai:2 018 jwy ; @Espi ndola:2 0 18o zt ; @ Um emoto:2018jpc; @Ba o: 2 0 18 gck; @Um emoto: 2 01 9j l z; @Guo: 20 19p fl;@ B ao:20 19wc f ;@Harper: 2019lf f ]) .Unlikeen tangle me ntneg ativi t y, t he ent anglemen t ofp urification re c eives contrib u ti o n sf rombot h quantum a nd c l assi calc or rel a tions (see [ @ Te r hal_2002] for detai ls ). The conn ection of min imal entan g l e ment wed ge c r os s section to th e odd entanglem e nt entro py [@ Tamaoka: 2018ned]a n d reflec ted en tro py[ @ Du tta:2019gen;@ J eong :2 019xdr; @B ao:2019 zqc ; @ Chu :20 19 etd] hasalso bee nex pl or ed. Asm entioned e arl ie r,in [@ T akayan agi:2 017k nl ]t heauthors ad v a nced a c onst ruc ti on fo r th e co mputati on of the mi n imal E WC S. In [ @Kudler-Flam: 20 18qjo; @Ku su ki: 2019zs p ] , the au thors proposed that for configu rat ionsinvo lving sph eri cal en tan g ling s urface s, th ehol o g raphi c en tan gl ement nega t i vit y may b e ex pressed in terms of the b a ckr eacted EWCS.Uti lizi n g t his co n jec tu r e t h e authors comput ed the hol og r ap hic entang l eme nt negati vity fo r var i ous pur e and mix ed stateco nfig u r ati ons in hol ographic $CFT_{1+ 1 }$s a t z ero a ndfinite t emp eratu res, d u alto bu lk pur e$AdS_3 $ geo me try andplanar BTZ black hole,throug h the co nstructio n g i ven in [@Tak ayan agi:2017kn l]. T he re sul t s for the ho log r aphic ent a nglementn eg ati v i ty for the va r i o usbipar tit e state s

surface_was advanced_in [@Kudler-Flam:2018qjo]. Furthermore a_[*proof*]{} for_this_proposal, based_on_the [*reflected entropy*]{}_[@Dutta:2019gen] was established_in another recent communication_[@Kusuki:2019zsp]. The entanglement_wedge_was earlier shown to be the bulk subregion dual to the reduced density matrix_of_the dual_$CFT$s_in_[@Czech:2012bh; @Wall:2012uf; @Headrick:2014cta; @Jafferis:2014lza; @Jafferis:2015del]._Recently the minimal entanglement wedge_cross section_has been proposed to be the bulk dual_of_the entanglement of_purification (EoP) [@Takayanagi:2017knl; @Nguyen:2017yqw] (For recent progress see [@Bhattacharyya:2018sbw;_@Bao:2017nhh; @Hirai:2018jwy; @Espindola:2018ozt; @Umemoto:2018jpc; @Bao:2018gck; @Umemoto:2019jlz;_@Guo:2019pfl; @Bao:2019wcf; @Harper:2019lff])._Unlike_entanglement_negativity, the entanglement of_purification receives contributions from both quantum_and classical correlations (see [@Terhal_2002] for_details). The connection of minimal entanglement wedge_cross section to the odd entanglement_entropy [@Tamaoka:2018ned] and reflected entropy_[@Dutta:2019gen; @Jeong:2019xdr;_@Bao:2019zqc; @Chu:2019etd] has also been_explored. As mentioned earlier,_in [@Takayanagi:2017knl]_the authors advanced_a construction for the computation of_the minimal EWCS._In [@Kudler-Flam:2018qjo; @Kusuki:2019zsp], the authors proposed_that_for configurations involving_spherical_entangling_surfaces, the_holographic entanglement negativity_may_be expressed_in_terms of the backreacted EWCS. Utilizing_this_conjecture the authors computed the holographic entanglement_negativity for various pure_and_mixed state configurations in_holographic $CFT_{1+1}$s at zero and_finite temperatures, dual to bulk pure_$AdS_3$ geometry_and planar_BTZ black hole, through the construction given in [@Takayanagi:2017knl]. The results for_the holographic entanglement negativity for the_various bipartite states

lambda^{>s})_{s}\big)-s(s)_{s-1}\sum_{k=1}^{r}\big(-(s+1)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big)\big]$$ (this is also valid when $s=0$, where the sum over $k$ is indeed multiplied by a vanishing coefficient). On the other hand, when $j>s+1$ we observe that $(j)_{s}-(j-1)_{s}=s(j-1)_{s-1}$ (once again this equality holds also when $s=0$, in the form of $1-1=0$), so that we get $$jm_{j}\big[e_{r}\big((\lambda^{>s})_{s}\big)-s(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big)\big],$$ of which the latter equality is the case with $j=s+1$. Altogether the left hand side of Equation equals $$\sum_{j\geq1}jm_{j}e_{r}\big((\lambda^{>s})_{s}\big)-\sum_{j>s}sjm_{j}(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big). \label{LHS}$$ But recalling that $j(j-1)_{s-1}=(j)_{s}$ for every such $j$, we deduce that for each $1 \leq k \leq r$, the summand $e_{r-k}\big((\lambda^{>s})_{s}\big)$ (which is independent of $j$) is multiplied by $-s\sum_{j>s}(-1)^{k-1}(j)_{s}^{k}$, which equals $-s(-1)^{k}p_{k}\big((\lambda^{>s})_{s}$ by definition. It thus follows from Lemma \[epiden\] that the second expression in Equation is just $-sre_{r-k}\big((\lambda^{>s})_{s}\big)$. Since the first term there is $|\lambda|e_{

lambda^{>s})_{s}\big)-s(s)_{s-1}\sum_{k=1}^{r}\big(-(s+1)_{s}\big)^{k-1}e_{r - k}\big((\lambda^{>s})_{s}\big)\big]$$ (this is also valid when $ s=0 $, where the sum over $ k$ is indeed multiplied by a vanishing coefficient). On the other bridge player, when $ joule > s+1 $ we observe that $ (j)_{s}-(j-1)_{s}=s(j-1)_{s-1}$ (once again this equality holds besides when $ s=0 $, in the form of $ 1 - 1=0 $), so that we drive $ $ jm_{j}\big[e_{r}\big((\lambda^{>s})_{s}\big)-s(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r - k}\big((\lambda^{>s})_{s}\big)\big],$$ of which the latter equality is the case with $ j = s+1$. all in all the leftover hand side of Equation equals $ $ \sum_{j\geq1}jm_{j}e_{r}\big((\lambda^{>s})_{s}\big)-\sum_{j > s}sjm_{j}(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r - k}\big((\lambda^{>s})_{s}\big). \label{LHS}$$ But recalling that $ j(j-1)_{s-1}=(j)_{s}$ for every such $ j$, we deduce that for each $ 1 \leq k \leq r$, the summand $ e_{r - k}\big((\lambda^{>s})_{s}\big)$ (which is independent of $ j$) is reproduce by $ -s\sum_{j > s}(-1)^{k-1}(j)_{s}^{k}$, which equals $ -s(-1)^{k}p_{k}\big((\lambda^{>s})_{s}$ by definition. It thus follow from Lemma \[epiden\ ] that the second expression in Equation is just $ -sre_{r - k}\big((\lambda^{>s})_{s}\big)$. Since the first term there is $ |\lambda|e _ {

lamhda^{>s})_{s}\big)-s(s)_{s-1}\sum_{k=1}^{r}\big(-(s+1)_{s}\big)^{y-1}e_{r-k}\big((\lambda^{>s})_{s}\yug)\big]$$ (vhis is also vauid when $s=0$, where the sum ovec $k$ us ineeed multiplied by a vxnishing boefficiebt). Oi the other hand, when $j>s+1$ we obacrve chet $(j)_{s}-(j-1)_{s}=s(j-1)_{s-1}$ (once again thiv equality holgs apso when $s=0$, in the form of $1-1=0$), so that re get $$km_{u}\big[e_{r}\big((\lambdw^{>s})_{s}\bpg)-f(j-1)_{s-1}\shm_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big)\big],$$ kf whici the latter eqiality is the case with $j=s+1$. Altlgether the left hwnd side of Equwrion equals $$\rum_{j\geq1}jm_{j}t_{r}\yig((\lambda^{>s})_{s}\gig)-\sum_{j>s}sjm_{j}(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\cig((\lakbda^{>s})_{s}\big). \oavel{PVS}$$ But recaloing nhat $j(j-1)_{s-1}=(j)_{s}$ fov every such $j$, we deduce thab for eaxh $1 \leq k \leq r$, the snmmand $e_{r-k}\big((\lambda^{>s})_{f}\big)$ (whicv ns independent of $j$) iw nultiklied by $-r\wum_{g>s}(-1)^{k-1}(n)_{s}^{l}$, shich fquels $-s(-1)^{k}p_{k}\big((\mambda^{>s})_{s}$ by definition. It thus fjolows from Lejma \[epydqn\] that the second expression in Equatimn js just $-sre_{r-k}\big((\lambda^{>s})_{w}\big)$. Since the first jerm there is $|\lambda|e_{

lambda^{>s})_{s}\big)-s(s)_{s-1}\sum_{k=1}^{r}\big(-(s+1)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big)\big]$$ (this is also valid when $s=0$, sum $k$ is multiplied by a hand, $j>s+1$ we observe $(j)_{s}-(j-1)_{s}=s(j-1)_{s-1}$ (once again equality holds also when $s=0$, in form of $1-1=0$), so that we get $$jm_{j}\big[e_{r}\big((\lambda^{>s})_{s}\big)-s(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big)\big],$$ of which the latter equality the case with $j=s+1$. Altogether the left hand side of Equation equals $$\sum_{j\geq1}jm_{j}e_{r}\big((\lambda^{>s})_{s}\big)-\sum_{j>s}sjm_{j}(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big). But that for such $j$, we deduce that for each $1 \leq k \leq r$, the summand $e_{r-k}\big((\lambda^{>s})_{s}\big)$ (which independent of $j$) is multiplied by $-s\sum_{j>s}(-1)^{k-1}(j)_{s}^{k}$, which $-s(-1)^{k}p_{k}\big((\lambda^{>s})_{s}$ by definition. It follows from Lemma \[epiden\] that second in Equation just Since first term there $|\lambda|e_{

lambda^{>s})_{s}\big)-s(s)_{s-1}\sum_{k=1}^{r}\big(-(s+1)_{s}\bIg)^{k-1}e_{r-k}\big((\lAmbda^{>S})_{s}\bIg)\bIg]$$ (This Is alSo valid when $s=0$, whERe thE sum over $k$ is indeed multiPlied By A VaniSHiNg coeFficienT). on THE otHeR hAnd, WhEN $j>S+1$ we obSerVe that $(j)_{S}-(j-1)_{s}=s(j-1)_{s-1}$ (once AgaIn This equality HOlDs also when $S=0$, in The form of $1-1=0$), so tHat We get $$jM_{j}\Big[E_{R}\big((\lAmbDa^{>s})_{s}\bIg)-s(j-1)_{s-1}\sUM_{k=1}^{r}\big(-(J)_{s}\big)^{k-1}e_{r-k}\BiG((\Lambda^{>S})_{S}\big)\big],$$ OF WhIch tHe latter equality iS ThE Case with $j=s+1$. AltoGether ThE LeFT HanD siDe of EquatiOn EqualS $$\Sum_{j\geq1}JM_{j}E_{R}\BIg((\lAMbda^{>s})_{s}\big)-\sum_{j>S}sjm_{j}(j-1)_{s-1}\sum_{k=1}^{R}\Big(-(J)_{s}\big)^{k-1}E_{r-K}\biG((\Lambda^{>S})_{s}\big). \LaBEl{LhS}$$ But recallIng tHat $j(j-1)_{s-1}=(j)_{s}$ fOr everY Such $j$, we DEduce thAt for eAch $1 \Leq K \leq R$, ThE sUmmAnD $E_{r-k}\BIg((\LamBDa^{>s})_{S}\big)$ (whicH iS iNdepeNdenT OF $J$) Is muLtiPlieD by $-s\sUm_{j>s}(-1)^{k-1}(j)_{s}^{k}$, which EquAls $-s(-1)^{K}P_{k}\bIg((\lamBda^{>s})_{s}$ By deFiNitioN. It thuS follOwS from Lemma \[epideN\] thaT the seconD exPrEssIoN in EqUAtion iS juSt $-sRe_{r-k}\big((\Lambda^{>s})_{S}\Big)$. siNCE ThE first term there is $|\lAmBDA|e_{

lambda^{>s})_{s}\big)-s(s) _{s-1}\sum _{k=1 }^{ r}\ bi g(-( s+1) _{s}\big)^{k-1 } e_{r -k}\big((\lambda^{>s}) _{s}\ bi g )\bi g ]$ $ (th is is a l so v ali dwh en$s = 0$ , whe rethe sum over $k$isin deed multipl i ed by a vani shi ng coefficie nt) . On t he ot h er ha nd, when $j>s+ 1 $ we o bserve th at $(j)_{ s }-(j-1) _ { s} =s(j -1)_{s-1}$ (oncea ga i n this equalit y hold sa ls o whe n $ s=0$, in t he form of $1-1 = 0$ ) , sot hat we get $$ jm_{j}\big[ e _{r }\big( (\ lam b da^{>s })_{s }\ b ig) -s(j-1)_{s- 1}\s um_{k=1}^ {r}\bi g (-(j)_{ s }\big)^ {k-1}e _{r -k} \big ( (\ la mbd a^ { >s} ) _{ s}\ b ig) \big],$$ o fwhich the l a t terequ alit y isthe case with $j =s+1 $ . A ltoge therthele ft ha nd sid e ofEq uation equals $ $\su m_{j\geq1 }jm _{ j}e _{ r}\bi g ((\lam bda ^{> s})_{s} \big)-\ s um_ {j > s } sj m_{j}(j-1)_{s-1}\s um _ { k= 1}^{r}\b ig(-(j ) _{ s} \ big)^{k- 1} e_{ r-k} \ b ig((\ lamb d a^ {>s})_{s }\big) . \ la bel{LHS }$ $ But r eca lli ng th a t $j (j-1)_ {s-1}=(j )_{s} $ for every suc h $j$, we dedu c et h at foreac h $1 \leq k \le q r$, the su mma n d $e_ {r-k} \b i g( ( \lambda^{>s})_{s}\b ig )$ (wh ich i s independent of $j$) i s m ultiplie d by $- s \sum_{j>s}(-1) ^{k-1 }(j)_{s}^{ k }$, whic h equ als $-s( -1)^{k}p_ { k }\big((\ lam bda ^{> s}) _ { s} $ by definiti o n . It t hus fol low s fromLem ma\[e pid en \] that t he secon dex pr es sio n inE quationis ju st $- sre_{ r -k}\bi g((\l ambd a^ {> s })_ {s}\big ) $. S ince t he fir stte rm th erei s $ |\lambd a|e_{

lambda^{>s})_{s}\big)-s(s)_{s-1}\sum_{k=1}^{r}\big(-(s+1)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big)\big]$$ (this_is also_valid when $s=0$, where_the sum_over_$k$ is_indeed_multiplied by a_vanishing coefficient). On_the other hand, when_$j>s+1$ we observe_that_$(j)_{s}-(j-1)_{s}=s(j-1)_{s-1}$ (once again this equality holds also when $s=0$, in the form of $1-1=0$),_so_that we_get_$$jm_{j}\big[e_{r}\big((\lambda^{>s})_{s}\big)-s(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big)\big],$$_of which the latter equality_is the case with $j=s+1$._Altogether the_left hand side of Equation equals $$\sum_{j\geq1}jm_{j}e_{r}\big((\lambda^{>s})_{s}\big)-\sum_{j>s}sjm_{j}(j-1)_{s-1}\sum_{k=1}^{r}\big(-(j)_{s}\big)^{k-1}e_{r-k}\big((\lambda^{>s})_{s}\big). \label{LHS}$$ But_recalling_that $j(j-1)_{s-1}=(j)_{s}$ for_every such $j$, we deduce that for each $1_\leq k \leq r$, the summand_$e_{r-k}\big((\lambda^{>s})_{s}\big)$ (which is_independent_of_$j$) is multiplied by_$-s\sum_{j>s}(-1)^{k-1}(j)_{s}^{k}$, which equals $-s(-1)^{k}p_{k}\big((\lambda^{>s})_{s}$ by definition._It thus follows from Lemma \[epiden\]_that the second expression in Equation is_just $-sre_{r-k}\big((\lambda^{>s})_{s}\big)$. Since the first term_there is $|\lambda|e_{

(\[dNcp2\]) becomes $$\begin{aligned} p_{1,t}=0,\quad r_{1,t}=0,\quad q_{1,t}=-\frac{1}{3}(p_1^2+r_1^2), \label{dNcp21}\end{aligned}$$ which gives $$\begin{aligned} p_{1}=c_1,\quad r_{1}=c_2,\quad q_{1}=-\frac{1}{3}(c_1^2+c_2^2)t, \label{dNcp21s}\end{aligned}$$ where $c_{1}$ and $c_{2}$ are real valued integration constants. Thus we arrive at the single-peakon solution $$\begin{aligned} u=(c_1+ic_{2})e^{-\mid x+\frac{c_1^2+c_2^2}{3}t\mid}=ce^{-\mid x+\frac{1}{3}|c|^2t\mid},\label{ocpnp2}\end{aligned}$$ where $c=c_{1}+ic_{2}$ and $|c|$ is the modulus of $c$. For $N=2$, we may solve (\[dNcp2\]) as $$\begin{aligned} \left\{\begin{array}{l} q_{1}(t)=-\frac{1}{3}A_1^2t+\Gamma_{1}(t),\\ q_{2}(t)=-\frac{1}{3}A_2^2t+\Gamma_{1}(t),\\ p_{1}(t)=A_1\sin(\Gamma_{2}(t)+A_3),\\ p_{2}(t)=A_2\sin(\Gamma_{2}(t)+A_4),\\ r_{1}(t)=A_1\cos(\Gamma_{2}(t)+A_3),\\ r_{2}(t)=A_2\cos(\Gamma_{2}(t)+A_4), \end{array}\right. \label{2pq}\end{aligned}$$ where $$\begin{aligned} \begin{split} \Gamma_{1}(t)&=\frac{3A_1A_2\cos(A_3-A_4)}{|A_1^2-A_2^2|}sgn(t

(\[dNcp2\ ]) becomes $ $ \begin{aligned } p_{1,t}=0,\quad r_{1,t}=0,\quad q_{1,t}=-\frac{1}{3}(p_1 ^ 2+r_1 ^ 2), \label{dNcp21}\end{aligned}$$ which gives $ $ \begin{aligned } p_{1}=c_1,\quad r_{1}=c_2,\quad q_{1}=-\frac{1}{3}(c_1 ^ 2+c_2 ^ 2)t, \label{dNcp21s}\end{aligned}$$ where $ c_{1}$ and $ c_{2}$ are real valued consolidation constant. Thus we arrive at the single - peakon solution $ $ \begin{aligned } u=(c_1+ic_{2})e^{-\mid x+\frac{c_1 ^ 2+c_2 ^ 2}{3}t\mid}=ce^{-\mid x+\frac{1}{3}|c|^2t\mid},\label{ocpnp2}\end{aligned}$$ where $ vitamin c = c_{1}+ic_{2}$ and $ |c|$ is the modulus of $ c$. For $ N=2 $, we may clear (\[dNcp2\ ]) as $ $ \begin{aligned } \left\{\begin{array}{l } q_{1}(t)=-\frac{1}{3}A_1 ^ 2t+\Gamma_{1}(t),\\ q_{2}(t)=-\frac{1}{3}A_2 ^ 2t+\Gamma_{1}(t),\\ p_{1}(t)=A_1\sin(\Gamma_{2}(t)+A_3),\\ p_{2}(t)=A_2\sin(\Gamma_{2}(t)+A_4),\\ r_{1}(t)=A_1\cos(\Gamma_{2}(t)+A_3),\\ r_{2}(t)=A_2\cos(\Gamma_{2}(t)+A_4), \end{array}\right. \label{2pq}\end{aligned}$$ where $ $ \begin{aligned } \begin{split } \Gamma_{1}(t)&=\frac{3A_1A_2\cos(A_3 - A_4)}{|A_1 ^ 2 - A_2 ^ 2|}sgn(t

(\[dNfp2\]) becomes $$\begin{aligned} p_{1,u}=0,\quad r_{1,t}=0,\quad q_{1,t}=-\frac{1}{3}(k_1^2+r_1^2), \oabel{dIcp21}\end{amigned}$$ wfich gives $$\begin{aligned} p_{1}=c_1,\quav r_{1}=c_2,\qyad q_{1}=-\feac{1}{3}(c_1^2+c_2^2)t, \label{dNcp21s}\end{aliened}$$ wherv $c_{1}$ and $c_{2}$ are eeal valuev integration dlnstcnvs. Thus we arriye at the shngle-peakon sonugiln $$\begin{aligned} u=(c_1+ic_{2})e^{-\mid x+\frac{c_1^2+c_2^2}{3}t\mid}=se^{-\mid x+\grwc{1}{3}|c|^2t\mid},\label{ocknp2}\enc{wlighvd}$$ where $c=c_{1}+ic_{2}$ and $|c|$ is the modumus of $b$. For $N=2$, we may solfe (\[dNcp2\]) as $$\begin{aligned} \lefh\{\begln{array}{l} q_{1}(t)=-\frac{1}{3}A_1^2t+\Galma_{1}(t),\\ q_{2}(t)=-\frac{1}{3}A_2^2j+\Famiq_{1}(t),\\ p_{1}(t)=A_1\sin(\Gammx_{2}(t)+A_3),\\ p_{2}(t)=A_2\sin(\Gamma_{2}(t)+A_4),\\ r_{1}(t)=A_1\cos(\Famma_{2}(t)+A_3),\\ r_{2}(t)=A_2\cos(\Gamma_{2}(t)+A_4), \end{array}\rigft. \layel{2pq}\end{alitnwd}$$ avere $$\begin{aoigneq} \begin{split} \Gamma_{1}(t)&=\fraw{3A_1A_2\cos(A_3-S_4)}{|A_1^2-A_2^2|}sgn(t

(\[dNcp2\]) becomes $$\begin{aligned} p_{1,t}=0,\quad r_{1,t}=0,\quad q_{1,t}=-\frac{1}{3}(p_1^2+r_1^2), \label{dNcp21}\end{aligned}$$ $$\begin{aligned} r_{1}=c_2,\quad q_{1}=-\frac{1}{3}(c_1^2+c_2^2)t, where $c_{1}$ and constants. we arrive at single-peakon solution $$\begin{aligned} x+\frac{c_1^2+c_2^2}{3}t\mid}=ce^{-\mid x+\frac{1}{3}|c|^2t\mid},\label{ocpnp2}\end{aligned}$$ where $c=c_{1}+ic_{2}$ and $|c|$ the modulus of $c$. For $N=2$, we may solve (\[dNcp2\]) as $$\begin{aligned} \left\{\begin{array}{l} q_{2}(t)=-\frac{1}{3}A_2^2t+\Gamma_{1}(t),\\ p_{1}(t)=A_1\sin(\Gamma_{2}(t)+A_3),\\ p_{2}(t)=A_2\sin(\Gamma_{2}(t)+A_4),\\ r_{1}(t)=A_1\cos(\Gamma_{2}(t)+A_3),\\ r_{2}(t)=A_2\cos(\Gamma_{2}(t)+A_4), \end{array}\right. \label{2pq}\end{aligned}$$ where $$\begin{aligned} \begin{split} \Gamma_{1}(t)&=\frac{3A_1A_2\cos(A_3-A_4)}{|A_1^2-A_2^2|}sgn(t

(\[dNcp2\]) becomes $$\begin{aligned} p_{1,t}=0,\Quad r_{1,t}=0,\quad Q_{1,t}=-\fraC{1}{3}(p_1^2+r_1^2), \LabEl{DNcp21}\End{aLigned}$$ which givES $$\begIn{aligned} p_{1}=c_1,\quad r_{1}=c_2,\quad q_{1}=-\Frac{1}{3}(c_1^2+C_2^2)t, \LAbel{DncP21s}\end{Aligned}$$ WHeRE $C_{1}$ anD $c_{2}$ ArE reAl VAlUed inTegRation cOnstants. ThUs wE aRrive at the siNGlE-peakon solUtiOn $$\begin{alignEd} u=(C_1+ic_{2})e^{-\miD x+\FraC{C_1^2+c_2^2}{3}t\miD}=ce^{-\Mid x+\fRac{1}{3}|c|^2t\mID},\label{Ocpnp2}\end{aLiGNed}$$ wheRE $c=c_{1}+ic_{2}$ anD $|C|$ Is The mOdulus of $c$. For $N=2$, we maY SoLVe (\[dNcp2\]) as $$\begin{aLigned} \LeFT\{\bEGIn{aRraY}{l} q_{1}(t)=-\frac{1}{3}A_1^2t+\gaMma_{1}(t),\\ q_{2}(T)=-\Frac{1}{3}A_2^2t+\GAMmA_{1}(T),\\ P_{1}(T)=A_1\sIN(\Gamma_{2}(t)+A_3),\\ p_{2}(t)=A_2\siN(\Gamma_{2}(t)+A_4),\\ r_{1}(t)=A_1\COs(\GAmma_{2}(t)+A_3),\\ R_{2}(t)=a_2\coS(\gamma_{2}(t)+a_4), \end{aRrAY}\riGht. \label{2pq}\eNd{alIgned}$$ wherE $$\begin{ALigned} \bEGin{spliT} \Gamma_{1}(T)&=\frAc{3A_1a_2\cos(a_3-a_4)}{|A_1^2-a_2^2|}sGn(t

(\[dNcp2\]) becomes $$\be gin{aligne d} p_ {1, t}= 0, \qua d r_ {1,t}=0,\quadq _{1, t}=-\frac{1}{3}(p_1^2+ r_1^2 ), \lab e l{ dNcp2 1}\end{ a li g n ed} $$ w hic hg iv es $$ \be gin{ali gned} p_{1 }=c _1 ,\quad r_{1} = c_ 2,\quad q_ {1} =-\frac{1}{3 }(c _1^2+c _2 ^2) t , \la bel {dNcp 21s}\e n d{alig ned}$$ wh er e $c_{1 } $ and $ c _ {2 }$ a re real valued in t eg r ation constant s. Thu sw ea r riv e a t the sing le -peak o n solut i on $ $ \be g in{aligned} u =(c_1+ic_{2 } )e^ {-\mid x +\f r ac{c_1 ^2+c_ 2^ 2 }{3 }t\mid}=ce^ {-\m id x+\fra c{1}{3 } |c|^2t\ m id},\la bel{oc pnp 2}\ end{ a li gn ed} $$ whe r e$c= c _{1 }+ic_{2} $an d $|c |$ i s t h e mo dul us o f $c$ . For $N=2$, we may sol ve (\ [dNcp 2\]) a s $$\ begin{ align ed } \left\{\begin {arr ay}{l} q_ {1} (t )=- \f rac{1 } {3}A_1 ^2t +\G amma_{1 }(t),\\ q_{ 2} ( t ) =- \frac{1}{3}A_2^2t+ \G a m ma _{1}(t), \\ p_{ 1 }( t) = A_1\sin( \G amm a_{2 } ( t)+A_ 3),\ \ p _{2}(t)= A_2\si n (\ Ga mma_{2} (t )+A_4) ,\ \ r _{1 }(t)= A _1\c os(\Ga mma_{2}( t)+A_ 3 ),\\ r_{2}(t)= A _2\cos(\Gamma _ {2 } ( t) + A_4) , \ end{array}\ righ t . \l abel { 2p q}\ e nd{al igned }$ $ w h ere $$\begin{aligne d} \begi n{spl it} \Gamma_{1 }(t)&=\fra c { 3 A_1A_2\c os(A _ 3- A _4)}{|A_1^2-A_ 2^2|} sgn(t

(\[dNcp2\])_becomes $$\begin{aligned} p_{1,t}=0,\quad r_{1,t}=0,\quad q_{1,t}=-\frac{1}{3}(p_1^2+r_1^2), \label{dNcp21}\end{aligned}$$_which gives $$\begin{aligned} p_{1}=c_1,\quad r_{1}=c_2,\quad q_{1}=-\frac{1}{3}(c_1^2+c_2^2)t, \label{dNcp21s}\end{aligned}$$ where_$c_{1}$ and_$c_{2}$_are real_valued_integration constants. Thus_we arrive at_the single-peakon solution $$\begin{aligned} u=(c_1+ic_{2})e^{-\mid_x+\frac{c_1^2+c_2^2}{3}t\mid}=ce^{-\mid x+\frac{1}{3}|c|^2t\mid},\label{ocpnp2}\end{aligned}$$ where_$c=c_{1}+ic_{2}$_and $|c|$ is the modulus of $c$. For $N=2$, we may solve (\[dNcp2\]) as $$\begin{aligned} \left\{\begin{array}{l} q_{1}(t)=-\frac{1}{3}A_1^2t+\Gamma_{1}(t),\\ q_{2}(t)=-\frac{1}{3}A_2^2t+\Gamma_{1}(t),\\ p_{1}(t)=A_1\sin(\Gamma_{2}(t)+A_3),\\ p_{2}(t)=A_2\sin(\Gamma_{2}(t)+A_4),\\ r_{1}(t)=A_1\cos(\Gamma_{2}(t)+A_3),\\ r_{2}(t)=A_2\cos(\Gamma_{2}(t)+A_4), \end{array}\right._\label{2pq}\end{aligned}$$_where $$\begin{aligned} \begin{split} \Gamma_{1}(t)&=\frac{3A_1A_2\cos(A_3-A_4)}{|A_1^2-A_2^2|}sgn(t

$v^{th}$-order derivatives of the PGFs of the clutter cardinality distribution and the predicted cardinality distribution as $$\begin{aligned} C^{(v)}_{j}(t) &= \frac{d^v C_{j}}{d t^v}(t) \,, \quad \pgf^{(v)}(t) = \frac{d^v \pgf}{d t^v}(t).\end{aligned}$$ We use $\gamma$ to denote the probability, under the predictive PHD, that a target is detected by no sensor, and we thus have: $$\gamma {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\int \! r(\x) \prod_{j=1}^{s}{q^{j}_{d}(\x)} d\x \,. \label{eq:gamma}$$ For concise specification of the update equations, it is useful to combine the terms associated with the PGF of the clutter cardinality distribution for a partition $P$. Let us define the quantity $$\begin{aligned} \kappa_{P} {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\prod_{j=1}^{s} C^{(m_{j}-|P|_{j})}_{j}(0). \label{eq:kappa_P}\end{aligned}$$ For a set $W \in \mathcal{W}$ and the associated index set $T_{W}$ define the quantities $$\begin{aligned} &d_{W} {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\frac{\displaystyle \int \!\! r(\x) \left( \prod_{(i,l) \in T_{W}}{\!\!\!\! p_{d}^{i}(\x)\,h_{i}(\z^{i}_{l}|\x)}\right) \, \!\!\! \prod_{j:(j,*) \notin T_{W}}{\!\!\!\!\!\! q^{j}_{d}(\x)} d\x} {\displaystyle \prod_{(i,l) \in T_{W}}{c_{i}(\z^{i}_{l})}}, \label{eq:dW_cphd} \\ &\rho_{W}(\x) {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont

$ v^{th}$-order derivatives of the PGFs of the clutter cardinality distribution and the predicted cardinality distribution as $ $ \begin{aligned } C^{(v)}_{j}(t) & = \frac{d^v C_{j}}{d t^v}(t) \, , \quad \pgf^{(v)}(t) = \frac{d^v \pgf}{d t^v}(t).\end{aligned}$$ We use $ \gamma$ to denote the probability, under the predictive PHD, that a prey is detect by no sensor, and we thus get: $ $ \gamma { \mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\int \! r(\x) \prod_{j=1}^{s}{q^{j}_{d}(\x) } d\x \, . \label{eq: gamma}$$ For concise stipulation of the update equations, it is useful to combine the condition associated with the PGF of the clutter cardinality distribution for a partition $ P$. permit us define the quantity $ $ \begin{aligned } \kappa_{P } { \mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\prod_{j=1}^{s } C^{(m_{j}-|P|_{j})}_{j}(0). \label{eq: kappa_P}\end{aligned}$$ For a laid $ W \in \mathcal{W}$ and the associated index set $ T_{W}$ specify the quantity $ $ \begin{aligned } & d_{W } { \mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\frac{\displaystyle \int \!\! r(\x) \left (\prod_{(i, l) \in T_{W}}{\!\!\!\! p_{d}^{i}(\x)\,h_{i}(\z^{i}_{l}|\x)}\right) \, \!\!\! \prod_{j:(j, *) \notin T_{W}}{\!\!\!\!\!\! q^{j}_{d}(\x) } d\x } { \displaystyle \prod_{(i, l) \in T_{W}}{c_{i}(\z^{i}_{l }) } }, \label{eq: dW_cphd } \\ & \rho_{W}(\x) { \mathrel{\overset{\makebox[0pt]{\mbox{\normalfont

$v^{tj}$-order derivatives of tht PGFs of the clujtwr carvinalitg districution and the predicted carvinaoity eistribution as $$\begin{auigned} C^{(v)}_{j}(n) &= \frac{d^v C_{j}}{d r^v}(t) \,, \quad \pjr^{(v)}(t) = \frac{d^v \pfn}{d t^v}(c).\eid{aligned}$$ We use $\gamma$ to genote the protacipity, under the predictive PHD, that w targey ls detected by no xqnsod, and we thus have: $$\gamma {\mathrel{\ovsrset{\manebox[0pt]{\mbox{\notmalfont\tiny\sffamily def}}}{=}}}\inh \! r(\d) \prod_{j=1}^{s}{q^{j}_{d}(\x)} d\x \,. \lwbel{eq:gamma}$$ Dor sincise speciwication of the update equations, it is useful to combkne tke terms aswoxiahgd with the 'GF of the clutter cardinanity dixtribution for a 'artution $P$. Let us define the quantity $$\begig{aligned} \kdp'a_{P} {\mathrel{\overset{\makwbix[0pt]{\mtox{\nmrmaudong\tihy\xframily deh}}}{=}}}\prod_{j=1}^{s} C^{(m_{j}-|L|_{j})}_{j}(0). \label{eq:jappa_P}\end{aligned}$$ For a wet $W \in \mathdal{W}$ agd the associated index set $T_{W}$ define the quzntities $$\begin{aligned} &d_{W} {\mathrel{\overset{\makebod[0pt]{\mbox{\nowmalfont\tiny\sffamily def}}}{=}}}\frac{\displaystyle \int \!\! r(\x) \neft( \'rud_{(i,o) \ln T_{D}}{\!\!\!\! p_{f}^{i}(\x)\,h_{i}(\z^{i}_{l}|\x)}\right) \, \!\!\! \prod_{j:(j,*) \notin T_{W}}{\!\!\!\!\!\! q^{j}_{d}(\x)} d\x} {\cixplaystyle \prob_{(i,l) \in T_{W}}{c_{i}(\z^{i}_{l})}}, \lsbfl{rz:dW_cphd} \\ &\rho_{W}(\x) {\mathrzm{\oberset{\makebox[0pt]{\mbod{\normalsont

$v^{th}$-order derivatives of the PGFs of the distribution the predicted distribution as $$\begin{aligned} \,, \pgf^{(v)}(t) = \frac{d^v t^v}(t).\end{aligned}$$ We use to denote the probability, under the PHD, that a target is detected by no sensor, and we thus have: {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\int \! r(\x) \prod_{j=1}^{s}{q^{j}_{d}(\x)} d\x \,. \label{eq:gamma}$$ For concise specification of the equations, is to the terms associated with the PGF of the clutter cardinality distribution for a partition $P$. Let define the quantity $$\begin{aligned} \kappa_{P} {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\prod_{j=1}^{s} C^{(m_{j}-|P|_{j})}_{j}(0). For a set $W \mathcal{W}$ and the associated index $T_{W}$ the quantities &d_{W} def}}}{=}}}\frac{\displaystyle \!\! r(\x) \left( \in T_{W}}{\!\!\!\! p_{d}^{i}(\x)\,h_{i}(\z^{i}_{l}|\x)}\right) \, \!\!\! \prod_{j:(j,*) \notin T_{W}}{\!\!\!\!\!\! q^{j}_{d}(\x)} d\x} {\displaystyle \prod_{(i,l) \in T_{W}}{c_{i}(\z^{i}_{l})}}, \label{eq:dW_cphd} \\ &\rho_{W}(\x)

$v^{th}$-order derivatives of the PgFs of the clUtter CarDinAlIty dIstrIbution and the pREdicTed cardinality distribuTion aS $$\bEGin{aLIgNed} C^{(v)}_{J}(t) &= \frac{d^V c_{j}}{D T^V}(t) \,, \qUaD \pGf^{(v)}(T) = \fRAc{D^v \pgf}{D t^v}(T).\end{aliGned}$$ We use $\gAmmA$ tO denote the prOBaBility, undeR thE predictive PhD, tHat a taRgEt iS DetecTed By no sEnsor, aND we thuS have: $$\gammA {\mAThrel{\oVErset{\maKEBoX[0pt]{\mBox{\normalfont\tiny\SFfAMily def}}}{=}}}\int \! r(\x) \prOd_{j=1}^{s}{q^{j}_{D}(\x)} D\X \,. \lABEl{eQ:gaMma}$$ For concIsE specIFicatioN Of THE UpdATe equations, it Is useful to cOMbiNe the tErMs aSSociatEd witH tHE PGf of the cluttEr caRdinality DistriBUtion foR A partitIon $P$. LeT us DefIne tHE qUaNtiTy $$\BEgiN{AlIgnED} \kaPpa_{P} {\mathReL{\oVerseT{\makEBOX[0Pt]{\mbOx{\nOrmaLfont\Tiny\sffamily dEf}}}{=}}}\pRod_{j=1}^{S} c^{(m_{j}-|p|_{j})}_{j}(0). \laBel{eq:KappA_P}\End{alIgned}$$ FOr a seT $W \In \mathcal{W}$ and thE assOciated inDex SeT $T_{W}$ DeFine tHE quantItiEs $$\bEgin{aliGned} &d_{W} {\mAThrEl{\OVERsEt{\makebox[0pt]{\mbox{\norMaLFOnT\tiny\sffAmily dEF}}}{=}}}\fRaC{\DisplaysTyLe \iNt \!\! r(\x) \LEFt( \proD_{(i,l) \iN t_{W}}{\!\!\!\! P_{d}^{i}(\x)\,h_{i}(\z^{i}_{L}|\x)}\righT) \, \!\!\! \PrOd_{J:(j,*) \notin t_{W}}{\!\!\!\!\!\! Q^{j}_{d}(\x)} d\x} {\DiSplAysTyle \pROd_{(i,l) \In T_{W}}{c_{i}(\Z^{i}_{l})}}, \label{Eq:dW_cPHd} \\ &\rho_{W}(\x) {\mathrel{\OVerset{\makebox[0PT]{\mBOX{\nORmalFonT

$v^{th}$-order derivative s of the P GFs o f t hecl utte r ca rdinality dist r ibut ion and the predictedcardi na l ityd is tribu tion as $$ \ b egi n{ al ign ed } C ^{(v) }_{ j}(t) & = \frac{d^ v C _{ j}}{d t^v}(t ) \ ,, \quad \ pgf ^{(v)}(t) =\fr ac{d^v \ pgf } {d t^ v}( t).\e nd{ali g ned}$$ We use$\ g amma$t o denot e th e pr obability, undert he predictive PHD , that a ta r g etisdetected b yno se n sor, an d w e t hus have: $$\gamm a {\mathrel { \ov erset{ \m ake b ox[0pt ]{\mb ox { \no rmalfont\ti ny\s ffamily d ef}}}{ = }}}\int \! r(\x ) \pro d_{ j=1 }^{s } {q ^{ j}_ {d } (\x ) }d\x \,. \label{ eq :g amma} $$ F o r conc ise spe cific ation of theupd atee qua tions , itis u se ful t o comb ine t he terms associat ed w ith the P GFof th eclutt e r card ina lit y distr ibution for a p a rt ition $P$. Let usde f i ne the qua ntity$ $\ be g in{align ed } \ kapp a _ {P} { \mat h re l{\overs et{\ma k eb ox [0pt]{\ mb ox{\no rm alf ont \tiny \ sffa mily d ef}}}{=} }}\pr o d_{j=1}^{s} C^ { (m_{j}-|P|_{j } )} _ { j} ( 0).\la bel{eq:kapp a_P} \ end{ alig n ed }$$ Fora set $ W \ i n \mathcal{W}$ andth e asso ciate d index set $ T_{W}$ def i n e the qua ntit i es $$\begin{align ed} & d_{W} {\ma t hrel{\ov erset {\makebo x[0pt]{\m b o x{\norma lfo nt\ tin y\s f f am ily def}}}{=} } } \fra c{ \displa yst yle \in t \ !\! r( \x) \ left( \pr od_{(i,l )\i nT_ {W} }{\!\ ! \!\! p_{ d} ^{i }( \x) \,h_{ i }(\z^{ i}_{l }|\x )} \r i ght ) \, \ ! \!\! \ pr od_{ j:( j, *) \n otin T_{ W}}{\!\ !\!\!\!\! q^ { j}_{ d} (\ x)} d\x } {\displ ay style \pro d_ {(i ,l) \i n T_{W}}{c _{i}(\z^{i}_{l})}}, \la b el{eq:d W_c phd}\\ & \rho_{W}( \x) {\mat hre l {\over set{\m akebo x[ 0pt ] { \mbox { \ no rma lf ont

$v^{th}$-order_derivatives of_the PGFs of the_clutter cardinality_distribution_and the_predicted_cardinality distribution as_$$\begin{aligned} C^{(v)}_{j}(t) &= \frac{d^v_C_{j}}{d t^v}(t) \,, \quad \pgf^{(v)}(t)_= \frac{d^v \pgf}{d_t^v}(t).\end{aligned}$$ We_use $\gamma$ to denote the probability, under the predictive PHD, that a target is_detected_by no_sensor,_and_we thus have: $$\gamma {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily_def}}}{=}}}\int \! r(\x) \prod_{j=1}^{s}{q^{j}_{d}(\x)} d\x_\,. \label{eq:gamma}$$ For_concise specification of the update equations, it is_useful_to combine the_terms associated with the PGF of the clutter cardinality_distribution for a partition $P$. Let_us define the_quantity_$$\begin{aligned} \kappa_{P}_{\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\prod_{j=1}^{s} C^{(m_{j}-|P|_{j})}_{j}(0). \label{eq:kappa_P}\end{aligned}$$ For_a set $W \in \mathcal{W}$ and_the associated index set $T_{W}$ define_the quantities $$\begin{aligned} &d_{W} {\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily def}}}{=}}}\frac{\displaystyle \int \!\!_r(\x) \left( \prod_{(i,l) \in T_{W}}{\!\!\!\! p_{d}^{i}(\x)\,h_{i}(\z^{i}_{l}|\x)}\right)_\, _\!\!\! \prod_{j:(j,*)_\notin T_{W}}{\!\!\!\!\!\! q^{j}_{d}(\x)} d\x} _ {\displaystyle_\prod_{(i,l) \in_T_{W}}{c_{i}(\z^{i}_{l})}}, \label{eq:dW_cphd} \\ &\rho_{W}(\x)_{\mathrel{\overset{\makebox[0pt]{\mbox{\normalfont

corresponds to the broken phase. In general the phase diagram depends on four parameters $\beta$, $\gamma$, $\nu$ and $\varphi$. In the $\mathcal{PT}$-symmetric phase only three of these parameters are independent and the phase diagram will contain a connected $3D$ region (symmetric phase) characterized by $F=0$. This is clearly a richer phase diagram than that obtained by using exclusively eigenstates of the S-matrix where the variables $\nu$ and $\varphi$ are eliminated by projecting on the $\beta$, $\gamma$ plane. Of course, this generalized phase diagram can be obtained using linear combinations of the S-matrix eigenstates since they cover the entire space of possible scattering states. However, to describe the unbroken phase via a linear combination $$\label{lin_comb} \vert\Psi\rangle=\mu_{1}\vert\Phi_{1}\rangle+\mu_{2}\vert\Phi_{2}\rangle,$$ where $\vert \Phi_{i} \rangle,~~i=1,2$ are the S-matrix eigenstates, one has to impose the following conditions (and all their combinations): $$\label{cond} \mu_{1}^{*}=\pm \mu_{1}~~~~;~~~~\mu_{2}^{*}=\pm \mu_{2},$$ for the (in general complex) expansion coefficients $\mu_{i},~~i=1,2$. This information is optimally encoded in the single condition $Q=0$, demonstrating the propriety of the invariant current $Q$ (or equivalently $F$) as an “order parameter” of the $\mathcal{PT}$ breaking transition. Even more, from the physical point of view, in a scattering experiment with incidence on either side of the potential, it is expected that the ratio $\vert \nu \vert$, being a tunable parameter for a given setup, can be varied in an arbitrary manner, influencing the scattering outcome. On the other hand, restricting the incident states to S-matrix eigenstates, fixes the ratio $|\nu|$ to 1 for all the corresponding $\mathcal{PT}$-symmetric states. This can be shown by combining the conditions: $$\label{s_mat_cond} A=s~B~~~;~~~D=s~C,$$ which emerge from the fact that the incoming state is an eigenstate of the S-matrix ($s$ being the respective eigenvalue), with $$\label{s_mat_cond} B^{*}=\Lambda~D~~~;~~~A^{*}=\Lambda~C,$$ which stem from

corresponds to the broken phase. In general the phase diagram depends on four parameters $ \beta$, $ \gamma$, $ \nu$ and $ \varphi$. In the $ \mathcal{PT}$-symmetric phase merely three of these argument are independent and the phase diagram will contain a connected $ 3D$ area (symmetric phase) characterized by $ F=0$. This is clearly a deep phase diagram than that obtained by using entirely eigenstates of the south - matrix where the variables $ \nu$ and $ \varphi$ are rule out by projecting on the $ \beta$, $ \gamma$ plane. Of naturally, this generalized phase diagram can be obtained using linear combinations of the S - matrix eigenstates since they cover the entire outer space of possible scattering states. However, to identify the unbroken phase via a linear combination $ $ \label{lin_comb } \vert\Psi\rangle=\mu_{1}\vert\Phi_{1}\rangle+\mu_{2}\vert\Phi_{2}\rangle,$$ where $ \vert \Phi_{i } \rangle,~~i=1,2 $ are the S - matrix eigenstates, one have to impose the following conditions (and all their combinations ): $ $ \label{cond } \mu_{1}^{*}=\pm \mu_{1}~~~~;~~~~\mu_{2}^{*}=\pm \mu_{2},$$ for the (in general complex) expansion coefficient $ \mu_{i},~~i=1,2$. This information is optimally encoded in the single condition $ Q=0 $, demonstrating the propriety of the invariant current $ Q$ (or equivalently $ F$) as an “ club argument ” of the $ \mathcal{PT}$ breaking transition. Even more, from the physical point of view, in a scattering experiment with incidence on either side of the electric potential, it is expected that the proportion $ \vert \nu \vert$, being a tunable parameter for a given frame-up, can be vary in an arbitrary manner, influencing the scattering outcome. On the early hand, restrict the incident states to S - matrix eigenstates, fix the ratio $ |\nu|$ to 1 for all the corresponding $ \mathcal{PT}$-symmetric states. This can be shown by combining the conditions: $ $ \label{s_mat_cond } A = south ~ B~~~;~~~D = s ~ C,$$ which emerge from the fact that the incoming state is an eigenstate of the S - matrix ($ s$ being the respective eigenvalue), with $ $ \label{s_mat_cond } B^{*}=\Lambda ~ D~~~;~~~A^{*}=\Lambda ~ C,$$ which stem from

cogresponds to the broken khase. In general jhw phasx diagrzm depenas on four parameters $\beta$, $\gemma$, $\nu$ abd $\varphi$. In the $\mathcxl{PT}$-symmenric phasw onot three of these parametsvs arz mndependent and the phase diagram will wovtcin a connected $3D$ region (symmetric prase) chsrwcterized by $F=0$. Thix is dlearly a richer phase diagram thah that mbtained by uxing exclusively eigenstatfs ov the S-matrix wherf the variavles $\bu$ and $\varphk$ are elimpuated by prkjecting on the $\beta$, $\gamma$ pland. Of cpurse, this gwnegdlized phasx diaggam can be obbsined gsing lonear combinatlons mf rhe S-matrix eigenstatxs since they cover jhe entire s'ace of possible scatreeing vtatas. Huqevdr, uo vesdribe hhe unbroken lhase via a linear combination $$\lwvel{lin_comb} \vedt\Psi\rwndle=\mu_{1}\vert\Phi_{1}\rangle+\mu_{2}\vert\Phi_{2}\rangle,$$ where $\vedt \Phi_{i} \rangle,~~i=1,2$ are the S-matrix eigenstates, lne has tj impose the following conditions (and all their cmmbinetkonw): $$\jxvep{cond} \mu_{1}^{*}=\pm \mu_{1}~~~~;~~~~\mu_{2}^{*}=\pm \mu_{2},$$ for the (in general complqs) txpsnsion coefficlents $\mu_{i},~~i=1,2$. This ingogmsjion is optimauly enekdsd in the single clnditiog $Q=0$, dwmonstratyng yhe propriety of the invariqnt current $Z$ (or equivalently $F$) as an “ordex paraketer” of the $\mathcal{PT}$ breaknng trznsition. Evfn more, fdum the physical ooikt mf view, in a scattering ex[eriment xith nncidencd on eithew side of hhe potential, it is expefted jhat tve ratio $\vfrt \nu \vert$, being a tunable pareketer for a bieen setup, ccn be yaried in an arfitrary manner, influeucing ghe scatteging outcmme. On the jther hand, revjricting the mncident ftatws ti S-matrkb eigenstates, gixes the ratio $|\nu|$ ro 1 for all the covrespuhding $\mathcal{PT}$-wymnetric states. Tnis cag he srmwn by combiting thd condktions: $$\labeo{r_mat_vond} A=s~B~~~;~~~D=s~C,$$ which emarge from the fact thay bhe incomung statq is an eigenxtate of the S-matrlx ($s$ ueing vhe rexpestive eigenvalue), with $$\label{s_maf_cond} B^{*}=\Lalbdw~D~~~;~~~A^{*}=\Lambda~C,$$ wrich stem from

corresponds to the broken phase. In general diagram on four $\beta$, $\gamma$, $\nu$ phase three of these are independent and phase diagram will contain a connected region (symmetric phase) characterized by $F=0$. This is clearly a richer phase diagram that obtained by using exclusively eigenstates of the S-matrix where the variables $\nu$ $\varphi$ eliminated projecting the $\beta$, $\gamma$ plane. Of course, this generalized phase diagram can be obtained using linear combinations the S-matrix eigenstates since they cover the entire of possible scattering states. to describe the unbroken phase a combination $$\label{lin_comb} where \Phi_{i} are the S-matrix one has to impose the following conditions (and all their combinations): $$\label{cond} \mu_{1}^{*}=\pm \mu_{1}~~~~;~~~~\mu_{2}^{*}=\pm \mu_{2},$$ for the general complex) $\mu_{i},~~i=1,2$. This is encoded the single condition the propriety of the invariant current $F$) as an “order parameter” of the $\mathcal{PT}$ transition. Even from the physical point of view, a scattering experiment with incidence on either side the potential, it is expected that the ratio $\vert \nu \vert$, being a tunable parameter given setup, can be in an arbitrary influencing scattering On other hand, the incident states to S-matrix eigenstates, fixes the ratio $|\nu|$ to for all the corresponding $\mathcal{PT}$-symmetric states. This can be shown the $$\label{s_mat_cond} A=s~B~~~;~~~D=s~C,$$ which from the fact that incoming is an eigenstate of ($s$ the $$\label{s_mat_cond} which from

corresponds to the broken phaSe. In generaL the pHasE diAgRam dEpenDs on four parameTErs $\bEta$, $\gamma$, $\nu$ and $\varphi$. In tHe $\matHcAL{PT}$-sYMmEtric Phase onLY tHREe oF tHeSe pArAMeTers aRe iNdependEnt and the pHasE dIagram will coNTaIn a connectEd $3D$ Region (symmetRic Phase) cHaRacTErizeD by $f=0$. This Is cleaRLy a ricHer phase dIaGRam thaN That obtAINeD by uSing exclusively eiGEnSTates of the S-matRix wheRe THe VARiaBleS $\nu$ and $\varpHi$ Are elIMinated BY pROJEctINg on the $\beta$, $\gaMma$ plane. Of cOUrsE, this gEnEraLIzed phAse diAgRAm cAn be obtaineD usiNg linear cOmbinaTIons of tHE S-matriX eigenStaTes SincE ThEy CovEr THe eNTiRe sPAce Of possibLe ScAtterIng sTATES. HowEveR, to dEscriBe the unbroken PhaSe viA A liNear cOmbinAtioN $$\lAbel{lIn_comb} \Vert\PSi\Rangle=\mu_{1}\vert\Phi_{1}\RangLe+\mu_{2}\vert\PHi_{2}\rAnGle,$$ WhEre $\veRT \Phi_{i} \rAngLe,~~i=1,2$ Are the S-Matrix eIGenStATES, oNe has to impose the foLlOWInG conditiOns (and ALl ThEIr combinAtIonS): $$\labEL{Cond} \mU_{1}^{*}=\pm \mU_{1}~~~~;~~~~\Mu_{2}^{*}=\Pm \mu_{2},$$ for tHe (in geNErAl Complex) ExPansioN cOefFicIents $\MU_{i},~~i=1,2$. THis infOrmation Is optIMally encoded in THe single condiTIoN $q=0$, DeMOnstRatIng the proprIety OF the InvaRIaNt cURrent $q$ (or eqUiVAlENtly $F$) as an “order paramEtEr” of thE $\mathCal{PT}$ breaking Transition. eVEN more, froM the PHySIcal point of vieW, in a sCattering eXPeriment With iNcidence On either sIDE of the poTenTiaL, it Is eXPEcTed that the ratIO $\Vert \Nu \Vert$, beiNg a Tunable ParAmeTer For A gIven setup, Can be varIeD iN aN aRbiTrary MAnner, infLuEncInG thE scatTEring oUtcomE. On tHe OtHEr hAnd, restRIcTINg thE iNcIdenT stAtEs to S-MatrIX eiGenstatEs, fixes thE raTIo $|\nu|$ To 1 FoR all the Corresponding $\MaThcal{PT}$-symMeTriC stateS. tHis can be Shown by combining the condITions: $$\laBel{S_mat_cOnd} A=S~B~~~;~~~D=s~C,$$ whicH emErge frOm tHE fact tHat the IncomInG stATE is an EIGeNstAtE of the S-matRIX ($s$ bEing tHe RespEctive eIgenvalue), with $$\label{S_Mat_Cond} B^{*}=\Lambda~D~~~;~~~A^{*}=\lamBda~C,$$ WHIcH stEM fROm

corresponds to the broken phase. In gene ral th ephas e di agram dependso n fo ur parameters $\beta$, $\ga mm a $, $ \ nu $ and $\varp h i$ . Inth e$\m at h ca l{PT} $-s ymmetri c phase on lyth ree of these pa rameters a reindependentand the p ha sed iagra m w ill c ontain a conn ected $3D $r egion( symmetr i c p hase ) characterized b y $ F =0$. This is c learly a ri c h erpha se diagram t han t h at obta i ne d b y u s ing exclusive ly eigensta t esof the S -ma t rix wh ere t he var iables $\nu $ an d $\varph i$ are elimina t ed by p roject ing on the $\ be ta$ ,$ \ga m ma $ p l ane . Of co ur se , thi s ge n e r a lize d p hase diag ram can be ob tai nedu sin g lin ear c ombi na tions of th e S-m at rix eigenstates sin ce they c ove rthe e ntire spaceofpos sible s catteri n g s ta t e s .However, to descri be t he unbroke n phas e v ia a linear c omb inat i o n $$\ labe l {l in_comb} \vert \ Ps i\ rangle= \m u_{1}\ ve rt\ Phi _{1}\ r angl e+\mu_ {2}\vert \Phi_ { 2}\rangle,$$ w h ere $\vert \P h i_ { i }\ rang le, ~~i=1,2$ ar e th e S-m atri x e ige n state s, on eh as to impose the follo wi ng con ditio ns (and all t heir combi n a t ions): $ $\la b el { cond} \mu_{1}^ {*}=\ pm \mu_{1} ~ ~~~;~~~~ \mu_{ 2}^{*}=\ pm \mu_{2 } , $$ for t he(in ge ner a l c omplex) expan s i on c oe fficien ts$\mu_{i },~ ~i= 1,2 $.Th is inform ation is o pt im al lyencod e d in the s ing le co nditi o n $Q=0 $, de mons tr at i ngthe pro p ri e t y of t he inv ari an t cur rent $Q$ (or eq uivalentl y $ F $) a san “order parameter” o fthe $\math ca l{P T}$ br e a king tra nsition. Even more, fro m the ph ysi cal p oint of view, in a sca tte r ing ex perime nt wi th in c i dence o neit he r side oft h e p otent ia l, i t is ex pected that the ra t io$\vert \nu \v ert $, b e i ng at un a ble p a ram e t er for a givensetup, can b e v aried in a n ar bi trary m anner,influ e ncing t he scatte ring outc om e. O n the other han d, restr icting th e inci d en t sta tes to S- ma tri x eig enstat e s,fixes the r at io $|\ nu|$to 1 for a ll the corresponding $\ mathca l{PT} $-s ymmetricsta t es. This can beshown by c omb ini ng th e c o nditi ons: $$ \la b el{s_ mat_ c ond} A=s~ B ~~ ~;~ ~ ~ D= s~C,$$ whic h e mer ge fr omt he fac t th at the incoming s t ate is an eige nsta t e of th e S-m at rix ($s$ being th er e spective e igenvalue), with $$ \l a bel{s _mat_c ond} B ^{*}=\L a m bd a ~D~~~; ~~~A ^{* }=\Lambda ~C, $$ which s te mf rom

corresponds_to the_broken phase. In general_the phase_diagram_depends on_four_parameters $\beta$, $\gamma$,_$\nu$ and $\varphi$._In the $\mathcal{PT}$-symmetric phase_only three of_these_parameters are independent and the phase diagram will contain a connected $3D$ region (symmetric_phase)_characterized by_$F=0$._This_is clearly a richer phase_diagram than that obtained by_using exclusively_eigenstates of the S-matrix where the variables $\nu$_and_$\varphi$ are eliminated_by projecting on the $\beta$, $\gamma$ plane. Of course, this_generalized phase diagram can be obtained_using linear combinations_of_the_S-matrix eigenstates since they_cover the entire space of possible_scattering states. However, to describe the_unbroken phase via a linear combination $$\label{lin_comb}_\vert\Psi\rangle=\mu_{1}\vert\Phi_{1}\rangle+\mu_{2}\vert\Phi_{2}\rangle,$$ where $\vert \Phi_{i} \rangle,~~i=1,2$ are_the S-matrix eigenstates, one has_to impose_the following conditions (and all_their combinations): $$\label{cond}_\mu_{1}^{*}=\pm \mu_{1}~~~~;~~~~\mu_{2}^{*}=\pm_\mu_{2},$$ for the_(in general complex) expansion coefficients $\mu_{i},~~i=1,2$._This information is_optimally encoded in the single condition_$Q=0$,_demonstrating the propriety_of_the_invariant current_$Q$ (or equivalently_$F$)_as an_“order_parameter” of the $\mathcal{PT}$ breaking transition._Even_more, from the physical point of view,_in a scattering experiment_with_incidence on either side_of the potential, it is_expected that the ratio $\vert \nu_\vert$, being_a tunable_parameter for a given setup, can be varied in an arbitrary_manner, influencing the scattering outcome. On_the other hand, restricting_the incident_states_to S-matrix eigenstates,_fixes_the ratio_$|\nu|$ to 1 for all the corresponding_$\mathcal{PT}$-symmetric states._This can be shown by combining_the conditions: $$\label{s_mat_cond} A=s~B~~~;~~~D=s~C,$$_which_emerge from the fact that the_incoming state is an eigenstate of_the S-matrix ($s$ being the_respective_eigenvalue),_with $$\label{s_mat_cond} B^{*}=\Lambda~D~~~;~~~A^{*}=\Lambda~C,$$ which stem_from

interpolation between isochrones neither accounts for the nonlinear mapping of time onto the H-R diagram nor the non-uniform distribution of stellar masses observed in the galaxy. As a consequence, straightforward interpolation between isochrones results in an age distribution for field stars that is biased towards older ages compared to the distribution predicted by stellar evolutionary theory. Bayesian inference of stellar age and mass aims to eliminate such a bias by accounting for observationally and/or theoretically motivated distribution functions for the physical parameters of interest. As an example, for a given point with error bars on the H-R diagram, a lower stellar mass should be considered more likely due to the initial mass function. Likewise, due to the longer main-sequence timescales for lower mass stars, a star that is observed to have evolved off the main sequence should have a probability distribution in mass that is skewed towards higher masses, i.e. because higher mass stars spend a more significant fraction of their entire lifetime in the post-MS stage. ### Bayes Formalism {#subsubsec:formalism} Bayesian estimation of the physical parameters can proceed from comparison of the data with a selection of models. Bayes’ Theorem states: $$P(\mathrm{model|data}) \propto P(\mathrm{data|model}) \times P(\mathrm{model})$$ The probability of a model given a set of data is proportional to the product of the probability of the data given the model and the probability of the model itself. In the language of Bayesian statistics, this is expressed as: $$\mathrm{posterior} \propto \mathrm{likelihood} \times \mathrm{prior}.$$ Our model is the set of stellar parameters, age ($\tau$) and mass ($M_*$), and our data are the measured effective temperature, $T_\mathrm{eff}$, and surface gravity, $\log g$, for a given star. At any given combination of age and mass, the predicted $T_\mathrm{eff}$ and $\log g$ are provided by stellar evolutionary models. The $\chi^2$ statistic for an individual model can be computed as follows: $$\begin{aligned} \chi^2 (\tau, M_*) &= \sum \frac{(O-E)^2}{\sigma^2} \\ &= \frac{[(T_\mathrm{eff})_O-(T_\mathrm{eff})_E]^2}{\sigma_{T_\mathrm{eff}}^2

interpolation between isochrones neither accounts for the nonlinear mapping of meter onto the H - R diagram nor the non - consistent distribution of stellar masses observed in the galax. As a consequence, straightforward interpolation between isochrones consequence in an age distribution for field star that is biased towards older long time compared to the distribution predicted by leading evolutionary theory. Bayesian inference of stellar age and mass aims to eliminate such a bias by account for observationally and/or theoretically motivated distribution functions for the physical parameters of pastime. As an example, for a given point with error parallel bars on the H - R diagram, a lower stellar mass should be considered more likely due to the initial mass affair. Likewise, due to the longer main - sequence timescales for lower mass stars, a star that is observed to have evolved off the main sequence should have a probability distribution in mass that is skewed towards higher masses, i.e. because high mass stars spend a more meaning fraction of their integral lifetime in the post - MS stage. # # # Bayes Formalism { # subsubsec: formalism } Bayesian estimation of the physical argument can proceed from comparison of the data with a selection of models. Bayes ’ Theorem country: $ $ P(\mathrm{model|data }) \propto P(\mathrm{data|model }) \times P(\mathrm{model})$$ The probability of a model given a set of data is proportional to the product of the probability of the data give the model and the probability of the model itself. In the language of Bayesian statistics, this is express as: $ $ \mathrm{posterior } \propto \mathrm{likelihood } \times \mathrm{prior}.$$ Our model is the stage set of stellar parameters, age ($ \tau$) and mass ($ M_*$), and our data are the measured effective temperature, $ T_\mathrm{eff}$, and open gravity, $ \log g$, for a given star. At any grant combination of age and mass, the predicted $ T_\mathrm{eff}$ and $ \log g$ are put up by stellar evolutionary model. The $ \chi^2 $ statistic for an individual model can be computed as be: $ $ \begin{aligned } \chi^2 (\tau, M _ *) & = \sum \frac{(O - E)^2}{\sigma^2 } \\ & = \frac{[(T_\mathrm{eff})_O-(T_\mathrm{eff})_E]^2}{\sigma_{T_\mathrm{eff}}^2

inherpolation between isocmrones neither aexounts for tge nonlivear mapping of time onto thx H-R diageam nor the non-uniform distribunion of sreller masses observxs in thc galzwy. As e consequence, sjraightforwasd interpolatimn bztween isochrones results in an age qistribitlon for field ftarx thaf is biased towards older ages comlared tm the distribition predicted by stellar evoputionary theory. Bajesian infetsncq of stellar xge and mass aims to emiminate such a bias by accountkng fpr observajninapny and/or thxoretibally motivatcc distsibutiom functions fov the phtsical parameters of mnterest. As an exampje, for a ciren point with error vaes on the H-R auagfam, a lkwer shelmar mass sgould be cobsidered more likelu qlr to the inifial mwsf function. Likewise, due to the longer mdin-aequence timescales for lower mass stars, a sjar that if observed to have evolved off the main sequence vhoulv favt q procqbllity distribution in mass that is skewed towwddx migher masses, i.e. because hibhfr iass stars spgnd a more significant fractlon of jheir wntire lisetike in the post-MS stage. ### Bayew Formalism {#fybsubsec:formalism} Bcyesian estioatipn of the physical parameterr cah proceed fgom compadkson of the data winh a selection of models. Bayes’ Theorem wtatzs: $$P(\mathro{modgl|data}) \[ropto P(\mahhrm{data|model}) \times P(\matjrm{mobel})$$ Tha probabillty of a model given a set of deva is proportootal to the 'roducb of the probabylity of the dcta giveu the oodel and nhe probauility of thq model itselx. In the langnage of Bwyesuan wtatistkzs, this is exptessed as: $$\mathrm{postwrior} \propto \mathrk{liyslihood} \times \mcuhrn{prior}.$$ Our model is thq dev of fdellar paramaterr, aer ($\tau$) and mass ($M_*$), xnd pur data are the meavures effective temperstmre, $T_\mathtm{eff}$, and surface gravoty, $\log g$, for a gigen svar. At any bivgn combination of age and mass, fhe predifteq $T_\mathrm{eff}$ and $\log g$ are pxovided by stellar evolutionary models. Tie $\chi^2$ statistic for an individual model ccn be computev as fjllows: $$\beghn{aligned} \chi^2 (\tau, M_*) &= \wum \frac{(O-E)^2}{\sigma^2} \\ &= \frac{[(T_\mathrm{eff})_O-(T_\mafhrm{efx})_E]^2}{\sigla_{T_\mathrm{eff}}^2

interpolation between isochrones neither accounts for the of onto the diagram nor the observed the galaxy. As consequence, straightforward interpolation isochrones results in an age distribution field stars that is biased towards older ages compared to the distribution predicted stellar evolutionary theory. Bayesian inference of stellar age and mass aims to eliminate a by for and/or theoretically motivated distribution functions for the physical parameters of interest. As an example, for a point with error bars on the H-R diagram, lower stellar mass should considered more likely due to initial function. Likewise, to longer timescales for lower stars, a star that is observed to have evolved off the main sequence should have a probability in mass skewed towards masses, because mass stars spend significant fraction of their entire lifetime stage. ### Bayes Formalism {#subsubsec:formalism} Bayesian estimation of physical parameters proceed from comparison of the data a selection of models. Bayes’ Theorem states: $$P(\mathrm{model|data}) P(\mathrm{data|model}) \times P(\mathrm{model})$$ The probability of a model given a set of data is proportional product of the probability the data given model the of model itself. the language of Bayesian statistics, this is expressed as: $$\mathrm{posterior} \propto \times \mathrm{prior}.$$ Our model is the set of stellar parameters, and ($M_*$), and our are the measured effective $T_\mathrm{eff}$, surface gravity, $\log g$, given At of and the predicted $T_\mathrm{eff}$ and g$ are provided by stellar models. The $\chi^2$ statistic be computed as follows: $$\begin{aligned} \chi^2 (\tau, M_*) \sum \frac{(O-E)^2}{\sigma^2} \\ &= \frac{[(T_\mathrm{eff})_O-(T_\mathrm{eff})_E]^2}{\sigma_{T_\mathrm{eff}}^2

interpolation between isochRones neithEr accOunTs fOr The nOnliNear mapping of tIMe onTo the H-R diagram nor the noN-unifOrM DistRIbUtion Of stellAR mASSes ObSeRveD iN ThE galaXy. AS a conseQuence, straIghTfOrward interpOLaTion betweeN isOchrones resuLts In an agE dIstRIbutiOn fOr fieLd starS That is Biased towArDS older AGes compAREd To thE distribution predICtED by stellar evolUtionaRy THeORY. BaYesIan inferenCe Of steLLar age aND mASS AimS To eliminate suCh a bias by acCOunTing foR oBseRVationAlly aNd/OR thEoretically MotiVated distRibutiON functiONs for thE physiCal ParAmetERs Of IntErESt. AS An ExaMPle, For a giveN pOiNt witH errOR BARs on The h-R diAgram, A lower stellar MasS shoULd bE consIdereD morE lIkely Due to tHe iniTiAl mass function. LIkewIse, due to tHe lOnGer MaIn-seqUEnce tiMesCalEs for loWer mass STarS, a STAR tHat is observed to havE eVOLvEd off the Main seQUeNcE Should haVe A prObabILIty diStriBUtIon in masS that iS SkEwEd towarDs Higher MaSseS, i.e. BecauSE higHer masS stars spEnd a mORe significant fRAction of their ENtIRE lIFetiMe iN the post-MS sTage. ### bAyes formALiSm {#sUBsubsEc:forMaLIsM} bayesian estimation oF tHe physIcal pArameters can pRoceed from COMParison oF the DAtA With a selection Of modEls. Bayes’ ThEOrem statEs: $$P(\maThrm{modeL|data}) \propTO p(\mathrm{dAta|ModEl}) \tImeS p(\MaThrm{model})$$ The pROBabiLiTy of a moDel Given a sEt oF daTa iS prOpOrtional tO the prodUcT oF tHe ProBabilITy of the dAtA giVeN thE modeL And the ProbaBiliTy Of THe mOdel itsELf. iN The lAnGuAge oF BaYeSian sTatiSTicS, this is Expressed As: $$\mAThrm{PoStErior} \prOpto \mathrm{likElIhood} \times \MaThrM{prior}.$$ oUR model is The set of stellar parameteRS, age ($\tau$) And Mass ($M_*$), And oUr data are The MeasurEd eFFectivE tempeRaturE, $T_\MatHRM{eff}$, aND SuRfaCe Gravity, $\log G$, FOr a Given StAr. At Any giveN combination of age aND maSs, the predicteD $T_\mAthrM{EFf}$ And $\LOg G$ Are PrOVidED By stellar evolutIonary modeLs. tHe $\Chi^2$ statistIC foR aN indiviDual modEl can BE computEd as folloWs: $$\begin{alIgNed} \cHI^2 (\Tau, m_*) &= \sum \frac{(O-E)^2}{\Sigma^2} \\ &= \fraC{[(T_\mathrm{eFF})_O-(T_\maTHrM{eff})_E]^2}{\SigMa_{T_\matHrM{efF}}^2

interpolation between iso chrones ne ither ac cou nt s fo r th e nonlinear ma p ping of time onto the H-Rdiagr am nort he non- uniform di s t rib ut io n o fs te llarmas ses obs erved in t hega laxy. As a c o ns equence, s tra ightforwardint erpola ti onb etwee n i sochr ones r e sultsin an age d i stribu t ion for f ie ld s tars that is bias e dt owards older a ges co mp a re d tothe distribut io n pre d icted b y s t e l lar evolutionarytheory. Ba y esi an inf er enc e of st ellar a g e a nd mass aim s to eliminat e such a biasb y accou ntingfor ob serv a ti on all ya nd/ o rthe o ret ically m ot iv ateddist r i b u tion fu ncti ons f or the physic alpara m ete rs of inte rest .As an examp le, f or a given pointwith error ba rson th eH-R d i agram, alow er stel lar mas s sh ou l d be considered more l ik e l ydue to t he ini t ia lm ass func ti on. Lik e w ise,duet othe long er mai n -s eq uence t im escale sfor lo wer m a ss s tars,a star t hat i s observed to h a ve evolved of f t h e m a in s equ ence should hav e a p roba b il ity distr ibuti on in mass that is skewed t owards high er masses, i. e. because h i gher mas s st a rs spend a more s ignif icant frac t ion of t heirentire l ifetime i n the post -MS st age . # # #Bayes Formali s m {#s ub subsec: for malism} B aye sia n e st imation o f the ph ys ic al p ara meter s can pro ce edfr omcompa r ison o f the dat awi t h a select i on o f mo de ls . Ba yes ’Theor em s t ate s: $$P (\mathrm{ mod e l|da ta }) \propt o P(\mathrm{d at a|model})\t ime s P(\m a t hrm{mode l})$$ The probabilityo f a mod elgiven a s et of dat a i s prop ort i onal t o theprodu ct of t he pr o b ab ili ty of the da t a gi ven t he mod el andthe probability of the model itself . I n th e la ngu a ge ofBa y esi a n statistics, th is is expr es s ed as: $$\m a thr m{ posteri or} \pr opto\ mathrm{ likelihoo d} \times \ math r m {pr ior}.$$ O ur model is the s e t ofs te llarpar ameter s, ag e ($\ tau$)a ndmass($M_*$ ), and o ur da ta are the measured effective tem peratu re, $ T_\ mathrm{ef f}$ , an d surface gra vity, $\lo g g $,for a gi v en st ar.A tany given com b ination o f a gea n dmass, the p r e d ict ed $T _\m a thrm{e ff}$ and $\log g$ are provided by st ella r evo lut i onar ymodels. The $\ chi ^2 $ statisti cfor an indi vidual m od e l can be co mputed as fol l o ws : $$\b egin {al igned} \c hi^ 2( \tau, M _* )& = \sum \fr ac {(O-E) ^2}{\s i gma^ 2 } \\ &= \frac{ [(T_\ m a thrm{ e ff} )_O-( T_ \mathrm { eff} )_E]^2}{\s igma_{T_\ma thrm{e ff}} ^2

interpolation_between isochrones_neither accounts for the_nonlinear mapping_of_time onto_the_H-R diagram nor_the non-uniform distribution_of stellar masses observed_in the galaxy._As_a consequence, straightforward interpolation between isochrones results in an age distribution for field stars_that_is biased_towards_older_ages compared to the distribution_predicted by stellar evolutionary theory. Bayesian_inference of_stellar age and mass aims to eliminate such_a_bias by accounting_for observationally and/or theoretically motivated distribution functions for the_physical parameters of interest. As an_example, for a_given_point_with error bars on_the H-R diagram, a lower stellar_mass should be considered more likely_due to the initial mass function. Likewise,_due to the longer main-sequence timescales_for lower mass stars, a_star that_is observed to have evolved_off the main_sequence should_have a probability_distribution in mass that is skewed_towards higher masses,_i.e. because higher mass stars spend_a_more significant fraction_of_their_entire lifetime_in the post-MS_stage. ###_Bayes Formalism_{#subsubsec:formalism} Bayesian_estimation of the physical parameters can_proceed_from comparison of the data with a_selection of models. Bayes’_Theorem_states: $$P(\mathrm{model|data}) \propto P(\mathrm{data|model}) \times_P(\mathrm{model})$$ The probability of a model_given a set of data is_proportional to_the product_of the probability of the data given the model and the_probability of the model itself. In_the language of Bayesian_statistics, this_is_expressed as: $$\mathrm{posterior} \propto_\mathrm{likelihood}_\times \mathrm{prior}.$$ Our_model is the set of stellar parameters,_age ($\tau$)_and mass ($M_*$), and our data_are the measured effective_temperature,_$T_\mathrm{eff}$, and surface gravity, $\log g$,_for a given star. At any_given combination of age and_mass,_the_predicted $T_\mathrm{eff}$ and $\log g$_are provided by stellar evolutionary models._The $\chi^2$ statistic_for an individual model can be computed_as_follows: $$\begin{aligned} \chi^2 (\tau, M_*) &= \sum \frac{(O-E)^2}{\sigma^2}_\\_ &= \frac{[(T_\mathrm{eff})_O-(T_\mathrm{eff})_E]^2}{\sigma_{T_\mathrm{eff}}^2

of this section. \[l:almost properness\] For each interval map $f : I \to I$ in ${\mathscr{A}}$ there is $\varepsilon > 0$ such that the following property holds. Let $J_0$ be an interval contained in $I$ satisfying $|J_0| \le \varepsilon$, let $n \ge 1$ be an integer, and let $J$ be a pull-back of $J_0$ by $f^n$ whose closure is contained in the interior of $I$. Suppose in addition that for each $j$ in $\{1, \ldots, n \}$ the pull-back of $J_0$ by $f^j$ containing $f^{n - j}(J)$ has length bounded from above by $\varepsilon$. Then $f^n(\partial J) \subset \partial J_0$. Note that each preimage of a point of $\partial I$ is either a point of $\partial I$, or a turning point in the interior of $I$. It follows that there is $\varepsilon$ in $(0, |I|)$ such that for every interval $J'$ contained in $I$ that shares an endpoint with $I$ and satisfies $|J'| \le \varepsilon$, the following property holds: For each pull-back $J$ of $J'$ by $f$, either $J$ shares an endpoint with $I$ and $f : J \to J'$ is a bijection, or both endpoints of $J$ are mapped to the endpoint of $J'$ that is not an endpoint of $I$. Reducing $\varepsilon$ if necessary, assume that for every pair of distinct elements $c$ and $c'$ of $\operatorname{Crit}(f) \cup \partial I$, we have $\varepsilon < \operatorname{dist}(c, c')$. We prove the lemma for this choice of $\varepsilon$. Let $J_0$, $n$, and $J$ be as in the statement of the lemma, and for each $j$ in $\{1, \ldots, n - 1 \}$ let $J_j$ be the pull-back of $J_0$ by 

of this section. \[l: almost properness\ ] For each interval map   $ farad: I \to I$ in   $ { \mathscr{A}}$ there be   $ \varepsilon > 0 $ such that the following property holds. get   $ J_0 $ be an interval contained in   $ I$ satisfy   $ |J_0| \le \varepsilon$, let   $ n \ge 1 $ be an integer, and let   $ J$ be a wrench - back of   $ J_0 $ by   $ f^n$ whose closure is contain in the interior of   $ I$. Suppose in summation that for each   $ j$ in   $ \{1, \ldots, n \}$ the wrench - back of   $ J_0 $ by   $ f^j$ containing   $ f^{n - j}(J)$ has length bounded from above by   $ \varepsilon$. Then   $ f^n(\partial J) \subset \partial J_0$. notice that each preimage of a point of   $ \partial I$ is either a point of   $ \partial I$, or a turning degree in the interior of   $ I$. It follows that there cost   $ \varepsilon$ in   $ (0, |I|)$ such that for every interval   $ J'$ incorporate in   $ I$ that shares an endpoint with   $ I$ and satisfies   $ |J'| \le \varepsilon$, the following property holds: For each wrench - back   $ J$ of   $ J'$ by   $ f$, either   $ J$ shares an endpoint with   $ I$ and   $ f: J \to J'$ is a bijection, or both endpoints of   $ J$ are mapped to the endpoint of   $ J'$ that is not an endpoint of   $ I$. Reducing   $ \varepsilon$ if necessary, assume that for every pair of distinct elements   $ c$ and   $ c'$ of   $ \operatorname{Crit}(f) \cup \partial I$, we have   $ \varepsilon < \operatorname{dist}(c, c')$. We prove the lemma for this choice of   $ \varepsilon$. Let   $ J_0 $, $ n$, and   $ J$ be as in the argument of the lemma, and for each   $ j$ in   $ \{1, \ldots, n - 1 \}$ let   $ J_j$ be the pull - back of   $ J_0 $ by

of this section. \[l:almost prokerness\] For each nbtervan map $f : I \to I$ in ${\mathscr{A}}$ there is $\varepsilln > 0$ suxh that the following oroperty jolds. Ler $J_0$ bt an interval convzined ik $N$ satjdfyiug $|O_0| \le \varepsilon$, let $n \ge 1$ te an integer, dna pet $J$ be a pull-back of $J_0$ by $f^n$ whose cjosure os contained in jhe imeerikg if $I$. Suppose in addition that ror eaci $j$ in $\{1, \ldots, n \}$ yhe pull-back of $J_0$ by $f^j$ contwinijg $f^{n - j}(J)$ has lengtj bounded ftkm wvove by $\vareprilon$. Then $f^n(\partial J) \aubset \partial J_0$. Note that each oreimcge of a pounr ov $\kartial I$ is eithvr a point of $\isrtial I$, or a turning point in thw interior of $I$. It folnows that there is $\darepsilot$ nn $(0, |I|)$ such that for evwrt intgrval $B'$ covraived ii $I$ fhat sjarxs an endpojnt with $I$ abd satisfies $|J'| \le \vate[wilon$, the folmowing pwoperty holds: For each pull-back $J$ of $J'$ by $x$, ejther $J$ shares an endpoibt with $I$ and $f : J \to J'$ is a bijqction, or both endpoints of $J$ are mapped to the engpoinv uf $J'$ tmat kw jot an endpoint of $I$. Reducing $\varepsilon$ if necqassrj, assume that for every pair ov cystinct elemevts $c$ aus $c'$ of $\operatorname{Crih}(f) \cup \kartiao I$, we hade $\vatepsilon < \operatorname{dist}(c, c')$. We prove nhe oemma for this chonce of $\varepsnlon$. Lej $J_0$, $n$, amd $J$ be as in the statemznt of the lemma, wnd for ezzh $j$ in $\{1, \ldots, n - 1 \}$ lvt $J_j$ be the kjll-back of $J_0$ by 

of this section. \[l:almost properness\] For each $f I \to in ${\mathscr{A}}$ there that following property holds. $J_0$ be an contained in $I$ satisfying $|J_0| \le let $n \ge 1$ be an integer, and let $J$ be a pull-back $J_0$ by $f^n$ whose closure is contained in the interior of $I$. Suppose addition for $j$ $\{1, \ldots, n \}$ the pull-back of $J_0$ by $f^j$ containing $f^{n - j}(J)$ has length from above by $\varepsilon$. Then $f^n(\partial J) \subset J_0$. Note that each of a point of $\partial is a point $\partial or turning point in interior of $I$. It follows that there is $\varepsilon$ in $(0, |I|)$ such that for every interval contained in shares an with and $|J'| \le \varepsilon$, property holds: For each pull-back $J$ $f$, either $J$ shares an endpoint with $I$ $f : \to J'$ is a bijection, or endpoints of $J$ are mapped to the endpoint $J'$ that is not an endpoint of $I$. Reducing $\varepsilon$ if necessary, assume that for of distinct elements $c$ $c'$ of $\operatorname{Crit}(f) \partial we $\varepsilon \operatorname{dist}(c, c')$. prove the lemma for this choice of $\varepsilon$. Let $J_0$, $n$, $J$ be as in the statement of the lemma, and $j$ $\{1, \ldots, n 1 \}$ let $J_j$ the of $J_0$ by

of this section. \[l:almost propeRness\] For eaCh intErvAl mAp $F : I \to i$ in ${\mAthscr{A}}$ there is $\VArepSilon > 0$ such that the followIng prOpERty hOLdS. Let $J_0$ Be an intERvAL ConTaInEd iN $I$ SAtIsfyiNg $|J_0| \Le \varepSilon$, let $n \gE 1$ be An Integer, and leT $j$ bE a pull-back Of $J_0$ By $f^n$ whose cloSurE is conTaIneD In the IntErior Of $I$. SupPOse in aDdition thAt FOr each $J$ In $\{1, \ldots, N \}$ THe Pull-Back of $J_0$ by $f^j$ contaiNInG $F^{n - j}(J)$ has length bOunded FrOM aBOVe bY $\vaRepsilon$. ThEn $F^n(\parTIal J) \subSEt \PARTiaL j_0$. Note that each Preimage of a POinT of $\parTiAl I$ IS eitheR a poiNt OF $\paRtial I$, or a tuRninG point in tHe inteRIor of $I$. IT Follows That thEre Is $\vArepSIlOn$ In $(0, |I|)$ SuCH thAT fOr eVEry Interval $j'$ cOnTaineD in $I$ THAT SharEs aN endPoint With $I$ and satisFieS $|J'| \le \VArePsiloN$, the fOlloWiNg proPerty hOlds: FOr Each pull-back $J$ of $j'$ by $f$, Either $J$ shAreS aN enDpOint wITh $I$ and $F : J \tO J'$ iS a bijecTion, or bOTh eNdPOINtS of $J$ are mapped to the EnDPOiNt of $J'$ thaT is not AN eNdPOint of $I$. REdUciNg $\vaREPsiloN$ if nECeSsary, assUme thaT FoR eVery paiR oF distiNcT elEmeNts $c$ aND $c'$ of $\OperatOrname{CrIt}(f) \cuP \Partial I$, we have $\VArepsilon < \operAToRNAmE{Dist}(C, c')$. WE prove the leMma fOR thiS choICe Of $\vARepsiLon$. LeT $J_0$, $N$, AnD $j$ be as in the statement Of The lemMa, and For each $j$ in $\{1, \ldoTs, n - 1 \}$ let $J_j$ be THE Pull-back Of $J_0$ bY 

of this section. \[l:alm ost proper ness\ ] F orea ch i nter val map $f : I \toI$ in ${\mathscr{A}}$there i s  $\v a re psilo n > 0$s uc h tha tth e f ol l ow ing p rop erty ho lds. Let $ J_0 $be an interv a lcontainedin$I$ satisfyi ng$|J_0| \ le\ varep sil on$,let $n \ge 1$ be an in te g er, an d let $J $ be a p ull-back of $J_0$ by $f^n$ whose cl osureis co n t ain edin the int er ior o f  $I$. S u pp o s e in addition that for each $ j $ i n $\{1 ,\ld o ts, n\}$ t he pul l-back of $ J_0$ by $f^j$ conta i ning $f ^ {n - j} (J)$ h aslen gthb ou nd edfr o m a b ov e b y  $\ varepsil on $. Then  $f^ n ( \ p arti alJ) \ subse t \partial J_ 0$. No t e t hat e ach p reim ag e ofa poin t of$\ partial I$ is e ithe r a point of  $ \pa rt ial I $ , or a tu rni ng poin t in th e in te r i o rof $I$. It follows t h a tthere is  $\var e ps il o n$ in $( 0, |I |)$s u ch th at f o revery in terval $J '$ contai ne d in $ I$ th atshare s anendpoi nt with$I$ a n d satisfies $| J '| \le \varep s il o n $, thefol lowing prop erty hold s: F o reac h pull -back  $ J $o f $J'$ by $f$, eith er  $J$ s hares an endpointwith $I$ a n d $f : J \ to J ' $i s a bijection, or b oth endpoi n ts of $J $ are mappedto the en d p oint of$J' $ t hat is n ot an endpointo f  $I$ .Reducin g $ \vareps ilo n$ifnec es sary, ass ume that f or e ve rypairo f distin ct el em ent s $c$ and $c '$ of  $\o pe ra t orn ame{Cri t }( f ) \cu p\p arti alI$ , wehave $\v arepsil on < \ope rat o rnam e{ di st}(c,c')$. We prov ethe lemmafo r t his ch o i ce of $\ varepsilon$. Let $J_0$ , $n$, a nd$J$ b e as in the s tat ementoft he lem ma, an d for e ach $ j$ in $ \{ 1,\l dots, n -1 \}$ let$J _j$be thepull-back of $J_0$ by

of_this section. \[l:almost_properness\] For each interval_map $f :_I_\to I$_in ${\mathscr{A}}$_there is $\varepsilon >_0$ such that_the following property holds._Let $J_0$ be an_interval_contained in $I$ satisfying $|J_0| \le \varepsilon$, let $n \ge 1$ be an integer, and let $J$ be_a_pull-back of $J_0$_by $f^n$_whose_closure is contained in the_interior of $I$. Suppose in addition_that for_each $j$ in $\{1, \ldots, n \}$ the pull-back of $J_0$_by $f^j$_containing $f^{n - j}(J)$_has length bounded from above by $\varepsilon$. Then $f^n(\partial J) \subset_\partial J_0$. Note that each preimage of_a point of $\partial_I$_is_either a point of $\partial_I$, or a turning point in_the interior of $I$. It follows that_there is $\varepsilon$ in $(0, |I|)$ such that for_every interval $J'$ contained in $I$ that shares_an endpoint with $I$ and satisfies $|J'|_\le \varepsilon$,_the following property holds: For_each pull-back $J$ of $J'$_by $f$, either $J$_shares an endpoint_with $I$ and $f : J \to J'$_is a bijection,_or both endpoints of $J$ are mapped_to_the endpoint of $J'$_that_is_not an_endpoint of $I$. Reducing $\varepsilon$_if_necessary, assume_that_for every pair of distinct elements $c$_and $c'$_of $\operatorname{Crit}(f) \cup \partial I$, we have $\varepsilon <_\operatorname{dist}(c, c')$. We prove_the_lemma for this choice_of $\varepsilon$. Let $J_0$, $n$, and $J$ be as_in the statement of the lemma,_and for_each $j$ in $\{1,_\ldots, n - 1 \}$ let $J_j$ be the pull-back of $J_0$ by 

the $d_{x^{2}-y^{2}}$ and $d_{xy}$ orbitals of Mn while the remaining Mn orbitals are delocalized. Strong quantum fluctuations, possibly related to an electronic instability that forms the Mn moment or to the one-dimensional character of [Ti$_{4}$MnBi$_{2}$]{}, nearly overcome magnetic order.' author: - Abhishek Pandey - Ping Miao - 'M. Klemm' - 'H. He' - 'H. Wang' - 'X. Qian' - 'J. W. Lynn' - 'M. C. Aronson' title: 'Correlations and incipient antiferromagnetic order within the linear Mn chains of metallic Ti$_4$MnBi$_2$' --- Introduction ============ Low dimensional magnetic systems continue to excite since strong quantum fluctuations suppress magnetic order, showcasing novel quantum effects at the lowest temperatures [@Giamarchi-2004]. Of particular importance is the extensive body of theoretical results on one-dimensional (1D) systems that can be directly tested by experiments like inelastic neutron scattering [@lake2009; @mourigal2013] and resonant inelastic x-ray scattering [@schlappa2018]. To date, most experiments have been carried out on spin $S = 1/2$ systems like KCuF$_3$ [@Hirakawa-1967; @Lake-2005; @Lee-2012; @Pavarini-2008], SrCuO$_2$ [@Kim-1996], Sr$_2$CuO$_3$ [@Schlappa-2012] and CuSO$_{4}$.5D$_2$O [@mourigal2013], which are strongly correlated insulators. Inelastic neutron scattering experiments demonstrate that the fundamental excitations are not the spin waves expected in three dimensional (3D) systems, but rather spinons, demonstrating that the electron spin and charge have become separated [@Wu-2016; @Jompol-2009; @Lake-2005]. It is less clear what happens to this scenario when electronic correlations are not strong enough to sustain an insulating gap. The first examples of metallic spin chains were realized in the organic conductors [@kanoda2011; @jerome2012; @ardavan2012], where the ratio of the Coulomb interaction $U$ and the electron bandwidth 4$t$ are approximately equal, $\it{i

the $ d_{x^{2}-y^{2}}$ and $ d_{xy}$ orbitals of Mn while the remaining Mn orbitals are delocalized. Strong quantum fluctuations, possibly relate to an electronic imbalance that forms the Mn moment or to the one - dimensional fictional character of [ Ti$_{4}$MnBi$_{2}$ ] { }, closely overcome magnetic club.' generator: - Abhishek Pandey - Ping Miao -' M. Klemm' -' H. He' -' H. Wang' -' X. Qian' -' J. W. Lynn' -' M. C. Aronson' title:' Correlations and incipient antiferromagnetic ordering within the linear Mn chains of metallic Ti$_4$MnBi$_2 $' --- initiation = = = = = = = = = = = = Low dimensional magnetic system continue to excite since strong quantum fluctuations suppress charismatic order, showcasing novel quantum effects at the lowest temperature   [ @Giamarchi-2004 ]. Of particular importance is the extensive torso of theoretical results on one - dimensional (1D) system that can be directly tested by experiments like inelastic neutron scattering   [ @lake2009; @mourigal2013 ] and resonant inelastic ten - ray scattering   [ @schlappa2018 ]. To date, most experiments have been carried out on spin $ S = 1/2 $ systems like KCuF$_3 $ [ @Hirakawa-1967; @Lake-2005; @Lee-2012; @Pavarini-2008 ], SrCuO$_2 $ [ @Kim-1996 ], Sr$_2$CuO$_3 $ [ @Schlappa-2012 ] and CuSO$_{4}$.5D$_2$O   [ @mourigal2013 ], which are powerfully correlate insulators. Inelastic neutron scattering experiments demonstrate that the fundamental excitations are not the spin waves expected in three dimensional (three-d) systems, but rather spinons, demonstrate that the electron tailspin and cathexis have become separated   [ @Wu-2016; @Jompol-2009; @Lake-2005 ]. It is less clear what happens to this scenario when electronic correlations are not strong enough to sustain an insulating gap. The inaugural examples of metallic spin chains were realized in the organic conductors   [ @kanoda2011; @jerome2012; @ardavan2012 ], where the proportion of the Coulomb interaction $ U$ and the electron bandwidth 4$t$ are approximately equal, $ \it{i

thf $d_{x^{2}-y^{2}}$ and $d_{xy}$ orbitals on Mn while the rgmqining Mn orgitals afe delocalized. Strong quantul dluctyations, possibly relatdd to an vlectronix inwrability tizt forms the Jk momznv or to the one-cimensionan character of [Tk$_{4}$MuBi$_{2}$]{}, nearly overcome magnetic order.' auehor: - Abnidhek Pandey - Pigg Mpaj - 'M. Ilemm' - 'H. He' - 'H. Wang' - 'X. Qian' - 'J. W. Lynn' - 'J. C. Aroison' title: 'Correkations and incipient antiverrlmagnetic order wihhin the libear Nn chains of metallic Ui$_4$KnBi$_2$' --- Introdoction ============ Low dimensional magnetic shstemx continue ti edwite since wtrond quantum flmbtuatiots supptess magnetic prdxr, syowcasing novel quantnm effects at the lorest temparctures [@Giamarchi-2004]. Of paetuculat impmrtavxe ks uhe estensige uody of thekretical rewults on one-dimensipnwo (1D) systems tgat cag fe directly tested by experiments like pnelzstic neutron scatterint [@lake2009; @mourigal2013] and redonant inqlastic x-ray scattering [@schlappa2018]. To date, most expesimenvs harc cwej carried out on spin $S = 1/2$ systems like KCuF$_3$ [@Hydalaea-1967; @Lake-2005; @Lee-2012; @Payarini-2008], SrCuO$_2$ [@Kim-1996], St$_2$CkO$_3$ [@Fchlappa-2012] and ZuSO$_{4}$.5D$_2$O [@moudigal2013], which are stgongly sorreoated instlatprs. Inelastic neutron scattwring experiiwnts demonstrate tkat the fundcmentak excotations are not the spnn wavss expected in three aimensional (3D) syrteks, but rather spinons, demonserating tiat tke electfon xpin agd charge jave nacome separated [@Wu-2016; @Uompop-2009; @Nake-2005]. It is pess clear what happens to this scenario when enecnronic coxrelatlons are not stwong enough to sustaiu an ivsulating fap. The first exam[les of metalnlc spin chaiis were rqalized un the ufganic conductprs [@kanoda2011; @jerome2012; @areavan2012], where the rabio ow the Coulomb inctraxtion $U$ and the eldctwoj uandwygth 4$t$ are ap[roxkmagrly edual, $\it{i

the $d_{x^{2}-y^{2}}$ and $d_{xy}$ orbitals of Mn remaining orbitals are Strong quantum fluctuations, instability forms the Mn or to the character of [Ti$_{4}$MnBi$_{2}$]{}, nearly overcome magnetic author: - Abhishek Pandey - Ping Miao - 'M. Klemm' - 'H. He' 'H. Wang' - 'X. Qian' - 'J. W. Lynn' - 'M. C. Aronson' 'Correlations incipient order the linear Mn chains of metallic Ti$_4$MnBi$_2$' --- Introduction ============ Low dimensional magnetic systems continue to since strong quantum fluctuations suppress magnetic order, showcasing quantum effects at the temperatures [@Giamarchi-2004]. Of particular importance the body of results one-dimensional systems that can directly tested by experiments like inelastic neutron scattering [@lake2009; @mourigal2013] and resonant inelastic x-ray scattering [@schlappa2018]. To most experiments carried out spin = systems like KCuF$_3$ @Lee-2012; @Pavarini-2008], SrCuO$_2$ [@Kim-1996], Sr$_2$CuO$_3$ [@Schlappa-2012] which are strongly correlated insulators. Inelastic neutron scattering demonstrate that fundamental excitations are not the spin expected in three dimensional (3D) systems, but rather demonstrating that the electron spin and charge have become separated [@Wu-2016; @Jompol-2009; @Lake-2005]. It is what happens to this when electronic correlations not enough sustain insulating gap. first examples of metallic spin chains were realized in the organic [@kanoda2011; @jerome2012; @ardavan2012], where the ratio of the Coulomb interaction the bandwidth 4$t$ are equal, $\it{i

the $d_{x^{2}-y^{2}}$ and $d_{xy}$ orbitals of Mn wHile the remAininG Mn OrbItAls aRe deLocalized. StronG QuanTum fluctuations, possiblY relaTeD To an ELeCtronIc instaBIlITY thAt FoRms ThE mn MomenT or To the onE-dimensionAl cHaRacter of [Ti$_{4}$MnbI$_{2}$]{}, nEarly overcOme Magnetic ordeR.' auThor: - AbHiSheK pandeY - PiNg MiaO - 'M. KlemM' - 'h. He' - 'H. WaNg' - 'X. Qian' - 'J. W. lyNN' - 'M. C. AroNSon' titlE: 'cOrRelaTions and incipient ANtIFerromagnetic oRder wiThIN tHE LinEar mn chains of MeTalliC ti$_4$MnBi$_2$' --- INTrODUCtiON ============ Low dimensionAl magnetic sYSteMs contInUe tO Excite Since StROng Quantum flucTuatIons supprEss magNEtic ordER, showcaSing noVel QuaNtum EFfEcTs aT tHE loWEsT teMPerAtures [@GiAmArChi-2004]. Of PartICULAr imPorTancE is thE extensive bodY of TheoREtiCal reSults On onE-dImensIonal (1D) SysteMs That can be directLy teSted by expEriMeNts LiKe ineLAstic nEutRon ScatterIng [@lake2009; @MOurIgAL2013] ANd Resonant inelastic x-RaY SCaTtering [@sChlappA2018]. to DaTE, most expErImeNts hAVE been CarrIEd Out on spiN $S = 1/2$ systEMs LiKe KCuF$_3$ [@HIrAkawa-1967; @LAkE-2005; @LeE-2012; @PaVarinI-2008], srCuo$_2$ [@Kim-1996], Sr$_2$cuO$_3$ [@SchlaPpa-2012] anD cuSO$_{4}$.5D$_2$O [@mourigal2013], WHich are strongLY cORReLAted InsUlators. InelAstiC NeutRon sCAtTerINg expErimeNtS DeMOnstrate that the fundAmEntal eXcitaTions are not thE spin waves EXPEcted in tHree DImENsional (3D) systemS, but rAther spinoNS, demonstRatinG that the Electron sPIN and charGe hAve BecOme SEPaRated [@Wu-2016; @Jompol-2009; @lAKe-2005]. It Is Less cleAr wHat happEns To tHis SceNaRio when elEctronic CoRrElAtIonS are nOT strong eNoUgh To SusTain aN InsulaTing gAp. ThE fIrST exAmples oF MeTALlic SpIn ChaiNs wErE realIzed IN thE organiC conductoRs [@kANoda2011; @JeRoMe2012; @ardavAn2012], where the ratIo Of the CouloMb IntEractiON $u$ and the eLectron bandwidth 4$t$ are appROximateLy eQual, $\iT{i

the $d_{x^{2}-y^{2}}$ and $d_{xy}$orbit als of M n wh ilethe remainingM n or bitals are delocalized . Str on g qua n tu m flu ctuatio n s, p oss ib ly re la t ed to a n e lectron ic instabi lit ythat forms t h eMn momentorto the one-d ime nsiona lcha r acter of [Ti$ _{4}$M n Bi$_{2 }$]{}, ne ar l y over c ome mag n e ti c or der.' author: - A b hi s hek Pandey - P ing Mi ao -' M . K lem m' - 'H. H e' - 'H . Wang'- ' X . Qia n ' - 'J. W. Ly nn' - 'M. C . Ar onson' t itl e : 'Cor relat io n s a nd incipien t an tiferroma gnetic order w i thin th e line arMnchai n sof me ta l lic Ti $_4 $ MnB i$_2$' - -- Intro duct i o n ==== === ==== = Lo w dimensional ma gnet i c s ystem s con tinu eto ex cite s incest rong quantum fl uctu ations su ppr es s m ag netic order, sh owc asing n ovel qu a ntu me f f ec ts at the lowest t em p e ra tures [@ Giamar c hi -2 0 04]. Ofpa rti cula r impor tanc e i s the ex tensiv e b od y of th eo retica lres ult s ono ne-d imensi onal (1D ) sys t ems that can b e directly tes t ed b ye xper ime nts like in elas t ic n eutr o nsca t terin g [@l ak e 20 0 9; @mourigal2013] a nd reson ant i nelastic x-ra y scatteri n g [@schlap pa20 1 8] . To date, most expe riments ha v e been c arrie d out on spin $S= 1/2$ sys tem s l ike KC u F $_ 3$ [@Hirakawa - 1 967; @ Lake-20 05; @Lee-2 012 ; @ Pav ari ni -2008], S rCuO$_2$ [ @K im -1 996 ], Sr $ _2$CuO$_ 3$ [@ Sc hla ppa-2 0 12] an d CuS O$_{ 4} $. 5 D$_ 2$O [@m o ur i g al20 13 ], whi char e str ongl y co rrelate d insulat ors . Ine la st ic neut ron scatterin gexperiment sdem onstra t e that th e fundamental excitatio n s are n otthe s pinwaves exp ect ed inthr e e dime nsiona l (3D )sys t e ms, b u t r ath er spinons,d e mon strat in g th at theelectron spin andc har ge have becom e s epar a t ed  [@ W u- 2 016 ;@ Jom p o l-2009; @Lake-2 005]. Itis le ss clear w h atha ppens t o thisscena r io when electron ic correl at ions a renot strong enoughto sustai n an i n su latin g g ap. Th efir st ex amples ofmetal lic sp in chain s wer erealized in the organic conduct ors [@ kanod a20 11; @jero me2 0 12; @ardavan 2012 ], where t herat io of th e Coul ombi nt era c tion$U$a nd the el e ct ron b an dwidth 4$t$ a r e a pprox ima t ely eq ual, $\it{i

the_$d_{x^{2}-y^{2}}$ and_$d_{xy}$ orbitals of Mn_while the_remaining_Mn orbitals_are_delocalized. Strong quantum_fluctuations, possibly related_to an electronic instability_that forms the_Mn_moment or to the one-dimensional character of [Ti$_{4}$MnBi$_{2}$]{}, nearly overcome magnetic order.' author: - Abhishek Pandey -_Ping_Miao - 'M._Klemm' -_'H._He' - 'H. Wang' - 'X. Qian' -_'J. W. Lynn' - 'M. C._Aronson' title: 'Correlations_and incipient antiferromagnetic order within the linear Mn_chains_of metallic Ti$_4$MnBi$_2$' --- Introduction ============ Low_dimensional magnetic systems continue to excite since strong quantum_fluctuations suppress magnetic order, showcasing novel_quantum effects at_the_lowest_temperatures [@Giamarchi-2004]. Of particular importance_is the extensive body of theoretical_results on one-dimensional (1D) systems that_can be directly tested by experiments like_inelastic neutron scattering [@lake2009; @mourigal2013] and resonant_inelastic x-ray scattering [@schlappa2018]. To date,_most experiments_have been carried out on_spin $S =_1/2$ systems_like KCuF$_3$ [@Hirakawa-1967;_@Lake-2005; @Lee-2012; @Pavarini-2008], SrCuO$_2$ [@Kim-1996], Sr$_2$CuO$_3$_[@Schlappa-2012] and CuSO$_{4}$.5D$_2$O [@mourigal2013],_which are strongly correlated insulators. Inelastic_neutron_scattering experiments demonstrate_that_the_fundamental excitations_are not the_spin_waves expected_in_three dimensional (3D) systems, but rather_spinons,_demonstrating that the electron spin and charge_have become separated [@Wu-2016; @Jompol-2009;_@Lake-2005]. It_is less clear what_happens to this scenario when_electronic correlations are not strong enough_to sustain_an insulating_gap. The first examples of metallic spin chains were realized in_the organic conductors [@kanoda2011; @jerome2012; @ardavan2012], where_the ratio of the_Coulomb interaction_$U$_and the electron_bandwidth_4$t$ are_approximately equal, $\it{i

&\equiv& k^2+a^2(\eta)(m_{\chi}^2-2H^2) \nonumber \\ &\sim& k^2-2a^2(\eta)H^2<0.\end{aligned}$$ And, since these modes turn out to be the ones which are most important in order to calculate the relevant quantities [@li05; @v83], we would finally conclude that the diagonalization method is not the most appropriate one to be used in this case. What is more, in order to compute the desired quantity, the authors calculate $\rho_{\chi}/m_{\chi}$ numerically. However, $\rho_{\chi}$ is known to be a divergent quantity, if the field is not conformally coupled (see, e.g., [@b80]). It must necessarily be renormalized previously, what renders utmost difficult to develop a rigorous procedure in order to perform such calculation numerically. This investigation has been supported in part by MICINN (Spain), projects MTM2011-27739-C04-01, FIS2006-02842 and FIS2010-15640, by CPAN Consolider Ingenio Project, and by AGAUR (Generalitat de Catalunya), contracts 2009SGR-345 and 2009SGR-994. [99]{} D.J.H. Chung, E.W. Kolb and A. Riotto, Phys. Rev. [**D59**]{}, 023501 (1998), hep-ph/9802238; Phys. Rev. Lett. [**81**]{}, 4048 (1998), hep-ph/9805473. V. Kuzmin and I. Tkachev, Phys. Rev. [**D59**]{}, 123006 (1999), hep-ph/9809547. G. Felder, L. Kovman and A. Linde, JHEP 0002:027 (2000), hep-ph/9909508. F. Finelli, G. Marozzi, A.A. Starobinsky, G.P. Vacca and G. Venturi, Phys. Rev. [**D79**]{}, 044007 (2009), hep-th/0808.1786. V. Mukhanov, [*Physical fundations of cosmology*]{} (Cambridge University Press, 2005). A. Linde, [*Particle Physics and Inflationary Cosmology*]{} (Harwood, Chur, Switzerland, 1990), hep-

& \equiv & k^2+a^2(\eta)(m_{\chi}^2 - 2H^2) \nonumber \\ & \sim & k^2 - 2a^2(\eta)H^2<0.\end{aligned}$$ And, since these modes turn out to be the ones which are most authoritative in decree to calculate the relevant quantities [ @li05; @v83 ], we would finally reason that the diagonalization method is not the most appropriate one to be used in this font. What is more, in order to compute the desire quantity, the authors count $ \rho_{\chi}/m_{\chi}$ numerically. However, $ \rho_{\chi}$ is known to be a divergent quantity, if the playing field is not conformally coupled (visualize, e.g., [ @b80 ]). It must necessarily be normalize previously, what renders utmost difficult to develop a rigorous routine in order to perform such calculation numerically. This probe has been supported in region by MICINN (Spain), projects MTM2011 - 27739 - C04 - 01, FIS2006 - 02842 and FIS2010 - 15640, by CPAN Consolider Ingenio Project, and by AGAUR (Generalitat de Catalunya), contracts 2009SGR-345 and 2009SGR-994. [ 99 ] { } D.J.H. Chung, E.W. Kolb and A. Riotto, Phys. Rev. [ * * D59 * * ] { }, 023501 (1998), hep - ph/9802238; Phys. Rev. Lett. [ * * 81 * * ] { }, 4048 (1998), hep - ph/9805473. V. Kuzmin and I. Tkachev, Phys. Rev. [ * * D59 * * ] { }, 123006 (1999), hep - ph/9809547. G. Felder, L. Kovman and A. Linde, JHEP 0002:027 (2000), hep - ph/9909508. F. Finelli, G. Marozzi, A.A. Starobinsky, G.P. Vacca and G. Venturi, Phys. Rev. [ * * D79 * * ] { }, 044007 (2009), hep - th/0808.1786. V. Mukhanov, [ * Physical fundations of cosmology * ] { } (Cambridge University Press, 2005). A. Linde, [ * Particle Physics and Inflationary Cosmology * ] { } (Harwood, Chur, Switzerland, 1990), hep-

&\equlv& k^2+a^2(\eta)(m_{\chi}^2-2H^2) \nonumber \\ &\rim& k^2-2a^2(\eta)H^2<0.\end{alntned}$$ Aid, sincs these oodes turn out to be the oned qhich are most important in order to calculare tie relevant quanvjties [@ll05; @v83], ws wounv finally conclode that the diagonalizatimn mzthod is not the most appropriate onq to be uded in this cafe. Wnwt ia more, in order to compute the desjred quentity, the authprs calculate $\rho_{\chi}/m_{\chi}$ nkmerlcally. However, $\rho_{\fhi}$ is knowb to ve a divergevt quantitj, if the figld is not conformally coupled (sde, e.g., [@b80]). It must nwcedvarily be rxnormajized previomxly, whdt rendrrs utmost difniculv to develop a rigorous pcocedure in order to perform vueh calculation numerixaoly. Thhs itvesgugagioh ias been dup'orted in pzrt by MICIBN (Spain), projects MUM2011-27739-C04-01, DIS2006-02842 and FIS2010-15640, bg CPAN Cjnsolider Ingenio Project, and by AGAUR (Censralitat de Catalunya), cintracts 2009SGR-345 and 2009SGR-994. [99]{} D.U.H. Chung, Q.W. Kolb and A. Riotto, Phys. Rev. [**D59**]{}, 023501 (1998), hep-ph/9802238; Phys. Rev. Nett. [**81**]{}, 4048 (1998), htp-im/9805473. V. Yyzlin and I. Tkachev, Phys. Rev. [**D59**]{}, 123006 (1999), hep-ph/9809547. G. Felder, M. Lonman and A. Linde, MHEP 0002:027 (2000), hep-ph/9909508. F. Finrlpi, D. Marozzi, A.A. Rtarobnhsiy, G.P. Vacca and G. Genturi, Phys. Rev. [**D79**]{}, 044007 (2009), rep-tn/0808.1786. V. Mukhanov, [*Physical fundarions of cosiilogy*]{} (Cambridge Unnversity Prers, 2005). S. Lince, [*Particle Physics and Infmationary Clsmology*]{} (Gxrwood, Chur, Switxerkatd, 1990), hep-

&\equiv& k^2+a^2(\eta)(m_{\chi}^2-2H^2) \nonumber \\ &\sim& k^2-2a^2(\eta)H^2<0.\end{aligned}$$ And, modes out to the ones which to the relevant quantities @v83], we would conclude that the diagonalization method is the most appropriate one to be used in this case. What is more, order to compute the desired quantity, the authors calculate $\rho_{\chi}/m_{\chi}$ numerically. However, $\rho_{\chi}$ known be divergent if the field is not conformally coupled (see, e.g., [@b80]). It must necessarily be renormalized previously, renders utmost difficult to develop a rigorous procedure order to perform such numerically. This investigation has been in by MICINN projects FIS2006-02842 FIS2010-15640, by CPAN Ingenio Project, and by AGAUR (Generalitat de Catalunya), contracts 2009SGR-345 and 2009SGR-994. [99]{} D.J.H. Chung, E.W. Kolb A. Riotto, [**D59**]{}, 023501 hep-ph/9802238; Rev. [**81**]{}, 4048 (1998), Kuzmin and I. Tkachev, Phys. Rev. hep-ph/9809547. G. Felder, L. Kovman and A. Linde, 0002:027 (2000), F. Finelli, G. Marozzi, A.A. Starobinsky, Vacca and G. Venturi, Phys. Rev. [**D79**]{}, 044007 hep-th/0808.1786. V. Mukhanov, [*Physical fundations of cosmology*]{} (Cambridge University Press, 2005). A. Linde, [*Particle Physics Cosmology*]{} (Harwood, Chur, Switzerland, hep-

&\equiv& k^2+a^2(\eta)(m_{\chi}^2-2H^2) \nonumber \\ &\siM& k^2-2a^2(\eta)H^2<0.\end{AlignEd}$$ ANd, sInCe thEse mOdes turn out to bE The oNes which are most importaNt in oRdER to cALcUlate The releVAnT QUanTiTiEs [@lI05; @v83], WE wOuld fInaLly concLude that thE diAgOnalization mEThOd is not the MosT appropriate One To be usEd In tHIs casE. WhAt is mOre, in oRDer to cOmpute the DeSIred quANtity, thE AUtHors Calculate $\rho_{\chi}/m_{\cHI}$ nUMerically. HowevEr, $\rho_{\cHi}$ IS kNOWn tO be A divergent QuAntitY, If the fiELd IS NOt cONformally coupLed (see, e.g., [@b80]). It MUst NecessArIly BE renorMalizEd PRevIously, what rEndeRs utmost dIfficuLT to deveLOp a rigoRous prOceDurE in oRDeR tO peRfORm sUCh CalCUlaTion numeRiCaLly. ThIs inVESTIgatIon Has bEen suPported in part By MiCINn (spaIn), proJects mTM2011-27739-C04-01, fIs2006-02842 and FiS2010-15640, by CPaN ConSoLider Ingenio ProJect, And by AGAUr (GeNeRalItAt de CATalunyA), coNtrActs 2009SGR-345 And 2009SGR-994. [99]{} D.j.h. ChUnG, e.w. koLb and A. Riotto, Phys. ReV. [**D59**]{}, 023501 (1998), HEP-pH/9802238; Phys. Rev. lett. [**81**]{}, 4048 (1998), heP-Ph/9805473. v. KUZmin and I. tkAchEv, PhYS. rev. [**D59**]{}, 123006 (1999), hEp-ph/9809547. g. feLder, L. KovMan and a. liNdE, JHEP 0002:027 (2000), heP-pH/9909508. F. FineLlI, G. MAroZzi, A.A. sTaroBinsky, g.P. Vacca aNd G. VeNTuri, Phys. Rev. [**D79**]{}, 044007 (2009), heP-Th/0808.1786. V. Mukhanov, [*PhYSiCAL fUNdatIonS of cosmologY*]{} (CamBRidgE UniVErSitY press, 2005). a. LindE, [*PARtICle Physics and InflatIoNary CoSmoloGy*]{} (Harwood, Chur, switzerlanD, 1990), HEP-

&\equiv& k^2+a^2(\eta)(m_{ \chi}^2-2H ^2) \no num be r \\ &\s im& k^2-2a^2(\ e ta)H ^2<0.\end{aligned}$$ A nd, s in c e th e se mode s turno ut t o b eth e o ne s w hichare most i mportant i n o rd er to calcul a te the relev ant quantities[@l i05; @ v8 3], we wo uld fina lly co n cludethat thedi a gonali z ation m e t ho d is not the most app r op r iate one to be usedin th i s ca se. What is m or e, in order t o c o m p ute the desired q uantity, th e au thorsca lcu l ate $\ rho_{ \c h i}/ m_{\chi}$ n umer ically. H owever , $\rho_ { \chi}$is kno wntobe a di ve rge nt qua n ti ty, ifthe fiel dis notconf o r m a llycou pled (see , e.g., [@b80 ]). Itm ust nece ssari ly b erenor malize d pre vi ously, what ren ders utmost d iff ic ult t o dev e lop arig oro us proc edure i n or de r t operform such calcu la t i on numeric ally.Th is investig at ion has b een s uppo r te d in par t by M I CI NN (Spain ), proje ct s M TM2 011-2 7 739- C04-01 , FIS200 6-028 4 2 and FIS2010- 1 5640, by CPAN Co n s ol i derIng enio Projec t, a n d by AGA U R(Ge n erali tat d eC at a lunya), contracts 2 00 9SGR-3 45 an d 2009SGR-994 . [99]{}D . J.H. Chu ng,E .W . Kolb and A. R iotto , Phys. Re v . [**D59 **]{} , 023501 (1998),h e p-ph/980 223 8;Phy s.R e v. Lett. [**81* * ] {},40 48 (199 8), hep-ph /98 054 73. V. K uzmin and I. Tkac he v, P hy s.Rev.[ **D59**] {} , 1 23 006 (199 9 ), hep -ph/9 8095 47 .G . F elder,L .K o vman a nd A.Lin de , JHE P 00 0 2:0 27 (200 0), hep-p h/9 9 0950 8. F . Finel li, G. Marozz i, A.A. Star ob ins ky, G. P . Vacca a nd G. Venturi, Phys. Re v . [**D7 9** ]{},0440 07 (2009) , h ep-th/ 080 8 .1786. V. Mu khano v, [* P h ysica l fu nda ti ons of cos m o log y*]{} ( Camb ridge U niversity Press, 2 0 05) . A. Linde, [ *Pa rtic l e P hys i cs and I n fla t i onary Cosmology *]{} (Harw oo d ,Chur, Swit z erl an d, 1990 ), hep-

&\equiv& k^2+a^2(\eta)(m_{\chi}^2-2H^2)_ \nonumber_\\ &\sim& k^2-2a^2(\eta)H^2<0.\end{aligned}$$ And, since_these modes_turn_out to_be_the ones which_are most important_in order to calculate_the relevant quantities_[@li05;_@v83], we would finally conclude that the diagonalization method is not the most appropriate_one_to be_used_in_this case. What is more,_in order to compute the_desired quantity,_the authors calculate $\rho_{\chi}/m_{\chi}$ numerically. However, $\rho_{\chi}$ is_known_to be a_divergent quantity, if the field is not conformally coupled_(see, e.g., [@b80]). It must necessarily_be renormalized previously,_what_renders_utmost difficult to develop_a rigorous procedure in order to_perform such calculation numerically. This investigation has_been supported in part by MICINN (Spain),_projects MTM2011-27739-C04-01, FIS2006-02842 and FIS2010-15640, by_CPAN Consolider Ingenio Project, and_by AGAUR_(Generalitat de Catalunya), contracts 2009SGR-345_and 2009SGR-994. [99]{} D.J.H. Chung,_E.W. Kolb_and A. Riotto,_Phys. Rev. [**D59**]{}, 023501 (1998), hep-ph/9802238;_Phys. Rev. Lett._[**81**]{}, 4048 (1998), hep-ph/9805473. V. Kuzmin_and_I. Tkachev, Phys._Rev._[**D59**]{},_123006 (1999),_hep-ph/9809547. G. Felder,_L._Kovman and_A._Linde, JHEP 0002:027 (2000), hep-ph/9909508. F._Finelli,_G. Marozzi, A.A. Starobinsky, G.P. Vacca and_G. Venturi, Phys. Rev._[**D79**]{},_044007 (2009), hep-th/0808.1786. V._Mukhanov, [*Physical fundations of cosmology*]{}_(Cambridge University Press, 2005). A. Linde,_[*Particle Physics_and Inflationary_Cosmology*]{} (Harwood, Chur, Switzerland, 1990), hep-

is always smaller than the lowest energy gap, $p_{0,1}$. Once $eB$ becomes bigger than the squared system-size inverse (that is, $\alpha\gtrsim 10$ from Fig. \[fig:p01\]), however, the energy gap is significantly reduced and the anomalous coupling between the magnetic field and the rotation is then manifested [@Chen:2015hfc; @Hattori:2016njk]. Integration Measure and Reweighted wave functions {#sec:measure} ================================================= Because the radial momenta are discretized, we replace the transverse momentum integration with the sums over the quantum numbers, $l$ and $k$, i.e. $$\int \frac{dp_xdp_y}{(2\pi)^2} \longrightarrow \frac{1}{\pi R^2} \sum_{l=-\infty}^\infty \sum_{k=1}^\infty \frac{1}{{N_{l,k}}^2}\;, \label{eq:phase-space}$$ where ${N_{l,k}}$ represents a weight factor which corresponds to the integration measure in finite-size systems. In the $B=0$ case, as discussed in Ref. [@Ebihara:2016fwa], the weight factor is deduced from the Bessel-Fourier expansion, that is, we know that in the limit of $\alpha\to 0$, $$\begin{split} \label{eq:Nlkzero} {N_{l,k}}^2 \;\;\longrightarrow & \;\; \frac{2}{R^2}\int_0^R r dr\, \bigl[J_l(p_{l,k}r)\bigr]^2 \end{split}$$ with the discretized momentum ${p_{l,k}}$ in Eq. . From the relation in Eq. , we extrapolate the above identification to nonzero $\alpha$ as $$\label{eq:Nlk} \begin{split} {N_{l,k}}^2 &= \frac{2}{R^2}\int_0^R r dr\, \bigl[\Phi_l({\lambda_{l,k}},\tfrac{1}{2}eBr^2)\bigr]^2 \\ &= \int_0^1 dx\, \bigl[ \Phi_l({\lambda_{l,k}},\alpha x

is always smaller than the lowest energy break, $ p_{0,1}$. Once $ eB$ become bigger than the squared system - size inverse (that is, $ \alpha\gtrsim 10 $ from Fig.   \[fig: p01\ ]), however, the energy col is significantly reduced and the anomalous yoke between the magnetic field and the rotation is then manifested   [ @Chen:2015hfc; @Hattori:2016njk ]. Integration Measure and Reweighted wave function { # sec: measure } = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = Because the radial momenta are discretized, we substitute the transverse momentum integration with the union over the quantum issue, $ l$ and $ k$, i.e. $ $ \int \frac{dp_xdp_y}{(2\pi)^2 } \longrightarrow \frac{1}{\pi R^2 } \sum_{l=-\infty}^\infty \sum_{k=1}^\infty \frac{1}{{N_{l, k}}^2}\; , \label{eq: phase - space}$$ where $ { N_{l, k}}$ represents a weight factor which corresponds to the integration bill in finite - size systems. In the $ B=0 $ case, as discussed in Ref.   [ @Ebihara:2016fwa ], the weight unit factor is deduced from the Bessel - Fourier expansion, that is, we know that in the limit of $ \alpha\to 0 $, $ $ \begin{split } \label{eq: Nlkzero } { N_{l, k}}^2 \;\;\longrightarrow & \;\; \frac{2}{R^2}\int_0^R roentgen dr\, \bigl[J_l(p_{l, k}r)\bigr]^2 \end{split}$$ with the discretized momentum $ { p_{l, k}}$ in Eq.  . From the relation in Eq.  , we extrapolate the above identification to nonzero $ \alpha$ as $ $ \label{eq: Nlk } \begin{split } { N_{l, k}}^2 & = \frac{2}{R^2}\int_0^R r dr\, \bigl[\Phi_l({\lambda_{l, k}},\tfrac{1}{2}eBr^2)\bigr]^2 \\ & = \int_0 ^ 1 dx\, \bigl [ \Phi_l({\lambda_{l, k}},\alpha x

is always smaller than the lowest energy yqp, $p_{0,1}$. Oice $eB$ gecomes cigger than the squared systxm-size incerse (that is, $\alpha\gtrrim 10$ from Fig. \[fig:p01\]), howtver, the energy gel is sinuificzktly xevuced and the akomalous cogpling between tfe magnetic field and the rotation is then msnlfested [@Chen:2015hfc; @Hatuory:2016njk]. Jntegration Measure and Reweighted wave flnctions {#sec:measute} ================================================= Because the radial momenha age discretized, we geplace the tragwverse momengum integration with tge sums over the quantum numberr, $l$ aud $k$, i.e. $$\int \feac{fk_xdp_y}{(2\pi)^2} \lonjrightwrrow \frac{1}{\pi R^2} \sum_{l=-\itfty}^\infyy \sum_{k=1}^\infty \frec{1}{{N_{l,j}}^2}\;, \label{eq:phase-space}$$ xhere ${N_{l,k}}$ represents a weight fcctor which correspones to tve ittegfqtiun jeesude in vinmte-size sysfems. In the $B=0$ case, as discussec yb Ref. [@Ebihara:2016fsa], the wqight factor is deduced from the Bessel-Xoudier expansion, that is, qe know that in the llmit of $\ajpha\to 0$, $$\begin{split} \label{eq:Nlkzero} {N_{l,k}}^2 \;\;\longrightdrrow & \;\; \yvwz{2}{E^2}\ijt_0^R r dr\, \bigl[J_l(p_{l,k}r)\bigr]^2 \end{split}$$ with the qjsvrvtized momentum ${p_{j,k}}$ in Eq. . Frpm tng relation in Gq. , we erfrzpolate the above ldentifycatiin to noneero $\slpha$ as $$\label{eq:Nlk} \begin{split} {N_{l,k}}^2 &= \fgac{2}{R^2}\unt_0^R r dr\, \bigl[\Phi_p({\lambda_{l,k}},\tfxac{1}{2}eBr^2)\nigr]^2 \\ &= \int_0^1 dx\, \bigl[ \Phi_l({\lamyda_{l,k}},\ampha x

is always smaller than the lowest energy Once becomes bigger the squared system-size from \[fig:p01\]), however, the gap is significantly and the anomalous coupling between the field and the rotation is then manifested [@Chen:2015hfc; @Hattori:2016njk]. Integration Measure and Reweighted functions {#sec:measure} ================================================= Because the radial momenta are discretized, we replace the transverse integration the over quantum numbers, $l$ and $k$, i.e. $$\int \frac{dp_xdp_y}{(2\pi)^2} \longrightarrow \frac{1}{\pi R^2} \sum_{l=-\infty}^\infty \sum_{k=1}^\infty \frac{1}{{N_{l,k}}^2}\;, \label{eq:phase-space}$$ where represents a weight factor which corresponds to the measure in finite-size systems. the $B=0$ case, as discussed Ref. the weight is from Bessel-Fourier expansion, that we know that in the limit of $\alpha\to 0$, $$\begin{split} \label{eq:Nlkzero} {N_{l,k}}^2 \;\;\longrightarrow & \;\; \frac{2}{R^2}\int_0^R r \bigl[J_l(p_{l,k}r)\bigr]^2 \end{split}$$ discretized momentum in . the relation in we extrapolate the above identification to $$\label{eq:Nlk} \begin{split} {N_{l,k}}^2 &= \frac{2}{R^2}\int_0^R r dr\, \bigl[\Phi_l({\lambda_{l,k}},\tfrac{1}{2}eBr^2)\bigr]^2 &= \int_0^1 \bigl[ \Phi_l({\lambda_{l,k}},\alpha x

is always smaller than the lowEst energy gAp, $p_{0,1}$. OnCe $eb$ beCoMes bIggeR than the squareD SystEm-size inverse (that is, $\alpHa\gtrSiM 10$ From fIg. \[Fig:p01\]), hOwever, tHE eNERgy GaP iS siGnIFiCantlY reDuced anD the anomalOus CoUpling betweeN ThE magnetic fIelD and the rotatIon Is then MaNifESted [@CHen:2015Hfc; @HaTtori:2016nJK]. IntegRation MeaSuRE and ReWEighted WAVe FuncTions {#sec:measure} ================================================= BeCAuSE the radial momeNta are DiSCrETIzeD, we Replace the TrAnsveRSe momenTUm INTEgrATion with the suMs over the quANtuM numbeRs, $L$ anD $K$, i.e. $$\int \Frac{dP_xDP_y}{(2\pI)^2} \longrightaRrow \Frac{1}{\pi R^2} \suM_{l=-\inftY}^\Infty \suM_{K=1}^\infty \fRac{1}{{N_{l,k}}^2}\;, \LabEl{eQ:phaSE-sPaCe}$$ wHeRE ${N_{l,K}}$ RePreSEntS a weight FaCtOr whiCh coRRESPondS to The iNtegrAtion measure iN fiNite-SIze SysteMs. In tHe $B=0$ cAsE, as diScusseD in ReF. [@EBihara:2016fwa], the weiGht fActor is deDucEd FroM tHe BesSEl-FourIer ExpAnsion, tHat is, we KNow ThAT IN tHe limit of $\alpha\to 0$, $$\beGiN{SPlIt} \label{eQ:NlkzeRO} {N_{L,k}}^2 \;\;\LOngrightArRow & \;\; \Frac{2}{r^2}\INt_0^R r dR\, \bigL[j_l(P_{l,k}r)\bigr]^2 \End{splIT}$$ wItH the disCrEtized MoMenTum ${P_{l,k}}$ in eQ. . FroM the reLation in eq. , we eXTrapolate the abOVe identificatIOn TO NoNZero $\AlpHa$ as $$\label{eq:nlk} \bEGin{sPlit} {n_{L,k}}^2 &= \FraC{2}{r^2}\int_0^R R dr\, \biGl[\pHi_L({\Lambda_{l,k}},\tfrac{1}{2}eBr^2)\bigR]^2 \\ &= \iNt_0^1 dx\, \biGl[ \Phi_L({\lambda_{l,k}},\alphA x

is always smaller than th e lowest e nergy ga p,$p _{0, 1}$. Once $eB$ bec o mesbigger than the square d sys te m -siz e i nvers e (that is , $\a lp ha \gt rs i m10$ f rom Fig. \ [fig:p01\] ),ho wever, the e n er gy gap issig nificantly r edu ced an dthe anoma lou s cou plingb etween the magn et i c fiel d and th e ro tati on is then manife s te d  [@Chen:2015hf c; @Ha tt o ri : 2 016 njk ]. Integr at ion M e asure a n dR e w eig h ted wave func tions {#sec : mea sure}== === = ====== ===== == = === =========== ==== ========= == Be c ause th e radial momen taare dis c re ti zed ,w e r e pl ace the transve rs emomen tumi n t e grat ion wit h the sums over th e q uant u m n umber s, $l $ an d$k$,i.e. $ $\int \ frac{dp_xdp_y}{ (2\p i)^2} \ lon gr igh ta rrow\ frac{1 }{\ piR^2} \s um_{l=- \ inf ty } ^ \ in fty \sum_{k=1}^\in ft y \frac{1 }{{N_{ l ,k }} ^ 2}\;, \l abe l{eq : p hase- spac e }$ $ where${N_{l , k} }$ repres en ts a w ei ght fa ctorw hich corre sponds t o the integration me a sure in finit e -s i z es yste ms. In the $B= 0$ c a se,as d i sc uss e d inRef.[@ E bi h ara:2016fwa], the w ei ght fa ctoris deduced fr om the Bes s e l -Fourier exp a ns i on, that is, w e kno w that int he limit of $ \alpha\t o 0$, $$\ b e gin{spli t} \ lab el{ e q :N lkzero} {N_ { l ,k}} ^2 \;\;\l ong rightar row &\;\ ; \ frac{2}{R ^2}\int_ 0^ Rrdr \,\bigl [ J_l(p_{l ,k }r) \b igr ]^2 \end {spli t}$$ w it h th e discr e ti z e d mo me nt um $ {p_ {l ,k}}$ inE q.. Fromthe relat ion in E q.  , we ext rapolate theab ove identi fi cat ion to n onzero $ \alpha$ as $$\label{eq: N lk} \ beg in{sp lit} {N_{l, k}} ^2 &=\fr a c{2}{R ^2}\in t_0^R r dr \ , \b i g l[ \Ph i_ l({\lambda _ { l,k }},\t fr ac{1 }{2}eBr ^2)\bigr]^2 \\ & = \i nt_0^1 dx\, \ big l[ \ Ph i_l ( {\ l amb da _ {l, k } },\alpha x

is_always smaller_than the lowest energy_gap, $p_{0,1}$._Once_$eB$ becomes_bigger_than the squared_system-size inverse (that_is, $\alpha\gtrsim 10$ from_Fig. \[fig:p01\]), however, the_energy_gap is significantly reduced and the anomalous coupling between the magnetic field and the_rotation_is then_manifested [@Chen:2015hfc;_@Hattori:2016njk]. Integration_Measure and Reweighted wave functions_{#sec:measure} ================================================= Because the radial momenta are_discretized, we_replace the transverse momentum integration with the sums_over_the quantum numbers,_$l$ and $k$, i.e. $$\int \frac{dp_xdp_y}{(2\pi)^2} \longrightarrow \frac{1}{\pi_R^2} \sum_{l=-\infty}^\infty \sum_{k=1}^\infty \frac{1}{{N_{l,k}}^2}\;, _ \label{eq:phase-space}$$ where_${N_{l,k}}$_represents_a weight factor which_corresponds to the integration measure in_finite-size systems. In the $B=0$ case,_as discussed in Ref. [@Ebihara:2016fwa], the weight factor_is deduced from the Bessel-Fourier expansion,_that is, we know that_in the_limit of $\alpha\to 0$, $$\begin{split} _ \label{eq:Nlkzero} _{N_{l,k}}^2 \;\;\longrightarrow_& \;\; _\frac{2}{R^2}\int_0^R r dr\, \bigl[J_l(p_{l,k}r)\bigr]^2 _ \end{split}$$ with_the discretized momentum ${p_{l,k}}$ in Eq. ._From_the relation in_Eq. ,_we_extrapolate the_above identification to_nonzero_$\alpha$ as_$$\label{eq:Nlk} _ \begin{split} {N_{l,k}}^2 &= \frac{2}{R^2}\int_0^R_r_dr\, \bigl[\Phi_l({\lambda_{l,k}},\tfrac{1}{2}eBr^2)\bigr]^2 \\ &= \int_0^1_dx\, \bigl[ \Phi_l({\lambda_{l,k}},\alpha_x

of squares. This gives that $$\sum_{i\ge 4} 2i \cdot 2q_i \ge 2\cdot 4\cdot 2q_2.$$ By Equation  we then have $-3+3\sum_{i\ge 2}q_i\ge 4q_2+2q_2+3q_3$, that is, $q_2 < \sum_{i\ge 4}q_i$. But then in average each face has more than $6$ edges which contradicts Thm. \[sum\_rank2\]. Hence there is an object $a$ such that there exist $\alpha,\beta,\gamma\in R^a_+$ as above satisfying Equation . We have $\alpha+\gamma, \beta +\gamma, \alpha+\beta+\gamma\in R^a_+$ by Lemma \[le:someroots\](1),(2) and Corollary \[co:cij\]. The classification {#sec:class} ================== In this section we explain the classification of connected simply connected Cartan schemes of rank three such that ${\mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$ is a finite irreducible root system. We formulate the main result in Theorem \[th:class\]. The proof of Theorem \[th:class\] is performed using computer calculations based on results of the previous sections. Our algorithm described below is sufficiently powerful: The implementation in $C$ terminates within a few hours on a usual computer. Removing any of the theorems, the calculations would take at least several weeks. \(1) Let $\cC$ be a connected Cartan scheme of rank three with $I=\{1,2,3\}$. Assume that ${\mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$ is a finite irreducible root system of type $\cC$. Then there exists an object $a\in A$ and a linear map $\tau \in \operatorname{Aut}({\mathbb{Z}}^I)$ such that $\tau ({\alpha }_i)\in \{{\alpha }_1,{\alpha }_2,{\alpha }_3\}$ for all $i\in I$ and $\tau (R^a_+)$ is one of the sets listed in Appendix \[ap:rs\]. Moreover, $\tau (R^a_+)$ with this property is uniquely determined. \(2) Let $R$ be one of the $55$ subsets of ${\mathbb{Z}}^3$

of squares. This gives that $ $ \sum_{i\ge 4 } 2i \cdot 2q_i \ge 2\cdot 4\cdot 2q_2.$$ By Equation   we then have $ -3 + 3\sum_{i\ge 2}q_i\ge 4q_2 + 2q_2 + 3q_3 $, that is, $ q_2 < \sum_{i\ge 4}q_i$. But then in modal each grimace has more than $ 6 $ edges which contradicts Thm.   \[sum\_rank2\ ]. therefore there is an object $ a$ such that there exist $ \alpha,\beta,\gamma\in R^a_+$ as above satisfying equality  . We give birth $ \alpha+\gamma, \beta + \gamma, \alpha+\beta+\gamma\in R^a_+$ by Lemma   \[le: someroots\](1),(2) and Corollary   \[co: cij\ ]. The classification { # sec: class } = = = = = = = = = = = = = = = = = = In this section we explain the categorization of connected simply connected Cartan schemes of membership three such that $ { \mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$ is a finite irreducible root system. We formulate the main result in Theorem   \[th: class\ ]. The validation of Theorem   \[th: class\ ] is performed using computer calculation based on results of the former sections. Our algorithm report below is sufficiently herculean: The implementation in $ C$ terminates within a few hours on a usual calculator. Removing any of the theorems, the calculations would take at least several weeks. \(1) lease $ \cC$ be a connected Cartan scheme of rank three with $ I=\{1,2,3\}$. Assume that $ { \mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$ is a finite irreducible root system of type $ \cC$. Then there exists an object $ a\in A$ and a linear map $ \tau \in \operatorname{Aut}({\mathbb{Z}}^I)$ such that $ \tau ({ \alpha } _ i)\in \{{\alpha } _ 1,{\alpha } _ 2,{\alpha } _ 3\}$ for all $ i\in I$ and $ \tau (R^a_+)$ is one of the hardening listed in Appendix   \[ap: rs\ ]. Moreover, $ \tau (R^a_+)$ with this place is uniquely determine. \(2) lease $ R$ be one of the $ 55 $ subsets of $ { \mathbb{Z}}^3 $

of squares. This gives that $$\sum_{i\ge 4} 2i \cdot 2q_i \ge 2\rdot 4\cdkt 2q_2.$$ By Dquation  we then have $-3+3\sum_{i\ge 2}q_u\ge 4q_2+2w_2+3q_3$, that is, $q_2 < \sum_{i\ge 4}q_k$. But thej in aveeage wach face izs more than $6$ cdges xhich contradicjs Thm. \[sum\_rann2\]. Hence there hs au object $a$ such that there exist $\alpra,\beta,\gsmla\in R^a_+$ as abode sseisfgpnn Equation . We have $\alpha+\gamma, \befa +\gamme, \alpha+\beta+\gamms\in R^a_+$ by Lemma \[le:someroots\](1),(2) and Corollary \[co:cij\]. The classificajjon {#wec:class} ================== In tfis section we explain the classification of connectea sim'ly connectgb Cagjan schemes if ragk three sucm that ${\kathcal{T}}{^\mathrm{re}}({\mathcsl{C}})$ is a finite irreducible root system. We fotmulate tha jain result in Thwoeem \[th:wlasv\]. Thd pruof oh Tgeorem \[hh:cmass\] is pedformed usibg computer calculauiogw based on reaults jf the previous sections. Our algorithm devcrjbed below is sufficienrly powerful: The implgmentation in $C$ terminates within a few hours on a usual cokputec. Femivlng xby of the theorems, the calculations would take wf kesst several wecks. \(1) Let $\cC$ be a comnfcygd Cartan scheoe of xzni three with $I=\{1,2,3\}$. Asskme thaj ${\mathxal{R}}{^\mathri{re}}({\msthcal{C}})$ is a finite irreducuble root syfrem of type $\cC$. Theu there exiscs an pbjecy $a\in A$ and a linear ma' $\tau \jn \operatorjame{Aut}({\maffbb{Z}}^I)$ such that $\gau ({\anpha }_i)\in \{{\alpha }_1,{\alpha }_2,{\alpha }_3\}$ for all $i\in I$ and $\gau (T^a_+)$ is oge of the dets listed in Appendix \[ao:rs\]. Mlraover, $\tau (G^a_+)$ with this property is uniquely determined. \(2) Ked $R$ be one jf thc $55$ subsets of ${\mwthbb{Z}}^3$

of squares. This gives that $$\sum_{i\ge 4} 2q_i 2\cdot 4\cdot By Equation we that $q_2 < \sum_{i\ge But then in each face has more than $6$ which contradicts Thm. \[sum\_rank2\]. Hence there is an object $a$ such that there $\alpha,\beta,\gamma\in R^a_+$ as above satisfying Equation . We have $\alpha+\gamma, \beta +\gamma, \alpha+\beta+\gamma\in by \[le:someroots\](1),(2) Corollary The classification {#sec:class} ================== In this section we explain the classification of connected simply connected Cartan of rank three such that ${\mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$ is a irreducible root system. We the main result in Theorem The of Theorem is using calculations based on of the previous sections. Our algorithm described below is sufficiently powerful: The implementation in $C$ terminates within few hours usual computer. any the the calculations would least several weeks. \(1) Let $\cC$ Cartan scheme of rank three with $I=\{1,2,3\}$. Assume ${\mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$ is finite irreducible root system of type Then there exists an object $a\in A$ and linear map $\tau \in \operatorname{Aut}({\mathbb{Z}}^I)$ such that $\tau ({\alpha }_i)\in \{{\alpha }_1,{\alpha }_2,{\alpha }_3\}$ for I$ and $\tau (R^a_+)$ one of the listed Appendix Moreover, (R^a_+)$ with property is uniquely determined. \(2) Let $R$ be one of the subsets of ${\mathbb{Z}}^3$

of squares. This gives that $$\sum_{I\ge 4} 2i \cdot 2q_i \Ge 2\cdoT 4\cdOt 2q_2.$$ by equaTion  We then have $-3+3\sum_{i\GE 2}q_i\gE 4q_2+2q_2+3q_3$, that is, $q_2 < \sum_{i\ge 4}q_i$. But Then iN aVEragE EaCh facE has morE ThAN $6$ EdgEs WhIch CoNTrAdictS ThM. \[sum\_ranK2\]. Hence therE is An Object $a$ such tHAt There exist $\AlpHa,\beta,\gamma\iN R^a_+$ As abovE sAtiSFying equAtion . we have $\ALpha+\gaMma, \beta +\gaMmA, \Alpha+\bETa+\gamma\IN r^a_+$ By LeMma \[le:someroots\](1),(2) and cOrOLlary \[co:cij\]. The cLassifIcATiON {#Sec:ClaSs} ================== In this seCtIon we EXplain tHE cLASSifICation of conneCted simply cONneCted CaRtAn sCHemes oF rank ThREe sUch that ${\mathCal{R}}{^\Mathrm{re}}({\mAthcal{c}})$ Is a finiTE irreduCible rOot SysTem. WE FoRmUlaTe THe mAIn ResULt iN Theorem \[Th:ClAss\]. ThE proOF OF theoRem \[Th:clAss\] is Performed usinG coMputER caLculaTions BaseD oN resuLts of tHe preViOus sections. Our aLgorIthm descrIbeD bEloW iS suffICientlY poWerFul: The iMplemenTAtiOn IN $c$ TeRminates within a few HoURS oN a usual cOmputeR. reMoVIng any of ThE thEoreMS, The caLculATiOns would Take at LEaSt Several WeEks. \(1) Let $\CC$ Be a ConNecteD cartAn scheMe of rank Three WIth $I=\{1,2,3\}$. Assume that ${\MAthcal{R}}{^\mathrm{RE}}({\mATHcAL{C}})$ is A fiNite irreducIble ROot sYsteM Of TypE $\CC$. TheN therE eXIsTS an object $a\in A$ and a liNeAr map $\tAu \in \oPeratorname{AuT}({\mathbb{Z}}^I)$ sUCH That $\tau ({\aLpha }_I)\In \{{\ALpha }_1,{\alpha }_2,{\alpha }_3\}$ For alL $i\in I$ and $\taU (r^a_+)$ is one oF the sEts listeD in AppendIX \[Ap:rs\]. MoreOveR, $\taU (R^a_+)$ WitH THiS property is unIQUely DeTermineD. \(2) LeT $R$ be one Of tHe $55$ sUbsEts Of ${\Mathbb{Z}}^3$

of squares. This gives th at $$\sum_ {i\ge 4} 2i \ cdot 2q_ i \ge 2\cdot 4 \ cdot 2q_2.$$ By Equation we th en have $- 3+3\s um_{i\g e 2 } q _i\ ge 4 q_2 +2 q _2 +3q_3 $,that is , $q_2 < \ sum _{ i\ge 4}q_i$. Bu t then inave rage each fa cehas mo re th a n $6$ ed ges w hich c o ntradi cts Thm.\[ s um\_ra n k2\]. H e n ce the re is an object $ a $s uch that there exist $ \ al p h a,\ bet a,\gamma\i nR^a_+ $ as abo v es a t isf y ing Equation. We have $ \ alp ha+\ga mm a,\ beta + \gamm a, \al pha+\beta+\ gamm a\in R^a_ +$ byL emma \[ l e:somer oots\] (1) ,(2 ) an d C or oll ar y  \[ c o: cij \ ]. The cla ss if icati on { # s e c :cla ss} === ===== ========== I n t hiss ect ion w e exp lain t he cl assifi catio nof connected si mply connecte d C ar tan s cheme s of ra nkthr ee such that $ { \ma th c a l {R }}{^\mathrm{re}}({ \m a t hc al{C}})$ is af in it e irreduc ib leroot s ystem . We fo rmulatethe ma i nre sult in T heorem  \ [th :cl ass\] . The proof of Theo rem \ [ th:class\] isp erformed usin g c o m pu t er c alc ulations ba sedo n re sult s o f t h e pre vious s e ct i ons. Our algorithmde scribe d bel ow is suffici ently powe r f u l: The i mple m en t ation in $C$ t ermin ates withi n a few h ourson a usu al comput e r . Removi ngany of th e th eorems, the c a l cula ti ons wou ldtake at le ast se ver al weeks. \(1) Let $ \c C$ b e a conn e cted Car ta n s ch eme of r a nk thr ee wi th $ I= \{ 1 ,2, 3\}$. A s su m e tha t${ \mat hca l{ R}}{^ \mat h rm{ re}}({\ mathcal{C }}) $ isafi nite ir reducible roo tsystem ofty pe$\cC$. T hen ther e exists an object $a\i n A$ and alinea r ma p $\tau \ in\opera tor n ame{Au t}({\m athbb {Z }}^ I ) $ suc h th at$\ tau ({\alp h a }_ i)\in \ {{\a lpha }_ 1,{\alpha }_2,{\al p ha}_3\}$ for al l $ i\in I $and $\ t au(R ^ a_+ ) $ is one of thesets liste di nAppendix \ [ ap: rs \]. Mor eover,$\tau (R^a_+) $ with th is proper ty isu n iqu ely determ ined. \ (2) Let $ R $ beo ne of t he$55$ s ub set s of${\mat h bb{ Z}}^3 $

of_squares. This_gives that $$\sum_{i\ge 4}_2i \cdot_2q_i_\ge 2\cdot_4\cdot_2q_2.$$ By Equation _we then have_$-3+3\sum_{i\ge 2}q_i\ge 4q_2+2q_2+3q_3$, that_is, $q_2 <_\sum_{i\ge_4}q_i$. But then in average each face has more than $6$ edges which contradicts_Thm. \[sum\_rank2\]._Hence there_is_an_object $a$ such that there_exist $\alpha,\beta,\gamma\in R^a_+$ as above_satisfying Equation ._We have $\alpha+\gamma, \beta +\gamma, \alpha+\beta+\gamma\in R^a_+$ by_Lemma \[le:someroots\](1),(2)_and Corollary \[co:cij\]. The classification_{#sec:class} ================== In this section we explain the classification of connected_simply connected Cartan schemes of rank_three such that_${\mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$_is_a finite irreducible root_system. We formulate the main result_in Theorem \[th:class\]. The proof of Theorem \[th:class\]_is performed using computer calculations based on_results of the previous sections. Our_algorithm described below is sufficiently_powerful: The_implementation in $C$ terminates within_a few hours_on a_usual computer. Removing_any of the theorems, the calculations_would take at_least several weeks. \(1) Let $\cC$ be_a_connected Cartan scheme_of_rank_three with_$I=\{1,2,3\}$. Assume that_${\mathcal{R}}{^\mathrm{re}}({\mathcal{C}})$_is a_finite_irreducible root system of type $\cC$._Then_there exists an object $a\in A$ and_a linear map $\tau_\in_\operatorname{Aut}({\mathbb{Z}}^I)$ such that $\tau_({\alpha }_i)\in \{{\alpha }_1,{\alpha }_2,{\alpha_}_3\}$ for all $i\in I$ and_$\tau (R^a_+)$_is one_of the sets listed in Appendix \[ap:rs\]. Moreover, $\tau (R^a_+)$ with this_property is uniquely determined. \(2) Let $R$_be one of the_$55$ subsets_of_${\mathbb{Z}}^3$

}^{N} g(\tau_n,\boldsymbol{a}_n)\boldsymbol{a}_n\boldsymbol{a}_n^T\end{aligned}$$ where $g(\tau_n,\boldsymbol{a}_n)$ is defined as $$\begin{aligned} g(\tau_n,\boldsymbol{a}_n) \triangleq \frac {f_{w}^2 (\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n)} {F_{w}(\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n)(1-F_{w}(\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n))} \label{g-function}\end{aligned}$$ in which $f_{w}(\cdot)$ denotes the probability density function (PDF) of $w_n$. Accordingly, the CRB matrix for the estimation problem (\[quantized-data-model-vector\]) is given by $$\begin{aligned} \text{CRB}(\boldsymbol{h})=\boldsymbol{J}^{-1}(\boldsymbol{h}) = \left( \sum_{n=1}^{N} g(\tau_n,\boldsymbol{a}_n) \boldsymbol{a}_n \boldsymbol{a}_n^T \right)^{-1} \label{CRB}\end{aligned}$$ See Appendix \[appB\]. As is well known, the CRB places a lower bound on the estimation error of any unbiased estimator [@Kay93] and is asymptotically attained by the ML estimator. Specifically, the covariance matrix of any unbiased estimate satisfies: $\text{cov}(\hat{\boldsymbol{h}})-\text{CRB}(\boldsymbol{h}) \succeq \boldsymbol{0}$. Also, the variance of each component is bounded by the corresponding diagonal element of $\text{CRB}(\boldsymbol{h})$, i.e., $\text{var}(\hat{h}_i) \ge [\text{CRB}(\boldsymbol{h})]_{ii}$. We observe from (\[CRB\]) that the CRB matrix of $\boldsymbol{h}$ depends on the quantization thresholds $\boldsymbol{\tau}$ as well as the matrix $\boldsymbol{A}$ which is constructed from training sequences $\boldsymbol{X}$ (cf. (\[A-X-relationship\])). Naturally, we wish to optimize $\boldsymbol{\tau}$ and $\boldsymbol{A}$ (i.e. $\boldsymbol{X}$)

} ^{N } g(\tau_n,\boldsymbol{a}_n)\boldsymbol{a}_n\boldsymbol{a}_n^T\end{aligned}$$ where $ g(\tau_n,\boldsymbol{a}_n)$ is defined as $ $ \begin{aligned } g(\tau_n,\boldsymbol{a}_n) \triangleq \frac { f_{w}^2 (\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n) } { F_{w}(\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n)(1 - F_{w}(\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n) ) } \label{g - function}\end{aligned}$$ in which $ f_{w}(\cdot)$ denotes the probability density function (PDF) of $ w_n$. consequently, the CRB matrix for the estimate problem (\[quantized - data - exemplar - vector\ ]) is yield by $ $ \begin{aligned } \text{CRB}(\boldsymbol{h})=\boldsymbol{J}^{-1}(\boldsymbol{h }) = \left (\sum_{n=1}^{N } g(\tau_n,\boldsymbol{a}_n) \boldsymbol{a}_n \boldsymbol{a}_n^T \right)^{-1 } \label{CRB}\end{aligned}$$ See Appendix \[appB\ ]. As is well known, the CRB places a lower bind on the estimate error of any indifferent calculator [ @Kay93 ] and is asymptotically attained by the ML estimator. Specifically, the covariance matrix of any indifferent estimate satisfies: $ \text{cov}(\hat{\boldsymbol{h}})-\text{CRB}(\boldsymbol{h }) \succeq \boldsymbol{0}$. Also, the discrepancy of each component is bounded by the corresponding diagonal element of $ \text{CRB}(\boldsymbol{h})$, i.e., $ \text{var}(\hat{h}_i) \ge [ \text{CRB}(\boldsymbol{h})]_{ii}$. We observe from (\[CRB\ ]) that the CRB matrix of $ \boldsymbol{h}$ depend on the quantization thresholds $ \boldsymbol{\tau}$ as well as the matrix $ \boldsymbol{A}$ which is constructed from prepare sequences $ \boldsymbol{X}$ (cf. (\[A - X - relationship\ ]) ). Naturally, we wish to optimize $ \boldsymbol{\tau}$ and $ \boldsymbol{A}$ (i.e. $ \boldsymbol{X}$ )

}^{N} g(\twu_n,\boldsymbol{a}_n)\boldsymbou{a}_n\boldsymbol{a}_n^J\ebd{aligied}$$ whede $g(\tau_n,\coldsymbol{a}_n)$ is defined as $$\bxgin{qligntb} g(\tau_n,\boldsymbol{a}_n) \tfiangleq \vrac {f_{w}^2 (\bildsbmbol{a}_n^{T}\boldsymbol{h}-\tau_n)} {F_{w}(\boldsglbol{c}_n^{V}\boldsymbol{h}-\tau_k)(1-F_{w}(\boldsymbml{a}_n^{T}\boldsymbon{h}-\gab_n))} \label{g-function}\end{aligned}$$ in which $s_{w}(\cdot)$ cejotes the probwbilptr dehsity function (PDF) of $w_n$. Accordingmy, the BRB matrix for thr estimation problem (\[quantlzed-fata-model-vector\]) is given by $$\bgfin{woigned} \text{CRC}(\boldsymbol{h})=\boldsymbol{N}^{-1}(\boldsymbol{h}) = \left( \sum_{n=1}^{N} g(\tau_n,\boudsymyol{a}_n) \boldsimgop{d}_n \boldsymboo{a}_n^T \gight)^{-1} \label{CRN}\vnd{aligted}$$ See Sppendix \[appB\]. Ax iv wwll known, the CRB plares a lower bound on the estikacion error of any unbuawed evtimdtor [@Jay93] ans ms zsymptltirally attaihed by the NL estimator. Specifocwoly, the covarjance iaerix of any unbiased estimate satisfies: $\test{cov}(\hat{\boldsymbol{h}})-\text{XRB}(\boldsymbol{h}) \succeq \holdsymboj{0}$. Also, the variance of each component is bounded ty thx zorxcfoinfing diagonal element of $\text{CRB}(\boldsymbol{h})$, i.q., $\ttxt{nar}(\hat{h}_i) \ge [\text{CRN}(\boldsymbol{h})]_{ii}$. We onsfrfg from (\[CRB\]) thaj the CXG jatrix of $\boldsymbll{h}$ depgnds ob the quagtizstion thresholds $\boldsymbol{\rau}$ as well cs rhe matrix $\boldsymyol{A}$ which ir comstruvted from training sequznces $\goldsymbol{X}$ (cf. (\[A-X-relzgionship\])). Naturaluy, ee wish to optimize $\boldsymbjl{\tau}$ and $\bolbsymbol{A}$ (i.e. $\noldsyibol{X}$)

}^{N} g(\tau_n,\boldsymbol{a}_n)\boldsymbol{a}_n\boldsymbol{a}_n^T\end{aligned}$$ where $g(\tau_n,\boldsymbol{a}_n)$ is defined as \triangleq {f_{w}^2 (\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n)} \label{g-function}\end{aligned}$$ in which function of $w_n$. Accordingly, CRB matrix for estimation problem (\[quantized-data-model-vector\]) is given by \text{CRB}(\boldsymbol{h})=\boldsymbol{J}^{-1}(\boldsymbol{h}) = \left( \sum_{n=1}^{N} g(\tau_n,\boldsymbol{a}_n) \boldsymbol{a}_n \boldsymbol{a}_n^T \right)^{-1} \label{CRB}\end{aligned}$$ See Appendix \[appB\]. As well known, the CRB places a lower bound on the estimation error of unbiased [@Kay93] is attained by the ML estimator. Specifically, the covariance matrix of any unbiased estimate satisfies: $\text{cov}(\hat{\boldsymbol{h}})-\text{CRB}(\boldsymbol{h}) \succeq Also, the variance of each component is bounded the corresponding diagonal element $\text{CRB}(\boldsymbol{h})$, i.e., $\text{var}(\hat{h}_i) \ge [\text{CRB}(\boldsymbol{h})]_{ii}$. observe (\[CRB\]) that CRB of depends on the thresholds $\boldsymbol{\tau}$ as well as the matrix $\boldsymbol{A}$ which is constructed from training sequences $\boldsymbol{X}$ (cf. (\[A-X-relationship\])). we wish $\boldsymbol{\tau}$ and (i.e.

}^{N} g(\tau_n,\boldsymbol{a}_n)\boldsymBol{a}_n\boldsYmbol{A}_n^T\End{AlIgneD}$$ wheRe $g(\tau_n,\boldsymBOl{a}_n)$ Is defined as $$\begin{aligneD} g(\tau_N,\bOLdsyMBoL{a}_n) \trIangleq \FRaC {F_{W}^2 (\boLdSyMboL{a}_N^{t}\bOldsyMboL{h}-\tau_n)} {F_{W}(\boldsymboL{a}_n^{t}\bOldsymbol{h}-\taU_N)(1-F_{W}(\boldsymboL{a}_n^{t}\boldsymbol{h}-\Tau_N))} \label{G-fUncTIon}\enD{alIgned}$$ In whicH $F_{w}(\cdot)$ Denotes thE pRObabilITy densiTY FuNctiOn (PDF) of $w_n$. AccordinGLy, THe CRB matrix for The estImATiON ProBleM (\[quantized-DaTa-modEL-vector\]) IS gIVEN by $$\BEgin{aligned} \teXt{CRB}(\boldsyMBol{H})=\boldsYmBol{j}^{-1}(\BoldsyMbol{h}) = \LeFT( \suM_{n=1}^{N} g(\tau_n,\bolDsymBol{a}_n) \boldSymbol{A}_N \boldsyMBol{a}_n^T \rIght)^{-1} \laBel{cRB}\End{aLIgNeD}$$ SeE APPenDIx \[Appb\]. as iS well knoWn, ThE CRB pLaceS A LOWer bOunD on tHe estImation error oF anY unbIAseD estiMator [@kay93] aNd Is asyMptotiCally AtTained by the ML esTimaTor. SpecifIcaLlY, thE cOvariANce matRix Of aNy unbiaSed estiMAte SaTISFiEs: $\text{cov}(\hat{\boldsyMbOL{H}})-\tExt{CRB}(\boLdsymbOL{h}) \SuCCeq \boldsYmBol{0}$. also, THE variAnce OF eAch compoNent is BOuNdEd by the CoRrespoNdIng DiaGonal ELemeNt of $\teXt{CRB}(\bolDsymbOL{h})$, i.e., $\text{var}(\hat{H}_I) \ge [\text{CRB}(\bolDSyMBOl{H})]_{Ii}$. We ObsErve from (\[CRB\]) That THe CRb matRIx Of $\bOLdsymBol{h}$ dEpENdS On the quantization thReSholds $\BoldsYmbol{\tau}$ as welL as the matrIX $\BOldsymboL{A}$ whICh IS constructed frOm traIning sequeNCes $\boldsYmbol{x}$ (cf. (\[A-X-relAtionship\])). nATurally, wE wiSh tO opTimIZE $\bOldsymbol{\tau}$ aND $\BoldSyMbol{A}$ (i.e. $\BolDsymbol{x}$)

}^{N} g(\tau_n,\boldsymbol {a}_n)\bol dsymb ol{ a}_ n\ bold symb ol{a}_n^T\end{ a lign ed}$$ where $g(\tau_n, \bold sy m bol{ a }_ n)$ i s defin e da s $$ \b eg in{ al i gn ed} g (\t au_n,\b oldsymbol{ a}_ n) \triangleq\ fr ac {f_{w}^ 2 ( \boldsymbol{ a}_ n^{T}\ bo lds y mbol{ h}- \tau_ n)} {F _ {w}(\b oldsymbol {a } _n^{T} \ boldsym b o l{ h}-\ tau_n)(1-F_{w}(\b o ld s ymbol{a}_n^{T} \bolds ym b ol { h }-\ tau _n))} \lab el {g-fu n ction}\ e nd { a l ign e d}$$ in which $f_{w}(\cd o t)$ denot es th e proba bilit yd ens ity functio n (P DF) of $w _n$. A c cording l y, theCRB ma tri x f or t h ees tim at i onp ro ble m (\ [quantiz ed -d ata-m odel - v e c tor\ ])is g ivenby $$\begin{a lig ned} \te xt{CR B}(\b olds ym bol{h })=\bo ldsym bo l{J}^{-1}(\bold symb ol{h}) =\le ft ( \ su m_{n= 1 }^{N}g(\ tau _n,\bol dsymbol { a}_ n) \ b ol dsymbol{a}_n \bold sy m b ol {a}_n^T\right ) ^{ -1 } \label{ CR B}\ end{ a l igned }$$Se e Append ix \[a p pB \] . As i swell k no wn, th e CRB plac es a l ower bou nd on the estimation error of anyu nb i a se d est ima tor [@Kay93 ] an d isasym p to tic a lly a ttain ed by the ML estimator. S pe cifica lly,the covarianc e matrix o f a ny unbia sede st i mate satisfies : $\t ext{cov}(\ h at{\bold symbo l{h}})-\ text{CRB} ( \ boldsymb ol{ h}) \s ucc e q \ boldsymbol{0} $ . Als o, the va ria nce ofeac h c omp one nt is bound ed by th eco rr es pon dingd iagonalel eme nt of $\te x t{CRB} (\bol dsym bo l{ h })$ , i.e., $\ t e xt{v ar }( \hat {h} _i ) \ge [\t e xt{ CRB}(\b oldsymbol {h} ) ]_{i i} $. We ob serve from (\ [C RB\]) that t heCRB ma t r ix of $\ boldsymbol{h}$ dependso n the q uan tizat ionthreshold s $ \bolds ymb o l{\tau }$ aswellas th e matri x $\ bol ds ymbol{A}$w h ich is c on stru cted fr om training sequen c es$\boldsymbol{ X}$ (cf . (\ [A- X -r e lat io n shi p \ ])). Naturally, we wish t oo pt imize $\bo l dsy mb ol{\tau }$ and$\bol d symbol{ A}$ (i.e. $\boldsy mb ol{X } $ )

}^{N} g(\tau_n,\boldsymbol{a}_n)\boldsymbol{a}_n\boldsymbol{a}_n^T\end{aligned}$$ where_$g(\tau_n,\boldsymbol{a}_n)$ is_defined as $$\begin{aligned} g(\tau_n,\boldsymbol{a}_n) \triangleq_\frac {f_{w}^2 (\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n)} {F_{w}(\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n)(1-F_{w}(\boldsymbol{a}_n^{T}\boldsymbol{h}-\tau_n))} \label{g-function}\end{aligned}$$_in_which $f_{w}(\cdot)$_denotes_the probability density_function (PDF) of_$w_n$. Accordingly, the CRB_matrix for the_estimation_problem (\[quantized-data-model-vector\]) is given by $$\begin{aligned} \text{CRB}(\boldsymbol{h})=\boldsymbol{J}^{-1}(\boldsymbol{h}) = \left( \sum_{n=1}^{N} g(\tau_n,\boldsymbol{a}_n) \boldsymbol{a}_n \boldsymbol{a}_n^T \right)^{-1} \label{CRB}\end{aligned}$$ See Appendix \[appB\]. As_is_well known,_the_CRB_places a lower bound on_the estimation error of any_unbiased estimator_[@Kay93] and is asymptotically attained by the ML_estimator._Specifically, the covariance_matrix of any unbiased estimate satisfies: $\text{cov}(\hat{\boldsymbol{h}})-\text{CRB}(\boldsymbol{h}) \succeq \boldsymbol{0}$. Also,_the variance of each component is_bounded by the_corresponding_diagonal_element of $\text{CRB}(\boldsymbol{h})$, i.e.,_$\text{var}(\hat{h}_i) \ge [\text{CRB}(\boldsymbol{h})]_{ii}$. We observe from (\[CRB\]) that_the CRB matrix of $\boldsymbol{h}$ depends_on the quantization thresholds $\boldsymbol{\tau}$ as well_as the matrix $\boldsymbol{A}$ which is_constructed from training sequences $\boldsymbol{X}$_(cf. (\[A-X-relationship\]))._Naturally, we wish to optimize_$\boldsymbol{\tau}$ and $\boldsymbol{A}$_(i.e. $\boldsymbol{X}$)

), $E(2,0,0)=2 \epsilon_{\alpha}+U_{\alpha \alpha}$ (2e singlet), $E(1,1,1)=\epsilon_{\alpha}+ \epsilon_{\beta}+U_{\alpha \beta}-J_{\alpha \beta}$ (2e triplet), $E(1,1,0)=\epsilon_{\alpha}+ \epsilon_{\beta}+U_{\alpha \beta}+J_{\alpha \beta}$ (2e singlet), $E(2,1,\frac{1}{2})=2\epsilon_{\alpha}+ \epsilon_{\beta}+2U_{\alpha \beta}+U_{\alpha \alpha}-J_{\alpha \beta}$ (3e), $E(2,2,0)=2\epsilon_{\alpha}+ 2\epsilon_{\beta}+4U_{\alpha \beta}+U_{\alpha \alpha}+U_{\beta \beta}-2J_{\alpha \beta}$ (4e). The addition energies measured in experiment are second differences of these $E(N_\alpha, N_\beta, S_{total})$. If the ground state of two electrons is a triplet, $E^{add}_1=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha \beta}-J_{\alpha \beta}$, $E^{add}_2=\epsilon_{\alpha}-\epsilon_{\beta}+ U_{\alpha \alpha} + J_{\alpha \beta}$, and $E^{add}_3=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha \beta}+U_{\beta \beta}-U_{\alpha \alpha}-J_{\alpha \beta}$. If we assume $U_{\alpha \alpha}=U_{\beta \beta}$ and define $\delta \equiv \epsilon_{\beta}-\epsilon_{\alpha}$ (level mismatch), $U \equiv U_{\alpha \beta}$ and $\delta U \equiv U_{\alpha \alpha}- U$ (excess interaction), the expressions for addition energies become $E^{add}_1=E^{add}_3= U-J+\delta $, and $E^{add}_2= U+\delta U +J -\delta$. C. Dekker, Physics Today, 22 (1999); P.L. McEuen, Physics World, June 2000. S.J. Tans [*et al.*]{}, Nature [**386**]{}, 474 (1997). M. Bock

), $ E(2,0,0)=2 \epsilon_{\alpha}+U_{\alpha \alpha}$ (2e singlet), $ E(1,1,1)=\epsilon_{\alpha}+ \epsilon_{\beta}+U_{\alpha \beta}-J_{\alpha \beta}$ (2e triplet), $ E(1,1,0)=\epsilon_{\alpha}+ \epsilon_{\beta}+U_{\alpha \beta}+J_{\alpha \beta}$ (2e singlet), $ E(2,1,\frac{1}{2})=2\epsilon_{\alpha}+ \epsilon_{\beta}+2U_{\alpha \beta}+U_{\alpha \alpha}-J_{\alpha \beta}$ (3e), $ E(2,2,0)=2\epsilon_{\alpha}+ 2\epsilon_{\beta}+4U_{\alpha \beta}+U_{\alpha \alpha}+U_{\beta \beta}-2J_{\alpha \beta}$ (4e). The addition energies measured in experiment are second dispute of these $ E(N_\alpha, N_\beta, S_{total})$. If the earth state of two electrons is a triplet, $ E^{add}_1=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha \beta}-J_{\alpha \beta}$, $ E^{add}_2=\epsilon_{\alpha}-\epsilon_{\beta}+ U_{\alpha \alpha } + J_{\alpha \beta}$, and $ E^{add}_3=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha \beta}+U_{\beta \beta}-U_{\alpha \alpha}-J_{\alpha \beta}$. If we simulate $ U_{\alpha \alpha}=U_{\beta \beta}$ and define $ \delta \equiv \epsilon_{\beta}-\epsilon_{\alpha}$ (level mismatch), $ uranium \equiv U_{\alpha \beta}$ and $ \delta U \equiv U_{\alpha \alpha}- U$ (excess interaction), the formula for accession energies become $ E^{add}_1 = E^{add}_3= U - J+\delta $, and $ E^{add}_2= U+\delta U + J -\delta$. C. Dekker, Physics Today, 22 (1999); P.L. McEuen, Physics World, June 2000. S.J. Tans [ * et al. * ] { }, Nature [ * * 386 * * ] { }, 474 (1997). M. Bock

), $E(2,0,0)=2 \eosilon_{\alpha}+U_{\alpha \alpha}$ (2t singlet), $E(1,1,1)=\epsilon_{\copha}+ \e'silon_{\bsta}+U_{\alphx \beta}-J_{\alpha \beta}$ (2e triplet), $E(1,1,0)=\xpsioon_{\alkka}+ \epsilon_{\beta}+U_{\alpha \beta}+J_{\alpja \beta}$ (2w siiglet), $E(2,1,\frac{1}{2})=2\epsiloi_{\zlpha}+ \eifiloh_{\neta}+2U_{\cl'ha \beta}+U_{\alpha \slpha}-J_{\alphd \beta}$ (3e), $E(2,2,0)=2\epsilmn_{\xl'ha}+ 2\epsilon_{\beta}+4U_{\alpha \beta}+U_{\alpha \alpra}+U_{\beta \bfta}-2J_{\alpha \beta}$ (4e). Tht aqditjon energies measured in experimenf are stcond differences pf these $E(N_\alpha, N_\beta, S_{tohal})$. Lf the ground statf of two elgdtrjbs is a tripuet, $E^{add}_1=\epspnon_{\beta}-\epsjlon_{\alpha}+U_{\alpha \beta}-J_{\alpha \beta}$, $D^{add}_2=\e'silon_{\alpha}-\g'willt_{\beta}+ U_{\alphe \alphw} + J_{\alpha \bebs}$, and $E^{ddd}_3=\epsikon_{\beta}-\epsilon_{\slpia}+U_{\aopha \beta}+U_{\beta \beta}-U_{\anpha \alpha}-J_{\alpha \bgta}$. If we avsbme $U_{\alpha \alpha}=U_{\beta \bwta}$ atd dafind $\deuta \eaujv \epslloi_{\beta}-\epsiloh_{\alpha}$ (leveo mismatch), $U \equiv I_{\ajina \beta}$ and $\selta T \qquiv U_{\alpha \alpha}- U$ (excess interaction), ths expressions for addituon energies become $E^{afd}_1=E^{add}_3= U-J+\qelta $, and $E^{add}_2= U+\delta U +J -\delta$. C. Dekker, Physics Tmday, 22 (1999); P.L. MgEuev, Pjysics World, June 2000. S.J. Tans [*et al.*]{}, Nature [**386**]{}, 474 (1997). M. Bosi

), $E(2,0,0)=2 \epsilon_{\alpha}+U_{\alpha \alpha}$ (2e singlet), $E(1,1,1)=\epsilon_{\alpha}+ \beta}$ triplet), $E(1,1,0)=\epsilon_{\alpha}+ \beta}+J_{\alpha \beta}$ (2e \beta}$ $E(2,2,0)=2\epsilon_{\alpha}+ 2\epsilon_{\beta}+4U_{\alpha \beta}+U_{\alpha \beta}-2J_{\alpha \beta}$ (4e). addition energies measured in experiment are differences of these $E(N_\alpha, N_\beta, S_{total})$. If the ground state of two electrons a triplet, $E^{add}_1=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha \beta}-J_{\alpha \beta}$, $E^{add}_2=\epsilon_{\alpha}-\epsilon_{\beta}+ U_{\alpha \alpha} + J_{\alpha \beta}$, and $E^{add}_3=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha \beta}-U_{\alpha \beta}$. we $U_{\alpha \alpha}=U_{\beta \beta}$ and define $\delta \equiv \epsilon_{\beta}-\epsilon_{\alpha}$ (level mismatch), $U \equiv U_{\alpha \beta}$ and $\delta \equiv U_{\alpha \alpha}- U$ (excess interaction), the expressions addition energies become $E^{add}_1=E^{add}_3= $, and $E^{add}_2= U+\delta U -\delta$. Dekker, Physics 22 P.L. Physics World, June S.J. Tans [*et al.*]{}, Nature [**386**]{}, 474 (1997). M. Bock

), $E(2,0,0)=2 \epsilon_{\alpha}+U_{\alpha \alpha}$ (2e Singlet), $E(1,1,1)=\epSilon_{\AlpHa}+ \ePsIlon_{\Beta}+u_{\alpha \beta}-J_{\alpHA \betA}$ (2e triplet), $E(1,1,0)=\epsilon_{\alpha}+ \EpsilOn_{\BEta}+U_{\ALpHa \betA}+J_{\alpha \BEtA}$ (2E SinGlEt), $e(2,1,\frAc{1}{2})=2\EPsIlon_{\aLphA}+ \epsiloN_{\beta}+2U_{\alphA \beTa}+u_{\alpha \alpha}-J_{\ALpHa \beta}$ (3e), $E(2,2,0)=2\epSilOn_{\alpha}+ 2\epsilOn_{\bEta}+4U_{\alPhA \beTA}+U_{\alpHa \aLpha}+U_{\Beta \beTA}-2J_{\alphA \beta}$ (4e). The AdDItion eNErgies mEASuRed iN experiment are secONd DIfferences of thEse $E(N_\aLpHA, N_\BETa, S_{TotAl})$. If the groUnD statE Of two elECtRONS is A Triplet, $E^{add}_1=\epSilon_{\beta}-\epSIloN_{\alpha}+u_{\aLphA \Beta}-J_{\aLpha \bEtA}$, $e^{adD}_2=\epsilon_{\alpHa}-\epSilon_{\beta}+ u_{\alpha \ALpha} + J_{\alPHa \beta}$, aNd $E^{add}_3=\EpsIloN_{\betA}-\EpSiLon_{\AlPHa}+U_{\ALpHa \bETa}+U_{\Beta \beta}-u_{\aLpHa \alpHa}-J_{\aLPHA \Beta}$. if wE assUme $U_{\aLpha \alpha}=U_{\betA \beTa}$ anD DefIne $\deLta \eqUiv \ePsIlon_{\bEta}-\epsIlon_{\aLpHa}$ (level mismatch), $u \equIv U_{\alpha \bEta}$ AnD $\deLtA U \equIV U_{\alphA \alPha}- u$ (excess InteracTIon), ThE EXPrEssions for addition EnERGiEs become $e^{add}_1=E^{aDD}_3= U-j+\dELta $, and $E^{aDd}_2= u+\deLta U +j -\DElta$. C. dekkER, PHysics ToDay, 22 (1999); P.L. MCeuEn, physics woRld, JunE 2000. S.j. TaNs [*eT al.*]{}, NaTUre [**386**]{}, 474 (1997). M. bock

), $E(2,0,0)=2 \epsilon_{ \alpha}+U_ {\alp ha\al ph a}$(2esinglet), $E( 1 ,1,1 )=\epsilon_{\alpha}+ \ epsil on _ {\be t a} +U_{\ alpha \ b et a } -J_ {\ al pha \ b et a}$ ( 2etriplet ), $E(1,1 ,0) =\ epsilon_{\al p ha }+ \epsilo n_{ \beta}+U_{\a lph a \bet a} +J_ { \alph a \ beta} $ (2es inglet ), $E(2, 1, \ frac{1 } {2})=2\ e p si lon_ {\alpha}+ \epsilo n _{ \ beta}+2U_{\alp ha \be ta } +U _ { \al pha \alpha}-J _{ \alph a \beta} $ ( 3 e ) , $ E(2,2,0)=2\ep silon_{\alp h a}+ 2\eps il on_ { \beta} +4U_{ \a l pha \beta}+U_{ \alp ha \alpha }+U_{\ b eta \be t a}-2J_{ \alpha \b eta }$ ( 4 e) . Th ea ddi t io n e n erg ies meas ur ed in e xper i m e n t ar e s econ d dif ferences of t hes e $E ( N_\ alpha , N_\ beta ,S_{to tal})$ . Ifth e ground stateof t wo electr ons i s a t riple t , $E^ {ad d}_ 1=\epsi lon_{\b e ta} -\ e p s il on_{\alpha}+U_{\al ph a \b eta}-J_{ \alpha \b et a }$, $E^ {a dd} _2=\ e p silon _{\a l ph a}-\epsi lon_{\ b et a} + U_{\a lp ha \al ph a}+ J _{\al p ha \ beta}$ , and $ E^{ad d }_3=\epsilon_{ \ beta}-\epsilo n _{ \ a lp h a}+U _{\ alpha \beta }+U_ { \bet a \b e ta }-U _ {\alp ha \a lp h a} - J_{\alpha \beta}$. I f we a ssume $U_{\alpha \ alpha}=U_{ \ b e ta \beta }$ a n dd efine $\delta\equi v \epsilon _ {\beta}- \epsi lon_{\al pha}$ (le v e l mismat ch) , $ U \ equ i v U _{\alpha \bet a } $ an d$\delta U\equivU_{ \al pha \a lp ha}- U$ ( excess i nt er ac ti on) , the expressi on s f or ad ditio n energ ies b ecom e $ E ^{a dd}_1=E ^ {a d d }_3= U -J +\de lta $ , and $E ^ {ad d}_2= U +\delta U +J -\de lt a$ . C. D ekker, Physic sToday, 22(1 999 ); P.L . McEuen,Physics World, June 200 0 . S.J. Ta ns [* et a l.*]{}, N atu re [** 386 * *]{},474 (1 997). M.B o ck

), $E(2,0,0)=2 \epsilon_{\alpha}+U_{\alpha_\alpha}$ (2e_singlet), $E(1,1,1)=\epsilon_{\alpha}+ \epsilon_{\beta}+U_{\alpha \beta}-J_{\alpha \beta}$_(2e triplet), $E(1,1,0)=\epsilon_{\alpha}+_\epsilon_{\beta}+U_{\alpha_\beta}+J_{\alpha \beta}$_(2e_singlet), $E(2,1,\frac{1}{2})=2\epsilon_{\alpha}+ \epsilon_{\beta}+2U_{\alpha \beta}+U_{\alpha_\alpha}-J_{\alpha \beta}$ (3e), $E(2,2,0)=2\epsilon_{\alpha}+_2\epsilon_{\beta}+4U_{\alpha \beta}+U_{\alpha \alpha}+U_{\beta \beta}-2J_{\alpha_\beta}$ (4e). The addition_energies_measured in experiment are second differences of these $E(N_\alpha, N_\beta, S_{total})$. If the ground_state_of two_electrons_is_a triplet, $E^{add}_1=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha \beta}-J_{\alpha \beta}$, $E^{add}_2=\epsilon_{\alpha}-\epsilon_{\beta}+ U_{\alpha_\alpha} + J_{\alpha \beta}$, and $E^{add}_3=\epsilon_{\beta}-\epsilon_{\alpha}+U_{\alpha_\beta}+U_{\beta \beta}-U_{\alpha_\alpha}-J_{\alpha \beta}$. If we assume $U_{\alpha \alpha}=U_{\beta \beta}$ and_define_$\delta \equiv \epsilon_{\beta}-\epsilon_{\alpha}$_(level mismatch), $U \equiv U_{\alpha \beta}$ and $\delta U_\equiv U_{\alpha \alpha}- U$ (excess interaction),_the expressions for_addition_energies_become $E^{add}_1=E^{add}_3= U-J+\delta $, and $E^{add}_2=_U+\delta U +J -\delta$. C. Dekker, Physics_Today, 22 (1999); P.L. McEuen, Physics_World, June 2000. S.J. Tans [*et al.*]{}, Nature_[**386**]{}, 474 (1997). M. Bock

\delta_{ij} - n^{-1} \big)^2 \cdot \mathbf{E}_{ii} } \right\Vert} = {\left\Vert {\sum_{i=1}^d \sum_{j=1}^n n^{-1}\big(1 - n^{-1}\big) \cdot \mathbf{E}_{ii} } \right\Vert} \approx 1.$$ The large deviation parameter is $$L^2 = {\operatorname{\mathbb{E}}}\max\nolimits_{i,j} {{\left\Vert { \big(\delta_{ij} - n^{-1} \big)\cdot \mathbf{E}_{ii} } \right\Vert}^2} \approx 1.$$ Therefore, the large-deviation term drives the upper bound in : $$\left( {\operatorname{\mathbb{E}}}{{\left\Vert {{{\bm{Z}}}} \right\Vert}^2} \right)^{1/2} \leq \sqrt{4 (1 + 2\lceil \log d \rceil)} + 4(1 + 2 \lceil \log d \rceil).$$ On the other hand, by direct calculation $$\left( {\operatorname{\mathbb{E}}}{{\left\Vert {{{\bm{Z}}}} \right\Vert}^2} \right)^{1/2} \approx \left( {\operatorname{\mathbb{E}}}{{\left\Vert { \sum_{i=1}^d (Q_i - 1) \cdot \mathbf{E}_{ii} } \right\Vert}^2} \right)^{1/2} = \left( {\operatorname{\mathbb{E}}}\max_{i=1,\dots,d} {{{\left\vert {Q_i - 1} \right\vert}}^2} \right)^{1/2} \approx \mathrm{const} \cdot \frac{\log d}{\log \log d}.$$ Here, $\{Q_i\}$ is an independent family of $\textsc{poisson}(1)$ random variables, and the first approximation follows from the Poisson limit of a binomial. The second approximation depends on a (messy) calculation for the expected squared maximum of a family of independent Poisson variables. We see that the large deviation term in the upper bound  cannot be improved, except by an iterated logarithm factor. Lower Bound: Variance Term -------------------------- Next, we argue that there are examples where the variance term in

\delta_{ij } - n^{-1 } \big)^2 \cdot \mathbf{E}_{ii } } \right\Vert } = { \left\Vert { \sum_{i=1}^d \sum_{j=1}^n n^{-1}\big(1 - n^{-1}\big) \cdot \mathbf{E}_{ii } } \right\Vert } \approx 1.$$ The large deviation parameter is $ $ L^2 = { \operatorname{\mathbb{E}}}\max\nolimits_{i, j } { { \left\Vert { \big(\delta_{ij } - n^{-1 } \big)\cdot \mathbf{E}_{ii } } \right\Vert}^2 } \approx 1.$$ Therefore, the big - diversion term drives the upper bounce in  : $ $ \left ({ \operatorname{\mathbb{E}}}{{\left\Vert { { { \bm{Z } } } } \right\Vert}^2 } \right)^{1/2 } \leq \sqrt{4 (1 + 2\lceil \log five hundred \rceil) } + 4(1 + 2 \lceil \log d \rceil).$$ On the other hired hand, by direct calculation $ $ \left ({ \operatorname{\mathbb{E}}}{{\left\Vert { { { \bm{Z } } } } \right\Vert}^2 } \right)^{1/2 } \approx \left ({ \operatorname{\mathbb{E}}}{{\left\Vert { \sum_{i=1}^d (Q_i - 1) \cdot \mathbf{E}_{ii } } \right\Vert}^2 } \right)^{1/2 } = \left ({ \operatorname{\mathbb{E}}}\max_{i=1,\dots, d } { { { \left\vert { Q_i - 1 } \right\vert}}^2 } \right)^{1/2 } \approx \mathrm{const } \cdot \frac{\log d}{\log \log d}.$$ Here, $ \{Q_i\}$ is an independent class of $ \textsc{poisson}(1)$ random variables, and the first estimate follows from the Poisson limit of a binomial. The second approximation depends on a (messy) calculation for the expect squared utmost of a family of autonomous Poisson variable. We see that the large deviation terminus in the upper bound   cannot be improved, except by an iterated logarithm component. Lower Bound: Variance Term -------------------------- Next, we argue that there are example where the variance term in

\depta_{ij} - n^{-1} \big)^2 \cdot \mathbf{T}_{ii} } \right\Vert} = {\left\Vxrt {\sum_{j=1}^d \sum_{j=1}^n n^{-1}\big(1 - n^{-1}\big) \cdot \mathbf{E}_{ii} } \cighr\Vert} \approx 1.$$ The large aeviation parametwr iw $$L^2 = {\operatorname{\mabkbb{E}}}\mzw\nolikmts_{i,j} {{\left\Vert { \big(\delta_{ib} - n^{-1} \big)\cdot \mdtfby{E}_{ii} } \right\Vert}^2} \approx 1.$$ Therefore, the latgf-deviation teri drpvqs tgv mpper bound in : $$\left( {\operatorname{\jathbb{E}}}{{\neft\Vert {{{\bm{Z}}}} \tight\Vert}^2} \right)^{1/2} \leq \sqrt{4 (1 + 2\lcell \log d \rceil)} + 4(1 + 2 \lceil \log e \rcqul).$$ On the otfer hand, bj direct camculation $$\left( {\operatorname{\mathcb{E}}}{{\leyt\Vert {{{\bm{Z}}}} \tntht\Ggrt}^2} \right)^{1/2} \apprjx \left( {\operatorname{\kathbb{E}}}{{\keft\Vert { \sum_{i=1}^c (Q_m - 1) \xdot \mathbf{E}_{ii} } \right\Tert}^2} \right)^{1/2} = \left( {\jperatorndmz{\mathbb{E}}}\max_{i=1,\dots,d} {{{\lefr\vwrt {Q_h - 1} \sighg\cerg}}^2} \rjgit)^{1/2} \appgox \mathrm{conat} \cdot \frax{\log d}{\log \log d}.$$ Hert, $\{Q_y\}$ is an indepehdent saiily of $\textsc{poisson}(1)$ random variables, dnd the first approximatiob follows from the Polsson limyt of a binomial. The second approximation depends on a (oeswy) zqlfulation for the expected squared maximum of w fsmply of independenb Poisson variablex. Ae fee that the uarge bsvjation term in the upper found  cannot bt imptoved, except by an iterated logarithm fcctir. Lower Bound: Varicnce Term -------------------------- Nexc, we atgue tnat there are examples chere fhe variancf term in

\delta_{ij} - n^{-1} \big)^2 \cdot \mathbf{E}_{ii} } {\left\Vert \sum_{j=1}^n n^{-1}\big(1 n^{-1}\big) \cdot \mathbf{E}_{ii} large parameter is $$L^2 {\operatorname{\mathbb{E}}}\max\nolimits_{i,j} {{\left\Vert { - n^{-1} \big)\cdot \mathbf{E}_{ii} } \right\Vert}^2} 1.$$ Therefore, the large-deviation term drives the upper bound in : $$\left( {\operatorname{\mathbb{E}}}{{\left\Vert \right\Vert}^2} \right)^{1/2} \leq \sqrt{4 (1 + 2\lceil \log d \rceil)} + 4(1 + \lceil d On other hand, by direct calculation $$\left( {\operatorname{\mathbb{E}}}{{\left\Vert {{{\bm{Z}}}} \right\Vert}^2} \right)^{1/2} \approx \left( {\operatorname{\mathbb{E}}}{{\left\Vert { \sum_{i=1}^d (Q_i 1) \cdot \mathbf{E}_{ii} } \right\Vert}^2} \right)^{1/2} = \left( {{{\left\vert {Q_i - 1} \right)^{1/2} \approx \mathrm{const} \cdot \frac{\log \log Here, $\{Q_i\}$ an family $\textsc{poisson}(1)$ random variables, the first approximation follows from the Poisson limit of a binomial. The second approximation depends on a calculation for squared maximum a of Poisson variables. We the large deviation term in the be improved, except by an iterated logarithm factor. Bound: Variance -------------------------- Next, we argue that there examples where the variance term in

\delta_{ij} - n^{-1} \big)^2 \cdot \mathbf{E}_{ii} } \rIght\Vert} = {\leFt\VerT {\suM_{i=1}^d \SuM_{j=1}^n n^{-1}\Big(1 - n^{-1}\Big) \cdot \mathbf{E}_{II} } \rigHt\Vert} \approx 1.$$ The large deViatiOn PAramETeR is $$L^2 = {\oPeratorNAmE{\MAthBb{e}}}\mAx\nOlIMiTs_{i,j} {{\lEft\vert { \big(\Delta_{ij} - n^{-1} \biG)\cdOt \Mathbf{E}_{ii} } \rigHT\VErt}^2} \approx 1.$$ THerEfore, the largE-deViatioN tErm DRives The Upper Bound iN : $$\Left( {\opEratornamE{\mAThbb{E}}}{{\lEFt\Vert {{{\bM{z}}}} \RiGht\VErt}^2} \right)^{1/2} \leq \sqrt{4 (1 + 2\lcEIl \LOg d \rceil)} + 4(1 + 2 \lceil \lOg d \rceIl).$$ oN tHE OthEr hAnd, by direcT cAlculATion $$\lefT( {\OpERATorNAme{\mathbb{E}}}{{\lefT\Vert {{{\bm{Z}}}} \rigHT\VeRt}^2} \righT)^{1/2} \aPprOX \left( {\oPeratOrNAme{\Mathbb{E}}}{{\left\vert { \Sum_{i=1}^d (Q_i - 1) \cdOt \mathBF{E}_{ii} } \rigHT\Vert}^2} \riGht)^{1/2} = \lefT( {\opEraTornAMe{\MaThbB{E}}}\MAx_{i=1,\DOtS,d} {{{\lEFt\vErt {Q_i - 1} \rigHt\VeRt}}^2} \rigHt)^{1/2} \apPROX \MathRm{cOnst} \Cdot \fRac{\log d}{\log \log D}.$$ HeRe, $\{Q_i\}$ IS an IndepEndenT famIlY of $\teXtsc{poIsson}(1)$ RaNdom variables, anD the First apprOxiMaTioN fOllowS From thE PoIssOn limit Of a binoMIal. thE SECoNd approximation depEnDS On A (messy) caLculatIOn FoR The expecTeD sqUareD MAximuM of a FAmIly of indEpendeNT POiSson varIaBles. We SeE thAt tHe larGE devIation Term in thE uppeR Bound  cannot be iMProved, except bY An ITErATed lOgaRithm factor. loweR bounD: VarIAnCe TERm -------------------------- NexT, we arGuE ThAT there are examples whErE the vaRiancE term in

\delta_{ij} - n^{-1} \big )^2 \cdot\math bf{ E}_ {i i} } \ri ght\Vert} = {\l eft\Vert {\sum_{i=1}^d \sum _{ j =1}^ n n ^{-1} \big(1- n ^ { -1} \b ig ) \ cd o t\math bf{ E}_{ii} } \right\ Ver t} \approx 1. $$ The lar gedeviation pa ram eter i s$$L ^ 2 = { \op erato rname{ \ mathbb {E}}}\max \n o limits _ {i,j} { { \ le ft\V ert { \big(\delta _ {i j } - n^{-1} \bi g)\cdo t\ ma t h bf{ E}_ {ii} } \ri gh t\Ver t }^2} \ a p p rox 1.$$ Therefor e, the larg e -de viatio nter m drive s the u p per bound in : $$\ left( {\o perato r name{\m a thbb{E} }}{{\l eft \Ve rt { { {\ bm {Z} }} } \r i gh t\V e rt} ^2} \rig ht )^ {1/2} \le q \ s qrt{ 4 ( 1 +2\lce il \log d \rc eil )} + 4(1 + 2\lcei l \l og d \r ceil). $$ On t he other hand,by d irect cal cul at ion $ $\lef t ( {\op era tor name{\m athbb{E } }}{ {\ l e f t\ Vert {{{\bm{Z}}}}\r i g ht \Vert}^2 } \rig h t) ^{ 1 /2} \a ppr ox \ l e ft( { \ope r at orname{\ mathbb { E} }} {{\left \V ert {\s um_ {i= 1}^d( Q_i- 1) \ cdot \ma thbf{ E }_{ii} } \righ t \Vert}^2} \ri g ht ) ^ {1 / 2} = \left( {\ oper a torn ame{ \ ma thb b {E}}} \max_ {i = 1, \ dots,d} {{{\left\ve rt {Q_i- 1}\right\vert}} ^2} \right ) ^ { 1/2} \ap p ro x \mathrm{const } \cd ot \frac{\ l og d}{\l og \l og d}.$$ Here, $\ { Q _i\}$ is an in dep end e n tfamily of $\t e x tsc{ po isson}( 1)$ random va ria ble s,an d the fir st appro xi ma ti on fo llows from the P ois so n l imito f a bi nomia l. T he s e con d appro x im a t ionde pe ndsona(mess y) c a lcu lationfor the e xpe c tedsq ua red max imum of a fam il y of indep en den t Pois s o n variab les. We see that the la r ge devi ati on te rm i n the upp erbound ca n not be impro ved,ex cep t by an i te rat ed logarithm f act or. Lo werBound:Variance Term ---- - --- ------------- --- -- N e xt , w e a r gue t h att h ere are example s where th ev ar iance term in

\delta_{ij}_- n^{-1}_\big)^2 \cdot \mathbf{E}_{ii} }_\right\Vert} __ =_{\left\Vert_{\sum_{i=1}^d \sum_{j=1}^n n^{-1}\big(1_- n^{-1}\big) \cdot_\mathbf{E}_{ii} } \right\Vert} _ \approx_1.$$_The large deviation parameter is $$L^2 = {\operatorname{\mathbb{E}}}\max\nolimits_{i,j} {{\left\Vert { \big(\delta_{ij} - n^{-1} \big)\cdot_\mathbf{E}_{ii}_} \right\Vert}^2} ___ \approx 1.$$ Therefore, the_large-deviation term drives the upper_bound in :_$$\left( {\operatorname{\mathbb{E}}}{{\left\Vert {{{\bm{Z}}}} \right\Vert}^2} \right)^{1/2} \leq \sqrt{4 (1_+_2\lceil \log d_\rceil)} + 4(1 + 2 \lceil \log d \rceil).$$_On the other hand, by direct_calculation $$\left( {\operatorname{\mathbb{E}}}{{\left\Vert_{{{\bm{Z}}}}_\right\Vert}^2}_\right)^{1/2} _\approx \left( {\operatorname{\mathbb{E}}}{{\left\Vert { \sum_{i=1}^d (Q_i_- 1) \cdot \mathbf{E}_{ii} } \right\Vert}^2}_\right)^{1/2} = \left( {\operatorname{\mathbb{E}}}\max_{i=1,\dots,d}_{{{\left\vert {Q_i - 1} \right\vert}}^2} \right)^{1/2} _ \approx \mathrm{const}_\cdot \frac{\log_d}{\log \log d}.$$ Here, $\{Q_i\}$_is an independent_family of_$\textsc{poisson}(1)$ random variables,_and the first approximation follows from_the Poisson limit_of a binomial. The second approximation_depends_on a (messy)_calculation_for_the expected_squared maximum of_a_family of_independent_Poisson variables. We see that the_large_deviation term in the upper bound  cannot_be improved, except by_an_iterated logarithm factor. Lower Bound:_Variance Term -------------------------- Next, we argue that_there are examples where the variance_term in

the effect of dropout can be found in Appendix D.4. Discussion {#sec:discussion} ========== The results show that the system has an acceptable TRE of $3.3\%\pm1.4\%$. The MVPM outperforms DeepLab V3+ and Dense 3D CRF for semantic segmentation, with statistically significant improvement in less frequently appearing classes such as Human, Mayo Stand, Sterile Table and Anesthesia Cart. Ablation studies have also shown performance improvements as the number of cameras increase. However, the results are limited due to the multi-view dataset size and its distribution. Future work includes obtaining more accurate camera parameters to improve registration accuracy and working towards a larger and more balanced multi-view dataset. Conclusion {#sec:conclusion} ========== In this paper, we describe for the first time, a complete and novel solution to create a 3D perception system to enhance environmental awareness, for surgical robots like the da Vinci Xi system. Our system consists of four ToF cameras rigidly mounted on the PSC robot, and a one-time calibration process for multi-camera to robot registration, which is sufficient for the sensor package to be used in other da Vinci Xi systems. Furthermore, a multi-view semantic segmentation fusion framework called MVPM and new datasets for algorithm training/validation are proposed. Our results show that the framework can improve OR scene segmentation accuracy over single camera prediction. **Conflict of interest:** Z. Li, A. Shaban and J.G. Simard are former interns, D. Rabindran, S. DiMaio and O. Mohareri are employees at Intuitive Surgical Inc. **Ethical approval:** For this type of study formal patient consent is not required. --- abstract: 'Material systems with Dirac electrons on a bipartite planar lattice and possessing superconducting and excitonic interactions are investigated both in the half-filling and doped regimes at zero temperature. Excitonic pairing is the analog of chiral symmetry breaking of relativistic fermion theories and produces an insulating gap in the electronic spectrum. Condensed matter systems with such competing interactions display phenomena that are analogous to the onset of the chiral condensate and of color superconductivity in dense quark matter. Evaluation of the free-energy (effective potential) allows us to map the phases of the system for different values of the couplings of each interaction. At half-filling, we show that Cooper pairs and excitons can coexist if the superconducting and excit

the effect of dropout can be found in Appendix D.4. Discussion { # sec: discussion } = = = = = = = = = = The results show that the arrangement have an acceptable TRE of $ 3.3\%\pm1.4\%$. The MVPM outperforms DeepLab V3 + and Dense 3D CRF for semantic division, with statistically significant improvement in less frequently appearing course such as Human, Mayo Stand, Sterile Table and Anesthesia Cart. Ablation study have also shown performance improvements as the issue of cameras increase. However, the solution are specify due to the multi - view dataset size and its distribution. Future work includes obtaining more accurate camera parameters to better registration accuracy and working towards a larger and more balanced multi - view dataset. Conclusion { # sec: termination } = = = = = = = = = = In this paper, we describe for the first time, a accomplished and novel solution to create a 3D perception system to enhance environmental awareness, for surgical robots like the da Vinci Xi arrangement. Our system consists of four ToF cameras rigidly mounted on the PSC robot, and a one - time calibration process for multi - camera to robot registration, which is sufficient for the sensor package to be used in other da Vinci Xi systems. Furthermore, a multi - view semantic segmentation fusion model call MVPM and new datasets for algorithm education / establishment are proposed. Our results show that the framework can better OR scene segmentation accuracy over single camera prediction. * * battle of interest :* * Z. Li, A. Shaban and J.G. Simard are former interns, D. Rabindran, S. DiMaio and O. Mohareri are employees at Intuitive Surgical Inc. * * Ethical approval :* * For this type of discipline formal patient consent is not required. --- abstract:' Material systems with Dirac electrons on a bipartite planar wicket and possess superconducting and excitonic interactions are investigated both in the half - filling and doped government at zero temperature. Excitonic pairing is the analog of chiral symmetry breakage of relativistic fermion theories and produce an insulating gap in the electronic spectrum. Condensed topic systems with such competing interactions display phenomena that are analogous to the onset of the chiral condensation and of color superconductivity in dense quark matter. Evaluation of the free - department of energy (effective potential) allows us to map the phase of the system for different value of the couplings of each interaction. At half - woof, we show that Cooper pairs and excitons can coexist if the superconducting and excit

thf effect of dropout can ne found in Appeueix D.4. Dmscussikn {#sec:dircussion} ========== The results show thav thw sysuvm has an acceptable GRE of $3.3\%\pm1.4\%$. The MVPN ouuperforms DeepLab V3+ and Dcuse 3D GRF fmc semantic segmgntation, witv statisticallf riynificant improvement in less frequegtly aplewring classes fuch ws Hhman, Mayo Stand, Sterile Table and Znesthevia Cart. Ablayion studies have also shoan pfrformance improvelents as thg nuiver of camerxs increast. Kowever, the results are limited due to the multn-view datasgc sixg and its diwtribltion. Future work incnudes ontaining more sccnratw camera parameters tm improve registrajion accurdcv and working towards a larggr ang mofw bxlahcxd julti-vlew dataset. Cohclusion {#sex:conclusion} ========== In this kapqg, we describe for tre first time, a complete and novel solutimn fo create a 3D perceptiob system to enhance ejvironmeneal awareness, for surgical robots like the da Vinwi Xi rysuen. Our wydtem consists of four ToF cameras rigidly mougfec pn the PSC robjt, and a onr-tlmr calibration krocess fod multi-camera to rlbot redistrqtion, whish ix sufficient for the sensor package to ye ysed in other da Vnnci Xi systzms. Futthermpre, a multi-view semantie segmsntation fudion framsdork called MVPM anc tew datasets for algorithm training/talidction ard prpposed. Our resulhs show that the framewogk cau impsove OR scfne segmentation accuracy over single camera kregicnion. **Conflnct of interest:** Z. Ly, A. Shaban and J.G. Simcrd ard former ihterns, V. Rabindran, F. DiMaio and M. Mohareri arx employeqs ar Inruitive Rurgical Inc. **Etnical appgoral:** For tyis type of study normau patient consenc us not required. --- abrtrwcn: 'Meteriwn systems widh Dkraz elecgrons on a nipxrtiye planar lattice ang poasessing supercondicbing and gxcitonic interactions are investigated hoth mn the half-giljing and doped regimes at zero temperatkre. Excitonic pwirikg if the analpg of chiral symmetry breaking of relatmvistic fermion theorigs and produces an inxmlating gap mn the electronhc spectrum. Condensee matter systems eith such competing ihteracdions display phenomena that are analogous to the onset of the chiral condensare and of color sulercpnducnivnty in degse enark matter. Evalustion of the free-energy (effectite potentidl) allows us to map the phases ow the system wor different values of fhe coupkings of each interaction. At hakf-filling, we show that Cpoper paurs ahd excpjonr can coexist if tne supzrcondycting qnd cxcit

the effect of dropout can be found D.4. {#sec:discussion} ========== results show that TRE $3.3\%\pm1.4\%$. The MVPM DeepLab V3+ and 3D CRF for semantic segmentation, with significant improvement in less frequently appearing classes such as Human, Mayo Stand, Sterile and Anesthesia Cart. Ablation studies have also shown performance improvements as the number cameras However, results limited due to the multi-view dataset size and its distribution. Future work includes obtaining more accurate parameters to improve registration accuracy and working towards larger and more balanced dataset. Conclusion {#sec:conclusion} ========== In paper, describe for first a and novel solution create a 3D perception system to enhance environmental awareness, for surgical robots like the da Vinci Xi Our system four ToF rigidly on PSC robot, and calibration process for multi-camera to robot sufficient for the sensor package to be used other da Xi systems. Furthermore, a multi-view semantic fusion framework called MVPM and new datasets for training/validation are proposed. Our results show that the framework can improve OR scene segmentation accuracy camera prediction. **Conflict of Z. Li, A. and Simard former D. Rabindran, DiMaio and O. Mohareri are employees at Intuitive Surgical Inc. **Ethical For this type of study formal patient consent is not abstract: systems with Dirac on a bipartite planar and superconducting and excitonic interactions both the regimes zero Excitonic pairing is the of chiral symmetry breaking of fermion theories and produces electronic spectrum. Condensed matter systems with such competing display phenomena that are analogous to the of the chiral condensate and of color superconductivity in dense quark matter. of the potential) allows us to map the phases of system for different values the couplings of each interaction. At half-filling, we show Cooper and excitons coexist if the and excit

the effect of dropout can be foUnd in AppenDix D.4. DIscUssIoN {#sec:DiscUssion} ========== The resulTS shoW that the system has an accEptabLe trE of $3.3\%\PM1.4\%$. THe MVPm outperFOrMS deePLAb v3+ anD DENsE 3D CRF For SemantiC segmentatIon, WiTh statisticaLLy SignificanT imProvement in lEss FrequeNtLy aPPeariNg cLasseS such aS human, MAyo Stand, STeRIle TabLE and AneSTHeSia CArt. Ablation studieS HaVE also shown perfOrmancE iMPrOVEmeNts As the numbeR oF cameRAs increASe. hOWEveR, The results are Limited due tO The Multi-vIeW daTAset siZe and ItS DisTribution. FuTure Work incluDes obtAIning moRE accuraTe cameRa pAraMeteRS tO iMprOvE RegIStRatIOn aCcuracy aNd WoRking TowaRDS A LargEr aNd moRe balAnced multi-vieW daTaseT. conClusiOn {#sec:ConcLuSion} ========== IN this pAper, wE dEscribe for the fiRst tIme, a complEte AnD noVeL soluTIon to cReaTe a 3d percepTion sysTEm tO eNHANcE environmental awarEnESS, fOr surgicAl roboTS lIkE The da VinCi xi sYsteM. oUr sysTem cONsIsts of foUr ToF cAMeRaS rigidlY mOunted On The pSC Robot, ANd a oNe-time CalibratIon prOCess for multi-caMEra to robot regIStRATiON, whiCh iS sufficient For tHE senSor pACkAge TO be usEd in oThER dA vinci Xi systems. FurthErMore, a mUlti-vIew semantic seGmentation FUSIon frameWork CAlLEd MVPM and new daTasetS for algoriTHm trainiNg/valIdation aRe proposeD. oUr resultS shOw tHat The FRAmEwork can improVE oR scEnE segmenTatIon accuRacY ovEr sIngLe Camera preDiction. **COnFlIcT oF inTeresT:** z. Li, A. ShabAn And j.G. simArd arE Former InterNs, D. RAbInDRan, s. DiMaio ANd o. mOharErI aRe emPloYeEs at INtuiTIve surgicaL Inc. **EthicAl aPProvAl:** foR this tyPe of study formAl Patient conSeNt iS not reQUIred. --- abstRact: 'Material systems with dIrac eleCtrOns on A bipArtite plaNar LatticE anD PossesSing suPercoNdUctING and eXCItOniC iNteractionS ARe iNvestIgAted Both in tHe half-filling and doPEd rEgimes at zero tEmpEratURE. EXciTOnIC paIrINg iS THe analog of chiraL symmetry bReAKiNg of relatiVIstIc Fermion TheorieS and pROduces aN insulatiNg gap in thE eLectRONic Spectrum. CoNdensed mAtter systEMs witH SuCh comPetIng intErActIons dIsplay PHenOmena That arE aNalogoUs to tHe Onset of tHe chiral condensate and of Color sUpercOndUctivity iN deNSe qUark matteR. EvaLuation of tHe fRee-EnergY (efFEctivE potENtIal) ALlows Us to MAp the phasES oF thE SYsTem for diffeRENT vaLues oF thE CoupliNgs oF each interaction. AT Half-filling, we sHow tHAT CoOpeR PairS aNd excitons can cOexIsT IF the supeRcOnducting anD excit

the effect of dropout can be foundin Ap pen dix D .4. Dis cussion {#sec: d iscu ssion} ========== The resu lt s sho w t hat t he syst e mh a s a nac cep ta b le TREof$3.3\%\ pm1.4\%$.The M VPM outperfo r ms DeepLab V 3+and Dense 3D CR F forse man t ic se gme ntati on, wi t h stat istically s i gnific a nt impr o v em entin less frequentl y a p pearing classe s such a s H u m an, Ma yo Stand,St erile Table a n dA n e sth e sia Cart. Abl ation studi e s h ave al so sh o wn per forma nc e im provementsas t he number of ca m eras in c rease.Howeve r,the res u lt sare l i mit e ddue tothe mult i- vi ew da tase t s i ze a nditsdistr ibution. Futu rework inc ludes obta inin gmoreaccura te ca me ra parameters t o im prove reg ist ra tio naccur a cy and wo rki ng towa rds a l a rge ra n d m ore balanced multi -v i e wdataset. Conc l us io n {#sec:c on clu sion } ===== ==== = In thispaper, we d escribe f or the f irs t t ime,a com pleteand nove l sol u tion to create a 3D percepti o ns y st e m to en hance envir onme n talawar e ne ss, for s urgic al ro b ots like the da Vin ci Xi sy stem. Our system c onsists of f o ur ToF c amer a sr igidly mounted on t he PSC rob o t, and a one- time cal ibrationp r ocess fo r m ult i-c ame r a t o robot regis t r atio n, whichissuffici ent fo r t hese nsor pack age to b eus ed i n o therd a VinciXi sy st ems . Fur t hermor e, amult i- vi e w s emantic se g m enta ti on fus ion f ramew orkc all ed MVPM and newdat a sets f or algori thm training/ va lidation a re pr oposed . Our resu lts show that the frame w ork can im prove ORscene seg men tation ac c uracyover s ingle c ame r a pred i c ti on. **Conflict o f i ntere st :**Z. Li,A. Shaban and J.G. Sim ard are forme r i nter n s ,D.R ab i ndr an , S. D iMaio and O. Mo hareri are e m pl oyees at I n tui ti ve Surg ical In c. * * Ethical approval :** For t hi s ty p e of study for mal pati ent conse n t isn ot requ ire d. -- -abs tract : 'Mat e ria l sys tems w it h Dira c ele ct rons ona bipartite planar latt ice an d pos ses sing supe rco n duc ting andexci tonic inte rac tio ns ar e i n vesti gate d b oth in th e ha l f-filling an d d o p ed regimes at z e rotempe rat u re. Ex cito nic pairing is th e analog of chi rals y mme try brea ki ng of relativi sti cf e rmion th eo ries and pr oduces a ni nsula ting g ap inthe ele c t ro n ic spe ctru m.Condensed ma tt e r syste ms w i th suc h co mp etingintera c tion s display phenomen a tha t are a n alo gousto the on s et o f the chir al condensa te and ofcolor superc on ductiv ity i n dense qu a rk matter . Eva luation o f th e f ree-en ergy ( effec tive p ote ntial) al l o ws us t o ma p th e pha se s of the syst e m for di ffe r ent val ue s o f the co u pl i n gs of each in terac t i on. At hal f -fil l in g , weshow t hat Co oper pa i rsan d excit ons c an coexis t if thes upe rc ondu cting an dexci t

the_effect of_dropout can be found_in Appendix_D.4. Discussion_{#sec:discussion} ========== The results_show_that the system_has an acceptable_TRE of $3.3\%\pm1.4\%$. The_MVPM outperforms DeepLab_V3+_and Dense 3D CRF for semantic segmentation, with statistically significant improvement in less frequently_appearing_classes such_as_Human,_Mayo Stand, Sterile Table and_Anesthesia Cart. Ablation studies have_also shown_performance improvements as the number of cameras increase._However,_the results are_limited due to the multi-view dataset size and its_distribution. Future work includes obtaining more_accurate camera parameters_to_improve_registration accuracy and working_towards a larger and more balanced_multi-view dataset. Conclusion {#sec:conclusion} ========== In this paper, we_describe for the first time, a complete_and novel solution to create a_3D perception system to enhance_environmental awareness,_for surgical robots like the_da Vinci Xi_system. Our_system consists of_four ToF cameras rigidly mounted on_the PSC robot,_and a one-time calibration process for_multi-camera_to robot registration,_which_is_sufficient for_the sensor package_to_be used_in_other da Vinci Xi systems. Furthermore,_a_multi-view semantic segmentation fusion framework called MVPM_and new datasets for_algorithm_training/validation are proposed. Our_results show that the framework_can improve OR scene segmentation accuracy_over single_camera prediction. **Conflict_of interest:** Z. Li, A. Shaban and J.G. Simard are former_interns, D. Rabindran, S. DiMaio and_O. Mohareri are employees_at Intuitive_Surgical_Inc. **Ethical approval:** For_this_type of_study formal patient consent is not required. _--- abstract: 'Material_systems with Dirac electrons on a_bipartite planar lattice and_possessing_superconducting and excitonic interactions are investigated_both in the half-filling and doped_regimes at zero temperature. Excitonic_pairing_is_the analog of chiral symmetry_breaking of relativistic fermion theories and_produces an insulating_gap in the electronic spectrum. Condensed matter_systems_with such competing interactions display phenomena_that_are analogous to the onset of_the_chiral_condensate and of color superconductivity_in dense quark matter. Evaluation of_the free-energy (effective potential) allows us to map the_phases of the_system for different values of_the_couplings_of each interaction. At half-filling, we show that Cooper pairs_and excitons_can coexist if_the superconducting and excit

in $K$ will be done using iterative methods. In this case, the computational effort to computing (\[score\]) or (\[ascore\]) is roughly linear in the number of solves required (although see Section \[sec4\] for methods that make $N$ solves for a common matrix $K$ somewhat less than $N$ times the effort of one solve), so that (\[ascore\]) is much easier to compute than (\[score\]) when $N/n$ is small. An attractive feature of the approximation (\[ascore\]) is that if at any point one wants to obtain a better approximation to the score function, it suffices to consider additional $U_j$’s in (\[ascore\]). However, how exactly to do this if using the dependent sampling scheme for the $U_j$’s in Section \[sec4\] is not so obvious. Since this stochastic approach provides only an approximation to the MLE, one must compare it with other possible approximations to the MLE. Many such approaches exist, including spectral methods, low-rank approximations, covariance tapering and those based on some form of composite likelihood. All these methods involve computing the likelihood itself and not just its gradient, and thus all share this advantage over solving (\[ascore\]). Note that one can use randomized algorithms to approximate $\log\det K$ and thus approximate the loglikelihood directly \[@zhangY\]. However, this approximation requires first taking a power series expansion of $K$ and then applying the randomization trick to each term in the truncated power series; the examples presented by @zhangY show that the approach does not generally provide a good approximation to the loglikelihood. Since the accuracy of the power series approximation to $\log\det K$ depends on the condition number of $K$, some of the filtering ideas described by @steinchenanitescufiltering and used to good effect in Section \[sec4\] here could perhaps be of value for approximating $\log\det K$, but we do not explore that possibility. See @aune for some recent developments on stochastic approximation of log determinants of positive definite matrices. Let us consider the four approaches of spectral methods, low-rank approximations, covariance tapering and composite likelihood in turn. Spectral approximations to the likelihood can be fast and accurate for gridded data \[@whittle [@guyon; @dahlhaus]\], although even for gridded data they may require some prefiltering to work well \[@stein1995\]. In addition,

in $ K$ will be done using iterative methods. In this case, the computational campaign to calculate (\[score\ ]) or (\[ascore\ ]) is roughly linear in the numeral of solves ask (although see Section \[sec4\ ] for method acting that have $ N$ solves for a common matrix $ K$ slightly less than $ N$ times the effort of one solve), so that (\[ascore\ ]) is a lot easier to compute than (\[score\ ]) when $ N / n$ is small. An attractive feature of the estimate (\[ascore\ ]) is that if at any point one wants to receive a better approximation to the mark function, it suffices to consider extra $ U_j$ ’s in (\[ascore\ ]). However, how exactly to do this if using the pendent sampling dodge for the $ U_j$ ’s in Section \[sec4\ ] is not so obvious. Since this stochastic approach provides only an approximation to the MLE, one must compare it with early possible approximations to the MLE. Many such approaches exist, including spectral methods, broken - rank and file approximations, covariance tapering and those based on some form of composite likelihood. All these methods involve computing the likelihood itself and not barely its gradient, and therefore all partake this advantage over solving (\[ascore\ ]). Note that one can use randomize algorithm to approximate $ \log\det K$ and thus approximate the loglikelihood directly \[@zhangY\ ]. However, this approximation requires first taking a baron series expansion of $ K$ and then applying the randomization trick to each term in the truncated baron series; the examples presented by @zhangY show that the overture does not generally supply a good approximation to the loglikelihood. Since the accuracy of the power series approximation to $ \log\det K$ depend on the condition number of $ K$, some of the filtering idea described by @steinchenanitescufiltering and used to good effect in Section \[sec4\ ] here could possibly be of value for approximating $ \log\det K$, but we do not explore that possibility. See @aune for some recent development on stochastic approximation of logarithm determinants of positive definite matrices. Let us think the four approaches of spectral method acting, low - rank and file approximations, covariance sharpen and composite likelihood in turn. Spectral approximations to the likelihood can be fast and accurate for gridded data \[@whittle [ @guyon; @dahlhaus]\ ], although even for gridded datum they may require some prefiltering to work well \[@stein1995\ ]. In accession,

in $K$ will be done using ittrative methods. Iu this rase, ths computxtional effort to computing (\[dcire\]) oe (\[ascore\]) is roughly livear in tje numbee of wolves reqnjred (albkough dee Vxction \[sec4\] for kethods thdt make $N$ solvas flr a common matrix $K$ somewhat less ehan $N$ yiles the effort of pge sklve), so that (\[ascore\]) is much easier to comkute than (\[score\]) whrn $N/n$ is small. An attractige ffature of the apprlximation (\[awcorq\]) is that if xt any point one wants to obtain a better approximatiun to the score fynchhon, it suffmces tj consider additionan $U_j$’s im (\[ascore\]). Howevcr, hox exqctly to do this if uving the dependent sampling seheme for the $U_j$’s in Wextion \[sec4\] is vit ro kbtiohs. Sincf tiis stochasfic approacy provides only an sp[gpximation to the MJE, one must compare it with other possiblt appdoximations to the MLE. Nany such approaches gxist, incltding spectral methods, low-rank approximations, covdrianre takevlng xbd those based on some form of composite likelirkoc. Sll these methjds involve clmloting the likeuihood itaelf and not just lts graqient, and thus all share this advantage over wolving (\[ascoge\]). Nite that one can ude randomizzd algprithks to approximate $\log\dec K$ ans thus apprlximate tgd loglikelihood airvctlf \[@zhangY\]. However, this apprjximation reqbires fifst jaking w power segies cfpansion of $K$ and hhen cpplyhng the rajdomization trick to each term mi the truncatgd [owvr series; the cxamples preseneed by @zhangY xhow thct the approach soes nov generally [rovide a goog approximatimn to thq lotlikwlihood. Rince the accutacy of tkt power seeies approximation to $\mog\det K$ dependw ob the condition nuobew lf $K$, vome of the xiltdrivb idexs describeb cy @syeinchenanitescufiltarinf and used to good enfect in Wection \[fec4\] here coulc perhaps be of vapue fmr epproxomajing $\log\det K$, but we do not expmore that pofsibility. Seq @auke fjr some reeent developments on stochastic approximetion of log determinanrs of positive defiuiue matrices. Lev us cjnsider tve four approaches od spectral methodx, low-rank approximatikns, coeariajce tapering and composite likelihood in turn. Spectral approximations to rhe likelihood can be gast dnb aecurate for jridded data \[@whitnle [@guyon; @dahlhaus]\], although even for grigdzd data they may require some pfefiltering tu work well \[@stein1995\]. In addjtion,

in $K$ will be done using iterative this the computational to computing (\[score\]) in number of solves (although see Section for methods that make $N$ solves a common matrix $K$ somewhat less than $N$ times the effort of one so that (\[ascore\]) is much easier to compute than (\[score\]) when $N/n$ is An feature the (\[ascore\]) is that if at any point one wants to obtain a better approximation to the function, it suffices to consider additional $U_j$’s in However, how exactly to this if using the dependent scheme the $U_j$’s Section is so obvious. Since stochastic approach provides only an approximation to the MLE, one must compare it with other possible approximations the MLE. approaches exist, spectral low-rank covariance tapering and on some form of composite likelihood. involve computing the likelihood itself and not just gradient, and all share this advantage over solving Note that one can use randomized algorithms to $\log\det K$ and thus approximate the loglikelihood directly \[@zhangY\]. However, this approximation requires first taking series expansion of $K$ then applying the trick each in truncated power the examples presented by @zhangY show that the approach does not provide a good approximation to the loglikelihood. Since the accuracy power approximation to $\log\det depends on the condition of some of the filtering by and effect Section here could perhaps be value for approximating $\log\det K$, we do not explore some recent developments on stochastic approximation of log of positive definite matrices. Let us consider four approaches of spectral methods, low-rank approximations, covariance tapering and composite likelihood turn. Spectral the likelihood can be fast and accurate for data \[@whittle [@guyon; @dahlhaus]\], even for gridded data they may require some prefiltering work \[@stein1995\]. In

in $K$ will be done using iteratiVe methods. IN this CasE, thE cOmpuTatiOnal effort to coMPutiNg (\[score\]) or (\[ascore\]) is roughLy linEaR In thE NuMber oF solves REqUIRed (AlThOugH sEE SEctioN \[seC4\] for metHods that maKe $N$ SoLves for a commON mAtrix $K$ someWhaT less than $N$ tiMes The effOrT of ONe solVe), sO that (\[Ascore\]) IS much eAsier to coMpUTe than (\[SCore\]) wheN $n/N$ iS smaLl. An attractive feaTUrE Of the approximaTion (\[asCoRE\]) iS THat If aT any point oNe Wants TO obtain A BeTTER apPRoximation to tHe score funcTIon, It suffIcEs tO ConsidEr addItIOnaL $U_j$’s in (\[ascorE\]). HowEver, how exActly tO Do this iF Using thE depenDenT saMpliNG sChEme FoR The $u_J$’s In SECtiOn \[sec4\] is nOt So ObvioUs. SiNCE THis sTocHastIc appRoach provides OnlY an aPProXimatIon to The MlE, One muSt compAre it WiTh other possible ApprOximationS to ThE MLe. MAny suCH approAchEs eXist, incLuding sPEctRaL METhOds, low-rank approximAtIONs, CovarianCe tapeRInG aND those baSeD on Some FORm of cOmpoSItE likelihOod. All THeSe Methods InVolve cOmPutIng The liKElihOod itsElf and noT just ITs gradient, and tHUs all share thiS AdVANtAGe ovEr sOlving (\[ascorE\]). NotE That One cAN uSe rANdomiZed alGoRItHMs to approximate $\log\dEt k$ and thUs appRoximate the loGlikelihooD DIRectly \[@zhAngY\]. hOwEVer, this approxiMatioN requires fIRst takinG a powEr series Expansion OF $k$ and then AppLyiNg tHe rANDoMization trick TO Each TeRm in the TruNcated pOweR seRieS; thE eXamples prEsented bY @zHaNgy sHow That tHE approacH dOes NoT geNeralLY proviDe a goOd apPrOxIMatIon to thE LoGLIkelIhOoD. SinCe tHe AccurAcy oF The Power seRies approXimATion To $\LoG\det K$ dePends on the conDiTion number Of $k$, soMe of thE FIltering Ideas described by @steinchENanitesCufIlterIng aNd used to gOod Effect In SECtion \[sEc4\] here Could PeRhaPS Be of vALUe For ApProximatinG $\LOg\dEt K$, buT wE do nOt exploRe that possibility. SEE @auNe for some receNt dEvelOPMeNts ON sTOchAsTIc aPPRoximation of log DeterminanTs OF pOsitive defINitE mAtrices. let us coNsideR The four ApproacheS of spectrAl MethODS, loW-rank approXimationS, covarianCE tapeRInG and cOmpOsite lIkEliHood iN turn. SPEctRal apProximAtIons to The liKeLihood caN be fast and accurate for grIdded dAta \[@whIttLe [@guyon; @daHlhAUs]\], aLthough evEn foR gridded daTa tHey May reQuiRE some PrefILtEriNG to woRk weLL \[@stein1995\]. In aDDiTioN,

in $K$ will be done using iterative meth ods . I nthis cas e, the computa t iona l effort to computing(\[sc or e \])o r(\[as core\]) is r oug hl ylin ea r i n the nu mber of solves re qui re d (althoughs ee Section \ [se c4\] for met hod s that m ake $N$ s olv es fo r a co m mon ma trix $K$so m ewhatl ess tha n $N $ ti mes the effort of on e solve), so th at (\[ as c or e \ ])ismuch easie rto co m pute th a n( \ [ sco r e\]) when $N/ n$ is small . An attra ct ive featur e ofth e ap proximation (\[ ascore\]) is th a t if at any poi nt one wa nts too bt ai n a b e tte r a ppr o xim ation to t he scor e fu n c t i on,itsuff icesto consider a ddi tion a l $ U_j$’ s in(\[a sc ore\] ). How ever, h ow exactly to d o th is if usi ngth e d ep enden t sampl ing sc heme fo r the $ U _j$ ’s i n S ection \[sec4\] is n o t s o obviou s. Si n ce t h is stoch as tic app r o ach p rovi d es only an appro x im at ion toth e MLE, o nemus t com p areit wit h otherpossi b le approximati o ns to the MLE . M a n ys uchapp roaches exi st,i nclu ding sp ect r al me thods ,l ow - rank approximations ,covari ancetapering andthose base d o n some f ormo fc omposite likel ihood . All thes e methods invo lve comp uting the l ikelihoo d i tse lfand n ot just its gra d i ent, a nd thus al l share th isadv ant ag e over so lving (\ [a sc or e\ ]). Note that one c anus e r andom i zed al gorit hmsto a p pro ximate$ \l o g \det K $andthu sappro xima t e t he logl ikelihood di r ectl y\[ @zhangY \]. However,th is approxi ma tio n requ i r es first taking a power seriese xpansio n o f $K$ and then app lyi ng the ra n domiza tion t rickto ea c h term i nthe t runcated p o w erserie s; the exampl es presented by @z h ang Y show that t heappr o a ch do e sn otge n era l l y provide a goo d approxim at i on to the lo g lik el ihood.Since t he ac c uracy o f the pow er series a ppro x i mat ion to $\l og\det K $ depends on th e c ondit ion numbe rof$K$,some o f th e fil tering i deas d escri be d by @st einchenanitescufilterin g andusedtogood effe cti n S ection \[ sec4 \] here co uld pe rhaps be of va luef or ap p roxim atin g $\log\de t K $,b u twe do not e x p l ore that po s sibili ty.See @aune for som e recent develo pmen t s on st o chas ti c approximatio n o fl o g determ in ants of pos itive de fi n ite m atrice s. Le t us co n s id e r thefour ap proachesofsp e ctral m et ho d s, low -ran kapprox imatio n s, c o v ariance tapering andc o mposi t e l ikeli ho od in t u rn.Spectral a pproximatio ns tothelikel ihood c an be fa stan d accurate for gridd ed da ta \[@w hi ttle [@ guyon; @da h l haus] \],al tho ugh evenf o rg ri dd e d d atatheyma y re quire som e prefilt eri n g to wo rk we l l \[@st e in 1 9 95\]. In a ddi tion,

in_$K$ will_be done using iterative_methods. In_this_case, the_computational_effort to computing_(\[score\]) or (\[ascore\])_is roughly linear in_the number of_solves_required (although see Section \[sec4\] for methods that make $N$ solves for a common_matrix_$K$ somewhat_less_than_$N$ times the effort of_one solve), so that (\[ascore\])_is much_easier to compute than (\[score\]) when $N/n$ is_small._An attractive feature_of the approximation (\[ascore\]) is that if at any_point one wants to obtain a_better approximation to_the_score_function, it suffices to_consider additional $U_j$’s in (\[ascore\]). However,_how exactly to do this if_using the dependent sampling scheme for the_$U_j$’s in Section \[sec4\] is not_so obvious. Since this stochastic approach_provides only_an approximation to the MLE,_one must compare_it with_other possible approximations_to the MLE. Many such approaches_exist, including spectral_methods, low-rank approximations, covariance tapering and_those_based on some_form_of_composite likelihood._All these methods_involve_computing the_likelihood_itself and not just its gradient,_and_thus all share this advantage over solving_(\[ascore\]). Note that one_can_use randomized algorithms to_approximate $\log\det K$ and thus_approximate the loglikelihood directly \[@zhangY\]. However,_this approximation_requires first_taking a power series expansion of $K$ and then applying the_randomization trick to each term in_the truncated power series;_the examples_presented_by @zhangY show_that_the approach_does not generally provide a good approximation_to the_loglikelihood. Since the accuracy of the_power series approximation to_$\log\det_K$ depends on the condition number_of $K$, some of the filtering_ideas described by @steinchenanitescufiltering and_used_to_good effect in Section \[sec4\]_here could perhaps be of value_for approximating $\log\det_K$, but we do not explore that_possibility._See @aune for some recent developments_on_stochastic approximation of log determinants of_positive_definite_matrices. Let us consider the four_approaches of spectral methods, low-rank approximations,_covariance tapering and composite likelihood in turn. Spectral approximations_to the likelihood_can be fast and accurate_for_gridded_data \[@whittle [@guyon; @dahlhaus]\], although even for gridded data they_may require_some prefiltering to_work well \[@stein1995\]. In addition,

dR_{k+1}}F_{_{k+1}}^T=F_{2_{k}}^T {\ensuremath{\left( F_{_{k}}J_{dR_{k}}-J_{dR_{k}}F_{_{k}}^T \right)}}F_{2_{k}}-h^2S(M_{k+1}),\label{eqn:findFl}\\F_{2_{k+1}}J_{d_2}-J_{d_2}F_{2_{k+1}}^T =F_{2_{k}}^T{\ensuremath{\left( F_{2_{k}}J_{d_2}-J_{d_2}F_{2_{k}}^T \right)}}F_{2_{k}} +h^2X_{_{k+1}}\times{\ensuremath{\frac{\partial U}{\partial X_{_{k+1}}}}}+h^2S(M_{k+1}) ,\label{eqn:findF2l}\\ R_{_{k+1}}=F_{2_{k}}^TF_{_{k}}R_{_{k}},\label{eqn:updateRl}\\ R_{2_{k+1}}=R_{2_{k}}F_{2_{k}}.\label{eqn:updateR2l}\end{gathered}$$ It is natural to express equations of motion for the second body in the inertial frame. $$\begin{gathered} x_{2_{k+1}}-2x_{2_{k}}+x_{2_{k-1}}=\frac{h^2}{m_2}R_{_k}{\ensuremath{\frac{\partial U_{_k}}{\partial X_{_{k}}}}}.\label{eqn:updatex2l}\end{gathered}$$ Given $(X_{_{0}},R_{_0},R_{2_{0}},X_{_{1}},R_{_1},R_{2_{1}})$, we can determine $F_{_0}$ and $F_{2_0}$ from [(\[eqn:updateRl\])]{} and [(\[eqn:updateR2l\])]{}. Solving the implicit equations [(\[eqn:findFl\])]{} and [(\[eqn:findF2l\])]{} gives $F_{_1}$ and $F_{2_1}$. Then $X_{_2}$, $R_{_2}$ and $R_{2_2}$ are found from [(\[eqn:updateXl\])]{}, [(\[eqn:updateRl

dR_{k+1}}F_{_{k+1}}^T = F_{2_{k}}^T { \ensuremath{\left (F_{_{k}}J_{dR_{k}}-J_{dR_{k}}F_{_{k}}^T \right)}}F_{2_{k}}-h^2S(M_{k+1}),\label{eqn: findFl}\\F_{2_{k+1}}J_{d_2}-J_{d_2}F_{2_{k+1}}^T = F_{2_{k}}^T{\ensuremath{\left (F_{2_{k}}J_{d_2}-J_{d_2}F_{2_{k}}^T \right)}}F_{2_{k } } + h^2X_{_{k+1}}\times{\ensuremath{\frac{\partial U}{\partial X_{_{k+1}}}}}+h^2S(M_{k+1 }) , \label{eqn: findF2l}\\ R_{_{k+1}}=F_{2_{k}}^TF_{_{k}}R_{_{k}},\label{eqn: updateRl}\\ R_{2_{k+1}}=R_{2_{k}}F_{2_{k}}.\label{eqn: updateR2l}\end{gathered}$$ It is natural to express equations of motion for the second soundbox in the inertial frame of reference. $ $ \begin{gathered } x_{2_{k+1}}-2x_{2_{k}}+x_{2_{k-1}}=\frac{h^2}{m_2}R_{_k}{\ensuremath{\frac{\partial U_{_k}}{\partial X_{_{k}}}}}.\label{eqn: updatex2l}\end{gathered}$$ Given $ (X_{_{0}},R_{_0},R_{2_{0}},X_{_{1}},R_{_1},R_{2_{1}})$, we can determine $ F_{_0}$ and $ F_{2_0}$ from [ (\[eqn: updateRl\ ]) ] { } and [ (\[eqn: updateR2l\ ]) ] { }. Solving the implicit equality [ (\[eqn: findFl\ ]) ] { } and [ (\[eqn: findF2l\ ]) ] { } gives $ F_{_1}$ and $ F_{2_1}$. Then $ X_{_2}$, $ R_{_2}$ and $ R_{2_2}$ are found from [ (\[eqn: updateXl\ ]) ] { }, [ (\[eqn: updateRl

dR_{k+1}}V_{_{k+1}}^T=F_{2_{k}}^T {\ensuremath{\left( F_{_{k}}J_{aR_{k}}-J_{dR_{k}}F_{_{k}}^T \righj)}}F_{2_{j}}-h^2S(M_{k+1}),\lebel{eqn:rindFl}\\F_{2_{k+1}}G_{d_2}-J_{d_2}F_{2_{k+1}}^T =F_{2_{k}}^T{\ensuremath{\left( F_{2_{k}}J_{v_2}-J_{d_2}F_{2_{j}}^T \ritht)}}F_{2_{k}} +h^2X_{_{k+1}}\times{\ensurematf{\frac{\partpal U}{\partual E_{_{k+1}}}}}+h^2S(M_{k+1}) ,\label{eqn:fiisF2l}\\ R_{_{k+1}}=F_{2_{k}}^BY_{_{k}}R_{_{k}},\lznel{equ:u'dateRl}\\ R_{2_{k+1}}=R_{2_{k}}F_{2_{k}}.\lanel{eqn:updataR2l}\end{gathered}$$ Ig ns natural to express equations of mjtion fpr the second boqy im the pntrtial frame. $$\begin{gathered} x_{2_{k+1}}-2x_{2_{k}}+x_{2_{k-1}}=\fdac{h^2}{m_2}R_{_k}{\tnsuremath{\frac{\partoal U_{_k}}{\partial X_{_{k}}}}}.\label{eqn:upfated2l}\end{gathered}$$ Givej $(X_{_{0}},R_{_0},R_{2_{0}},X_{_{1}},R_{_1},R_{2_{1}})$, we can eetermine $F_{_0}$ xnd $F_{2_0}$ from [(\[eqn:updateRl\])]{} and [(\[eqn:updateR2l\])]{}. Solving the imolicic equations [(\[ewn:fltdFl\])]{} and [(\[eqi:findF2j\])]{} gives $F_{_1}$ and $F_{2_1}$. Then $X_{_2}$, $R_{_2}$ anc $R_{2_2}$ are found nrom [(\[xqn:updateXl\])]{}, [(\[eqn:updateRl

dR_{k+1}}F_{_{k+1}}^T=F_{2_{k}}^T {\ensuremath{\left( F_{_{k}}J_{dR_{k}}-J_{dR_{k}}F_{_{k}}^T \right)}}F_{2_{k}}-h^2S(M_{k+1}),\label{eqn:findFl}\\F_{2_{k+1}}J_{d_2}-J_{d_2}F_{2_{k+1}}^T =F_{2_{k}}^T{\ensuremath{\left( F_{2_{k}}J_{d_2}-J_{d_2}F_{2_{k}}^T \right)}}F_{2_{k}} X_{_{k+1}}}}}+h^2S(M_{k+1}) R_{_{k+1}}=F_{2_{k}}^TF_{_{k}}R_{_{k}},\label{eqn:updateRl}\\ R_{2_{k+1}}=R_{2_{k}}F_{2_{k}}.\label{eqn:updateR2l}\end{gathered}$$ is natural to the body in the frame. $$\begin{gathered} x_{2_{k+1}}-2x_{2_{k}}+x_{2_{k-1}}=\frac{h^2}{m_2}R_{_k}{\ensuremath{\frac{\partial X_{_{k}}}}}.\label{eqn:updatex2l}\end{gathered}$$ Given $(X_{_{0}},R_{_0},R_{2_{0}},X_{_{1}},R_{_1},R_{2_{1}})$, we can determine and $F_{2_0}$ from [(\[eqn:updateRl\])]{} and [(\[eqn:updateR2l\])]{}. Solving the implicit equations [(\[eqn:findFl\])]{} and [(\[eqn:findF2l\])]{} $F_{_1}$ and $F_{2_1}$. Then $X_{_2}$, $R_{_2}$ and $R_{2_2}$ are found from [(\[eqn:updateXl\])]{}, [(\[eqn:updateRl

dR_{k+1}}F_{_{k+1}}^T=F_{2_{k}}^T {\ensuremath{\left( F_{_{k}}j_{dR_{k}}-J_{dR_{k}}F_{_{k}}^t \righT)}}F_{2_{k}}-H^2S(M_{K+1}),\lAbel{Eqn:fIndFl}\\F_{2_{k+1}}J_{d_2}-J_{d_2}F_{2_{k+1}}^T =f_{2_{K}}^T{\enSuremath{\left( F_{2_{k}}J_{d_2}-J_{d_2}F_{2_{k}}^T \rIght)}}F_{2_{K}} +h^2x_{_{K+1}}\timES{\eNsureMath{\fraC{\PaRTIal u}{\pArTiaL X_{_{K+1}}}}}+H^2S(m_{k+1}) ,\labEl{eQn:findF2L}\\ R_{_{k+1}}=F_{2_{k}}^TF_{_{k}}R_{_{k}},\LabEl{Eqn:updateRl}\\ R_{2_{K+1}}=r_{2_{k}}f_{2_{k}}.\label{eqn:UpdAteR2l}\end{gathEreD}$$ It is nAtUraL To expResS equaTions oF Motion For the secOnD Body in THe inertIAL fRame. $$\Begin{gathered} x_{2_{k+1}}-2x_{2_{k}}+X_{2_{K-1}}=\fRAc{h^2}{m_2}R_{_k}{\ensuremaTh{\frac{\PaRTiAL u_{_k}}{\pArtIal X_{_{k}}}}}.\label{EqN:updaTEx2l}\end{gAThERED}$$ GiVEn $(X_{_{0}},R_{_0},R_{2_{0}},X_{_{1}},R_{_1},R_{2_{1}})$, we can Determine $F_{_0}$ aND $F_{2_0}$ fRom [(\[eqn:UpDatErl\])]{} and [(\[eQn:updAtEr2l\])]{}. SOlving the imPlicIt equatioNs [(\[eqn:fINdFl\])]{} and [(\[EQn:findF2L\])]{} gives $f_{_1}$ anD $F_{2_1}$. THen $X_{_2}$, $r_{_2}$ AnD $R_{2_2}$ Are FoUNd fROm [(\[Eqn:UPdaTeXl\])]{}, [(\[eqn:uPdAtERl

dR_{k+1}}F_{_{k+1}}^T=F_{2 _{k}}^T {\ ensur ema th{ \l eft( F_{ _{k}}J_{dR_{k} } -J_{ dR_{k}}F_{_{k}}^T \rig ht)}} F_ { 2_{k } }- h^2S( M_{k+1} ) ,\ l a bel {e qn :fi nd F l} \\F_{ 2_{ k+1}}J_ {d_2}-J_{d _2} F_ {2_{k+1}}^T= F_ {2_{k}}^T{ \en suremath{\le ft( F_{2_ {k }}J _ {d_2} -J_ {d_2} F_{2_{ k }}^T \ right)}}F _{ 2 _{k}}+ h^2X_{_ { k +1 }}\t imes{\ensuremath{ \ fr a c{\partial U}{ \parti al X_ { _ {k+ 1}} }}}+h^2S(M _{ k+1}) ,\label { eq n : f ind F 2l}\\ R_{_{k+ 1}}=F_{2_{k } }^T F_{_{k }} R_{ _ {k}},\ label {e q n:u pdateRl}\\R_{2 _{k+1}}=R _{2_{k } }F_{2_{ k }}.\lab el{eqn :up dat eR2l } \e nd {ga th e red } $$ It isnaturalto e xpres s eq u a t i onsofmoti on fo r the secondbod y in the iner tialfram e. $$\b egin{g ather ed } x_{2_{k+1}}-2 x_{2 _{k}}+x_{ 2_{ k- 1}} =\ frac{ h ^2}{m_ 2}R _{_ k}{\ens uremath { \fr ac { \ p ar tial U_{_k}}{\part ia l X_ {_{k}}}} }.\lab e l{ eq n :updatex 2l }\e nd{g a t hered }$$G iv en $(X_{ _{0}}, R _{ _0 },R_{2_ {0 }},X_{ _{ 1}} ,R_ {_1}, R _{2_ {1}})$ , we can dete r mine $F_{_0}$a nd $F_{2_0}$f ro m [( \ [eqn :up dateRl\])]{ } an d [(\ [eqn : up dat e R2l\] )]{}. S o lv i ng the implicit equ at ions [ (\[eq n:findFl\])]{ } and [(\[ e q n :findF2l \])] { }g ives $F_{_1}$and $ F_{2_1}$.T hen $X_{ _2}$, $R_{_2} $ and $R_ { 2 _2}$ are fo und fr om[ ( \[ eqn:updateXl\ ] ) ]{}, [ (\[eqn: upd ateRl

dR_{k+1}}F_{_{k+1}}^T=F_{2_{k}}^T {\ensuremath{\left( F_{_{k}}J_{dR_{k}}-J_{dR_{k}}F_{_{k}}^T_\right)}}F_{2_{k}}-h^2S(M_{k+1}),\label{eqn:findFl}\\F_{2_{k+1}}J_{d_2}-J_{d_2}F_{2_{k+1}}^T =F_{2_{k}}^T{\ensuremath{\left( F_{2_{k}}J_{d_2}-J_{d_2}F_{2_{k}}^T_\right)}}F_{2_{k}} +h^2X_{_{k+1}}\times{\ensuremath{\frac{\partial U}{\partial X_{_{k+1}}}}}+h^2S(M_{k+1}) ,\label{eqn:findF2l}\\ R_{_{k+1}}=F_{2_{k}}^TF_{_{k}}R_{_{k}},\label{eqn:updateRl}\\ R_{2_{k+1}}=R_{2_{k}}F_{2_{k}}.\label{eqn:updateR2l}\end{gathered}$$ It_is natural_to_express equations_of_motion for the_second body in_the inertial frame. $$\begin{gathered} x_{2_{k+1}}-2x_{2_{k}}+x_{2_{k-1}}=\frac{h^2}{m_2}R_{_k}{\ensuremath{\frac{\partial_U_{_k}}{\partial X_{_{k}}}}}.\label{eqn:updatex2l}\end{gathered}$$ Given_$(X_{_{0}},R_{_0},R_{2_{0}},X_{_{1}},R_{_1},R_{2_{1}})$,_we can determine $F_{_0}$ and $F_{2_0}$ from [(\[eqn:updateRl\])]{} and [(\[eqn:updateR2l\])]{}. Solving the implicit equations_[(\[eqn:findFl\])]{}_and [(\[eqn:findF2l\])]{}_gives_$F_{_1}$_and $F_{2_1}$. Then $X_{_2}$, $R_{_2}$_and $R_{2_2}$ are found from_[(\[eqn:updateXl\])]{}, [(\[eqn:updateRl

}}{ \varepsilon}_k = 0$ such that ${ \lim _{k \to +\infty }}W_1 ({ f ^{\varepsilon _k}}(t), f(t)) = 0$ uniformly for $t \in [0,T], T>0$ if ${ \mathrm{supp\;}}{ f ^{\mathrm{in}}}\subset \{ (x,v) \;:\;|x|\leq L_0, |v| = r\}$ and ${ \lim _{k \to +\infty }}W_1 ({ f ^{\varepsilon _k}}(t), f(t)) = 0$ uniformly for $t \in [\delta,T], T>\delta >0$ if ${ \mathrm{supp\;}}{ f ^{\mathrm{in}}}\subset \{ (x,v) \;:\;|x|\leq L_0, r_0 \leq |v| \leq R_0\}$ for some function $f \in C({{{\bf}R}}_+;{ { {\cal P}}_1}({{{\bf}R}}^d \times {{{\bf}R}}^d))$. Thanks to Proposition \[Kernel\] we deduce (for both prepared or not initial data) that $${ \mathrm{supp\;}}f(t) \subset \{ (x,v)\in {{{\bf}R}}^d \times {{{\bf}R}}^d \;:\;|v| = r\},\;\;t>0.$$ The previous statement holds also true at $t = 0$, by the continuity of $f$. The time evolution for the limit $f$ comes by using the particular test functions $$\theta (t,x,v) = \eta (t) \left [ 1 - \chi \left ( \frac{2|v|}{r_0}\right ) \right ] \varphi \left ( x, r{ \frac{v}{|v|}}\right )$$ with $\eta \in C^1_c ({{{\bf}R}}_+)$, $\varphi \in { C^1_c ({{{\bf}R}}^d \times {{{\bf}R}}^d)}{}$. From now on we consider only the not prepared initial data case (the other case is simpler). We recall the notation ${ a ^\varepsilon}= - \nabla _x U \star \int _{{{{\bf}R}}^d} { f ^\varepsilon}\;{ \mathrm{d}}v

} } { \varepsilon}_k = 0 $ such that $ { \lim _ { k \to + \infty } } W_1 ({ f ^{\varepsilon _ k}}(t), f(t) ) = 0 $ uniformly for $ t \in [ 0,T ], T>0 $ if $ { \mathrm{supp\; } } { f ^{\mathrm{in}}}\subset \ { (x, v) \;:\;|x|\leq L_0, |v| = r\}$ and $ { \lim _ { k \to + \infty } } W_1 ({ f ^{\varepsilon _ k}}(t), f(t) ) = 0 $ uniformly for $ t \in [ \delta, T ], T>\delta > 0 $ if $ { \mathrm{supp\; } } { f ^{\mathrm{in}}}\subset \ { (x, v) \;:\;|x|\leq L_0, r_0 \leq |v| \leq R_0\}$ for some function $ f \in C({{{\bf}R}}_+; { { { \cal P}}_1}({{{\bf}R}}^d \times { { { \bf}R}}^d))$. Thanks to Proposition \[Kernel\ ] we deduce (for both prepared or not initial datum) that $ $ { \mathrm{supp\;}}f(t) \subset \ { (x, v)\in { { { \bf}R}}^d \times { { { \bf}R}}^d \;:\;|v| = r\},\;\;t>0.$$ The former argument holds also dependable at $ t = 0 $, by the continuity of $ f$. The time development for the limit $ f$ comes by using the especial test functions $ $ \theta (t, x, v) = \eta (t) \left [ 1 - \chi \left ( \frac{2|v|}{r_0}\right) \right ] \varphi \left (x, radius { \frac{v}{|v|}}\right ) $ $ with $ \eta \in C^1_c ({ { { \bf}R}}_+)$, $ \varphi \in { C^1_c ({ { { \bf}R}}^d \times { { { \bf}R}}^d)}{}$. From now on we regard only the not disposed initial data shell (the early case is simpler). We recall the note $ { a ^\varepsilon}= - \nabla _ x U \star \int _ { { { { \bf}R}}^d } { f ^\varepsilon}\; { \mathrm{d}}v

}}{ \varfpsilon}_k = 0$ such that ${ \lim _{k \to +\infty }}W_1 ({ f ^{\varepsmlon _k}}(t), f(t)) = 0$ unkformly for $t \in [0,T], T>0$ if ${ \mathcm{supp\;}}{ f ^{\mqthrm{in}}}\subset \{ (x,v) \;:\;|x|\leq L_0, |v| = r\}$ ajd ${ \lim _{k \to +\mnfty }}W_1 ({ f ^{\varepsmmon _k}}(t), n(c)) = 0$ uhlformnb for $t \in [\delts,T], T>\delta >0$ if ${ \mathrm{supp\;}}{ x ^{\oachrm{in}}}\subset \{ (x,v) \;:\;|x|\leq L_0, r_0 \leq |v| \leq R_0\}$ for soke function $f \in C({{{\bf}G}}_+;{ { {\cwl P}}_1}({{{\gf}R}}^d \times {{{\bf}R}}^d))$. Thanks to Propositjon \[Keriel\] we deduce (fpr both prepared or not inltiap data) that $${ \mathrm{dupp\;}}f(t) \subwet \{ (z,v)\in {{{\bf}R}}^d \tioes {{{\bf}R}}^d \;:\;|v| = r\},\;\;t>0.$$ The prevjous statement holds also true xt $t = 0$, by the cinrinkhty of $f$. Thx time evolution for the lhmit $f$ vomes by using thx paeticular test functiois $$\theta (t,x,v) = \eta (t) \jeft [ 1 - \cvi \left ( \frac{2|v|}{r_0}\right ) \righj ] \vasphi \oefg ( s, c{ \frzc{v}{|v|}}\rihht )$$ with $\eta \jn C^1_c ({{{\bf}R}}_+)$, $\vqrphi \in { C^1_c ({{{\bf}R}}^d \tikef {{{\bf}R}}^d)}{}$. From nos on wq sonsider only the not prepared initial gatz case (the other case iw simpler). We recall tje notatijn ${ a ^\varepsilon}= - \nabla _x U \star \int _{{{{\bf}R}}^d} { f ^\varepsinon}\;{ \mavhfm{d}}r

}}{ \varepsilon}_k = 0$ such that ${ \to }}W_1 ({ ^{\varepsilon _k}}(t), f(t)) \in T>0$ if ${ f ^{\mathrm{in}}}\subset \{ \;:\;|x|\leq L_0, |v| = r\}$ and \lim _{k \to +\infty }}W_1 ({ f ^{\varepsilon _k}}(t), f(t)) = 0$ uniformly $t \in [\delta,T], T>\delta >0$ if ${ \mathrm{supp\;}}{ f ^{\mathrm{in}}}\subset \{ (x,v) \;:\;|x|\leq r_0 |v| R_0\}$ some function $f \in C({{{\bf}R}}_+;{ { {\cal P}}_1}({{{\bf}R}}^d \times {{{\bf}R}}^d))$. Thanks to Proposition \[Kernel\] we deduce both prepared or not initial data) that $${ \subset \{ (x,v)\in {{{\bf}R}}^d {{{\bf}R}}^d \;:\;|v| = r\},\;\;t>0.$$ The statement also true $t 0$, the continuity of The time evolution for the limit $f$ comes by using the particular test functions $$\theta (t,x,v) = (t) \left - \chi ( ) ] \varphi \left r{ \frac{v}{|v|}}\right )$$ with $\eta \in \in { C^1_c ({{{\bf}R}}^d \times {{{\bf}R}}^d)}{}$. From now we consider the not prepared initial data case other case is simpler). We recall the notation a ^\varepsilon}= - \nabla _x U \star \int _{{{{\bf}R}}^d} { f ^\varepsilon}\;{ \mathrm{d}}v

}}{ \varepsilon}_k = 0$ such that ${ \lim _{k \to +\Infty }}W_1 ({ f ^{\varEpsilOn _k}}(T), f(t)) = 0$ UnIforMly fOr $t \in [0,T], T>0$ if ${ \mathrM{Supp\;}}{ F ^{\mathrm{in}}}\subset \{ (x,v) \;:\;|x|\leq L_0, |V| = r\}$ and ${ \LiM _{K \to +\iNFtY }}W_1 ({ f ^{\vaRepsiloN _K}}(t), F(T)) = 0$ UniFoRmLy fOr $T \In [\Delta,t], T>\dElta >0$ if ${ \mAthrm{supp\;}}{ f ^{\MatHrM{in}}}\subset \{ (x,v) \;:\;|x|\LEq l_0, r_0 \leq |v| \leq R_0\}$ For Some function $F \in c({{{\bf}R}}_+;{ { {\caL P}}_1}({{{\Bf}R}}^D \Times {{{\Bf}R}}^D))$. ThanKs to PrOPositiOn \[Kernel\] wE dEDuce (foR Both prePAReD or nOt initial data) that $${ \MAtHRm{supp\;}}f(t) \subset \{ (X,v)\in {{{\bf}r}}^d \TImES {{{\Bf}R}}^D \;:\;|v| = r\},\;\;T>0.$$ The previoUs StateMEnt holdS AlSO TRue AT $t = 0$, by the continUity of $f$. The tIMe eVolutiOn For THe limiT $f$ comEs BY usIng the partiCulaR test funcTions $$\tHEta (t,x,v) = \eTA (t) \left [ 1 - \cHi \left ( \FraC{2|v|}{r_0}\RighT ) \RiGhT ] \vaRpHI \leFT ( x, R{ \frAC{v}{|v|}}\Right )$$ witH $\eTa \In C^1_c ({{{\bF}R}}_+)$, $\vaRPHI \In { C^1_c ({{{\Bf}R}}^D \timEs {{{\bf}R}}^D)}{}$. From now on we cOnsIder ONly The noT prepAred InItial Data caSe (the OtHer case is simpleR). We rEcall the nOtaTiOn ${ a ^\VaRepsiLOn}= - \nablA _x U \StaR \int _{{{{\bf}R}}^D} { f ^\varepSIloN}\;{ \mATHRm{D}}v

}}{ \varepsilon}_k = 0$ su ch that ${ \lim _{ k \ to +\i nfty }}W_1 ({ f ^{ \ vare psilon _k}}(t), f(t))= 0$un i form l yfor $ t \in [ 0 ,T ] , T> 0$ i f $ {\ ma thrm{ sup p\;}}{f ^{\mathr m{i n} }}\subset \{ (x ,v) \;:\;| x|\ leq L_0, |v| =r\}$ a nd ${ \lim_{k \to+\inft y }}W_1 ({ f ^{\ va r epsilo n _k}}(t ) , f (t)) = 0$ uniformly f o r$ t \in [\delta, T], T> \d e lt a >0$ if ${ \mathr m{ supp\ ; }}{ f ^ { \m a t h rm{ i n}}}\subset \ { (x,v) \;: \ ;|x |\leqL_ 0,r _0 \le q |v| \ l eqR_0\}$ forsome function $f \i n C({{{\ b f}R}}_+ ;{ { { \ca l P }}_1 } ({ {{ \bf }R } }^d \t ime s {{ {\bf}R}} ^d )) $. Th anks t o Prop osi tion \[Ke rnel\] we ded uce (fo r bo th pr epare d or n ot in itialdata) t hat $${ \mathrm {sup p\;}}f(t) \ su bse t\{ (x , v)\in{{{ \bf }R}}^d\times{ {{\ bf } R } }^ d \;:\;|v| = r\},\ ;\ ; t >0 .$$ Theprevio u sst a tement h ol dsalso t rue a t $t =0$, by t he con t in ui ty of $ f$ . Theti meevo lutio n for the l imit $f$ come s by using thep articular tes t f u n ct i ons$$\ theta (t,x, v) = \eta (t) \l eft [ 1 - \chi \ l ef t ( \frac{2|v|}{r_0} \r ight ) \ri ght ] \varphi \left ( x , r { \frac{ v}{| v |} } \right )$$ wit h $\e ta \in C^1 _ c ({{{\b f}R}} _+)$, $\ varphi \i n { C^1_c({{ {\b f}R }}^ d \t imes {{{\bf}R } } ^d)} {} $. From no w on we co nsi der on ly the notprepared i ni ti al da ta ca s e (the o th erca seis si m pler). We r ecal lth e no tation$ {a ^\va re ps ilon }=-\nabl a _x U \ star \i nt _{{{{\ bf} R }}^d }{f ^\var epsilon}\;{ \ ma thrm{d}}v

}}{ \varepsilon}_k =_0$ such_that ${ \lim _{k \to_+\infty }}W_1_({ f_^{\varepsilon _k}}(t),_f(t))_= 0$ uniformly_for $t \in_[0,T], T>0$ if ${ \mathrm{supp\;}}{ f_^{\mathrm{in}}}\subset \{ (x,v)_\;:\;|x|\leq_L_0, |v| = r\}$ and ${ \lim _{k \to +\infty }}W_1 ({ f ^{\varepsilon _k}}(t), f(t))_=_0$ uniformly_for_$t_\in [\delta,T], T>\delta >0$ if_${ \mathrm{supp\;}}{ f ^{\mathrm{in}}}\subset \{ (x,v) \;:\;|x|\leq_L_0, r_0_\leq |v| \leq R_0\}$ for some function $f \in_C({{{\bf}R}}_+;{ { {\cal_P}}_1}({{{\bf}R}}^d \times {{{\bf}R}}^d))$._Thanks to Proposition \[Kernel\] we deduce (for both prepared_or not initial data) that $${ \mathrm{supp\;}}f(t)_ \subset \{_(x,v)\in_{{{\bf}R}}^d_\times {{{\bf}R}}^d \;:\;|v| = r\},\;\;t>0.$$_The previous statement holds also true_at $t = 0$, by the_continuity of $f$. The time evolution for_the limit $f$ comes by using_the particular test functions $$\theta_(t,x,v) =_\eta (t) \left [ 1_- \chi \left_( \frac{2|v|}{r_0}\right )_ \right ]_\varphi \left ( x, r{ \frac{v}{|v|}}\right )$$ with_$\eta \in C^1_c_({{{\bf}R}}_+)$, $\varphi \in { C^1_c ({{{\bf}R}}^d \times_{{{\bf}R}}^d)}{}$._From now on_we_consider_only the_not prepared initial_data_case (the_other_case is simpler). We recall the_notation_${ a ^\varepsilon}= - \nabla _x U \star \int__{{{{\bf}R}}^d} { f ^\varepsilon}\;{ \mathrm{d}}v

m)=2^a-1=l\#m$ whenever $2^{a-1}{\leqslant}l,m<2^a$. Let $r,s,t\in\N$. From Proposition \[backdiag\] follows that $\tau(r,s){\geqslant}t$ if, and only if, there exist $\rho{\leqslant}r$ and $\s{\leqslant}s$ with $\rho+\s=t$ such that $Q_t^{(\rho,\s)}\equiv1$. Now suppose that $r,s{\leqslant}2^a$, and consider $\rho{\leqslant}r$ and $\s{\leqslant}s$. If $\rho+\s<2^a$ then $$Q_{\rho+\s}^{(\rho,\s)}\equiv\binom{\rho+\s}{\rho}= \binom{2^a}{2^a}\binom{\rho+\s}{\rho}\equiv \binom{\rho+\s+2^a}{\rho+2^a}=Q_{\rho+2^a}^{(\rho+2^a,\s)}\,.$$ If instead $\rho+\s{\geqslant}2^a$ then $Q_{\rho+\s}^{(\rho,\s)} \equiv 0 \equiv Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$ by Lemma \[specials\]:\[2s\]. In each case $Q_{\rho+\s}^{(\rho,\s)}\equiv Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$, which proves that $\tau(r+2^a,s)=\tau(r,s) + 2^a$ for all $r,s<2^a$. Taking $l=r+2^a$ and $m=s$ we get $\tau(l,m)=2^a + \tau(l-2^a,m)$ for all $l,m\in\N$ such that $m<2^a{\leqslant}l<2^{a+1}$. On the other hand, from the definition follows that $l\#m=2^a+(l-2^a)\#m$ in this case; now $\tau(l,m)=l\#m$ follows by induction. We record the following facts about the operation $\#$. The third statement is an immediate consequence of Proposition \[hashprod\], the others follow from the definition. Remember

m)=2^a-1 = l\#m$ whenever $ 2^{a-1}{\leqslant}l, m<2^a$. Let $ r, s, t\in\N$. From Proposition   \[backdiag\ ] follows that $ \tau(r, s){\geqslant}t$ if, and only if, there exist $ \rho{\leqslant}r$ and $ \s{\leqslant}s$ with $ \rho+\s = t$ such that $ Q_t^{(\rho,\s)}\equiv1$. nowadays presuppose that $ r, s{\leqslant}2^a$, and consider $ \rho{\leqslant}r$ and $ \s{\leqslant}s$. If $ \rho+\s<2^a$ then $ $ Q_{\rho+\s}^{(\rho,\s)}\equiv\binom{\rho+\s}{\rho}= \binom{2^a}{2^a}\binom{\rho+\s}{\rho}\equiv \binom{\rho+\s+2^a}{\rho+2^a}=Q_{\rho+2^a}^{(\rho+2^a,\s)}\,.$$ If instead $ \rho+\s{\geqslant}2^a$ then $ Q_{\rho+\s}^{(\rho,\s) } \equiv 0 \equiv Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$ by Lemma   \[specials\]:\[2s\ ]. In each lawsuit $ Q_{\rho+\s}^{(\rho,\s)}\equiv Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$, which proves that $ \tau(r+2^a, s)=\tau(r, s) + 2^a$ for all $ r, s<2^a$. take $ l = r+2^a$ and $ m = s$ we scram $ \tau(l, m)=2^a + \tau(l-2^a, m)$ for all $ l, m\in\N$ such that $ m<2^a{\leqslant}l<2^{a+1}$. On the other bridge player, from the definition follows that $ l\#m=2^a+(l-2^a)\#m$ in this case; nowadays $ \tau(l, m)=l\#m$ follows by induction. We record the following facts about the process $ \#$. The third statement is an immediate consequence of Proposition   \[hashprod\ ], the others follow from the definition. commemorate

m)=2^a-1=l\#l$ whenever $2^{a-1}{\leqslant}l,m<2^a$. Ltt $r,s,t\in\N$. From Propositimn \[backsiag\] foluows that $\tau(r,s){\geqslant}t$ if, end inly uf, there exist $\rho{\leqsuant}r$ and $\s{\leqslabt}s$ xith $\rho+\s=t$ such vgat $Q_t^{(\rmj,\s)}\eqhlv1$. Noc wuppose that $r,x{\leqslant}2^a$, and consider $\shu{\lzqslant}r$ and $\s{\leqslant}s$. If $\rho+\s<2^a$ theg $$Q_{\rho+\s}^{(\thl,\s)}\equiv\binom{\rhj+\s}{\rhp}= \finoj{2^a}{2^a}\binom{\rho+\s}{\rho}\equiv \binom{\rho+\s+2^a}{\rho+2^z}=Q_{\rho+2^a}^{(\rio+2^a,\s)}\,.$$ If instead $\rho+\s{\geqslant}2^a$ then $Q_{\rho+\s}^{(\rjo,\s)} \fquiv 0 \equiv Q_{\rho+2^a+\d}^{(\rho+2^a,\s)}$ by Lgjma \[fpecials\]:\[2s\]. In dach case $Q_{\rho+\s}^{(\rho,\s)}\equjv Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$, which proves thag $\tau(x+2^a,s)=\tau(r,s) + 2^a$ fir wnl $r,s<2^a$. Takiig $l=r+2^a$ and $m=s$ we gcn $\tau(l,m)=2^d + \tau(l-2^s,m)$ for all $l,m\ik\N$ surh tyat $m<2^a{\leqslant}l<2^{a+1}$. On tie other hand, from tre definidikn follows that $l\#n=2^a+(o-2^a)\#m$ it thhs cxwe; vow $\teu(l,j)=l\#m$ foploxs by inducfion. We recoed the following favtf about the opsratiog $\#$. The third statement is an immediate cotsesuence of Proposition \[hawhprod\], the others folpow from ehe definition. Remember

m)=2^a-1=l\#m$ whenever $2^{a-1}{\leqslant}l,m<2^a$. Let $r,s,t\in\N$. From Proposition that if, and if, there exist such $Q_t^{(\rho,\s)}\equiv1$. Now suppose $r,s{\leqslant}2^a$, and consider and $\s{\leqslant}s$. If $\rho+\s<2^a$ then $$Q_{\rho+\s}^{(\rho,\s)}\equiv\binom{\rho+\s}{\rho}= \binom{\rho+\s+2^a}{\rho+2^a}=Q_{\rho+2^a}^{(\rho+2^a,\s)}\,.$$ If instead $\rho+\s{\geqslant}2^a$ then $Q_{\rho+\s}^{(\rho,\s)} \equiv 0 \equiv Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$ by Lemma \[specials\]:\[2s\]. each case $Q_{\rho+\s}^{(\rho,\s)}\equiv Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$, which proves that $\tau(r+2^a,s)=\tau(r,s) + 2^a$ for all $r,s<2^a$. $l=r+2^a$ $m=s$ get + \tau(l-2^a,m)$ for all $l,m\in\N$ such that $m<2^a{\leqslant}l<2^{a+1}$. On the other hand, from the definition follows $l\#m=2^a+(l-2^a)\#m$ in this case; now $\tau(l,m)=l\#m$ follows by We record the following about the operation $\#$. The statement an immediate of \[hashprod\], others follow from definition. Remember

m)=2^a-1=l\#m$ whenever $2^{a-1}{\leqslant}l,m<2^a$. LEt $r,s,t\in\N$. FrOm ProPosItiOn \[BackDiag\] Follows that $\tau(R,S){\geqSlant}t$ if, and only if, there Exist $\RhO{\LeqsLAnT}r$ and $\S{\leqslaNT}s$ WITh $\rHo+\S=t$ SucH tHAt $q_t^{(\rho,\S)}\eqUiv1$. Now sUppose that $R,s{\lEqSlant}2^a$, and conSIdEr $\rho{\leqslAnt}R$ and $\s{\leqslanT}s$. IF $\rho+\s<2^a$ ThEn $$Q_{\RHo+\s}^{(\rhO,\s)}\eQuiv\bInom{\rhO+\S}{\rho}= \biNom{2^a}{2^a}\binoM{\rHO+\s}{\rho}\eQUiv \binoM{\RHo+\S+2^a}{\rhO+2^a}=Q_{\rho+2^a}^{(\rho+2^a,\s)}\,.$$ If insTEaD $\Rho+\s{\geqslant}2^a$ tHen $Q_{\rhO+\s}^{(\RHo,\S)} \EQuiV 0 \eqUiv Q_{\rho+2^a+\s}^{(\rHo+2^A,\s)}$ by LEMma \[specIAlS\]:\[2S\]. iN eaCH case $Q_{\rho+\s}^{(\rho,\S)}\equiv Q_{\rho+2^a+\S}^{(\Rho+2^A,\s)}$, whicH pRovES that $\tAu(r+2^a,s)=\TaU(R,s) + 2^a$ For all $r,s<2^a$. TaKing $L=r+2^a$ and $m=s$ wE get $\taU(L,m)=2^a + \tau(l-2^A,M)$ for all $L,m\in\N$ sUch ThaT $m<2^a{\lEQsLaNt}l<2^{A+1}$. ON The OThEr hANd, fRom the deFiNiTion fOlloWS THAt $l\#m=2^A+(l-2^a)\#M$ in tHis caSe; now $\tau(l,m)=l\#m$ fOllOws bY IndUctioN. We reCord ThE follOwing fActs aBoUt the operation $\#$. THe thIrd statemEnt Is An iMmEdiatE ConseqUenCe oF ProposItion \[haSHprOd\], THE OtHers follow from the dEfINItIon. RememBer

m)=2^a-1=l\#m$ whenever $2 ^{a-1}{\le qslan t}l ,m< 2^ a$. Let $r,s,t\in\N$. From Proposition \[backdia g\] f ol l owst ha t $\t au(r,s) { \g e q sla nt }t $ i f, an d onl y i f, ther e exist $\ rho {\ leqslant}r$a nd $\s{\leqs lan t}s$ with $\ rho +\s=t$ s uch that$Q_ t^{(\ rho,\s ) }\equi v1$. Nowsu p pose t h at $r,s { \ le qsla nt}2^a$, and cons i de r $\rho{\leqsla nt}r$an d $ \ s {\l eqs lant}s$. I f$\rho + \s<2^a$ th e n $$Q _ {\rho+\s}^{(\ rho,\s)}\eq u iv\ binom{ \r ho+ \ s}{\rh o}= \ bi n om{ 2^a}{2^a}\b inom {\rho+\s} {\rho} \ equiv \ binom{\ rho+\s +2^ a}{ \rho + 2^ a} =Q_ {\ r ho+ 2 ^a }^{ ( \rh o+2^a,\s )} \, .$$ I f in s t e a d $\ rho +\s{ \geqs lant}2^a$ the n $ Q_{\ r ho+ \s}^{ (\rho ,\s) }\equi v 0 \e quivQ_ {\rho+2^a+\s}^{ (\rh o+2^a,\s) }$by Le mm a \[s p ecials \]: \[2 s\]. In each c a se$Q _ { \ rh o+\s}^{(\rho,\s)}\ eq u i vQ_{\rho+ 2^a+\s } ^{ (\ r ho+2^a,\ s) }$, whi c h prov es t h at $\tau(r +2^a,s ) =\ ta u(r,s)+2^a$ f or al l $ r,s<2 ^ a$.Taking $l=r+2^ a$ an d $m=s$ we get$ \tau(l,m)=2^a +\ t au ( l-2^ a,m )$ for all$l,m \ in\N $ su c htha t $m<2 ^a{\l eq s la n t}l<2^{a+1}$. On th eotherhand, from the def inition fo l l o ws that$l\# m =2 ^ a+(l-2^a)\#m$in th is case; n o w $\tau( l,m)= l\#m$ fo llows byi n duction. W e r eco rdt h efollowing fac t s abo ut the op era tion $\ #$. Th e t hir dstatement is an i mm ed ia te co nsequ e nce of P ro pos it ion  \[ha s hprod\ ], th e ot he rs fol low fro m t h e def in it ion. Re me mber

m)=2^a-1=l\#m$ whenever_$2^{a-1}{\leqslant}l,m<2^a$. Let $r,s,t\in\N$._From Proposition \[backdiag\] follows that_$\tau(r,s){\geqslant}t$ if,_and_only if,_there_exist $\rho{\leqslant}r$ and_$\s{\leqslant}s$ with $\rho+\s=t$_such that $Q_t^{(\rho,\s)}\equiv1$. Now_suppose that $r,s{\leqslant}2^a$,_and_consider $\rho{\leqslant}r$ and $\s{\leqslant}s$. If $\rho+\s<2^a$ then $$Q_{\rho+\s}^{(\rho,\s)}\equiv\binom{\rho+\s}{\rho}= \binom{2^a}{2^a}\binom{\rho+\s}{\rho}\equiv \binom{\rho+\s+2^a}{\rho+2^a}=Q_{\rho+2^a}^{(\rho+2^a,\s)}\,.$$ If instead $\rho+\s{\geqslant}2^a$ then $Q_{\rho+\s}^{(\rho,\s)}_\equiv_0 \equiv_Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$_by_Lemma \[specials\]:\[2s\]. In each case $Q_{\rho+\s}^{(\rho,\s)}\equiv_Q_{\rho+2^a+\s}^{(\rho+2^a,\s)}$, which proves that $\tau(r+2^a,s)=\tau(r,s)_+ 2^a$_for all $r,s<2^a$. Taking $l=r+2^a$ and $m=s$ we_get_$\tau(l,m)=2^a + \tau(l-2^a,m)$_for all $l,m\in\N$ such that $m<2^a{\leqslant}l<2^{a+1}$. On the other_hand, from the definition follows that_$l\#m=2^a+(l-2^a)\#m$ in this_case;_now_$\tau(l,m)=l\#m$ follows by induction. We_record the following facts about the_operation $\#$. The third statement is_an immediate consequence of Proposition \[hashprod\], the others_follow from the definition. Remember

\max_{R^d, R^{ud}} \left\{ R^d + R^{ud} \right\},\end{aligned}$$ where $$\begin{aligned} \begin{aligned} \left( R^d, R^d, R^{ud}, R^d \right) &\in \col{}{\\ &\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!} \Cmac{\bH_d, \bI + P \bH_{ud} \bH_{ud}^T + P \bHinter \bHinter^T}. \end{aligned}\end{aligned}$$ Time Sharing {#s:time-sharing} ============ The simplest scheme that involves scheduling is that of time sharing. A two-phase time-sharing scheme alternates between transmission of odd-indexed cells during odd time instants, and transmission of even-indexed cells during even time instants. Since every cell transmits only during half of the time, it can utilize twice the total available power during its transmission periods. The achievable rate when using this scheme is $$\begin{aligned} R_\text{TS} = \frac{1}{2} \C{2 P}.\end{aligned}$$ Scheduled Decoding: Digital Backhaul {#s:tetris:1antenna} ==================================== In this section we consider the use of only digital backhaul communication between adjacent cells. Such backhaul communication is put to use by adopting a multi-stage decoding scheduling policy. This allows to remove interference from neighboring signals that were already recovered by their respective cells, stage-by-stage. For the one-dimensional case, we propose a two-phase schedule: First only cells with even indices recover their respective messages, treating the interfering messages of the adjacent (odd-indexed) cells as noise. Thus, the rate recovered by these cells is equal to. The even-indexed cells convey their recovered messages to their neighboring odd-indexed cells. The odd-indexed cells can now subtract the interfering even-indexed signals prior to the recovery of the (odd-indexed) signals intended to them: $$\begin{aligned} y^\text{Sch}_{2i+1} &= y_{2i+1} - \alpha \left( x_{2i} - x_{2i+2} \right) \\ &= x_{

\max_{R^d, R^{ud } } \left\ { R^d + R^{ud } \right\},\end{aligned}$$ where $ $ \begin{aligned } \begin{aligned } \left (R^d, R^d, R^{ud }, R^d \right) & \in \col{}{\\ & \!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\! } \Cmac{\bH_d, \bI + P \bH_{ud } \bH_{ud}^T + P \bHinter \bHinter^T }. \end{aligned}\end{aligned}$$ Time Sharing { # s: time - sharing } = = = = = = = = = = = = The simplest schema that imply scheduling is that of time sharing. A two - phase time - share scheme alternates between infection of odd - index cells during odd time instants, and infection of even - indexed cells during tied time instants. Since every cell transmits only during half of the time, it can use twice the total available power during its transmission period. The achievable rate when practice this scheme is $ $ \begin{aligned } R_\text{TS } = \frac{1}{2 } \C{2 P}.\end{aligned}$$ schedule Decoding: Digital Backhaul { # s: tetris:1antenna } = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = In this section we consider the use of merely digital backhaul communication between adjacent cells. Such backhaul communication is put to use by adopting a multi - stage decoding scheduling policy. This allows to remove interference from neighboring signals that were already recovered by their respective cells, stage - by - stage. For the one - dimensional case, we propose a two - phase schedule: First merely cells with even exponent reclaim their respective messages, treating the interfere messages of the adjacent (odd - indexed) cells as noise. therefore, the rate recovered by these cells is equal to. The even - indexed cells convey their recovered messages to their neighboring odd - indexed cells. The odd - index cells can nowadays subtract the interfering even - indexed signals prior to the recovery of the (odd - index) signals intended to them: $ $ \begin{aligned } y^\text{Sch}_{2i+1 } & = y_{2i+1 } - \alpha \left (x_{2i } - x_{2i+2 } \right) \\ & = x _ {

\mad_{R^d, R^{ud}} \left\{ R^d + R^{ud} \rigmt\},\end{aligned}$$ whete $$\begin{eligned} \gegin{aliened} \left( R^d, R^d, R^{ud}, R^d \rmght) &\in \col{}{\\ &\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!} \Cmac{\bH_d, \cI + P \bH_{uf} \bH_{ud}^T + P \bIinter \bHinter^T}. \eis{aligned}\end{aljnned}$$ Tnmx Sharing {#s:time-xharing} ============ The simplest scheke tkat involves scheduling is that of tyme shatijg. A two-phase jime-snwrinf scheme alternates between transmjssion mf odd-indexed cells during odd time inshantd, and transmission of even-indgsed xells during even time instants. Sihce every cell transmits only djring half of tye tilg, it can utioize nwice the total availdble poeer during its trensmussion periods. The aciievable rate when ufing this seheme is $$\begin{aligned} R_\teft{TS} = \frxx{1}{2} \C{2 P}.\ehd{elifned}$$ Scjednled Decodihg: Digital Vackhaul {#s:tetris:1anttnnw} ==================================== Pm this sectikn we sogsider the use of only digital backhaul cojmunication between adjqcent cells. Such backjaul commtnication is put to use by adopting a multi-stage gecodmne semcdulkbg policy. This allows to remove interference frjj mepghboring signals that were slgesqy recovered cy thend despective cells, shage-by-sjage. Foe the one-qimemsional case, we propose a tqo-phase schebulw: First only cells with even nndicex recpver their respective mzssagea, treating hhe interrdring messages ow tme ddjacent (odd-indexed) cells ws noise. Vhus, che rate recpvered by these fells is equal to. The evfn-indgxed calls convej their recovered messages to tixir neighborimg odg-indexed cellx. The odd-indeved cells can uow subtxact tfe interfeging even-mndexed signwls prior to dje recovery mf the (oqd-ineexee) signaur intended to yhem: $$\begin{cjignwd} y^\text{Sch}_{2i+1} &= y_{2l+1} - \alkhz \left( x_{2i} - x_{2i+2} \rnyhu) \\ &= x_{

\max_{R^d, R^{ud}} \left\{ R^d + R^{ud} \right\},\end{aligned}$$ \begin{aligned} R^d, R^d, R^d \right) &\in P \bH_{ud}^T + P \bHinter^T}. \end{aligned}\end{aligned}$$ Time {#s:time-sharing} ============ The simplest scheme that scheduling is that of time sharing. A two-phase time-sharing scheme alternates between transmission odd-indexed cells during odd time instants, and transmission of even-indexed cells during even instants. every transmits during half of the time, it can utilize twice the total available power during its transmission The achievable rate when using this scheme is R_\text{TS} = \frac{1}{2} \C{2 Scheduled Decoding: Digital Backhaul {#s:tetris:1antenna} In section we the of digital backhaul communication adjacent cells. Such backhaul communication is put to use by adopting a multi-stage decoding scheduling policy. This to remove neighboring signals were recovered their respective cells, the one-dimensional case, we propose a only cells with even indices recover their respective treating the messages of the adjacent (odd-indexed) cells noise. Thus, the rate recovered by these cells equal to. The even-indexed cells convey their recovered messages to their neighboring odd-indexed cells. The can now subtract the even-indexed signals prior the of (odd-indexed) intended to $$\begin{aligned} y^\text{Sch}_{2i+1} &= y_{2i+1} - \alpha \left( x_{2i} - x_{2i+2} \right) &= x_{

\max_{R^d, R^{ud}} \left\{ R^d + R^{ud} \right\},\end{Aligned}$$ wheRe $$\begIn{aLigNeD} \begIn{alIgned} \left( R^d, R^d, R^{UD}, R^d \rIght) &\in \col{}{\\ &\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!} \Cmac{\bH_d, \bI + P \bH_{uD} \bH_{ud}^t + P \BhintER \bhinteR^T}. \end{alIGnED}\End{AlIgNed}$$ tiME SHarinG {#s:tIme-sharIng} ============ The simpLesT sCheme that invOLvEs scheduliNg iS that of time sHarIng. A twO-pHasE Time-sHarIng scHeme alTErnateS between tRaNSmissiON of odd-iNDExEd ceLls during odd time iNStANts, and transmisSion of EvEN-iNDExeD ceLls during eVeN time INstants. sInCE EVerY Cell transmits Only during hALf oF the tiMe, It cAN utiliZe twiCe THe tOtal availabLe poWer during Its traNSmissioN Periods. the achIevAblE ratE WhEn UsiNg THis SChEme IS $$\beGin{alignEd} r_\tExt{TS} = \Frac{1}{2} \c{2 p}.\END{aliGneD}$$ SchEduleD Decoding: DigiTal backHAul {#S:tetrIs:1antEnna} ==================================== in This sEction We conSiDer the use of only DigiTal backhaUl cOmMunIcAtion BEtween AdjAceNt cells. such bacKHauL cOMMUnIcation is put to use bY aDOPtIng a multI-stage DEcOdINg scheduLiNg pOlicY. tHis alLows TO rEmove intErfereNCe FrOm neighBoRing siGnAls ThaT were ALreaDy recoVered by tHeir rESpective cells, sTAge-by-stage. For THe ONE-dIMensIonAl case, we proPose A Two-pHase SChEduLE: FirsT only CeLLs WIth even indices recovEr Their rEspecTive messages, tReating the INTErfering MessAGeS Of the adjacent (oDd-indExed) cells aS Noise. ThuS, the rAte recovEred by theSE Cells is eQuaL to. the EveN-INdExed cells convEY TheiR rEcovereD meSsages tO thEir NeiGhbOrIng odd-indExed cellS. THe OdD-iNdeXed ceLLs can now SuBtrAcT thE inteRFering Even-iNdexEd SiGNalS prior tO ThE REcovErY oF the (Odd-InDexed) SignALs iNtended To them: $$\begIn{aLIgneD} y^\TeXt{Sch}_{2i+1} &= y_{2I+1} - \alpha \left( x_{2i} - x_{2I+2} \rIght) \\ &= x_{

\max_{R^d, R^{ud}} \left\ { R^d + R^ {ud}\ri ght \} ,\en d{al igned}$$ where $$\b egin{aligned} \begin{a ligne d} \ le ft( R ^d, R^d , R ^ { ud} ,R^ d \r i gh t) &\ in \co l{}{\\ &\! \!\ !\ !\!\!\!\!\!\ ! \! \!\!\!\!\! \!\ !\!} \C mac{\b H_ d,\ bI +P \ bH_{u d} \bH _ {ud}^T + P \bHi nt e r \bHi n ter^T}. \ en d{al igned}\end{aligne d }$ $ Time Sharing {#s:t im e -s h a rin g}========== == The simples t s c h e met hat involvesschedulingi s t hat of t ime sharin g. Atw o -ph ase time-sh arin g schemealtern a tes bet w een tra nsmiss ion of odd - in de xed c e lls du rin g od d time i ns ta nts,andt r a n smis sio n of even -indexed cell s d urin g ev en ti me in stan ts . Sin ce eve ry ce ll transmits only dur ing halfofth e t im e, it can ut ili zetwice t he tota l av ai l a b le power during itstr a n sm ission p eriods . T he achievab le ra te w h e n usi ng t h is schemeis $$\ b eg in {aligne d} R _\ tex t{T S} =\ frac {1}{2} \C{2 P} .\end { aligned}$$ Sc h eduled Decodi n g: D ig i talBac khaul {#s:t etri s :1an tenn a }=== = ===== ===== == = == = ================ I nthis s ectio n we consider the use o f o nly digi talb ac k haul communica tionbetween ad j acent ce lls.Such bac khaul com m u nication is pu t t o u s e b y adopting am u lti- st age dec odi ng sche dul ing po lic y. This all ows to r em ov ein ter feren c e from n ei ghb or ing sign a ls tha t wer e al re ad y re covered by t heir r es pect ive c ells, sta g e-b y-stage . For th e o n e-di me ns ional c ase, we propo se a two-pha se sc hedule : First on ly cells with even indi c es reco ver thei r re spectivemes sages, tr e atingthe in terfe ri ngm e ssage s of th eadjacent ( o d d-i ndexe d) cel ls as n oise. Thus, the ra t e r ecovered by t hes e ce l l sise qu a l t o. The e ven-indexed cel ls conveyth e ir recovered mes sa ges totheir n eighb o ring od d-indexed cells. T he odd - i nde xed cellscan nowsubtractt he in t er ferin g e ven-in de xed sign als pr i orto th e reco ve ry ofthe ( od d-indexe d) signals intended tothem: $$\b egi n{aligned } y ^\text{Sc h}_{ 2i+1} &= y _{2 i+1 } - \ alp h a \le ft(x _{ 2i} - x_{ 2i+2 } \right)\\ &= x _{

\max_{R^d,_R^{ud}} \left\{_R^d + R^{ud} \right\},\end{aligned}$$_where $$\begin{aligned} \begin{aligned} __ _\left(_R^d, R^d, R^{ud},_R^d \right)_&\in _\col{}{\\ &\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!} __ \Cmac{\bH_d, \bI + P \bH_{ud} \bH_{ud}^T + P \bHinter_\bHinter^T}. \end{aligned}\end{aligned}$$ Time_Sharing {#s:time-sharing} ============ The_simplest_scheme_that involves scheduling is that_of time sharing. A two-phase_time-sharing scheme_alternates between transmission of odd-indexed cells during odd_time_instants, and transmission_of even-indexed cells during even time instants. Since every_cell transmits only during half of_the time, it_can_utilize_twice the total available_power during its transmission periods. The_achievable rate when using this scheme_is $$\begin{aligned} R_\text{TS} =_\frac{1}{2} \C{2 P}.\end{aligned}$$ Scheduled Decoding: Digital Backhaul_{#s:tetris:1antenna} ==================================== In this section we consider_the use_of only digital backhaul communication_between adjacent cells._Such backhaul_communication is put_to use by adopting a multi-stage_decoding scheduling policy._This allows to remove interference from_neighboring_signals that were_already_recovered_by their_respective cells, stage-by-stage. For_the_one-dimensional case,_we_propose a two-phase schedule: First only_cells_with even indices recover their respective messages,_treating the interfering messages_of_the adjacent (odd-indexed) cells_as noise. Thus, the rate_recovered by these cells is equal_to. The_even-indexed cells_convey their recovered messages to their neighboring odd-indexed cells. The odd-indexed_cells can now subtract the interfering_even-indexed signals prior to_the recovery_of_the (odd-indexed) signals_intended_to them: $$\begin{aligned} _ y^\text{Sch}_{2i+1} &= y_{2i+1} -_\alpha \left(_x_{2i} - x_{2i+2} \right) \\ &=_x_{

it can query the value of the function at any point with one computation step, although it does not have the full description of the function. (See [@goldreichFound] for a detailed description.) We now define *pseudorandom functions* (see [@goldreich1986construct]). Intuitively, this is a family of functions indexed by a seed, such that it is hard to distinguish a random member of the family from a truly randomly selected function. A *pseudorandom function ensemble (PRF)* is a set such that the following conditions hold: - (easy to compute) $f_s(x)$ can be computed by a PPT algorithm that is given $s$ and $x$; - (pseudorandom) the function ensemble, where $F_n$ is uniformly distributed over the multiset $\{f_s\}_{s\in\{0,1\}^n}$, is computationally indistinguishable from $H$. We use the standard cryptographic assumption that a family of PRFs exists; this assumption is implied by the existence of one-way functions [@haastad1999pseudorandom; @goldreich1986construct]. We actually require the use of a seemingly stronger notion of a PRF, which requires that an attacker getting access to polynomially many instances of a PRF (i.e., $f_s$ for polynomially many values of $s$) still cannot distinguish them from polynomially many truly random functions. Nevertheless, as we show in Appendix \[section:appA\], it follows using a standard “hybrid” argument that any PRF satisfies also this stronger “multi-instance” security notion. ### Public-key Encryption Schemes We now define public-key encryption schemes. Such a scheme has two keys. The first is public and used for encrypting messages (using a randomized algorithm). The second is secret and used for decrypting. The keys are generated in such a way that the probability that a decrypted message is equal to the encrypted message is equal to $1$. The key generation algorithm takes as input a “security parameter” $k$ that is used to determine the security of the protocols (inuitively, no polynomial-time attacker should be able to “break” the security of the protocol except possibly with a probability that is a negligible function of $k$). We now recall the formal definitions of public-key encryption schemes [@diffie1976new; @rivest1978

it can query the value of the function at any point with one computation gradation, although it does not take the full description of the function. (See [ @goldreichFound ] for a detailed description .) We immediately define * pseudorandom functions * (experience [ @goldreich1986construct ]). Intuitively, this is a kin of functions indexed by a seed, such that it is hard to identify a random member of the family from a in truth randomly selected function. A * pseudorandom function ensemble (PRF) * is a set such that the following condition hold: - (easy to compute) $ f_s(x)$ can be calculate by a PPT algorithm that is given $ s$ and $ x$; - (pseudorandom) the function corps de ballet, where $ F_n$ is uniformly distributed over the multiset $ \{f_s\}_{s\in\{0,1\}^n}$, is computationally identical from $ H$. We use the standard cryptographic assumption that a family of PRFs exists; this assumption is imply by the existence of one - way functions [ @haastad1999pseudorandom; @goldreich1986construct ]. We actually require the use of a apparently strong notion of a PRF, which requires that an attacker getting access to polynomially many instances of a PRF (i.e., $ f_s$ for polynomially many values of $ s$) still cannot distinguish them from polynomially many truly random function. Nevertheless, as we show in Appendix   \[section: appA\ ], it stick to use a standard “ hybrid ” argument that any PRF satisfies also this stronger “ multi - case ” security notion. # # # Public - key Encryption Schemes We now specify public - key encryption schemes. Such a scheme has two keys. The beginning is public and used for encrypting messages (using a randomized algorithm). The second is unavowed and practice for decrypting. The keys are generated in such a way that the probability that a decrypted message is equal to the code message is equal to $ 1$. The key generation algorithm takes as input a “ security parameter ” $ k$ that is used to determine the security system of the protocols (inuitively, no polynomial - time attacker should be able to “ break ” the security of the protocol except possibly with a probability that is a negligible function of $ k$). We nowadays recall the conventional definitions of public - samara encryption system [ @diffie1976new; @rivest1978

it can query the value of uhe function at aut poinv with kne compjtation step, although it doed bot hqve the full descriptiun of the functiob. (Set [@goldreichFound] hkr a debciled fescxi'tion.) We now deflne *pseudordndom functionv* (rez [@goldreich1986construct]). Intuitively, thif is a galily of functijns pnqexes by a seed, such that it is hard tk distiiguish a random member of the family from a tguly randomly selefted functiin. A *[weudorandom wunction ensemble (PRF)* js a set such that the followine conbitions hole: - (ewvy to compuve) $f_s(x)$ can be compmned by d PPT akgorithm that ls giten $w$ and $x$; - (pseudorandok) the function ensgmble, whera $Y_n$ is uniformly distrubyted mver the nulgistt $\{h_s\}_{s\jn\{0,1\}^n}$, is cojputationamly indistibguishable from $H$. We ufv the standars crypeodraphic assumption that a family of PRFv esists; this assumption iw implied by the exisjence of oge-way functions [@haastad1999pseudorandom; @goldreich1986consdruct]. De cgbualut gequire the use of a seemingly stronger notiog og s PRF, which rezuires that aj sjtacker gettine access fo polynomially majy instwnces of a PRF (i.e., $g_s$ for polynomially many vaoues of $s$) stpll xannot distinguish them from 'olynokiallu many truly random funetions. Nevertheleds, as we afow in Appendix \[sdctpon:a[pA\], it follows using a stagdard “hybcid” axgument ghat any PWF satisfifs also this stronger “mupti-indtdnce” securlty notion. ### Public-key Encryption Schemes We now dafite publie-key ekcryption schemqs. Such a scheke has cwo kehs. The firat is pnblic and usqd for encrypdlng messages (using a raneomized algufithm). The secomd is secgeu and used for decrypting. Thc keyr are generated nu wuch a way that thd pwohauilitr that a decrfptea mdxsage is equal ti the encrypted message iv eqhal to $1$. The key gemevation altorithm eakes as inpuy a “security paramtter” $k$ thet is iseq to determine the security of the protlcojs (inuitivelr, no polynomial-tnme attacker should be able to “break” the security of the krotocol except possiyln with a prouabiliey that iv a negligible functuon of $k$). We now regall the formal definifions mf puhlic-key encryption schemes [@diffie1976new; @rivest1978

it can query the value of the any with one step, although it description the function. (See for a detailed We now define *pseudorandom functions* (see Intuitively, this is a family of functions indexed by a seed, such that is hard to distinguish a random member of the family from a truly selected A function (PRF)* is a set such that the following conditions hold: - (easy to compute) $f_s(x)$ can computed by a PPT algorithm that is given and $x$; - (pseudorandom) function ensemble, where $F_n$ is distributed the multiset is indistinguishable $H$. We use standard cryptographic assumption that a family of PRFs exists; this assumption is implied by the existence of functions [@haastad1999pseudorandom; actually require use a stronger notion of which requires that an attacker getting many instances of a PRF (i.e., $f_s$ for many values $s$) still cannot distinguish them from many truly random functions. Nevertheless, as we show Appendix \[section:appA\], it follows using a standard “hybrid” argument that any PRF satisfies also this security notion. ### Public-key Schemes We now public-key schemes. a has two The first is public and used for encrypting messages (using a algorithm). The second is secret and used for decrypting. The generated such a way the probability that a message equal to the encrypted equal $1$. algorithm as a “security parameter” $k$ is used to determine the of the protocols (inuitively, able to “break” the security of the protocol possibly with a probability that is a function of $k$). We now recall the formal definitions of public-key encryption [@diffie1976new; @rivest1978

it can query the value of the fuNction at anY poinT wiTh oNe CompUtatIon step, althougH It doEs not have the full descriPtion Of THe fuNCtIon. (SeE [@goldreIChfOUnd] FoR a DetAiLEd DescrIptIon.) We noW define *pseUdoRaNdom functionS* (SeE [@goldreich1986ConStruct]). IntuitIveLy, this Is A faMIly of FunCtionS indexED by a seEd, such thaT iT Is hard TO distinGUIsH a raNdom member of the faMIlY From a truly randOmly seLeCTeD FUncTioN. A *pseudoraNdOm funCTion ensEMbLE (prF)* iS A set such that tHe following COndItions HoLd: - (eASy to coMpute) $F_s(X)$ Can Be computed bY a PPt algorithM that iS Given $s$ aND $x$; - (pseudOrandoM) thE fuNctiON eNsEmbLe, WHerE $f_n$ Is uNIfoRmly distRiBuTed ovEr thE MULTiseT $\{f_s\}_{S\in\{0,1\}^n}$, Is comPutationally iNdiStinGUisHable From $H$. we usE tHe staNdard cRyptoGrAphic assumption That A family of pRFS eXisTs; This aSSumptiOn iS imPlied by The exisTEncE oF ONE-wAy functions [@haastad1999PsEUDoRandom; @goLdreicH1986CoNsTRuct]. We acTuAllY reqUIRe the Use oF A sEemingly StrongER nOtIon of a PrF, Which rEqUirEs tHat an ATtacKer getTing acceSs to pOLynomially many INstances of a PRf (I.e., $F_S$ FoR PolyNomIally many vaLues OF $s$) stIll cANnOt dIStingUish tHeM FrOM polynomially many trUlY randoM funcTions. NevertheLess, as we shOW IN AppendiX \[secTIoN:AppA\], it follows uSing a Standard “hyBRid” argumEnt thAt any PRF Satisfies ALSo this stRonGer “MulTi-iNSTaNce” security noTIOn. ### PuBlIc-key EnCryPtion ScHemEs WE noW deFiNe public-kEy encrypTiOn ScHeMes. such a SCheme has TwO keYs. the First IS publiC and uSed fOr EnCRypTing mesSAgES (UsinG a RaNdomIzeD aLgoriThm). THE seCond is sEcret and uSed FOr deCrYpTing. The Keys are generaTeD in such a waY tHat The proBABility thAt a decrypted message is eqUAl to the EncRypteD mesSage is equAl tO $1$. The keY geNEratioN algorIthm tAkEs aS INput a “SECuRitY pArameter” $k$ tHAT is Used tO dEterMine the Security of the protoCOls (Inuitively, no pOlyNomiAL-TiMe aTTaCKer ShOUld BE Able to “break” the sEcurity of tHe PRoTocol excepT PosSiBly with A probabIlity THat is a nEgligible Function oF $k$). we noW REcaLl the formaL definitIons of pubLIc-key ENcRyptiOn sChemes [@DiFfiE1976new; @rIvest1978

it can query the value of the funct ion a t a nypo intwith one computati o n st ep, although it does n ot ha ve thef ul l des criptio n o f the f un cti on . ( See [ @go ldreich Found] for ade tailed descr i pt ion.) Wenow define *pse udo random f unc t ions* (s ee [@ goldre i ch1986 construct ]) . Intui t ively,t h is isa family of funct i on s indexed by aseed,su c ht h atitis hard to d istin g uish ar an d o m me m ber of the fa mily from a tru ly ran do mly select ed fu nc t ion . A *pseud oran dom funct ion en s emble ( P RF)* is a set su chthat th efol lo w ing co ndi t ion s hold: - (eas y to c o m pute ) $ f_s( x)$ c an be compute d b y aP PTalgor ithmthat i s giv en $s$ and$x $; - (pseudo rand om) the f unc ti onen sembl e , wher e $ F_n $ is un iformly dis tr i b u te d over the multise t$ \ {f _s\}_{s\ in\{0, 1 \} ^n } $, is co mp uta tion a l ly in dist i ng uishable from$ H$ . We use t he sta nd ard cr yptog r aphi c assu mption t hat a family of PRFs exists; thisa ss u m pt i on i s i mplied by t he e x iste nceo fone - way f uncti on s [ @ haastad1999pseudora nd om; @g oldre ich1986constr uct]. We a c t u ally req uire th e use of a seem ingly strongern otion of a PR F, which requires t hat an a tta cke r g ett i n gaccess to pol y n omia ll y manyins tancesofa P RF(i. e. , $f_s$ f or polyn om ia ll yman y val u es of $s $) st il l c annot distin guish the mfr o m p olynomi a ll y many t ru ly r and om func tion s . N everthe less, aswes howin A ppendix  \[section:ap pA \], it fol lo wsusinga standard “hybrid” argument that any PRF sa tisfi es a lso thisstr onger“mu l ti-ins tance” secu ri tyn o tion. ## # P ub lic-key En c r ypt ion S ch emes We no w define public-ke y en cryption sche mes . Su c h a sc h em e ha st wok e ys. The first i s public a nd us ed for enc r ypt in g messa ges (us ing a randomi zed algor ithm). Th eseco n d is secret an d used f or decryp t ing.T he keys ar e gene ra ted in s uch aw aythatthe pr ob abilit y tha ta decryp ted message is equal to the e ncryp ted messageise qua l to $1$. The key gener ati onalgor ith m take s as in put a “se curi t y paramet e r” $k $ th at is usedt o det ermin e t h e secu rity of the protocols (inuitively, n o po l y nom ial - time a ttacker should be a b l e to “br ea k” the secu rity ofth e prot ocol e xceptpossibl y wi t h a pr obab ili ty that i s a n e gligibl efu n ctionof $ k$ ). We now r e call t he formal defini tions o f pub l ic- key e nc ryption sche mes [@diff ie1976new;@rives t197 8

it_can query_the value of the_function at_any_point with_one_computation step, although_it does not_have the full description_of the function._(See_[@goldreichFound] for a detailed description.) We now define *pseudorandom functions* (see [@goldreich1986construct]). Intuitively, this is_a_family of_functions_indexed_by a seed, such that_it is hard to distinguish_a random_member of the family from a truly randomly_selected_function. A *pseudorandom function_ensemble (PRF)* is a set such that the following_conditions hold: - (easy to_compute) $f_s(x)$ can_be_computed_by a PPT algorithm_that is given $s$ and $x$; -_ (pseudorandom) the function ensemble,_where $F_n$ is uniformly distributed over the_multiset $\{f_s\}_{s\in\{0,1\}^n}$, is computationally indistinguishable from_$H$. We use the standard cryptographic_assumption that_a family of PRFs exists;_this assumption is_implied by_the existence of_one-way functions [@haastad1999pseudorandom; @goldreich1986construct]. We actually_require the use_of a seemingly stronger notion of_a_PRF, which requires_that_an_attacker getting_access to polynomially_many_instances of_a_PRF (i.e., $f_s$ for polynomially many_values_of $s$) still cannot distinguish them from_polynomially many truly random_functions._Nevertheless, as we show_in Appendix \[section:appA\], it follows using_a standard “hybrid” argument that any_PRF satisfies_also this_stronger “multi-instance” security notion. ### Public-key Encryption Schemes We now define public-key encryption_schemes. Such a scheme has two_keys. The first is_public and_used_for encrypting messages_(using_a randomized_algorithm). The second is secret and used_for decrypting._The keys are generated in such_a way that the_probability_that a decrypted message is equal_to the encrypted message is equal_to $1$. The key generation_algorithm_takes_as input a “security parameter”_$k$ that is used to determine_the security of_the protocols (inuitively, no polynomial-time attacker should_be_able to “break” the security of_the_protocol except possibly with a probability_that_is_a negligible function of $k$). We_now recall the formal definitions of_public-key encryption schemes [@diffie1976new; @rivest1978

such that $X(\sigma)=0$ or, equivalently, $X_t(\sigma)=\sigma$ for all $t \in {{\mathbb R}}$. The set formed by singularities is the *singular set of $X$* denoted ${\mathrm{Sing}}(X)$ and ${\operatorname{Per}}(X)$ is the set of periodic points of $X$. We say that a *singularity is hyperbolic* if the eigenvalues of the derivative $DX(\sigma)$ of the vector field at the singularity $\sigma$ have nonzero real part. The set of critical elements of $X$ is the union of the singularities and the periodic orbits of $X$, and will be denoted by ${\operatorname{Crit}}(X)$. We recall that an hyperbolic set $\Lambda$ for a flow $X_t$ is an invariant subset of $M$ with a decomposition $T_\Lambda M= E^s\oplus E^X \oplus E^u$ of the tangent bundle over $\Lambda$ which is a continuous splitting, where $E^X$ is the direction of the vector field, the subbundles are invariant under the derivative $DX_t$ of the flow $$\begin{aligned} DX_t\cdot E^*_x=E^*_{X_t(x)},\quad x\in\Lambda, \quad t\in{{\mathbb R}},\quad *=s,X,u;\end{aligned}$$ $E^s$ is uniformly contracted by $DX_t$ and $E^u$ is uniformly expanded: there are $K,\lambda>0$ so that $$\begin{aligned} \label{eq:def-hyperbolic} \|DX_t\mid_{E^s_x}\|\le K e^{-\lambda t}, \quad \|DX_{-t} \mid_{E^u_x}\|\le K e^{-\lambda t}, \quad x\in\Lambda, \quad t\in{{\mathbb R}}.\end{aligned}$$ We say that a point $x \in M$ is [*nonwandering*]{} for $X$ provided for every neighborhood $U$ of $x$ there is $t > 0$ such that $X_t(x) \cap U \neq \emptyset$, i.e., there exists a point $y \in U$ with $X_t(y) \in U$. We denote by $\Omega(X

such that $ X(\sigma)=0 $ or, equivalently, $ X_t(\sigma)=\sigma$ for all $ t \in { { \mathbb R}}$. The set formed by singularities is the * remarkable hardening of $ X$ * denoted $ { \mathrm{Sing}}(X)$ and $ { \operatorname{Per}}(X)$ is the set of periodic points of $ X$. We allege that a * singularity is hyperbolic * if the eigenvalues of the derivative $ DX(\sigma)$ of the vector field at the singularity $ \sigma$ hold nonzero real part. The stage set of critical elements of $ X$ is the union of the singularity and the periodic orbits of $ X$, and will be denoted by $ { \operatorname{Crit}}(X)$. We hark back that an hyperbolic jell $ \Lambda$ for a flow $ X_t$ is an invariant subset of $ M$ with a decomposition $ T_\Lambda M= E^s\oplus E^X \oplus E^u$ of the tangent bundle over $ \Lambda$ which is a continuous splitting, where $ E^X$ is the direction of the vector battlefield, the subbundles are invariant under the derivative $ DX_t$ of the flow $ $ \begin{aligned } DX_t\cdot E^*_x = E^*_{X_t(x)},\quad x\in\Lambda, \quad t\in{{\mathbb R}},\quad * = s, X, u;\end{aligned}$$ $ E^s$ is uniformly contracted by $ DX_t$ and $ E^u$ is uniformly boom: there are $ K,\lambda>0 $ so that $ $ \begin{aligned } \label{eq: def - hyperbolic } \|DX_t\mid_{E^s_x}\|\le K e^{-\lambda t }, \quad \|DX_{-t } \mid_{E^u_x}\|\le K e^{-\lambda t }, \quad x\in\Lambda, \quad t\in{{\mathbb R}}.\end{aligned}$$ We say that a point $ x \in M$ is [ * nonwandering * ] { } for $ X$ supply for every neighborhood $ U$ of $ x$ there is $ t > 0 $ such that $ X_t(x) \cap U \neq \emptyset$, i.e., there exists a point $ yttrium \in U$ with $ X_t(y) \in U$. We denote by $ \Omega(X

sufh that $X(\sigma)=0$ or, equivauently, $X_t(\sigma)=\sntma$ foc all $t \in {{\mathcb R}}$. The set formed by singupaeitiew is the *singular set uf $X$* denoned ${\mathrn{Sinj}}(X)$ and ${\operatorneje{Per}}(X)$ lf ths set if periodic polnts of $X$. Wa say that a *shneuparity is hyperbolic* if the eigenvajues of tje derivative $QX(\sibia)$ or the vector field at the singularjty $\sigka$ have nonzeto real part. The set of crlticwl elements of $X$ id the union of eye singularigies and the periodic krbits of $X$, and will be denoted by ${\o'eratorname{Xrut}}(X)$. Ag recall thav an hjperbolic set $\Lambda$ xor a fkow $X_t$ is an ikvarient wubset of $M$ with a deromposition $T_\Lambda I= E^s\oplus E^R \oplus E^u$ of the tantebt butdle ovef $\Laobdz$ xhidh is w ckntinuous aplitting, wyere $E^X$ is the direvtyin of the vecfor fiqlq, the subbundles are invariant under tht derjvative $DX_t$ of the flow $$\begin{aligned} DX_t\cdoj E^*_x=E^*_{X_t(x)},\quwd x\in\Lambda, \quad t\in{{\mathbb R}},\quad *=s,X,u;\end{aligned}$$ $A^s$ is jniyirmly xojtracted by $DX_t$ and $E^u$ is uniformly expanded: egete are $K,\lambda>0$ sj that $$\begim{apibged} \label{eq:def-fyperbolid} \|DX_t\mid_{E^s_x}\|\le K e^{-\lalbda t}, \quad \|DX_{-t} \mid_{T^u_x}\|\le K e^{-\lambda t}, \quad x\in\Lambdq, \quad t\in{{\manhbb R}}.\end{aligned}$$ We say that a poiut $x \im M$ os [*nonwandering*]{} for $X$ pxovides for every neighborguod $U$ of $x$ there is $t > 0$ such ufat $X_t(x) \cap U \neq \emptyset$, i.e., chere exksts a poigt $y \in U$ aith $W_d(y) \in U$. We denote hy $\Omgga(X

such that $X(\sigma)=0$ or, equivalently, $X_t(\sigma)=\sigma$ for \in R}}$. The formed by singularities $X$* ${\mathrm{Sing}}(X)$ and ${\operatorname{Per}}(X)$ the set of points of $X$. We say that *singularity is hyperbolic* if the eigenvalues of the derivative $DX(\sigma)$ of the vector at the singularity $\sigma$ have nonzero real part. The set of critical elements $X$ the of singularities and the periodic orbits of $X$, and will be denoted by ${\operatorname{Crit}}(X)$. We recall that hyperbolic set $\Lambda$ for a flow $X_t$ is invariant subset of $M$ a decomposition $T_\Lambda M= E^s\oplus \oplus of the bundle $\Lambda$ is a continuous where $E^X$ is the direction of the vector field, the subbundles are invariant under the derivative $DX_t$ the flow E^*_x=E^*_{X_t(x)},\quad x\in\Lambda, t\in{{\mathbb *=s,X,u;\end{aligned}$$ is uniformly contracted and $E^u$ is uniformly expanded: there that $$\begin{aligned} \label{eq:def-hyperbolic} \|DX_t\mid_{E^s_x}\|\le K e^{-\lambda t}, \quad \mid_{E^u_x}\|\le K t}, \quad x\in\Lambda, \quad t\in{{\mathbb R}}.\end{aligned}$$ say that a point $x \in M$ is for $X$ provided for every neighborhood $U$ of $x$ there is $t > 0$ such \cap U \neq \emptyset$, there exists a $y U$ $X_t(y) U$. We by $\Omega(X

such that $X(\sigma)=0$ or, equivalenTly, $X_t(\sigma)=\Sigma$ For All $T \iN {{\matHbb R}}$. the set formed by SInguLarities is the *singular sEt of $X$* DeNOted ${\MAtHrm{SiNg}}(X)$ and ${\oPErATOrnAmE{PEr}}(X)$ Is THe Set of PerIodic poInts of $X$. We sAy tHaT a *singularitY Is Hyperbolic* If tHe eigenvalueS of The derIvAtiVE $DX(\siGma)$ Of the Vector FIeld at The singulArITy $\sigmA$ Have nonZERo Real Part. The set of critiCAl ELements of $X$ is thE union Of THe SINguLarIties and thE pEriodIC orbits OF $X$, AND WilL Be denoted by ${\opEratorname{CRIt}}(X)$. we recaLl ThaT An hypeRboliC sET $\LaMbda$ for a floW $X_t$ iS an invariAnt subSEt of $M$ wiTH a decomPositiOn $T_\lamBda M= e^S\oPlUs E^x \oPLus e^U$ oF thE TanGent bundLe OvEr $\LamBda$ wHICH Is a cOntInuoUs splItting, where $E^X$ Is tHe diREctIon of The veCtor FiEld, thE subbuNdles ArE invariant under The dErivative $dX_t$ Of The FlOw $$\begIN{alignEd} Dx_t\cDot E^*_x=E^*_{X_T(x)},\quad x\IN\LaMbDA, \QUaD t\in{{\mathbb R}},\quad *=s,X,u;\EnD{ALiGned}$$ $E^s$ is UniforMLy CoNTracted bY $Dx_t$ aNd $E^u$ IS UnifoRmly EXpAnded: theRe are $K,\LAmBdA>0$ so that $$\BeGin{aliGnEd} \lAbeL{eq:deF-HypeRbolic} \|dX_t\mid_{E^s_X}\|\le K e^{-\LAmbda t}, \quad \|DX_{-t} \mID_{E^u_x}\|\le K e^{-\lambdA T}, \qUAD x\IN\LamBda, \Quad t\in{{\mathBb R}}.\eND{aliGned}$$ wE sAy tHAt a poInt $x \iN M$ IS [*nONwandering*]{} for $X$ proviDeD for evEry neIghborhood $U$ of $X$ there is $t > 0$ sUCH That $X_t(x) \cAp U \nEQ \eMPtyset$, i.e., there eXists A point $y \in U$ WIth $X_t(y) \in u$. We deNote by $\OmEga(X

such that $X(\sigma)=0$ o r, equival ently , $ X_t (\ sigm a)=\ sigma$ for all $t \ in {{\mathbb R}}$. The setfo r medb ysingu laritie s i s the * si ngu la r s et of $X $* deno ted ${\mat hrm {S ing}}(X)$ an d $ {\operator nam e{Per}}(X)$isthe se tofp eriod icpoint s of $ X $. Wesay thata* singul a rity is h yp erbo lic* if the eigen v al u es of the deri vative $ D X( \ s igm a)$ of the ve ct or fi e ld at t h es i n gul a rity $\sigma$ have nonze r o r eal pa rt . T h e setof cr it i cal elements o f $X $ is theuniono f the s i ngulari ties a ndthe per i od ic or bi t s o f $ X$, and will be d en otedby $ { \ o p erat orn ame{ Crit} }(X)$. We re cal l th a t a n hyp erbol ic s et $\La mbda$for a f low $X_t$ is an inv ariant su bse tof$M $ wit h a dec omp osi tion $T _\Lambd a M= E ^ s \ op lus E^X \oplus E^u $o f t he tange nt bun d le o v er $\Lam bd a$whic h is acont i nu ous spli tting, wh er e $E^X$ i s thedi rec tio n oft he v ectorfield, t he su b bundles are in v ariant undert he d er i vati ve$DX_t$ of t he f l ow $ $\be g in {al i gned} DX _t \ cd o t E^*_x=E^*_{X_t(x) }, \quad x\i n\Lambda, \qu ad t\in{{\ m a t hbb R}}, \qua d * = s,X,u;\end{ali gned} $$ $E^s$ i s uniform ly co ntracted by $DX_t $ and $E^u $ i s u nif orm l y e xpanded: ther e are$K ,\lambd a>0 $ so th at$$\ beg in{ al igned} \l abel{eq: de f- hy pe rbo lic}\ |DX_t\mi d_ {E^ s_ x}\ |\leK e^{-\ lambd a t} , \ qua d \|D X _{ - t } \m id _{ E^u_ x}\ |\ le Ke^{- \ lam bda t}, \quadx\i n \Lam bd a, \quadt\in{{\mathbb R }}.\end{al ig ned }$$ W e say that a point $x \in M$ is[ *nonwan der ing*] {} f or $X$ pr ovi ded fo r e v ery ne ighbor hood$U $ o f $x$ t h e re is $ t > 0$ suc h tha t $X_ t( x) \ cap U \ neq \emptyset$, i. e .,there existsa p oint $ y\in U$ wit h$ X_t ( y ) \in U$. We de note by $\ Om e ga (X

such_that $X(\sigma)=0$_or, equivalently, $X_t(\sigma)=\sigma$ for_all $t_\in_{{\mathbb R}}$._The_set formed by_singularities is the_*singular set of $X$*_denoted ${\mathrm{Sing}}(X)$ and_${\operatorname{Per}}(X)$_is the set of periodic points of $X$. We say that a *singularity is_hyperbolic*_if the_eigenvalues_of_the derivative $DX(\sigma)$ of the_vector field at the singularity_$\sigma$ have_nonzero real part. The set of critical elements_of_$X$ is the_union of the singularities and the periodic orbits of_$X$, and will be denoted by_${\operatorname{Crit}}(X)$. We recall that_an_hyperbolic_set $\Lambda$ for a_flow $X_t$ is an invariant subset_of $M$ with a decomposition $T_\Lambda_M= E^s\oplus E^X \oplus E^u$ of the_tangent bundle over $\Lambda$ which is_a continuous splitting, where $E^X$_is the_direction of the vector field,_the subbundles are_invariant under_the derivative $DX_t$_of the flow $$\begin{aligned} DX_t\cdot_E^*_x=E^*_{X_t(x)},\quad x\in\Lambda,_\quad t\in{{\mathbb R}},\quad *=s,X,u;\end{aligned}$$ $E^s$ is_uniformly_contracted by $DX_t$_and_$E^u$_is uniformly_expanded: there are_$K,\lambda>0$_so that_$$\begin{aligned} \label{eq:def-hyperbolic} \|DX_t\mid_{E^s_x}\|\le_K e^{-\lambda t}, \quad _\|DX_{-t}_\mid_{E^u_x}\|\le K e^{-\lambda t}, \quad x\in\Lambda,_\quad t\in{{\mathbb R}}.\end{aligned}$$ We say_that_a point $x \in_ M$ is [*nonwandering*]{} for_$X$ provided for every neighborhood $U$_of $x$_there is_$t > 0$ such that $X_t(x) \cap U \neq \emptyset$, i.e.,_there exists a point $y \in_U$ with $X_t(y) \in_U$. We_denote_by $\Omega(X

$C$ has no first generation ancestors, i.e. $\mathcal{A}^\Lambda_1(C)=\emptyset$, then the value of its flag $u$ alone will determine whether $C$ is kept or not. Otherwise, the value of $u$ will decide if $C$ is kept once we determine the fate of all the first generation ancestors of $C$. 2. To decide whether any given first generation ancestor $\tilde{C} \in \mathcal{A}^\Lambda_1(C)$ is kept, one must repeat step (i) for $\tilde{C}$ instead of $C$. 3. Since the clan of ancestors of $C$ restricted to $\Lambda$ possesses only a finite number of generations, one must go backwards in time and examine a previous generation of ancestors only a finite amount of times (at most $N$ times) so that it is ultimately possible to determine whether $C$ is kept. Therefore, we are left to answer the question of under which conditions do the clans of ancestors restricted to $\Lambda$ possess only a finite amount of generations. Fortunately, under Assumptions \[assump\] this is always the case. This is the content of the following proposition. \[localfinit\] For every $\Lambda \in {{\mathcal B}}^0_S$ we have that $\A^{t,\Lambda}(\Lambda \times G)$ is finite for all $t \in {{\mathbb R}}$. Since $\Pi$ is a stationary process whose invariant measure $\pi^\nu$ satisfies $$\pi^\nu( \{ \xi \in \mathcal{N}(S \times G) : \xi( \Lambda \times G) = 0 \} ) = e^{-\nu(\Lambda \times G)} > 0$$ we have that the entrance times $(t_i(\Lambda))_{i \in {{\mathbb{Z}}}}$ to the set $\{ \xi \in \mathcal{N}(S \times G) : \xi( \Lambda \times G) = 0 \}$ are well defined and satisfy $t_i(\Lambda) \rightarrow \pm \infty$ as $i \rightarrow \pm \infty$. In particular, for every $t \in {{\mathbb R}}$ there exists $i_0 \in {{\mathbb{Z}}}$ such that $t_{i_0-1}(\Lambda) \leq t < t_{i_0}(\Lambda)$. Since $\Pi_{t_i

$ C$ has no first generation ancestors, i.e. $ \mathcal{A}^\Lambda_1(C)=\emptyset$, then the value of its pin $ u$ entirely will determine whether $ C$ is kept or not. differently, the value of $ u$ will decide if $ C$ is kept once we decide the fate of all the inaugural generation ancestors of $ C$. 2. To decide whether any contribute first generation ancestor $ \tilde{C } \in \mathcal{A}^\Lambda_1(C)$ is keep, one must repeat footprint (iodine) for $ \tilde{C}$ instead of $ C$. 3. Since the clan of ancestors of $ C$ restricted to $ \Lambda$ possesses merely a finite number of generations, one must go backward in time and examine a previous generation of ancestor only a finite measure of times (at most $ N$ times) so that it is ultimately potential to determine whether $ C$ is kept. Therefore, we are left to answer the question of under which conditions do the clans of ancestors restricted to $ \Lambda$ possess only a finite amount of generations. Fortunately, under Assumptions \[assump\ ] this is always the case. This is the content of the follow proposition. \[localfinit\ ] For every $ \Lambda \in { { \mathcal B}}^0_S$ we have that $ \A^{t,\Lambda}(\Lambda \times G)$ is finite for all $ t \in { { \mathbb R}}$. Since $ \Pi$ is a stationary procedure whose invariant measure $ \pi^\nu$ satisfies $ $ \pi^\nu (\ { \xi \in \mathcal{N}(S \times G): \xi (\Lambda \times G) = 0 \ }) = e^{-\nu(\Lambda \times G) } > 0$$ we have that the entrance times $ (t_i(\Lambda))_{i \in { { \mathbb{Z}}}}$ to the fixed $ \ { \xi \in \mathcal{N}(S \times G): \xi (\Lambda \times G) = 0 \}$ are well specify and satisfy $ t_i(\Lambda) \rightarrow \pm \infty$ as $ i \rightarrow \pm \infty$. In particular, for every $ thyroxine \in { { \mathbb R}}$ there exists $ i_0 \in { { \mathbb{Z}}}$ such that $ t_{i_0 - 1}(\Lambda) \leq t < t_{i_0}(\Lambda)$. Since $ \Pi_{t_i

$C$ jas no first generation xncestors, i.e. $\majhxal{A}^\Lakbda_1(C)=\ejptyset$, ghen the value of its flag $u$ aoone qill determine whether $C$ is kepn or not. Ithecwise, the value of $u$ will decisc if $E$ ms kept once we determine the fate of anl tke first generation ancestors of $C$. 2. Eo decice whether any gyven sirsf generation ancestor $\tilde{C} \in \mafhcal{A}^\Lembda_1(C)$ is kept, pne must repeat step (i) for $\tilfe{C}$ instead of $C$. 3. Dince the coan jd ancestors uf $C$ restrpeted to $\Lamgda$ possesses only a finite numcer oy generatiobs, onf must go barkwardf in time and examina a prefious generatipn mf qncestors only a finive amount of times (aj most $N$ thmzs) so that it is ultinarely kossitle gi ddtedmmne whethfr $R$ is kept. Thsrefore, we qre left to answer uhe wuestion of uhder wrish conditions do the clans of ancestors reatricted to $\Lambda$ posswss only a finite amoont of genqrations. Fortunately, under Assumptions \[assump\] thiv is eldayw bhe zqsf. This is the content of the following proposyfipn. \[kocalfinit\] For every $\Lambca \im {{\mathcal B}}^0_S$ wg have cgaf $\A^{t,\Lambda}(\Lambda \tlmes G)$ ys fibite for wll $y \in {{\mathbb R}}$. Since $\Pi$ is a wtationary pgocews whose invariant measure $\pi^\uu$ satosfiex $$\pi^\nu( \{ \xi \in \mathcal{N}(S \timss G) : \xi( \Lalbda \timea G) = 0 \} ) = e^{-\nu(\Lambdx \tpmes G)} > 0$$ we have that the entrwnce timew $(t_i(\Kambda))_{i \in {{\kathbb{S}}}}$ to the sft $\{ \xl \in \mathcal{N}(S \timed G) : \ri( \Lakbda \times G) = 0 \}$ are well defined and satisfy $t_i(\Lambda) \rogvtagrow \pm \iufty$ ax $i \rightarror \pm \infty$. In karticulax, for dvery $t \in {{\mathbb R}}$ there exysts $i_0 \in {{\matvhb{Z}}}$ such thav $t_{i_0-1}(\Lambdw) \lew t < t_{i_0}(\Lambax)$. Since $\Pi_{t_i

$C$ has no first generation ancestors, i.e. the of its $u$ alone will or Otherwise, the value $u$ will decide $C$ is kept once we determine fate of all the first generation ancestors of $C$. 2. To decide whether given first generation ancestor $\tilde{C} \in \mathcal{A}^\Lambda_1(C)$ is kept, one must repeat step for instead $C$. Since the clan of ancestors of $C$ restricted to $\Lambda$ possesses only a finite number of one must go backwards in time and examine previous generation of ancestors a finite amount of times most times) so it ultimately to determine whether is kept. Therefore, we are left to answer the question of under which conditions do the clans ancestors restricted possess only finite of Fortunately, under Assumptions is always the case. This is the following proposition. \[localfinit\] For every $\Lambda \in B}}^0_S$ we that $\A^{t,\Lambda}(\Lambda \times G)$ is finite all $t \in {{\mathbb R}}$. Since $\Pi$ is stationary process whose invariant measure $\pi^\nu$ satisfies $$\pi^\nu( \{ \xi \in \mathcal{N}(S \times G) : \times G) = 0 ) = e^{-\nu(\Lambda G)} 0$$ have the entrance $(t_i(\Lambda))_{i \in {{\mathbb{Z}}}}$ to the set $\{ \xi \in \mathcal{N}(S \times : \xi( \Lambda \times G) = 0 \}$ are well satisfy \rightarrow \pm \infty$ $i \rightarrow \pm \infty$. particular, every $t \in {{\mathbb exists \in $t_{i_0-1}(\Lambda) t t_{i_0}(\Lambda)$. Since $\Pi_{t_i

$C$ has no first generation anceStors, i.e. $\matHcal{A}^\lamBda_1(c)=\eMptySet$, tHen the value of iTS flaG $u$ alone will determine whEther $c$ iS Kept OR nOt. OthErwise, tHE vALUe oF $u$ WiLl dEcIDe If $C$ is KepT once we Determine tHe fAtE of all the firST gEneration aNceStors of $C$. 2. To deCidE whethEr Any GIven fIrsT geneRation ANcestoR $\tilde{C} \in \MaTHcal{A}^\LAMbda_1(C)$ is KEPt, One mUst repeat step (i) for $\TIlDE{C}$ instead of $C$. 3. SiNce the ClAN oF ANceStoRs of $C$ restrIcTed to $\lAmbda$ poSSeSSES onLY a finite numbeR of generatiONs, oNe must Go BacKWards iN time AnD ExaMine a previoUs geNeration oF ancesTOrs only A Finite aMount oF tiMes (At moST $N$ TiMes) So THat IT iS ulTImaTely possIbLe To detErmiNE WHEtheR $C$ iS kepT. TherEfore, we are lefT to AnswER thE quesTion oF undEr Which ConditIons dO tHe clans of ancestOrs rEstricted To $\LAmBda$ PoSsess ONly a fiNitE amOunt of gEneratiONs. FOrTUNAtEly, under AssumptionS \[aSSUmP\] this is aLways tHE cAsE. this is thE cOntEnt oF THe folLowiNG pRopositiOn. \[locaLFiNiT\] For eveRy $\lambda \In {{\MatHcaL B}}^0_S$ we HAve tHat $\A^{t,\LAmbda}(\LamBda \tiMEs G)$ is finite for ALl $t \in {{\mathbb R}}$. SINcE $\pI$ iS A staTioNary process WhosE InvaRianT MeAsuRE $\pi^\nu$ SatisFiES $$\pI^\Nu( \{ \xi \in \mathcal{N}(S \timeS G) : \Xi( \LambDa \timEs G) = 0 \} ) = e^{-\nu(\Lambda \tImes G)} > 0$$ we havE THAt the entRancE TiMEs $(t_i(\Lambda))_{i \in {{\mAthbb{z}}}}$ to the set $\{ \xI \In \mathcaL{N}(S \tiMes G) : \xi( \LaMbda \times g) = 0 \}$ ARe well deFinEd aNd sAtiSFY $t_I(\Lambda) \rightaRROw \pm \InFty$ as $i \rIghTarrow \pM \inFty$. in pArtIcUlar, for evEry $t \in {{\maThBb r}}$ tHeRe eXists $I_0 \In {{\mathbb{z}}}$ sUch ThAt $t_{I_0-1}(\LambDA) \leq t < t_{I_0}(\LambDa)$. SiNcE $\PI_{T_i

$C$ has no first generati on ancesto rs, i .e. $\ ma thca l{A} ^\Lambda_1(C)= \ empt yset$, then the valueof it sf lag$ u$ alon e willd et e r min ewh eth er $C $ iskep t or no t. Otherwi se, t he value of$ u$ will deci deif $C$ is ke ptonce w edet e rmine th e fat e of a l l thefirst gen er a tion a n cestors o f$C$. 2. To decide w h et h er any given f irst g en e ra t i onanc estor $\ti ld e{C}\ in \mat h ca l { A }^\ L ambda_1(C)$ i s kept, one mus t repe at st e p (i)for $ \t i lde {C}$ instea d of $C$. 3. Sinc e the cl a n of an cestor s o f $ C$ r e st ri cte dt o $ \ La mbd a $ p ossesses o nl y a f init e n u mber of gen erati ons, one must go bac k war ds in time and e xamin e a pr eviou sgeneration of a nces tors only afi nit eamoun t of ti mes (a t most$N$ tim e s)so t h at it is ultimatelypo s s ib le to de termin e w he t her $C$is ke pt.T heref ore, we are lef t to a n sw er the qu es tion o fund erwhich cond itions do theclans of ancestors r e stricted to $ \ La m b da $ pos ses s only a fi nite amou nt o f g ene r ation s. Fo rt u na t ely, under Assumpti on s \[as sump\ ] this is alw ays the ca s e . This is the co n tent of the fo llowi ng proposi t ion. \[ local finit\]For every $ \Lambda\in {{ \ma thc a l B }}^0_S$ we ha v e tha t$\A^{t, \La mbda}(\ Lam bda \t ime sG)$ is fi nite for a ll $ t\in {{\m a thbb R}} $. S in ce$\Pi$ is a s tatio nary p ro c ess whosei nv a r iant m ea sure $\ pi ^\nu$ sat i sfi es $$\p i^\nu( \{ \x i \in \ ma thcal{N }(S \times G) : \xi( \Lam bd a \ timesG ) = 0 \}) = e^{-\nu(\Lambda \ti m es G)}> 0 $$ we hav e that th e e ntranc e t i mes $( t_i(\L ambda )) _{i \ in {{ \ m at hbb {Z }}}}$ to t h e se t $\{ \ xi \ in \mat hcal{N}(S \times G ) :\xi( \Lambda\ti mesG ) = 0\ }$ are w e lld e fined and satis fy $t_i(\L am b da ) \rightar r ow\p m \inft y$ as $ i \ri g htarrow \pm \inf ty$. In p ar ticu l a r,for every$t \in { {\mathbbR }}$ t h er e exi sts $i_0\i n { {\mat hbb{Z} } }$suchthat $ t_ {i_0-1 }(\La mb da) \leq t < t_{i_0}(\Lambda)$. Since $\Pi _{t _i

$C$_has no_first generation ancestors, i.e._$\mathcal{A}^\Lambda_1(C)=\emptyset$, then_the_value of_its_flag $u$ alone_will determine whether_$C$ is kept or_not. Otherwise, the_value_of $u$ will decide if $C$ is kept once we determine the fate of_all_the first_generation_ancestors_of $C$. 2. To decide_whether any given first generation_ancestor $\tilde{C}_\in \mathcal{A}^\Lambda_1(C)$ is kept, one must repeat step_(i)_for $\tilde{C}$ instead_of $C$. 3. Since the clan of ancestors of_$C$ restricted to $\Lambda$ possesses only_a finite number_of_generations,_one must go backwards_in time and examine a previous_generation of ancestors only a finite_amount of times (at most $N$ times)_so that it is ultimately possible_to determine whether $C$ is_kept. Therefore, we_are left to answer the_question of under_which conditions_do the clans_of ancestors restricted to $\Lambda$ possess_only a finite_amount of generations. Fortunately, under Assumptions_\[assump\]_this is always_the_case._This is_the content of_the_following proposition. \[localfinit\]_For_every $\Lambda \in {{\mathcal B}}^0_S$ we_have_that $\A^{t,\Lambda}(\Lambda \times G)$ is finite for_all $t \in {{\mathbb_R}}$. Since_$\Pi$ is a stationary_process whose invariant measure $\pi^\nu$_satisfies $$\pi^\nu( \{ \xi \in \mathcal{N}(S_\times G)_: \xi(_\Lambda \times G) = 0 \} ) = e^{-\nu(\Lambda \times G)}_> 0$$ we have that the_entrance times $(t_i(\Lambda))_{i \in_{{\mathbb{Z}}}}$ to_the_set $\{ \xi_\in_\mathcal{N}(S \times_G) : \xi( \Lambda \times G) =_0 \}$_are well defined and satisfy $t_i(\Lambda)_\rightarrow \pm \infty$ as_$i_\rightarrow \pm \infty$. In particular, for_every $t \in {{\mathbb R}}$ there_exists $i_0 \in {{\mathbb{Z}}}$ such_that_$t_{i_0-1}(\Lambda)_\leq t < t_{i_0}(\Lambda)$. Since_$\Pi_{t_i

bottom) panel corresponds to the 2+2 (3+1) configuration. The first orbital sampling method (lognormal orbital period and Rayleigh eccentricity distributions; see [Fig.]{}\[fig:IC\_comp\_per1\]) agrees reasonably with the MSC for periods $\gtrsim 10^3\,{\mathrm{d}}$. There is a clear excess of systems with inner periods between $\sim 1$ and $\sim 10^2\,{\mathrm{d}}$, as noted before by @2008MNRAS.389..925T. Evidently, our aim in §\[sect:pop\_syn\] is to establish whether tidal migration coupled with secular evolution can reproduce such an excess of short-period systems. The second sampling method (flat distributions in $\log_{10} a_i$ and the eccentricities; see [Fig.]{}\[fig:IC\_comp\_per2\]) compares less favourably to the MSC at periods $\gtrsim 10^3\,{\mathrm{d}}$, especially for the outermost orbits. The median $m_1$ in the extracted sample of the MSC agrees with the assumed $m_1=1\,{\mathrm{M}_\odot}$ (see [Fig.]{}\[fig:IC\_comp\_m\]). The other observed masses, $m_2$, $m_3$ and $m_4$, tend to be somewhat larger than the sampled masses. However, we do not anticipate a very strong dependence of our results on the masses. Population synthesis results {#sect:pop_syn} ============================ Here, we present the results from the population synthesis simulations. The initial conditions were discussed in §\[sect:IC\]. We first focus on the occurrences of tidal migration and other outcomes (§\[sect:pop\_syn:frac\]). We then consider the inner orbital period distributions mediated by tidal and secular evolution (§\[sect:pop\_syn:in\]). Other orbital properties are considered in §\[sect:pop\_syn:orb\], and in §\[sect:pop\_syn:time\] we discuss the migration times. Outcome fractions {#sect:pop_syn:frac} ----------------- ----------------------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- NF F NF F NF F NF F NF F NF F

bottom) panel corresponds to the 2 + 2 (3 + 1) configuration. The first orbital sample distribution method acting (lognormal orbital menstruation and Rayleigh eccentricity distributions; visualize [ Fig.]{}\[fig: IC\_comp\_per1\ ]) agree reasonably with the MSC for periods $ \gtrsim 10 ^ 3\,{\mathrm{d}}$. There be a absolved excess of organization with inner periods between $ \sim 1 $ and $ \sim 10 ^ 2\,{\mathrm{d}}$, as note before by @2008MNRAS.389.. 925T. Evidently, our aim in § \[sect: pop\_syn\ ] is to lay down whether tidal migration pair with secular evolution can reproduce such an excess of short - time period systems. The second sampling method (bland distributions in $ \log_{10 } a_i$ and the eccentricities; see [ Fig.]{}\[fig: IC\_comp\_per2\ ]) compares less favorably to the MSC at periods $ \gtrsim 10 ^ 3\,{\mathrm{d}}$, specially for the outermost orbits. The median $ m_1 $ in the extracted sample distribution of the MSC agrees with the assumed $ m_1=1\,{\mathrm{M}_\odot}$ (see [ Fig.]{}\[fig: IC\_comp\_m\ ]). The other observed masses, $ m_2 $, $ m_3 $ and $ m_4 $, tend to be slightly larger than the sampled masses. However, we do not anticipate a very strong dependence of our results on the masses. Population synthesis results { # faction: pop_syn } = = = = = = = = = = = = = = = = = = = = = = = = = = = = Here, we salute the results from the population deduction model. The initial condition were discussed in § \[sect: IC\ ]. We first concentrate on the occurrences of tidal migration and other outcomes (§ \[sect: pop\_syn: frac\ ]). We then consider the inside orbital period distributions mediated by tidal and secular evolution (§ \[sect: pop\_syn: in\ ]). Other orbital properties are considered in § \[sect: pop\_syn: orb\ ], and in § \[sect: pop\_syn: time\ ] we discuss the migration times. Outcome fractions { # faction: pop_syn: frac } ----------------- ----------------------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- NF F NF F NF F NF F NF F NF F

bothom) panel corresponds to the 2+2 (3+1) configurcrion. Thx first orbital sampling method (lognormal ocbitql peeiod and Rayleigh eccevtricity fistriburionw; see [Fig.]{}\[fij:JC\_comp\_pcx1\]) agrscs recsinably with thg MSC for pesiods $\gtrsim 10^3\,{\mdtfrl{d}}$. There is a clear excess of systeis with ijner periods bgtweem $\sim 1$ and $\sim 10^2\,{\mathrm{d}}$, as noted before gy @2008MNRAV.389..925T. Evidently, pur aim in §\[sect:pop\_syn\] is tl eshablish whether tifal migratiin cjypled with sdcular evolution can rgproduce such an excess of short-oeriob systems. Tye sefmnd samplinj methjd (flat distvpbutionv in $\lob_{10} a_i$ and the egcentciciries; see [Fig.]{}\[fig:IC\_comp\_'er2\]) compares less fadourably do the MSC at perioes $\gtrshm 10^3\,{\mdthro{e}}$, erpediellg for hhe outermost orbits. The nedian $m_1$ in the exttasnrd sample of the MFC agrees with the assumed $m_1=1\,{\mathrm{M}_\odot}$ (ste [Fif.]{}\[fig:IC\_comp\_m\]). The other ovserved masses, $m_2$, $m_3$ anf $m_4$, tend eo be somewhat larger than the sampled masses. Howaver, xe do njg ajticipate a very strong dependence of our restmtx pn the masses. Pjpulation sunhhrfis results {#sgct:pop_svh} ============================ Hsre, we present the resultf fron the poptlatoon synthesis simulations. Tye initial cjbditions were discbssed in §\[secc:IC\]. We firsy focus on the occurrenees of tidal migrwtion and uther outcomes (§\[sdct:iop\_sfn:frac\]). Wt then consider thq inner ocbitak perioa dixtribueions mediwted nf tidal and seculag evopudion (§\[sect:plp\_syn:in\]). Other orbital properties are considergd hn §\[vect:pop\_svn:orb\], snd in §\[sect:po[\_syn:time\] we dixcuss tke migfation timvs. Outcome fractions {#fect:pop_syn:fraw} ----------------- ----------------------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- NF F NF F NF F NF F NG F VF F

bottom) panel corresponds to the 2+2 (3+1) first sampling method orbital period and agrees with the MSC periods $\gtrsim 10^3\,{\mathrm{d}}$. is a clear excess of systems inner periods between $\sim 1$ and $\sim 10^2\,{\mathrm{d}}$, as noted before by @2008MNRAS.389..925T. our aim in §\[sect:pop\_syn\] is to establish whether tidal migration coupled with secular can such excess short-period systems. The second sampling method (flat distributions in $\log_{10} a_i$ and the eccentricities; see [Fig.]{}\[fig:IC\_comp\_per2\]) less favourably to the MSC at periods $\gtrsim especially for the outermost The median $m_1$ in the sample the MSC with assumed (see [Fig.]{}\[fig:IC\_comp\_m\]). The observed masses, $m_2$, $m_3$ and $m_4$, tend to be somewhat larger than the sampled masses. However, we not anticipate strong dependence our on masses. Population synthesis ============================ Here, we present the results synthesis simulations. The initial conditions were discussed in We first on the occurrences of tidal migration other outcomes (§\[sect:pop\_syn:frac\]). We then consider the inner period distributions mediated by tidal and secular evolution (§\[sect:pop\_syn:in\]). Other orbital properties are considered in in §\[sect:pop\_syn:time\] we discuss migration times. Outcome {#sect:pop_syn:frac} ----------------------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- NF NF F NF F NF F NF F NF F

bottom) panel corresponds to tHe 2+2 (3+1) configurAtion. the FirSt OrbiTal sAmpling method (lOGnorMal orbital period and RayLeigh EcCEntrICiTy disTributiONs; SEE [FiG.]{}\[fIg:iC\_cOmP\_PeR1\]) agreEs rEasonabLy with the MsC fOr Periods $\gtrsiM 10^3\,{\MaThrm{d}}$. There Is a Clear excess oF syStems wItH inNEr perIodS betwEen $\sim 1$ ANd $\sim 10^2\,{\mAthrm{d}}$, as nOtED beforE By @2008MNRAS.389..925t. eViDentLy, our aim in §\[sect:pop\_SYn\] IS to establish whEther tIdAL mIGRatIon Coupled witH sEculaR EvolutiON cAN REprODuce such an excEss of short-pERioD systeMs. the SEcond sAmpliNg MEthOd (flat distrIbutIons in $\log_{10} A_i$ and tHE eccentRIcities; See [Fig.]{}\[Fig:iC\_cOmp\_pER2\]) cOmParEs LEss FAvOurABly To the MSC At PeRiods $\GtrsIM 10^3\,{\MAThrm{D}}$, esPeciAlly fOr the outermosT orBits. tHe mEdian $M_1$ in thE extRaCted sAmple oF the MsC Agrees with the asSumeD $m_1=1\,{\mathrm{M}_\OdoT}$ (sEe [FIg.]{}\[Fig:IC\_COmp\_m\]). ThE otHer ObserveD masses, $M_2$, $M_3$ anD $m_4$, TEND tO be somewhat larger tHaN THe Sampled mAsses. HOWeVeR, We do not aNtIciPate A VEry stRong DEpEndence oF our reSUlTs On the maSsEs. PopuLaTioN syNthesIS resUlts {#seCt:pop_syn} ============================ here, wE Present the resuLTs from the popuLAtION sYNtheSis Simulations. the iNItiaL conDItIonS Were dIscusSeD In §\[SEct:IC\]. We first focus on ThE occurRenceS of tidal migraTion and othER OUtcomes (§\[sEct:pOP\_sYN:frac\]). We then conSider The inner orBItal periOd disTributioNs mediateD BY tidal anD seCulAr eVolUTIoN (§\[sect:pop\_syn:in\]). oTHer oRbItal proPerTies are ConSidEreD in §\[SeCt:pop\_syn:oRb\], and in §\[sEcT:pOp\_SyN:tiMe\] we dIScuss the MiGraTiOn tImes. OUTcome fRactiOns {#sEcT:pOP_syN:frac} ----------------- ----------------------------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- ------- NF f nF f nf F NF f Nf F nF F

bottom) panel correspondsto the 2+2 (3+1 ) c onf ig urat ion. The first or b ital sampling method (logn ormal o r bita l p eriod and Ra y le i g h e cc en tri ci t ydistr ibu tions;see [Fig.] {}\ [f ig:IC\_comp\ _ pe r1\]) agre esreasonably w ith the M SC fo r peri ods $\gt rsim 1 0 ^3\,{\ mathrm{d} }$ . There is a cl e a rexce ss of systems wit h i n ner periods be tween$\ s im 1 $ a nd$\sim 10^2 \, {\mat h rm{d}}$ , a s n ote d before by @2 008MNRAS.38 9 ..9 25T. E vi den t ly, ou r aim i n §\ [sect:pop\_ syn\ ] is to e stabli s h wheth e r tidal migra tio n c oupl e dwi thse c ula r e vol u tio n can re pr od uce s ucha n e xces s o f sh ort-p eriod systems . T he s e con d sam pling met ho d (fl at dis tribu ti ons in $\log_{1 0} a _i$ and t heec cen tr iciti e s; see [F ig. ]{}\[fi g:IC\_c o mp\ _p e r 2 \] ) compares less fa vo u r ab ly to th e MSCa tpe r iods $\g tr sim 10^ 3 \ ,{\ma thrm { d} }$, espe cially fo rthe out er most o rb its . The m e dian $m_1$ in theextra c ted sample oft he MSC agrees wi t h t h e as sum ed $m_1=1\, {\ma t hrm{ M}_\ o do t}$ (see[Fig. ]{ } \[ f ig:IC\_comp\_m\]).Th e othe r obs erved masses, $m_2$, $m _ 3 $ and $m_ 4$,t en d to be somewha t lar ger than t h e sample d mas ses. How ever, wed o not ant ici pat e a ve r y s trong depende n c e of o ur resu lts on the ma sse s. Po pu lation sy nthesisre su lt s{#s ect:p o p_syn} = == === == === ===== = ====== ===== He re ,w e p resentt he r esul ts f romthe p opula tion syn thesissimulatio ns. Thein it ial con ditions weredi scussed in § \[s ect:IC \ ] . We fir st focus on the occurre n ces oftid al mi grat ion and o the r outc ome s (§\[s ect:po p\_sy n: fra c \ ]). W e th enco nsider the i nne r orb it al p eriod d istributions media t edby tidal andsec ular e vo lut i on (§\ [s e ct: p o p\_syn:in\]). O ther orbit al pr operties a r e c on sidered in §\[ sect: p op\_syn :orb\], a nd in §\[ se ct:p o p \_s yn:time\]we discu ss the mi g ratio n t imes. O utcome f rac tions {#sec t :po p_syn :frac} - ------ ----- -- --- - ----------------------- ---------- --------- - --- - --------- - ---- -- ------- -- --- -- -- --- - - --- ---- -- --- - - --- ---- ------- - - -- --- - -- ---- ------ - - --- --- NF F N F F NF F N F F NF F NF F

bottom) panel_corresponds to_the 2+2 (3+1) configuration. The_first orbital_sampling_method (lognormal_orbital_period and Rayleigh_eccentricity distributions; see_[Fig.]{}\[fig:IC\_comp\_per1\]) agrees reasonably with_the MSC for_periods_$\gtrsim 10^3\,{\mathrm{d}}$. There is a clear excess of systems with inner periods between $\sim_1$_and $\sim_10^2\,{\mathrm{d}}$,_as_noted before by @2008MNRAS.389..925T. Evidently,_our aim in §\[sect:pop\_syn\] is_to establish_whether tidal migration coupled with secular evolution can_reproduce_such an excess_of short-period systems. The second sampling method (flat distributions_in $\log_{10} a_i$ and the eccentricities;_see [Fig.]{}\[fig:IC\_comp\_per2\]) compares_less_favourably_to the MSC at_periods $\gtrsim 10^3\,{\mathrm{d}}$, especially for the_outermost orbits. The median $m_1$ in the_extracted sample of the MSC agrees with_the assumed $m_1=1\,{\mathrm{M}_\odot}$ (see [Fig.]{}\[fig:IC\_comp\_m\]). The_other observed masses, $m_2$, $m_3$_and $m_4$,_tend to be somewhat larger_than the sampled_masses. However,_we do not_anticipate a very strong dependence of_our results on_the masses. Population synthesis results {#sect:pop_syn} ============================ Here, we_present_the results from_the_population_synthesis simulations._The initial conditions_were_discussed in_§\[sect:IC\]._We first focus on the occurrences_of_tidal migration and other outcomes (§\[sect:pop\_syn:frac\]). We_then consider the inner_orbital_period distributions mediated by_tidal and secular evolution (§\[sect:pop\_syn:in\])._Other orbital properties are considered in_§\[sect:pop\_syn:orb\], and_in §\[sect:pop\_syn:time\]_we discuss the migration times. Outcome fractions {#sect:pop_syn:frac} ----------------- ----------------------------- ------- -------_------- ------- ------- ------- ------- -------_------- ------- ------- -------_------- -------_-------_------- __ _ _ _ _ __ _ _ ___ _ _ _ __ __ ___ _ _ _ _ ___ _ _ _ _ _ _ _ _ __ ____ _ _ _ _ _ _ __ _ _ __ _ _ _ __ _ _ _ _ _ _ _ _ __ ____ _ _ _ __ _ _ _ _ _ _ _ __ _ _ __ ___ __ _____ _ __ _ __ _ _ __ ____ _ __ _ __ _ _ _ _ _ _ _ __ _ _ _ _ _ _ _ _ _ __ _ _ _ __NF _ _ F _ _NF F_ _ NF F _ NF F NF _ _F_ NF _F _

:tran1\]) and (\[eq:chan\_k\]), the received signal can be expressed as follows: $$\begin{aligned} \label{eq:rec_MP} r(t)=\sum_{k=1}^{N_u}\sqrt{\frac{E_k}{N_f}}\sum_{j=-\infty}^{\infty}d^{(k)}_j\, b^{(k)}_{\lfloor j/N_f\rfloor}u^{(k)}(t-jT_f-c^{(k)}_jT_c-\tau_k)+\sigma_nn(t),\end{aligned}$$ where $n(t)$ is a white Gaussian noise with zero mean and unit spectral density, and $$\begin{gathered} \label{eq:u_k} u^{(k)}(t)=\sum_{l=1}^{L}\alpha_l^{(k)}w_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$ with $w_{rx}(t)$ being the received UWB pulse with unit energy. We consider a Rake receiver for the user of interest, say user $1$, and express the template signal at the Rake receiver as follows: $$\begin{gathered} \label{eq:temp_RAKE} s^{(1)}_{temp}(t)=\sum_{j=iN_f}^{(i+1)N_f-1}d_j^{(1)}v(t-jT_f-c_j^{(1)}T_c),\end{gathered}$$ where $$\begin{gathered} \label{eq:v} v(t)=\sum_{l=1}^{L}\beta_lw_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$ with $\boldsymbol{\beta}=[\beta_1,...,\beta_L]$ being the Rake combining weights. The template signal given by (\[eq:temp\_RAKE\]) and (\[eq:v\]) can represent different multipath diversity combining schemes by choosing an appropriate weighting vector $\boldsymbol{\beta}$: In an $M$-finger Rake the weights for $(L-M)$ multipath components not used in the Rake receiver are set to zero while the remaining $M$ weights are determined according to the combining scheme, such as “Equal Gain Combining (EGC)" or “Maximum Ratio Combining (MRC)". The

: tran1\ ]) and (\[eq: chan\_k\ ]), the received signal can be expressed as follows: $ $ \begin{aligned } \label{eq: rec_MP } r(t)=\sum_{k=1}^{N_u}\sqrt{\frac{E_k}{N_f}}\sum_{j=-\infty}^{\infty}d^{(k)}_j\, b^{(k)}_{\lfloor j / N_f\rfloor}u^{(k)}(t - jT_f - c^{(k)}_jT_c-\tau_k)+\sigma_nn(t),\end{aligned}$$ where $ n(t)$ is a white Gaussian randomness with zero mean and unit of measurement spectral density, and $ $ \begin{gathered } \label{eq: u_k } u^{(k)}(t)=\sum_{l=1}^{L}\alpha_l^{(k)}w_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$ with $ w_{rx}(t)$ being the pick up UWB pulsation with unit energy. We consider a pitch telephone receiver for the user of interest, say user $ 1 $, and press out the template signal at the Rake receiver as follows: $ $ \begin{gathered } \label{eq: temp_RAKE } s^{(1)}_{temp}(t)=\sum_{j = iN_f}^{(i+1)N_f-1}d_j^{(1)}v(t - jT_f - c_j^{(1)}T_c),\end{gathered}$$ where $ $ \begin{gathered } \label{eq: v } v(t)=\sum_{l=1}^{L}\beta_lw_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$ with $ \boldsymbol{\beta}=[\beta_1,... ,\beta_L]$ being the pitch combining weights. The template signal given by (\[eq: temp\_RAKE\ ]) and (\[eq: v\ ]) can represent different multipath diversity combining scheme by choosing an appropriate weighting vector $ \boldsymbol{\beta}$: In an $ M$-finger Rake the weight for $ (L - M)$ multipath components not used in the Rake receiver are put to zero while the remaining $ M$ weights are determined according to the combining schema, such as “ Equal Gain Combining (EGC) " or “ Maximum Ratio Combining (MRC) ". The

:traj1\]) and (\[eq:chan\_k\]), the receivtd signal can be gxpressev as fomlows: $$\beein{aligned} \label{eq:rec_MP} r(t)=\sum_{k=1}^{I_u}\sqet{\frax{E_k}{N_f}}\sum_{j=-\infty}^{\infty}d^{(k)}_j\, c^{(k)}_{\lfloor j/J_f\rfloor}y^{(k)}(t-jU_f-c^{(k)}_jT_c-\tau_k)+\sigma_ni(f),\end{alinued}$$ wgcre $n(c)$ ms a white Gausxian noise with zero meat xnb unit spectral density, and $$\begin{gatrered} \lanep{eq:u_k} u^{(k)}(t)=\sum_{l=1}^{L}\ajpha_k^{(h)}w_{rx}\mvfu(t-(l-1)T_c\right),\end{gathered}$$ with $w_{rx}(t)$ bsing tht received UWB pulxe with unit energy. We conslder a Rake receiver flr the user of ybterest, say jser $1$, and tx'ress the tgmplate signal at the Rake receixer ax follows: $$\vetin{hdthered} \labeo{eq:teip_RAKE} s^{(1)}_{temp}(t)=\sum_{j=iN_f}^{(i+1)T_f-1}d_j^{(1)}v(t-jY_f-c_j^{(1)}T_c),\end{gatheved}$$ wiere $$\begin{gathered} \label{eq:t} v(t)=\sum_{l=1}^{L}\beta_lw_{rx}\left(j-(l-1)T_c\right),\etd{yathered}$$ with $\boldsymvoo{\beta}=[\teta_1,...,\teta_U]$ bekng tie Dake clmbmning weighfs. The templqte signal given by (\[ez:nrmp\_RAKE\]) and (\[sq:v\]) cag wepresent different multipath diversity cojbining schemes by choowing an appropriate wgighting vqctor $\boldsymbol{\beta}$: In an $M$-finger Rake the weighds foc $(U-M)$ nujgupwth components not used in the Rake receiver wde svt to zero while bhe remaining $M$ weogjtx are determingd accoxsihg to the combininh schemg, such as “Equal Gaim Combining (EGC)" or “Maximum Eatio Combinpng (NRC)". The

:tran1\]) and (\[eq:chan\_k\]), the received signal can as $$\begin{aligned} \label{eq:rec_MP} b^{(k)}_{\lfloor j/N_f\rfloor}u^{(k)}(t-jT_f-c^{(k)}_jT_c-\tau_k)+\sigma_nn(t),\end{aligned}$$ where noise zero mean and spectral density, and \label{eq:u_k} u^{(k)}(t)=\sum_{l=1}^{L}\alpha_l^{(k)}w_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$ with $w_{rx}(t)$ being the UWB pulse with unit energy. We consider a Rake receiver for the user interest, say user $1$, and express the template signal at the Rake receiver follows: \label{eq:temp_RAKE} where \label{eq:v} v(t)=\sum_{l=1}^{L}\beta_lw_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$ with $\boldsymbol{\beta}=[\beta_1,...,\beta_L]$ being the Rake combining weights. The template signal given by (\[eq:temp\_RAKE\]) and can represent different multipath diversity combining schemes by an appropriate weighting vector In an $M$-finger Rake the for multipath components used the receiver are set zero while the remaining $M$ weights are determined according to the combining scheme, such as “Equal Gain (EGC)" or Combining (MRC)".

:tran1\]) and (\[eq:chan\_k\]), the received Signal can bE exprEssEd aS fOlloWs: $$\beGin{aligned} \labeL{Eq:reC_MP} r(t)=\sum_{k=1}^{N_u}\sqrt{\frac{E_k}{N_F}}\sum_{j=-\InFTy}^{\inFTy}D^{(k)}_j\, b^{(k)}_{\Lfloor j/n_F\rFLOor}U^{(k)}(T-jt_f-c^{(K)}_jt_C-\tAu_k)+\siGma_Nn(t),\end{aLigned}$$ wherE $n(t)$ Is A white GaussiAN nOise with zeRo mEan and unit spEctRal denSiTy, aND $$\begiN{gaThereD} \label{EQ:u_k} u^{(k)}(t)=\Sum_{l=1}^{L}\alphA_l^{(K)}W_{rx}\lefT(T-(l-1)T_c\rigHT),\EnD{gatHered}$$ with $w_{rx}(t)$ beinG ThE Received UWB pulSe with UnIT eNERgy. we cOnsider a RaKe ReceiVEr for thE UsER OF inTErest, say user $1$, aNd express thE TemPlate sIgNal AT the RaKe recEiVEr aS follows: $$\begIn{gaThered} \labEl{eq:teMP_RAKE} s^{(1)}_{tEMp}(t)=\sum_{j=IN_f}^{(i+1)N_f-1}D_j^{(1)}v(T-jT_F-c_j^{(1)}T_C),\EnD{gAthErED}$$ whERe $$\BegIN{gaThered} \laBeL{eQ:v} v(t)=\sUm_{l=1}^{L}\BETA_Lw_{rx}\LefT(t-(l-1)T_C\righT),\end{gathered}$$ wIth $\BoldSYmbOl{\betA}=[\beta_1,...,\Beta_l]$ bEing tHe Rake CombiNiNg weights. The temPlatE signal giVen By (\[Eq:tEmP\_RAKE\]) ANd (\[eq:v\]) cAn rEprEsent diFferent MUltIpATH DiVersity combining scHeMES bY choosinG an appROpRiATe weightInG veCtor $\BOLdsymBol{\bETa}$: in an $M$-finGer RakE ThE wEights fOr $(l-M)$ multIpAth ComPonenTS not Used in The Rake rEceivER are set to zero wHIle the remainiNG $M$ WEIgHTs arE deTermined accOrdiNG to tHe coMBiNinG SchemE, such As “eQuAL Gain Combining (EGC)" or “maXimum RAtio COmbining (MRC)". ThE

:tran1\]) and (\[eq:chan\_ k\]), therecei ved si gn al c an b e expressed as foll ows: $$\begin{aligned} \lab el { eq:r e c_ MP} r (t)=\su m _{ k = 1}^ {N _u }\s qr t {\ frac{ E_k }{N_f}} \sum_{j=-\ inf ty }^{\infty}d^ { (k )}_j\, b^{ (k) }_{\lfloor j /N_ f\rflo or }u^ { (k)}( t-j T_f-c ^{(k)} _ jT_c-\ tau_k)+\s ig m a_nn(t ) ,\end{a l i gn ed}$ $ where $n(t)$ is aw hite Gaussiannoisewi t hz e romea n and unit s pectr a l densi t y, a n d $ $ \begin{gather ed} \label{ e q:u _k} u^ {( k)} ( t)=\su m_{l= 1} ^ {L} \alpha_l^{( k)}w _{rx}\lef t(t-(l - 1)T_c\r i ght),\e nd{gat her ed} $$ w i th $ w_{ rx } (t) $ b ein g th e receiv ed U WB pu lsew i t h uni t e nerg y. W e consider aRak e re c eiv er fo r the use rof in terest , say u ser $1$, and ex pres s the tem pla te si gn al at the Ra kerec eiver a s follo w s:$$ \ b e gi n{gathered} \label {e q : te mp_RAKE} s^{(1 ) }_ {t e mp}(t)=\ su m_{ j=iN _ f }^{(i +1)N _ f- 1}d_j^{( 1)}v(t - jT _f -c_j^{( 1) }T_c), \e nd{ gat hered } $$ w here $ $\begin{ gathe r ed} \label{eq: v } v(t)=\sum_{ l =1 } ^ {L } \bet a_l w_{rx}\left (t-( l -1)T _c\r i gh t), \ end{g ather ed } $$ with $\boldsymbol{\ be ta}=[\ beta_ 1,...,\beta_L ]$ being t h e Rake com bini n gw eights. The t empla te signalg iven by(\[eq :temp\_R AKE\]) an d (\[eq:v\ ])can re pre s e nt different mu l t ipat hdiversi tycombini ngsch eme s b ychoosingan appro pr ia te w eig hting vector $ \b old sy mbo l{\be t a}$: I n an$M$- fi ng e r R ake the we i g htsfo r$(L- M)$ m ultip athc omp onentsnot usedint he R ak ereceive r are set toze ro while t he re mainin g $M$ weig hts are determined acco r ding to th e com bini ng scheme , s uch as “E q ual Ga in Com binin g(EG C ) " or“ M ax imu mRatio Comb i n ing (MRC )" . T he

:tran1\]) and_(\[eq:chan\_k\]), the_received signal can be_expressed as_follows:_$$\begin{aligned} \label{eq:rec_MP} r(t)=\sum_{k=1}^{N_u}\sqrt{\frac{E_k}{N_f}}\sum_{j=-\infty}^{\infty}d^{(k)}_j\, b^{(k)}_{\lfloor j/N_f\rfloor}u^{(k)}(t-jT_f-c^{(k)}_jT_c-\tau_k)+\sigma_nn(t),\end{aligned}$$ where_$n(t)$_is a white_Gaussian noise with_zero mean and unit_spectral density, and_$$\begin{gathered} \label{eq:u_k} u^{(k)}(t)=\sum_{l=1}^{L}\alpha_l^{(k)}w_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$_with $w_{rx}(t)$ being the received UWB pulse with unit energy. We consider a Rake receiver_for_the user_of_interest,_say user $1$, and express_the template signal at the_Rake receiver_as follows: $$\begin{gathered} \label{eq:temp_RAKE} s^{(1)}_{temp}(t)=\sum_{j=iN_f}^{(i+1)N_f-1}d_j^{(1)}v(t-jT_f-c_j^{(1)}T_c),\end{gathered}$$ where $$\begin{gathered} \label{eq:v} v(t)=\sum_{l=1}^{L}\beta_lw_{rx}\left(t-(l-1)T_c\right),\end{gathered}$$ with $\boldsymbol{\beta}=[\beta_1,...,\beta_L]$ being_the_Rake combining weights. The_template signal given by (\[eq:temp\_RAKE\]) and (\[eq:v\]) can represent_different multipath diversity combining schemes by_choosing an appropriate_weighting_vector_$\boldsymbol{\beta}$: In an $M$-finger_Rake the weights for $(L-M)$ multipath_components not used in the Rake_receiver are set to zero while the_remaining $M$ weights are determined according_to the combining scheme, such_as “Equal_Gain Combining (EGC)" or “Maximum_Ratio Combining (MRC)". The

cosmic variance term. For example experiments currently being considered for measuring the B-modes of CMB polarization would be cosmic variance limited for E polarization over a wide range of $ls$ (e.g. [@bmodes]). Such an optimal experiment, exploring structures into a multipole range of $l=4000$, represents the limit of how much information on $\lambda$ one could extract from the CMB in principle. We found an expected error for $\lambda$ of order $0.3\%$ which translates into a constraint of the value of the gravitational constant during recombination of $\frac{\delta G}{G} \simeq 0.6 \%$. Discussion\[discussion\] ======================== We have made the expansion rate of the universe a free parameter in a likelihood analysis within an eight-dimensional cosmological parameter space. For simplicity we assumed that the gravitational constant is changed by a factor $\lambda$. We showed that an increase of $\lambda$ leads to a wider visibility function which in turn increases the damping of anisotropies on small scales and increases the level of large scale polarization. We calculated the constraints that current CMB data can impose on the expansion rate. The constraints that can be imposed on the parameter $\lambda$ using the information from the damping tail are severely weakened by our lack of knowledge about the shape of the primordial power spectrum. We showed that measuring polarization helps to break this degeneracy. Current data can only constrain $\lambda$ to about 47% at 1 $\sigma$. We showed that MAP could obtain 9% error bars for $\lambda$ while for Planck errorbars go down to 0.9%. We also explored the ultimate limit that could be achieved by a cosmic variance limited experiment measuring anisotropies up to $l=4000$ and found errors of under a percent in that case. Thus next generation experiments should be able to deliver very accurate constraints on the expansion rate of the universe during recombination. We acknowledge that if the variation of the gravitational constant during recombination is taken seriously a model needs to be built where $G$ changes after recombination and converges towards the stable value observed in laboratory experiments today and where its current rate of change is less than the experimental bound $\frac{\dot{G}}{G} \simeq 10^{-12} \text{yr}^{-1}$[@Uzan:2002vq; @Guenther:1998]. If we introduce a scalar field to control the value of $G$ we would also have to require that this field does not lead to an unacceptably large fifth force and does not

cosmic variance term. For example experiments presently being view for measuring the boron - mode of CMB polarization would be cosmic variance limited for vitamin e polarization over a wide range of $ ls$ (for example [ @bmodes ]). Such an optimum experiment, exploring structures into a multipole stove of $ l=4000 $, represents the limit of how much information on $ \lambda$ one could press out from the CMB in principle. We found an expected error for $ \lambda$ of order $ 0.3\%$ which translates into a restraint of the value of the gravitational constant during recombination of $ \frac{\delta G}{G } \simeq 0.6 \%$. Discussion\[discussion\ ] = = = = = = = = = = = = = = = = = = = = = = = = We have made the expansion rate of the universe a detached parameter in a likelihood analysis within an eight - dimensional cosmologic parameter outer space. For simplicity we assumed that the gravitational constant is changed by a factor $ \lambda$. We showed that an addition of $ \lambda$ leads to a wider visibility function which in turn increases the damping of anisotropies on belittled scales and increases the level of large scale polarization. We calculated the constraints that current CMB data can impose on the expansion rate. The constraint that can be imposed on the parameter $ \lambda$ use the data from the damping tail are severely weakened by our lack of knowledge about the shape of the primordial power spectrum. We indicate that measuring polarization helps to break this degeneracy. Current data can only stiffen $ \lambda$ to about 47% at 1 $ \sigma$. We showed that MAP could obtain 9% error bar for $ \lambda$ while for Planck errorbars go down to 0.9% . We besides explore the ultimate limit that could be achieved by a cosmic variance limited experiment measure anisotropies up to $ l=4000 $ and found errors of under a percentage in that case. therefore next generation experiments should be able to deliver very accurate constraints on the expansion pace of the universe during recombination. We acknowledge that if the variation of the gravitational constant during recombination is taken seriously a exemplar needs to be build up where $ G$ changes after recombination and converges towards the static value observed in laboratory experiments today and where its current rate of change is less than the experimental tie down $ \frac{\dot{G}}{G } \simeq 10^{-12 } \text{yr}^{-1}$[@Uzan:2002vq; @Guenther:1998 ]. If we introduce a scalar field to control the value of $ G$ we would also have to require that this field does not lead to an unacceptably bombastic fifth force and does not

codmic variance term. For ewample experimenjs curreitly bejng conskdered for measuring the B-moves if CMV polarization would bd cosmic nariance oimiued for E polarizefion ovcx a wjfe rcnje of $ls$ (e.g. [@bmoces]). Such at optimal expesioeut, exploring structures into a multi[ole ramgf of $l=4000$, represegts uhe limjn if how much information on $\lajbda$ ont could extract frpm the CMB in principle. We foujd an expected errlr for $\lambea$ os order $0.3\%$ whicf translatts into a cohstraint of the value of the grxvitacional consjcbt foring recombmnatiog of $\frac{\delbs G}{G} \shmeq 0.6 \%$. Doscussion\[discuxsimn\] ======================== Ww have made the expanvion rate of the ugiverse a fxee parameter in a lijeoihoog andlyskw wkthjn ah eighh-dijensional dosmologicao parameter space. Fpr wimplicity we assumqd that the gravitational constant is chatges by a factor $\lambda$. We showed that an increwse of $\laibda$ leads to a wider visibility function which it turi kncxcwrws the damping of anisotropies on small scales whd ikcreases the levcl of large scale lopatyzation. We calzulateb tge constraints thah curregt CMV data cag imlose on the expansion rate. Rhe constraiuts that can be imposzd on the paxametet $\lambca$ using the informatiou from the dampinh tail ars severely weakendd ny mur lack of knowledge aboue the sha'e of the prkmorcial pjwer spectgum. Wc showed that measuging kolarisation helos to break this degeneracy. Curcxnt data can pnny bonstrain $\lambca$ to about 47% wt 1 $\sigma$. We skowed thct MAP could obtzin 9% ercor bars for $\lambda$ while vor Planck ecrorbars do diwn ro 0.9%. We xuso explored tne ultimane limut that could be aghievgd by a cosmic vaxnabce limited exprrioene leesurigc anisotropias uo tu $l=4000$ ana found errifs og under a percent in thaf case. Thus next grncration ezperimenes should be sble to deliver vegy acrurate consyraynts on the expansion rate of fhe univegse during recoibinqtion. We acknpwledge that if the variation of the grevitational constant doring recombination ix taken seripusly a model teeds to be built whwre $G$ changes afttr recombination and cohvergev towwrds the stable value observed in laboratory experiments today and where uts cucrqnt rate of chamge iv jzss thag thx xxperimental bounc $\frac{\dot{G}}{G} \simeq 10^{-12} \text{yr}^{-1}$[@Uzan:2002vq; @Juenther:1998]. Ix ce introduce a scalar field tp zontrol the vxlue of $G$ we would also gave to tequire that this field does noy lead to an unacceptabln large fufth rorce and voes not

cosmic variance term. For example experiments currently for the B-modes CMB polarization would E over a wide of $ls$ (e.g. Such an optimal experiment, exploring structures a multipole range of $l=4000$, represents the limit of how much information on one could extract from the CMB in principle. We found an expected error $\lambda$ order which into a constraint of the value of the gravitational constant during recombination of $\frac{\delta G}{G} \simeq \%$. Discussion\[discussion\] ======================== We have made the expansion of the universe a parameter in a likelihood analysis an cosmological parameter For we that the gravitational is changed by a factor $\lambda$. We showed that an increase of $\lambda$ leads to a wider function which increases the of on scales and increases of large scale polarization. We calculated current CMB data can impose on the expansion The constraints can be imposed on the parameter using the information from the damping tail are weakened by our lack of knowledge about the shape of the primordial power spectrum. We measuring polarization helps to this degeneracy. Current can constrain to 47% at $\sigma$. We showed that MAP could obtain 9% error bars for while for Planck errorbars go down to 0.9%. We also ultimate that could be by a cosmic variance experiment anisotropies up to $l=4000$ errors under that Thus generation experiments should be to deliver very accurate constraints the expansion rate of acknowledge that if the variation of the gravitational during recombination is taken seriously a model to be built where $G$ changes after recombination and converges towards the value observed experiments today and where its current rate of is less than the bound $\frac{\dot{G}}{G} \simeq 10^{-12} \text{yr}^{-1}$[@Uzan:2002vq; @Guenther:1998]. If we introduce scalar to control value of $G$ would also have require that this not lead an large does not

cosmic variance term. For examPle experimEnts cUrrEntLy BeinG conSidered for measURing The B-modes of CMB polarizaTion wOuLD be cOSmIc varIance liMItED For e pOlAriZaTIoN over A wiDe range Of $ls$ (e.g. [@bmodEs]). SUcH an optimal exPErIment, exploRinG structures iNto A multiPoLe rANge of $L=4000$, rePreseNts the LImit of How much inFoRMation ON $\lambda$ ONE cOuld Extract from the CMB IN pRInciple. We found An expeCtED eRROr fOr $\lAmbda$ of ordEr $0.3\%$ Which TRanslatES iNTO A coNStraint of the vAlue of the grAVitAtionaL cOnsTAnt durIng reCoMBinAtion of $\frac{\DeltA G}{G} \simeq 0.6 \%$. DIscussIOn\[discuSSion\] ======================== We hAve madE thE exPansIOn RaTe oF tHE unIVeRse A FreE parametEr In A likeLihoOD ANAlysIs wIthiN an eiGht-dimensionaL coSmolOGicAl parAmeteR spaCe. for siMpliciTy we aSsUmed that the gravItatIonal consTanT iS chAnGed by A Factor $\LamBda$. we showeD that an INcrEaSE OF $\lAmbda$ leads to a wider ViSIBiLity funcTion whICh In TUrn increAsEs tHe daMPIng of AnisOTrOpies on sMall scALeS aNd increAsEs the lEvEl oF laRge scALe poLarizaTion. We caLculaTEd the constrainTS that current Cmb dATA cAN impOse On the expansIon rATe. ThE conSTrAinTS that Can be ImPOsED on the parameter $\lambDa$ Using tHe infOrmation from tHe damping tAIL Are severEly wEAkENed by our lack of KnowlEdge about tHE shape of The prImordial Power specTRUm. We showEd tHat MeaSurING pOlarization heLPS to bReAk this dEgeNeracy. CUrrEnt DatA caN oNly constrAin $\lambdA$ tO aBoUt 47% At 1 $\sIgma$. WE Showed thAt mAP CoUld ObtaiN 9% Error bArs foR $\lamBdA$ wHIle For PlanCK eRROrbaRs Go Down To 0.9%. WE aLso exPlorED thE ultimaTe limit thAt cOUld bE aChIeved by A cosmic varianCe Limited expErImeNt measURIng anisoTropies up to $l=4000$ and found errORs of undEr a PerceNt in That case. THus Next geNerATion exPerimeNts shOuLd bE ABle to DELiVer VeRy accurate CONstRaintS oN the ExpansiOn rate of the universE DurIng recombinatIon. we acKNOwLedGE tHAt iF tHE vaRIAtion of the graviTational coNsTAnT during recOMbiNaTion is tAken serIouslY A model nEeds to be bUilt where $g$ cHangES AftEr recombinAtion and Converges TOwardS ThE stabLe vAlue obSeRveD in laBoratoRY exPerimEnts toDaY and whEre itS cUrrent raTe of change is less than the ExperiMentaL boUnd $\frac{\doT{G}}{G} \SImeQ 10^{-12} \text{yr}^{-1}$[@UzAn:2002vq; @guenther:1998]. If We iNtrOduce A scALar fiEld tO CoNtrOL the vAlue OF $G$ we would ALsO haVE To Require that THIS fiEld doEs nOT lead tO an uNacceptably large fIFth force and doeS not

cosmic variance term. For example e xperi men tscu rren tlybeing consider e d fo r measuring the B-mode s ofCM B pol a ri zatio n would be c osm ic v ari an c elimit edfor E p olarizatio n o ve r a wide ran g eof $ls$ (e .g. [@bmodes]). Su ch anop tim a l exp eri ment, explo r ing st ructuresin t o a mu l tipoler a ng e of $l=4000$, repres e nt s the limit ofhow mu ch in f o rma tio n on $\lam bd a$ on e coulde xt r a c t f r om the CMB in principle. Wefoundan ex p ectederror f o r $ \lambda$ of ord er $0.3\% $ whic h transl a tes int o a co nst rai nt o f t he va lu e of th e g r avi tational c on stant dur i n g reco mbi nati on of $\frac{\delt a G }{G} \si meq 0 .6 \% $. Di scuss ion\[d iscus si on\] ========== ==== ========= = We ha ve made the ex pan sio n rateof theu niv er s e afree parameter inal i ke lihood a nalysi s w it h in an ei gh t-d imen s i onalcosm o lo gical pa ramete r s pa ce. For s implic it y w e a ssume d tha t thegravitat ional constant is ch a nged by a fac t or $ \l a mbda $.We showed t hata n in crea s eof$ \lamb da$ l ea d st o a wider visibilit yfuncti on wh ich in turn i ncreases t h e dampingof a n is o tropies on sma ll sc ales and i n creasesthe l evel oflarge sca l e polariz ati on. W e c a l cu lated the con s t rain ts that c urr ent CMB da tacan im po se on the expansi on r at e. Th e con s traintsth atca n b e imp o sed on thepara me te r $\ lambda$ us i n g th ein form ati on from the dam ping ta il are se ver e ly w ea ke ned byour lack of k no wledge abo ut th e shap e of the p rimordial power spectru m . We sh owe d tha t me asuring p ola rizati onh elps t o brea k thi sdeg e n eracy . Cu rre nt data cano n lyconst ra in $ \lambda $ to about 47% at1 $\ sigma$. We sh owe d th a t M APc ou l d o bt a in9 % error bars for $\lambda$ w h il e for Plan c k e rr orbarsgo down to 0 . 9%. Wealso expl ored theul tima t e li mit that c ould beachievedb y a c o sm ic va ria nce li mi ted expe riment mea surin g anis ot ropies up t o$l=4000$ and found errors of un der aperce ntin that c ase . Th us next g ener ation expe rim ent s sho uld be ab le t o d eli v er ve ry a c curate co n st rai n t son the expa n s i onrateoft he uni vers e during recombin a tion. We ackn owle d g e t hat if t he variation ofthe g r a vitation al constant d uring re co m binat ion is taken seriou s l ya model nee dsto be bui ltwh e re $G$ch an g es aft er r ec ombina tion a n d co n v erges towards th e sta b l e val u e o bserv ed in lab o rato ry experim ents todayand wh ereits c urrentra te ofcha ng e is lesst han the e xperi mentalbo und$\f rac{\d ot{G } } {G} \ sime q10^ {-12} \te x t {y r }^ {- 1 }$[ @Uza n:200 2v q; @ Guenther: 1 998]. If we introdu ce as c alar f i el d to control th e val u e of $G$ we woul d a l so ha ve torequir e thatt his f ield do esn o t lead to an unacc e pta bl y la rge fift hforc eand d oes not

cosmic_variance term._For example experiments currently_being considered_for_measuring the_B-modes_of CMB polarization_would be cosmic_variance limited for E_polarization over a_wide_range of $ls$ (e.g. [@bmodes]). Such an optimal experiment, exploring structures into a multipole_range_of $l=4000$,_represents_the_limit of how much information_on $\lambda$ one could extract_from the_CMB in principle. We found an expected error_for_$\lambda$ of order_$0.3\%$ which translates into a constraint of the value_of the gravitational constant during recombination_of $\frac{\delta G}{G}_\simeq_0.6_\%$. Discussion\[discussion\] ======================== We have made the_expansion rate of the universe a_free parameter in a likelihood analysis_within an eight-dimensional cosmological parameter space. For_simplicity we assumed that the gravitational_constant is changed by a_factor $\lambda$._We showed that an increase_of $\lambda$ leads_to a_wider visibility function_which in turn increases the damping_of anisotropies on_small scales and increases the level_of_large scale polarization. We_calculated_the_constraints that_current CMB data_can_impose on_the_expansion rate. The constraints that can_be_imposed on the parameter $\lambda$ using the_information from the damping_tail_are severely weakened by_our lack of knowledge about_the shape of the primordial power_spectrum. We_showed that_measuring polarization helps to break this degeneracy. Current data can only_constrain $\lambda$ to about 47% at_1 $\sigma$. We showed_that MAP_could_obtain 9% error_bars_for $\lambda$_while for Planck errorbars go down to_0.9%. We_also explored the ultimate limit that_could be achieved by_a_cosmic variance limited experiment measuring anisotropies_up to $l=4000$ and found errors_of under a percent in_that_case._Thus next generation experiments should_be able to deliver very accurate_constraints on the_expansion rate of the universe during recombination. We_acknowledge_that if the variation of the_gravitational_constant during recombination is taken seriously_a_model_needs to be built where_$G$ changes after recombination and converges_towards the stable value observed in laboratory experiments today_and where its_current rate of change is_less_than_the experimental bound $\frac{\dot{G}}{G} \simeq 10^{-12} \text{yr}^{-1}$[@Uzan:2002vq; @Guenther:1998]. If we introduce_a scalar_field to control_the value of $G$ we would also have to require_that this field does not lead to_an unacceptably large fifth force and does not

ach, J. V. 1983,, 202, 113 Gunn, J. E., & Gott, J. R. I. 1972,, 176, 1 Jørgensen, I., Franx, M., & Kjærgard, P. 1995,, 273, 1097 Jørgensen, I., Franx, M., & Kjærgard, P. 1996,, 280, 167 Kormendy, J., & Freeman, K. C. 2004, in IAU Symp. 220, Dark Matter in Galaxies, ed. S. D. Ryder et al. (San Francisco: ASP), 377 Kronawitter, A., Saglia, R. P., Gerhard, O. E., Bender, R. 2000,, 144, 53 Mehlert, D., Saglia, R. P., Bender, R., Wegner, G. 2000,, 141, 449 Mehlert, D., Thomas, D., Saglia, R. P., Bender, R., Wegner, G. 2003,, 407, 423 Navarro, J. F., Frenk, C. S., & White, S. D. M. 1996,, 462, 563 Persic, M., Salucci, P., & Stel, F. 1996a,, 281, 27 Persic, M., Salucci, P., & Stel, F. 1996b,, 283, 1102 Press, W. H., Teukoslky, S. A., Vetterling, W. T., Flannery, B. P. 1992, Numerical Recipes in FORTRAN 77, 2nd edn. Cambridge Univ. Press, Cambridge Saglia, R. P., Burstein, D., Baggley, G., Davies, R. L., Bertschinger, E., Colless, M. M., McMahan, R. K., Jr., & Wegner, G. 1997,, 292, 499 Schwarzschild, M. 1979,, 232, 236 Simien, F., & De Vaucouleurs, G. 1986,, 302, 564 Springel, V., et al. 2005,, 435, 629 Terlevich, A. I., & Forbes, D. A

ach, J.   V. 1983, , 202, 113 Gunn, J.   E., & Gott, J.   R.   I.   1972, , 176, 1 Jørgensen, I., Franx, M., & Kjærgard, P. 1995, , 273, 1097 Jørgensen, I., Franx, M., & Kjærgard, P. 1996, , 280, 167 Kormendy, J., & Freeman, K.   C. 2004, in IAU Symp. 220, Dark Matter in Galaxies, ed. S.   D. Ryder et al. (San Francisco: ASP), 377 Kronawitter, A., Saglia, R.   P., Gerhard, O.   E., Bender, R. 2000, , 144, 53 Mehlert, D., Saglia, R.   P., Bender, R., Wegner, G. 2000, , 141, 449 Mehlert, D., Thomas, D., Saglia, R.   P., Bender, R., Wegner, G. 2003, , 407, 423 Navarro, J.   F., Frenk, C.   S., & White, S.   D.   M. 1996, , 462, 563 Persic, M., Salucci, P., & Stel, F. 1996a, , 281, 27 Persic, M., Salucci, P., & Stel, F. 1996b, , 283, 1102 Press, W.   H., Teukoslky, S.   A., Vetterling, W.   T., Flannery, B.   P. 1992, Numerical Recipes in FORTRAN 77, 2nd edn. Cambridge Univ. Press, Cambridge Saglia, R.   P., Burstein, D., Baggley, G., Davies, R.   L., Bertschinger, E., Colless, M.   M., McMahan, R.   K., Jr., & Wegner, G.   1997, , 292, 499 Schwarzschild, M. 1979, , 232, 236 Simien, F., & De Vaucouleurs, G. 1986, , 302, 564 Springel, V., et al. 2005, , 435, 629 Terlevich, A.   I., & Forbes, D.   A

ach, J. V. 1983,, 202, 113 Gunn, J. E., & Gott, J. R. I. 1972,, 176, 1 Jørgensen, I., Ftabx, M., & Njærgars, P. 1995,, 273, 1097 Jøfgensen, I., Franx, M., & Kjærgard, P. 1996,, 280, 167 Kornendy, J., & Freeman, K. C. 2004, iv IAU Symi. 220, Dark Mqttec in Galaxies, ed. S. D. Rydev et zp. (Sau Hrancisco: ASP), 377 Lronawittes, A., Saglia, R. P., Cefhcrd, O. E., Bender, R. 2000,, 144, 53 Mehlert, D., Saglia, W. P., Bendrr, R., Wegner, G. 2000,, 141, 449 Mehkqrt, S., Thomas, D., Saglia, R. P., Bender, R., Wegnsr, G. 2003,, 407, 423 Navarro, J. F., Ftenk, C. S., & White, S. D. M. 1996,, 462, 563 Perslc, M., Salucci, P., & Stel, F. 1996a,, 281, 27 Persic, M., Swoucci, P., & Steu, F. 1996b,, 283, 1102 Prtsv, W. H., Teukoalky, S. A., Vetterling, W. T., Flannery, C. P. 1992, Nbmerical Rexipes hn FORTRAN 77, 2nd eqn. Cambridge Univ. Prass, Camnridge Saglia, V. P., Bucsteun, D., Baggley, G., Davies, R. L., Bertschinger, E., Colless, K. M., McMahan, R. K., Jr., & Wwgber, G. 1997,, 292, 499 Swhwafzscfils, K. 1979,, 232, 236 Simlen, F., & De Vaudouleurs, G. 1986,, 302, 564 Springel, V., et al. 2005,, 435, 629 Terlevich, A. I., & Forbqs, D. A

ach, J. V. 1983,, 202, 113 Gunn, & J. R. 1972,, 176, 1 Kjærgard, 1995,, 273, 1097 I., Franx, M., Kjærgard, P. 1996,, 280, 167 Kormendy, & Freeman, K. C. 2004, in IAU Symp. 220, Dark Matter in Galaxies, S. D. Ryder et al. (San Francisco: ASP), 377 Kronawitter, A., Saglia, R. Gerhard, E., R. 144, 53 Mehlert, D., Saglia, R. P., Bender, R., Wegner, G. 2000,, 141, 449 Mehlert, D., D., Saglia, R. P., Bender, R., Wegner, G. 407, 423 Navarro, J. Frenk, C. S., & White, D. 1996,, 462, Persic, Salucci, & Stel, F. 281, 27 Persic, M., Salucci, P., & Stel, F. 1996b,, 283, 1102 Press, W. H., Teukoslky, S. Vetterling, W. B. P. Numerical in 77, 2nd edn. Press, Cambridge Saglia, R. P., Burstein, Davies, R. L., Bertschinger, E., Colless, M. M., R. K., & Wegner, G. 1997,, 292, 499 M. 1979,, 232, 236 Simien, F., & De G. 1986,, 302, 564 Springel, V., et al. 2005,, 435, 629 Terlevich, A. I., & A

ach, J. V. 1983,, 202, 113 Gunn, J. E., & Gott, J. R. I. 1972,, 176, 1 JørgensEn, I., Franx, M., & KJærgaRd, P. 1995,, 273, 1097 jørGeNsen, i., FraNx, M., & Kjærgard, P. 1996,, 280, 167 KoRMendY, J., & Freeman, K. C. 2004, in IAU Symp. 220, DaRk MatTeR In GaLAxIes, ed. s. D. Ryder ET aL. (sAn FRaNcIscO: Asp), 377 KRonawIttEr, A., SaglIa, R. P., GerharD, O. E., beNder, R. 2000,, 144, 53 Mehlert, d., saGlia, R. P., BendEr, R., wegner, G. 2000,, 141, 449 MehleRt, D., thomas, d., SAglIA, R. P., BeNdeR, R., WegNer, G. 2003,, 407, 423 NaVArro, J. F., frenk, C. S., & WhItE, s. D. M. 1996,, 462, 563 PerSIc, M., SaluCCI, P., & stel, f. 1996a,, 281, 27 Persic, M., Salucci, P., & sTeL, f. 1996b,, 283, 1102 Press, W. H., TeukoSlky, S. A., veTTeRLIng, w. T., FLannery, B. P. 1992, NUmEricaL recipes IN FortrAN 77, 2ND edn. Cambridge univ. Press, CaMBriDge SagLiA, R. P., bUrsteiN, D., BagGlEY, G., DAvies, R. L., BertSchiNger, E., CollEss, M. M., MCmahan, R. K., jR., & Wegner, g. 1997,, 292, 499 SchwaRzsChiLd, M. 1979,, 232, 236 SIMiEn, f., & De vaUCouLEuRs, G. 1986,, 302, 564 sPriNgel, V., et aL. 2005,, 435, 629 TErLevicH, A. I., & FORBES, D. A

ach, J. V. 1983,, 202, 113 Gunn, J.E., & Go tt, J . R.  I.1972,, 176, 1J ørge nsen, I., Franx, M., & Kjær ga r d, P . 1 995,, 273, 1 0 97 J ørg en se n,I. , F ranx, M. , & Kjæ rgard, P.199 6, , 280, 167 K o rm endy, J.,& F reeman, K. C . 2 004, i nIAU Symp. 22 0, Da rk Mat t er inGalaxies, e d . S. D . Rydere t a l. ( San Francisco: AS P ), 377 Kronawitte r, A., S a gl i a , R . P ., Gerhard ,O. E. , Bender , R . 2 000 , , 144, 53 Meh lert, D., S a gli a, R.P. , B e nder,R., W eg n er, G. 2000,,141, 449 Mehl ert, D . , Thoma s , D., S aglia, R.  P. , Be n de r, R. ,W egn e r, G. 200 3,, 407, 4 23 Nava rro, J . F.,Fre nk,C. S. , & White, S.  D.  M.1 996 ,, 46 2, 56 3 Pe rs ic, M ., Sal ucci, P ., & Stel, F. 1 996a ,, 281, 2 7 P er sic ,M., S a lucci, P. , & Stel,F. 1996 b ,,28 3 , 11 02 Press, W. H., T eu k o sl ky, S. A ., Vet t er li n g, W. T. ,Fla nner y , B. P . 19 9 2, Numeric al Rec i pe sin FORT RA N 77,2n d e dn. Camb r idge Univ. Press,Cambr i dge Saglia, R. P., Burstein, D. , Ba g gley , G ., Davies,R. L . , Be rtsc h in ger , E.,Colle ss , M .  M., McMahan, R. K. ,Jr., & Wegn er, G. 1997,, 292, 499S c h warzschi ld,M .1 979,, 232, 236 Simi en, F., &D e Vaucou leurs , G. 198 6,, 302,5 6 4 Spring el, V. , e t a l . 2 005,, 435, 62 9 Terl ev ich, A.  I. , & For bes , D . A

ach, J. V._1983,, 202,_113 Gunn, J. E., &_Gott, J. R. I. 1972,,_176,_1 Jørgensen,_I.,_Franx, M., &_Kjærgard, P. 1995,,_273, 1097 Jørgensen, I.,_Franx, M., &_Kjærgard,_P. 1996,, 280, 167 Kormendy, J., & Freeman, K. C. 2004, in IAU Symp. 220,_Dark_Matter in_Galaxies,_ed._S. D. Ryder et al. (San_Francisco: ASP), 377 Kronawitter, A.,_Saglia, R. P.,_Gerhard, O. E., Bender, R. 2000,, 144, 53 Mehlert,_D.,_Saglia, R. P., Bender,_R., Wegner, G. 2000,, 141, 449 Mehlert, D., Thomas,_D., Saglia, R. P., Bender, R., Wegner,_G. 2003,, 407,_423_Navarro,_J. F., Frenk, C. S., &_White, S. D. M. 1996,, 462, 563 Persic,_M., Salucci, P., & Stel, F._1996a,, 281, 27 Persic, M., Salucci, P.,_& Stel, F. 1996b,, 283, 1102_Press, W. H., Teukoslky, S. A., Vetterling,_W. T., Flannery,_B. P. 1992, Numerical Recipes in_FORTRAN 77, 2nd_edn. Cambridge_Univ. Press, Cambridge_Saglia, R. P., Burstein, D., Baggley, G.,_Davies, R. L., Bertschinger,_E., Colless, M. M., McMahan, R. K., Jr.,_&_Wegner, G. 1997,, 292,_499_Schwarzschild,_M. 1979,,_232, 236 Simien,_F.,_& De_Vaucouleurs,_G. 1986,, 302, 564 Springel, V.,_et_al. 2005,, 435, 629 Terlevich, A. I., &_Forbes, D. A

^*S\to E$ is an isomorphism of DG ${{\mathcal A}}^0$-modules such that the following holds: there exists an isomorphism of graded ${{\mathcal A}}^0 _{{{\mathcal R}}}$-modules $\eta :(E\otimes {{\mathcal R}})^{\gr} \to S^{\gr}$ so that the composition $$E= i^*(E\otimes {{\mathcal R}}) \stackrel{i^*(\eta)}{\to} i^*S\stackrel{\sigma}{\to}E$$ is the identity. Given objects $(S,\sigma),(S^\prime,\sigma ^\prime)\in {\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ a map $f:(S,\sigma)\to (S^\prime,\sigma ^\prime)$ is an isomorphism $f:S\to S^\prime$ such that $\sigma ^\prime \cdot i^*f=\sigma$. An allowable homotopy between maps $f,g$ is a homotopy $h:f\to g$ such that $i^*(h)=0$. We define morphisms in ${\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ to be classes of maps modulo allowable homotopies. Note that a homomorphism of artinian DG algebras $\phi :{{\mathcal R}}\to {{\mathcal Q}}$ induces the functor $\phi ^*:{\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)\to {\operatorname{Def}}_{{{\mathcal Q}}}^{\h}(E)$. This defines the 2-functor $${\operatorname{Def}}{^h}(E):{\operatorname{dgart}}\to {\bf Gpd}.$$ We refer to objects of ${\operatorname{Def}}_{{{\mathcal R}}}^{\h} (E)$ as homotopy ${{\mathcal R}}$-deformations of $E$. We call $(p^*E,{\operatorname{id}})\in {\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ the trivial ${{\mathcal R}}$-deformation of $E$. Denote by ${\operatorname{Def}}_+^{\h}(E)$, ${\operatorname{Def}}_-^{\h}(E)$, ${\operatorname{Def}}_0^{\h}(E)$, ${\operatorname{Def}}^{\h}_{{\operatorname{cl}}}(E)$ the restrictions of the 2-functor ${\operatorname{Def}}^{\h}(E)$ to subcategories ${\operatorname{dgart}}_+$, ${\operatorname{dgart}}_-$, ${\operatorname{art}}$, ${\operatorname{cart}}$ respectively. Let us give an alternative description

^*S\to E$ is an isomorphism of DG $ { { \mathcal A}}^0$-modules such that the following holds: there exists an isomorphism of graded $ { { \mathcal A}}^0 _ { { { \mathcal R}}}$-modules $ \eta :( E\otimes { { \mathcal R}})^{\gr } \to S^{\gr}$ so that the composing $ $ E= i^*(E\otimes { { \mathcal radius } }) \stackrel{i^*(\eta)}{\to } i^*S\stackrel{\sigma}{\to}E$$ is the identity. Given objects $ (S,\sigma),(S^\prime,\sigma ^\prime)\in { \operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ a map $ f:(S,\sigma)\to (S^\prime,\sigma ^\prime)$ is an isomorphism $ fluorine: S\to S^\prime$ such that $ \sigma ^\prime \cdot i^*f=\sigma$. An allowable homotopy between maps $ f, g$ is a homotopy $ heat content: f\to g$ such that $ i^*(h)=0$. We specify morphisms in $ { \operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ to be classes of map modulo allowable homotopies. Note that a homomorphism of artinian DG algebras $ \phi: { { \mathcal R}}\to { { \mathcal Q}}$ induce the functor $ \phi ^*:{\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)\to { \operatorname{Def}}_{{{\mathcal Q}}}^{\h}(E)$. This defines the 2 - functor $ $ { \operatorname{Def}}{^h}(E):{\operatorname{dgart}}\to { \bf Gpd}.$$ We refer to object of $ { \operatorname{Def}}_{{{\mathcal R}}}^{\h } (E)$ as homotopy $ { { \mathcal R}}$-deformations of $ E$. We call $ (p^*E,{\operatorname{id}})\in { \operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ the trivial $ { { \mathcal R}}$-deformation of $ E$. Denote by $ { \operatorname{Def}}_+^{\h}(E)$, $ { \operatorname{Def}}_-^{\h}(E)$, $ { \operatorname{Def}}_0^{\h}(E)$, $ { \operatorname{Def}}^{\h}_{{\operatorname{cl}}}(E)$ the restrictions of the 2 - functor $ { \operatorname{Def}}^{\h}(E)$ to subcategories $ { \operatorname{dgart}}_+$, $ { \operatorname{dgart}}_-$, $ { \operatorname{art}}$, $ { \operatorname{cart}}$ respectively. Let us give an alternative description

^*S\to E$ is an isomorphism of AG ${{\mathcal A}}^0$-modolws suci that fhe folluwing holds: there exists an msomirphiwm of graded ${{\mathcal A}}^0 _{{{\mathcal G}}}$-modules $\wta :(T\otimes {{\mathcal R}})^{\gc} \to S^{\gr}$ so thzb the romposition $$E= i^*(G\otimes {{\mathwal R}}) \stackrel{i^*(\atx)}{\tl} i^*S\stackrel{\sigma}{\to}E$$ is the identitr. Given pbuects $(S,\sigma),(S^\ptime,\spgia ^\pdpmt)\in {\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ a jap $f:(S,\spgma)\to (S^\prime,\sigms ^\prime)$ is an isomorphism $v:S\to S^\prime$ such that $\digma ^\prime \cdoe u^*f=\sigma$. An aulowable homotopy betwgen maps $f,g$ is a homotopy $h:f\to g$ such that $i^*(h)=0$. Wg sevhne morphisns in ${\operatornamc{Cef}}_{{{\matvcal R}}}^{\h}(R)$ to be classex oh maps modulo allowable hmmotopies. Note that a homomospkism of artinian DG aogwbras $\phi :{{\matfxal R}}\tk {{\kafhcal E}}$ iiduces the runctor $\phi ^*:{\operatorname{Def}}_{{{\matncwo R}}}^{\h}(E)\to {\operafornamq{Dqf}}_{{{\mathcal Q}}}^{\h}(E)$. This defines the 2-functor $${\mpedatorname{Def}}{^h}(E):{\operatornqme{dgart}}\to {\bf Gpd}.$$ We rgfer to obtects of ${\operatorname{Def}}_{{{\mathcal R}}}^{\h} (E)$ as homotopy ${{\kathcel R}}$-dtfirmatkind of $E$. We call $(p^*E,{\operatorname{id}})\in {\operatorname{Qsf}}_{{{\kanhcal R}}}^{\h}(E)$ the triyial ${{\mathcal R}}$-defotmwtojn of $E$. Denote by ${\opzdaforname{Def}}_+^{\h}(E)$, ${\operahorname{Qef}}_-^{\h}(E)$, ${\operatorgame{Cef}}_0^{\h}(E)$, ${\operatorname{Def}}^{\h}_{{\operarorname{cl}}}(E)$ tke eestrictions of thz 2-functor ${\opzratormame{Drf}}^{\h}(E)$ to subcategories ${\o'eratodname{dgart}}_+$, ${\lperatornzoe{dgart}}_-$, ${\operatorvamv{art}}$, ${\operatorname{cart}}$ respectidely. Let uw gire an algernstive qescriptioj

^*S\to E$ is an isomorphism of DG such the following there exists an _{{{\mathcal $\eta :(E\otimes {{\mathcal \to S^{\gr}$ so the composition $$E= i^*(E\otimes {{\mathcal R}}) i^*S\stackrel{\sigma}{\to}E$$ is the identity. Given objects $(S,\sigma),(S^\prime,\sigma ^\prime)\in {\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ a map $f:(S,\sigma)\to ^\prime)$ is an isomorphism $f:S\to S^\prime$ such that $\sigma ^\prime \cdot i^*f=\sigma$. An homotopy maps is homotopy $h:f\to g$ such that $i^*(h)=0$. We define morphisms in ${\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ to be classes of modulo allowable homotopies. Note that a homomorphism of DG algebras $\phi :{{\mathcal {{\mathcal Q}}$ induces the functor ^*:{\operatorname{Def}}_{{{\mathcal {\operatorname{Def}}_{{{\mathcal Q}}}^{\h}(E)$. defines 2-functor {\bf Gpd}.$$ We to objects of ${\operatorname{Def}}_{{{\mathcal R}}}^{\h} (E)$ as homotopy ${{\mathcal R}}$-deformations of $E$. We call $(p^*E,{\operatorname{id}})\in {\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)$ trivial ${{\mathcal $E$. Denote ${\operatorname{Def}}_+^{\h}(E)$, ${\operatorname{Def}}_0^{\h}(E)$, the restrictions of ${\operatorname{Def}}^{\h}(E)$ to subcategories ${\operatorname{dgart}}_+$, ${\operatorname{dgart}}_-$, ${\operatorname{art}}$, us give an alternative description

^*S\to E$ is an isomorphism of DG ${{\maThcal A}}^0$-moduLes suCh tHat ThE folLowiNg holds: there exISts aN isomorphism of graded ${{\maThcal a}}^0 _{{{\mAThcaL r}}}$-mOduleS $\eta :(E\otIMeS {{\MAthCaL R}})^{\Gr} \tO S^{\GR}$ sO that The ComposiTion $$E= i^*(E\otiMes {{\MaThcal R}}) \stackrEL{i^*(\Eta)}{\to} i^*S\staCkrEl{\sigma}{\to}E$$ is The IdentiTy. givEN objeCts $(s,\sigmA),(S^\primE,\Sigma ^\pRime)\in {\opeRaTOrname{dEf}}_{{{\mathcAL r}}}^{\h}(e)$ a maP $f:(S,\sigma)\to (S^\prime,\sIGmA ^\Prime)$ is an isomoRphism $F:S\TO S^\PRIme$ SucH that $\sigma ^\PrIme \cdOT i^*f=\sigmA$. an ALLOwaBLe homotopy betWeen maps $f,g$ iS A hoMotopy $H:f\To g$ SUch thaT $i^*(h)=0$. We DeFIne Morphisms in ${\OperAtorname{DEf}}_{{{\mathCAl R}}}^{\h}(E)$ to BE classeS of mapS moDulO allOWaBlE hoMoTOpiES. NOte THat A homomorPhIsM of arTiniAN dg AlgeBraS $\phi :{{\MathcAl R}}\to {{\mathcal Q}}$ IndUces THe fUnctoR $\phi ^*:{\oPeraToRname{def}}_{{{\matHcal R}}}^{\H}(E)\To {\operatorname{DEf}}_{{{\maThcal Q}}}^{\h}(E)$. THis DeFinEs The 2-fuNCtor $${\opEraTorName{Def}}{^H}(E):{\operaTOrnAmE{DGArT}}\to {\bf Gpd}.$$ We refer to oBjECTs Of ${\operatOrname{dEf}}_{{{\MaTHcal R}}}^{\h} (E)$ aS hOmoTopy ${{\MAThcal r}}$-defORmAtions of $e$. We calL $(P^*E,{\OpEratornAmE{id}})\in {\oPeRatOrnAme{DeF}}_{{{\MathCal R}}}^{\h}(E)$ The triviAl ${{\matHCal R}}$-deformatioN Of $E$. Denote by ${\opERaTORnAMe{DeF}}_+^{\h}(E)$, ${\OperatornamE{Def}}_-^{\H}(e)$, ${\opeRatoRNaMe{DEF}}_0^{\h}(E)$, ${\opEratoRnAMe{dEf}}^{\h}_{{\operatorname{cl}}}(E)$ tHe RestriCtionS of the 2-functor ${\OperatornaME{dEf}}^{\h}(E)$ to suBcatEGoRIes ${\operatornamE{dgarT}}_+$, ${\operatornAMe{dgart}}_-$, ${\oPeratOrname{arT}}$, ${\operatorNAMe{cart}}$ reSpeCtiVelY. LeT US gIve an alternatIVE desCrIption

^*S\to E$ is an isomorphis m of DG ${ {\mat hca l A }} ^0$- modu les such thatt he f ollowing holds: thereexist sa n is o mo rphis m of gr a de d ${{ \m at hca lA }} ^0 _{ {{\ mathcal R}}}$-mod ule s$\eta :(E\ot i me s {{\mathc alR}})^{\gr} \ toS^{\gr }$ so thatthe comp ositio n $$E=i^*(E\oti me s {{\ma t hcal R} } ) \ stac krel{i^*(\eta)}{\ t o} i^*S\stackrel{ \sigma }{ \ to } E $$isthe identi ty . Gi v en obje c ts $ ( S,\ s igma),(S^\pri me,\sigma ^ \ pri me)\in { \op e ratorn ame{D ef } }_{ {{\mathcalR}}} ^{\h}(E)$ a map $f:(S,\ s igma)\t o (S^\ pri me, \sig m a^\ pri me ) $ i s a n i s omo rphism $ f: S\ to S^ \pri m e $ such th at $ \sigm a ^\prime \cd oti^*f = \si gma$. An a llow ab le ho motopy betw ee n maps $f,g$ is a h omotopy $ h:f \t o g $sucht hat $i ^*( h)= 0$. Wedefinem orp hi s m s i n ${\operatorname{ De f } }_ {{{\math cal R} } }^ {\ h }(E)$ to b e c lass e s of m apsm od ulo allo wableh om ot opies. N ote th at ahom omorp h ismof art inian DG alge b ras $\phi :{{\ m athcal R}}\to {{ \ m at h calQ}} $ induces t he f u ncto r $\ p hi ^* : {\ope rator na m e{ D ef}}_{{{\mathcal R} }} ^{\h}( E)\to {\operatorna me{Def}}_{ { { \ mathcalQ}}} ^ {\ h }(E)$. This de fines the 2-fun c tor $${\ opera torname{ Def}}{^h} ( E ):{\oper ato rna me{ dga r t }} \to {\bf Gpd} . $ $ W erefer t o o bjectsof${\ ope rat or name{Def} }_{{{\ma th ca lR} }}^ {\h}( E)$ as h om oto py ${ {\mat h cal R} }$-de form at io n s o f $E$.We c all$( p^ *E,{ \op er atorn ame{ i d}} )\in {\ operatorn ame { Def} }_ {{ {\mathc al R}}}^{\h}( E) $ the triv ia l $ {{\mat h c al R}}$- deformation of $E$. De n ote by${\ opera torn ame{Def}} _+^ {\h}(E )$, ${\ope ratorn ame{D ef }}_ - ^ {\h}( E ) $, ${ \o peratornam e { Def }}_0^ {\ h}(E )$, ${\ operatorname{Def}} ^ {\h }_{{\operator nam e{cl } } }( E)$ th e re st r ict i o ns of the 2-fun ctor ${\op er a to rname{Def} } ^{\ h} (E)$ to subcat egori e s ${\op eratornam e{dgart}} _+ $, $ { \ ope ratorname{ dgart}}_ -$, ${\op e rator n am e{art }}$ , ${\o pe rat ornam e{cart } }$respe ctivel y. Letus gi ve an alte rnative description

^*S\to E$_is an_isomorphism of DG ${{\mathcal_A}}^0$-modules such_that_the following_holds:_there exists an_isomorphism of graded_${{\mathcal A}}^0 _{{{\mathcal R}}}$-modules_$\eta :(E\otimes {{\mathcal R}})^{\gr}_\to_S^{\gr}$ so that the composition $$E= i^*(E\otimes {{\mathcal R}}) \stackrel{i^*(\eta)}{\to} i^*S\stackrel{\sigma}{\to}E$$ is the identity. Given objects_$(S,\sigma),(S^\prime,\sigma_^\prime)\in {\operatorname{Def}}_{{{\mathcal_R}}}^{\h}(E)$_a_map $f:(S,\sigma)\to (S^\prime,\sigma ^\prime)$ is_an isomorphism $f:S\to S^\prime$ such_that $\sigma_^\prime \cdot i^*f=\sigma$. An allowable homotopy between maps $f,g$_is_a homotopy $h:f\to_g$ such that $i^*(h)=0$. We define morphisms in ${\operatorname{Def}}_{{{\mathcal_R}}}^{\h}(E)$ to be classes of maps_modulo allowable homotopies. Note_that_a_homomorphism of artinian DG_algebras $\phi :{{\mathcal R}}\to {{\mathcal Q}}$_induces the functor $\phi ^*:{\operatorname{Def}}_{{{\mathcal R}}}^{\h}(E)\to_{\operatorname{Def}}_{{{\mathcal Q}}}^{\h}(E)$. This defines the 2-functor $${\operatorname{Def}}{^h}(E):{\operatorname{dgart}}\to_{\bf Gpd}.$$ We refer to objects of_${\operatorname{Def}}_{{{\mathcal R}}}^{\h} (E)$ as homotopy_${{\mathcal R}}$-deformations_of $E$. We call $(p^*E,{\operatorname{id}})\in {\operatorname{Def}}_{{{\mathcal_R}}}^{\h}(E)$ the trivial_${{\mathcal R}}$-deformation_of $E$. Denote by_${\operatorname{Def}}_+^{\h}(E)$, ${\operatorname{Def}}_-^{\h}(E)$, ${\operatorname{Def}}_0^{\h}(E)$, ${\operatorname{Def}}^{\h}_{{\operatorname{cl}}}(E)$ the restrictions_of the 2-functor_${\operatorname{Def}}^{\h}(E)$ to subcategories ${\operatorname{dgart}}_+$, ${\operatorname{dgart}}_-$, ${\operatorname{art}}$,_${\operatorname{cart}}$_respectively. Let us give_an_alternative_description

@{-->}[rr]^-{\displaystyle{{A}{}^{{\circlearrowleft}}}} && \operatorname{W}^*_Z({Y},L_2)\,. }$$ By Corollary \[coro:desc\] for $\bar A=\bar A_2\circ\bar A_1{^{-1}}$, there exists $A:L_1{\leadsto}L_2$ such that $f^!A\circ\bar A_1\simeq \bar A_2$. By Proposition \[prop:gen-func\], ${{A}{}^{{\circlearrowleft}}}\circ{\operatorname{Push}_{f,\bar A_1}}={\operatorname{Push}_{f,f^!A\circ \bar A_1}}={\operatorname{Push}_{f,\bar A_2}}$. \[rema:feel-good\]Combined with Corollary \[coro:reasy\], the last two propositions tell us that, for a morphism $f:\bar {Y}\to {Y}$ in ${{\mathcal S}}_{X}$, it is not so important to know where a lax pull-back ${\operatorname{Pull}_{\bar A,f}}$ or a lax push-forward ${\operatorname{Push}_{f,\bar A}}$ exactly lands, as long as we keep track of classes in $\operatorname{\mathrm{Pic}}_{X}(?)/2$. Different choices can always be “realigned". Relative alignments and lax module structure {#se:mock} ============================================ Now that we have a stable understanding of alignments, we introduce a relative version of this notion, allowing a line bundle on the base ${X}$ to intervene. \[defi:X-aligned\]Let $\pi_{Y}:{Y}\to {X}$ be a scheme over ${X}$. We say that two line bundles $L_1$ and $L_2$ over ${Y}$ are *(quadratically) ${X}$-aligned* if $[L_1]=[L_2]\in\operatorname{\mathrm{Pic}}_{X}({Y})=\operatorname{\mathrm{coker}}(\pi_{Y}^*:\operatorname{\mathrm{Pic}}({X})\to \operatorname{\mathrm{Pic}}({Y})/2)$. This amounts to the existence of a line bundle $K$ over ${X}$ and an alignment $\pi_{Y}^*K\otimes L_1{\leadsto}L_2$ as in Definition \[def

@{-->}[rr]^-{\displaystyle{{A}{}^{{\circlearrowleft } } } } & & \operatorname{W}^*_Z({Y},L_2)\, . } $ $ By Corollary   \[coro: desc\ ] for $ \bar A=\bar A_2\circ\bar A_1{^{-1}}$, there exists $ A: L_1{\leadsto}L_2 $ such that $ f^!A\circ\bar A_1\simeq \bar A_2$. By Proposition   \[prop: gen - func\ ], $ { { A}{}^{{\circlearrowleft}}}\circ{\operatorname{Push}_{f,\bar A_1}}={\operatorname{Push}_{f, f^!A\circ \bar A_1}}={\operatorname{Push}_{f,\bar A_2}}$. \[rema: feel - good\]Combined with Corollary   \[coro: reasy\ ], the last two proposition assure us that, for a morphism $ f:\bar { Y}\to { Y}$ in   $ { { \mathcal S}}_{X}$, it is not so important to know where a lax puff - back $ { \operatorname{Pull}_{\bar A, f}}$ or a lax push - fore $ { \operatorname{Push}_{f,\bar A}}$ precisely lands, as long as we keep track of course in $ \operatorname{\mathrm{Pic}}_{X}(?)/2$. Different choices can constantly be “ realign ". Relative alignments and lax faculty structure { # se: mock } = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = nowadays that we suffer a stable understanding of alignments, we precede a relative version of this notion, allowing a argumentation bundle on the base   $ { X}$ to intervene. \[defi :X - aligned\]Let $ \pi_{Y}:{Y}\to { X}$ be a scheme over   $ { X}$. We say that two line pack $ L_1 $ and $ L_2 $ over   $ { Y}$ are * (quadratically) $ { X}$-aligned * if $ [ L_1]=[L_2]\in\operatorname{\mathrm{Pic}}_{X}({Y})=\operatorname{\mathrm{coker}}(\pi_{Y}^*:\operatorname{\mathrm{Pic}}({X})\to \operatorname{\mathrm{Pic}}({Y})/2)$. This amounts to the existence of a line bundle $ K$ over   $ { X}$ and an alignment $ \pi_{Y}^*K\otimes L_1{\leadsto}L_2 $ as in Definition   \[def

@{-->}[rr]^-{\dlsplaystyle{{A}{}^{{\circlearrowltft}}}} && \operatorname{W}^*_E({Y},O_2)\,. }$$ By Cocollary \[doro:desc\] for $\bar A=\bar A_2\circ\bar A_1{^{-1}}$, thece ezists $A:L_1{\leadsto}L_2$ such that $w^!A\circ\bar A_1\simeq \bqr A_2$. Vy Proposivjon \[prop:nzn-fund\], ${{A}{}^{{\cixcoearrowleft}}}\cirg{\operatornake{Push}_{f,\bar A_1}}={\oparxtlrname{Push}_{f,f^!A\circ \bar A_1}}={\operatorname{Ptsh}_{f,\bar A_2}}$. \[gema:feel-good\]Coibintd rith Bovollary \[coro:reasy\], the last two prkpositimns tell us tnat, for a morphism $f:\bar {Y}\tl {Y}$ ln ${{\mathcal S}}_{X}$, it is not so impirtagr to know whdre a lax kunl-back ${\opetatorname{Pull}_{\bar A,f}}$ or a lax pusf-forwcrd ${\operatotuqme{Oosh}_{f,\bar A}}$ exectly jands, as lonn as we keep ttack of classex ii $\opwratorname{\mathrm{Pic}}_{X}(?)/2$. Vifferent choices cag always te “realigned". Relativw qlignkentv ana lab mkdnle struchurx {#se:mock} ============================================ Now that we hace a stable understsnqpmg of alignmsnts, wq yntroduce a relative version of this nouion, zllowing a line bundle in the base ${X}$ to intergene. \[defi:X-wligned\]Let $\pi_{Y}:{Y}\to {X}$ be a scheme over ${X}$. We say that two mkne bmndldw $P_1$ and $L_2$ over ${Y}$ are *(quadratically) ${X}$-aligned* if $[L_1]=[J_2]\jn\ppvratorname{\mathrm{Plc}}_{X}({Y})=\operatorname{\mayhgm{vjker}}(\pi_{Y}^*:\operaturname{\matgrm{Pic}}({X})\to \operatornwme{\mathtm{Pic}}({Y})/2)$. This amotnts to the existence of a line bundle $K$ ovvr ${X}$ qnd an alignment $\pn_{Y}^*K\otimes L_1{\lzadsto}K_2$ as on Definition \[def

@{-->}[rr]^-{\displaystyle{{A}{}^{{\circlearrowleft}}}} && \operatorname{W}^*_Z({Y},L_2)\,. }$$ By Corollary \[coro:desc\] A=\bar A_1{^{-1}}$, there $A:L_1{\leadsto}L_2$ such that Proposition ${{A}{}^{{\circlearrowleft}}}\circ{\operatorname{Push}_{f,\bar A_1}}={\operatorname{Push}_{f,f^!A\circ \bar A_2}}$. \[rema:feel-good\]Combined with \[coro:reasy\], the last two propositions tell that, for a morphism $f:\bar {Y}\to {Y}$ in ${{\mathcal S}}_{X}$, it is not important to know where a lax pull-back ${\operatorname{Pull}_{\bar A,f}}$ or a lax push-forward A}}$ lands, long we keep track of classes in $\operatorname{\mathrm{Pic}}_{X}(?)/2$. Different choices can always be “realigned". Relative alignments and module structure {#se:mock} ============================================ Now that we have stable understanding of alignments, introduce a relative version of notion, a line on base to intervene. \[defi:X-aligned\]Let {X}$ be a scheme over ${X}$. We say that two line bundles $L_1$ and $L_2$ over ${Y}$ *(quadratically) ${X}$-aligned* \operatorname{\mathrm{Pic}}({Y})/2)$. This to existence a line bundle ${X}$ and an alignment $\pi_{Y}^*K\otimes L_1{\leadsto}L_2$ \[def

@{-->}[rr]^-{\displaystyle{{A}{}^{{\circlearroWleft}}}} && \operaTornaMe{W}^*_z({Y},L_2)\,. }$$ by coroLlarY \[coro:desc\] for $\baR a=\bar a_2\circ\bar A_1{^{-1}}$, there exists $A:L_1{\LeadsTo}l_2$ Such THaT $f^!A\ciRc\bar A_1\sIMeQ \BAr A_2$. by prOpoSiTIoN \[prop:Gen-Func\], ${{A}{}^{{\ciRclearrowlEft}}}\CiRc{\operatornaME{PUsh}_{f,\bar A_1}}={\opEraTorname{Push}_{f,F^!A\cIrc \bar a_1}}={\oPerATornaMe{PUsh}_{f,\bAr A_2}}$. \[remA:Feel-goOd\]CombineD wITh CoroLLary \[corO:REaSy\], thE last two propositiONs TEll us that, for a mOrphisM $f:\BAr {y}\TO {Y}$ iN ${{\maThcal S}}_{X}$, it iS nOt so iMPortant TO kNOW WheRE a lax pull-back ${\OperatornamE{pulL}_{\bar A,f}}$ Or A laX Push-foRward ${\OpERatOrname{Push}_{f,\Bar A}}$ Exactly laNds, as lONg as we kEEp track Of clasSes In $\oPeraTOrNaMe{\mAtHRm{PIC}}_{X}(?)/2$. difFEreNt choiceS cAn AlwayS be “rEALIGned". relAtivE aligNments and lax mOduLe stRUctUre {#se:Mock} ============================================ NOw thAt We havE a stabLe undErStanding of alignMentS, we introdUce A rElaTiVe verSIon of tHis NotIon, alloWing a liNE buNdLE ON tHe base ${X}$ to intervene. \[DeFI:x-aLigned\]LeT $\pi_{Y}:{Y}\tO {x}$ bE a SCheme oveR ${X}$. we sAy thAT Two liNe buNDlEs $L_1$ and $L_2$ oVer ${Y}$ arE *(QuAdRaticalLy) ${x}$-alignEd* If $[L_1]=[l_2]\in\OperaTOrnaMe{\mathRm{Pic}}_{X}({Y})=\oPeratORname{\mathrm{cokER}}(\pi_{Y}^*:\operatornAMe{\MAThRM{Pic}}({x})\to \OperatornamE{\matHRm{PiC}}({Y})/2)$. ThIS aMouNTs to tHe exiStENcE Of a line bundle $K$ over ${X}$ AnD an aliGnmenT $\pi_{Y}^*K\otimes L_1{\lEadsto}L_2$ as iN dEFinition \[Def

@{-->}[rr]^-{\displaystyle {{A}{}^{{\ circl ear row le ft}} }} & & \operatornam e {W}^ *_Z({Y},L_2)\,. }$$ B y Cor ol l ary\ [c oro:d esc\] f o r$ \ bar A =\ bar A _ 2\ circ\ bar A_1{^{ -1}}$, the reex ists $A:L_1{ \ le adsto}L_2$ su ch that $f^! A\c irc\ba rA_1 \ simeq \b ar A_ 2$. By Propos ition \[p ro p :gen-f u nc\], $ { { A} {}^{ {\circlearrowleft } }} \ circ{\operator name{P us h }_ { f ,\b arA_1}}={\op er atorn a me{Push } _{ f , f ^!A \ circ \bar A_1 }}={\operat o rna me{Pus h} _{f , \bar A _2}}$ .\[r ema:feel-go od\] Combinedwith C o rollary \[coro: reasy\ ],the las t t wo pr op o sit i on s t e llus that, f or a mo rphi s m $ f:\b ar{Y}\ to {Y }$ in ${{\mat hca l S} } _{X }$, i t isnotso impo rtantto kn ow where a lax pu ll-b ack ${\op era to rna me {Pull } _{\bar A, f}} $ or alax pus h -fo rw a r d $ {\operatorname{Pus h} _ { f, \bar A}} $ exac t ly l a nds, aslo ngas w e keeptrac k o f classe s in $ \ op er atornam e{ \mathr m{ Pic }}_ {X}(? ) /2$. Diffe rent cho icesc an always be “ r ealigned". R e la t i ve alig nme nts and lax mod u le s truc t ur e { # se:mo ck} = == = == = =================== == ====== ===== ===== Now th at we have a stable u nder s ta n ding of alignm ents, we introd u ce a rel ative version of thisn o tion, al low ing alin e bu ndle on the b a s e ${ X} $ to in ter vene. \[d efi :X- ali gn ed\]Let $ \pi_{Y}: {Y }\ to { X}$ be a scheme o ve r $ {X }$. We s a y that twoline b un d les $L_1$a nd $ L_2$ o ve r ${ Y}$ a re *( quad r ati cally)${X}$-ali gne d * if $ [L _1]=[L_ 2]\in\operato rn ame{\mathr m{ Pic }}_{X} ( { Y})=\ope ratorname{\mathrm{coker } }(\pi_{ Y}^ *:\op erat orname{\m ath rm{Pic }}( { X})\to \oper atorn am e{\ m a thrm{ P i c} }({ Y} )/2)$. Thi s amo untsto the existe nce of a line bund l e $ K$ over ${X}$ an d an a li gnm e nt $\p i_ { Y}^ * K \otimes L_1{\le adsto}L_2$ a s i n Definiti o n \ [d ef

@{-->}[rr]^-{\displaystyle{{A}{}^{{\circlearrowleft}}}} && \operatorname{W}^*_Z({Y},L_2)\,. }$$ By_Corollary \[coro:desc\] for_$\bar A=\bar A_2\circ\bar A_1{^{-1}}$,_there exists_$A:L_1{\leadsto}L_2$_such that_$f^!A\circ\bar_A_1\simeq \bar A_2$. By_Proposition \[prop:gen-func\], ${{A}{}^{{\circlearrowleft}}}\circ{\operatorname{Push}_{f,\bar A_1}}={\operatorname{Push}_{f,f^!A\circ_\bar A_1}}={\operatorname{Push}_{f,\bar A_2}}$. \[rema:feel-good\]Combined with Corollary \[coro:reasy\],_the last two_propositions_tell us that, for a morphism $f:\bar {Y}\to {Y}$ in ${{\mathcal S}}_{X}$, it is not_so_important to_know_where_a lax pull-back ${\operatorname{Pull}_{\bar A,f}}$_or a lax push-forward ${\operatorname{Push}_{f,\bar_A}}$ exactly_lands, as long as we keep track of_classes_in $\operatorname{\mathrm{Pic}}_{X}(?)/2$. Different_choices can always be “realigned". Relative alignments and lax module_structure {#se:mock} ============================================ Now that we have a_stable understanding of_alignments,_we_introduce a relative version_of this notion, allowing a line_bundle on the base ${X}$ to intervene. \[defi:X-aligned\]Let_$\pi_{Y}:{Y}\to {X}$ be a scheme over ${X}$. We_say that two line bundles $L_1$_and $L_2$ over ${Y}$ are *(quadratically)_${X}$-aligned* if_$[L_1]=[L_2]\in\operatorname{\mathrm{Pic}}_{X}({Y})=\operatorname{\mathrm{coker}}(\pi_{Y}^*:\operatorname{\mathrm{Pic}}({X})\to \operatorname{\mathrm{Pic}}({Y})/2)$. This amounts to the_existence of a_line bundle_$K$ over ${X}$ and_an alignment $\pi_{Y}^*K\otimes L_1{\leadsto}L_2$ as in_Definition \[def

I \rangle^{{ 2k }}+ \langle v_*, I_* \rangle^{{ 2k }} \right) \mathcal{B}(v,v_*,I,I_*,r,R,\sigma) \\ \times \varphi_\alpha(r) \, (1-R) R^{1/2} \psi_\alpha(R) \, \mathrm{d} \sigma\, \mathrm{d}r \, \mathrm{d}R \, \mathrm{d}I_* \mathrm{d} v_* \mathrm{d}I \mathrm{d}v, \end{gathered}$$ so that $$\label{W} \mathcal{W}=\mathcal{W}^+-\mathcal{W}^-.$$ We treat each term separately. For the gain part, we use the bound from above stated in, $$\begin{gathered} \mathcal{W}^+ \leq \frac{1}{2} \int_{\Delta^2 \times {K}} f f_* \left( \langle v', I' \rangle^{{ 2k }}+ \langle v'_*, I'_* \rangle^{{ 2k }} \right) b(\hat{u} \cdot \sigma) \, {d_{\gamma}^{{ub}}(r)}\, {e_{\gamma}^{{ub}}(R)}\, { \tilde{\mathcal{B}}}(v,v_*,I,I_*) \\ \times \varphi_\alpha(r) \, (1-R) R^{1/2} \psi_\alpha(R) \, \mathrm{d} \sigma\, \mathrm{d}r \, \mathrm{d}R \, \mathrm{d}I_* \mathrm{d} v_* \mathrm{d}I \mathrm{d}v, \\ = \frac{1}{2} {C_{\gamma,\alpha}^{{ub}}}\int_{\Delta^2 } \left( \int_{S^2 \times [0,1]^2 } \left( \langle v', I' \rangle^{{ 2k }}+ \langle v'_*, I'_* \rangle^{{ 2k }} \right) b(\hat{u}\cdot\sigma) \, {d_{\gamma}^{{ub}}(r)}\, \varphi_\alpha(r) \, \

I \rangle^ { { 2k } } + \langle v _ *, I _ * \rangle^ { { 2k } } \right) \mathcal{B}(v, v_*,I, I_*,r, R,\sigma) \\ \times \varphi_\alpha(r) \, (1 - R) R^{1/2 } \psi_\alpha(R) \, \mathrm{d } \sigma\, \mathrm{d}r \, \mathrm{d}R \, \mathrm{d}I _ * \mathrm{d } v _ * \mathrm{d}I \mathrm{d}v, \end{gathered}$$ so that $ $ \label{W } \mathcal{W}=\mathcal{W}^+-\mathcal{W}^-.$$ We treat each term separately. For the gain character, we practice the bound from above stated in, $ $ \begin{gathered } \mathcal{W}^+ \leq \frac{1}{2 } \int_{\Delta^2 \times { K } } f f _ * \left (\langle five', I' \rangle^ { { 2k } } + \langle vanadium' _ *, I' _ * \rangle^ { { 2k } } \right) b(\hat{u } \cdot \sigma) \, { d_{\gamma}^{{ub}}(r)}\, { e_{\gamma}^{{ub}}(R)}\, { \tilde{\mathcal{B}}}(v, v_*,I, I _ *) \\ \times \varphi_\alpha(r) \, (1 - R) R^{1/2 } \psi_\alpha(R) \, \mathrm{d } \sigma\, \mathrm{d}r \, \mathrm{d}R \, \mathrm{d}I _ * \mathrm{d } v _ * \mathrm{d}I \mathrm{d}v, \\ = \frac{1}{2 } { C_{\gamma,\alpha}^{{ub}}}\int_{\Delta^2 } \left (\int_{S^2 \times [ 0,1]^2 } \left (\langle v', I' \rangle^ { { 2k } } + \langle volt' _ *, I' _ * \rangle^ { { 2k } } \right) b(\hat{u}\cdot\sigma) \, { d_{\gamma}^{{ub}}(r)}\, \varphi_\alpha(r) \, \

I \gangle^{{ 2k }}+ \langle v_*, I_* \rannle^{{ 2k }} \right) \mathcan{B}(v,v_*,I,I_*,d,R,\sigma) \\ \times \varphi_\alpha(r) \, (1-R) R^{1/2} \psi_\elphq(R) \, \mqthrm{d} \sigma\, \mathrm{d}r \, \mathrm{d}R \, \mathrm{e}I_* \methrm{d} v_* \mathrm{d}M \mathrm{d}v, \skd{gatkeced}$$ so that $$\labgl{W} \mathcdl{W}=\mathcal{W}^+-\matvcxl{C}^-.$$ We treat each term separately. For ehe gaim oart, we use thg boumq frkm above stated in, $$\begin{gathered} \mafhcal{W}^+ \ltq \frac{1}{2} \int_{\Delta^2 \yimes {K}} f f_* \left( \langle v', I' \ranhle^{{ 2k }}+ \langle v'_*, I'_* \gangle^{{ 2k }} \rught) v(\hat{u} \cdot \skgma) \, {d_{\gamma}^{{ub}}(r)}\, {e_{\gamma}^{{ob}}(R)}\, { \tilde{\mathcal{B}}}(v,v_*,I,I_*) \\ \times \xarphn_\alpha(r) \, (1-T) D^{1/2} \ovi_\alpha(R) \, \methrm{d} \sigma\, \mathrm{d}r \, \matvrm{d}R \, \kathrm{d}I_* \mathrk{d} t_* \marhrm{d}I \mathrm{d}v, \\ = \frac{1}{2} {C_{\gamma,\alpha}^{{ub}}}\int_{\Qelta^2 } \neyt( \int_{S^2 \times [0,1]^2 } \left( \lqntle v', I' \rdngld^{{ 2k }}+ \lahgke v'_*, I'_* \rwngme^{{ 2k }} \righf) b(\hat{u}\cdot\wigma) \, {d_{\gamma}^{{ub}}(r)}\, \vatprp_\slpha(r) \, \

I \rangle^{{ 2k }}+ \langle v_*, I_* }} \mathcal{B}(v,v_*,I,I_*,r,R,\sigma) \\ \varphi_\alpha(r) \, (1-R) \mathrm{d}r \mathrm{d}R \, \mathrm{d}I_* v_* \mathrm{d}I \mathrm{d}v, so that $$\label{W} \mathcal{W}=\mathcal{W}^+-\mathcal{W}^-.$$ We treat term separately. For the gain part, we use the bound from above stated $$\begin{gathered} \mathcal{W}^+ \leq \frac{1}{2} \int_{\Delta^2 \times {K}} f f_* \left( \langle v', I' 2k \langle I'_* 2k }} \right) b(\hat{u} \cdot \sigma) \, {d_{\gamma}^{{ub}}(r)}\, {e_{\gamma}^{{ub}}(R)}\, { \tilde{\mathcal{B}}}(v,v_*,I,I_*) \\ \times \varphi_\alpha(r) \, (1-R) \psi_\alpha(R) \, \mathrm{d} \sigma\, \mathrm{d}r \, \mathrm{d}R \, \mathrm{d} v_* \mathrm{d}I \mathrm{d}v, = \frac{1}{2} {C_{\gamma,\alpha}^{{ub}}}\int_{\Delta^2 } \left( \times } \left( v', \rangle^{{ }}+ \langle v'_*, \rangle^{{ 2k }} \right) b(\hat{u}\cdot\sigma) \, {d_{\gamma}^{{ub}}(r)}\, \varphi_\alpha(r) \, \

I \rangle^{{ 2k }}+ \langle v_*, I_* \rangle^{{ 2k }} \riGht) \mathcal{b}(v,v_*,I,I_*,R,R,\sIgmA) \\ \tImes \VarpHi_\alpha(r) \, (1-R) R^{1/2} \psi_\aLPha(R) \, \Mathrm{d} \sigma\, \mathrm{d}r \, \maThrm{d}r \, \mAThrm{D}i_* \mAthrm{D} v_* \mathrM{D}I \MAThrM{d}V, \eNd{gAtHErEd}$$ so tHat $$\Label{W} \mAthcal{W}=\matHcaL{W}^+-\Mathcal{W}^-.$$ We trEAt Each term seParAtely. For the gAin Part, we UsE thE Bound FroM abovE stateD In, $$\begiN{gathered} \MaTHcal{W}^+ \lEQ \frac{1}{2} \inT_{\dElTa^2 \tiMes {K}} f f_* \left( \langle v', i' \RaNGle^{{ 2k }}+ \langle v'_*, I'_* \raNgle^{{ 2k }} \rIgHT) b(\HAT{u} \cDot \Sigma) \, {d_{\gammA}^{{uB}}(r)}\, {e_{\gaMMa}^{{ub}}(R)}\, { \tiLDe{\MATHcaL{b}}}(v,v_*,I,I_*) \\ \times \varPhi_\alpha(r) \, (1-R) R^{1/2} \PSi_\aLpha(R) \, \mAtHrm{D} \Sigma\, \mAthrm{D}r \, \MAthRm{d}R \, \mathrm{d}i_* \matHrm{d} v_* \mathRm{d}I \maTHrm{d}v, \\ = \frAC{1}{2} {C_{\gamma,\Alpha}^{{uB}}}\inT_{\DeLta^2 } \lEFt( \InT_{S^2 \tImES [0,1]^2 } \leFT( \lAngLE v', I' \Rangle^{{ 2k }}+ \lAnGlE v'_*, I'_* \raNgle^{{ 2K }} \RIGHt) b(\hAt{u}\Cdot\Sigma) \, {D_{\gamma}^{{ub}}(r)}\, \varpHi_\aLpha(R) \, \

I \rangle^{{ 2k }}+ \lang le v_*, I_ * \ra ngl e^{ {2k } } \right) \m a thca l{B}(v,v_*,I,I_*,r,R,\ sigma )\ \ \t i me s \va rphi_\a l ph a ( r)\, ( 1-R )R ^{ 1/2}\ps i_\alph a(R) \, \m ath rm {d} \sigma\, \m athrm{d}r\,\mathrm{d}R\,\mathr m{ d}I _ * \ma thr m{d}v_* \m a thrm{d }I \mathr m{ d }v, \end{g a t he red} $$ so that $$\lab e l{ W } \mathcal {W}=\m at h ca l { W}^ +-\ mathcal{W} ^- .$$ W e treate ac h t erm separately. F or the gain par t, weus e t h e boun d fro ma bov e stated in , $$ \begin{ga thered } \mathc a l{W}^+\leq \ fra c{1 }{2} \ in t_{ \D e lta ^ 2\ti m es{K}} f f _* \ left( \la n g l e v', I' \ra ngle^ {{ 2k }}+ \la ngl e v' _ *,I'_*\rang le^{ {2k }} \rig ht) b( \hat{u} \cdot \ sigm a) \, {d_ {\g am ma} ^{ {ub}} ( r)}\,{e_ {\g amma}^{ {ub}}(R ) }\, { \ ti lde{\mathcal{B}}}( v, v _ *, I,I_*) \ \ \tim e s \ v arphi_\a lp ha( r) \ , (1- R) R ^ {1 /2} \psi _\alph a (R )\, \mat hr m{d} \ si gma \,\math r m{d} r \, \ mathrm{d }R \, \mathrm{d}I_*\ mathrm{d} v_* \m a t hr m {d}I \m athrm{d}v,\\ = \fra c{1} { 2} { C_{\g amma, \a l ph a }^{{ub}}}\int_{\Del ta ^2 } \lef t( \int_{S^2\times [0, 1 ] ^ 2 } \lef t( \ l an g le v', I' \ran gle^{ { 2k }}+ \ l angle v' _*, I '_* \ran gle^{{ 2k } } \righ t) b( \ha t{u } \ cd ot\sigma) \,{ d _{\g am ma}^{{u b}} (r)}\,\va rph i_\ alp ha (r) \, \

I_\rangle^{{ 2k_}}+ \langle v_*, I_*_\rangle^{{ 2k_}}_ __ \right) \mathcal{B}(v,v_*,I,I_*,r,R,\sigma)_\\ \times \varphi_\alpha(r)_\, (1-R) R^{1/2} \psi_\alpha(R)_\, \mathrm{d} \sigma\,_\mathrm{d}r_\, \mathrm{d}R \, \mathrm{d}I_* \mathrm{d} v_* \mathrm{d}I \mathrm{d}v, \end{gathered}$$ so that_$$\label{W} _ __\mathcal{W}=\mathcal{W}^+-\mathcal{W}^-.$$_We treat each term separately._For the gain part, we_use the_bound from above stated in, $$\begin{gathered} \mathcal{W}^+ \leq \frac{1}{2} _\int_{\Delta^2_\times {K}} f_f_* \left( \langle v', I' \rangle^{{ 2k }}+ \langle v'_*,_I'_* \rangle^{{ 2k }} \right) b(\hat{u}_\cdot \sigma) \,_{d_{\gamma}^{{ub}}(r)}\,_{e_{\gamma}^{{ub}}(R)}\,_{ \tilde{\mathcal{B}}}(v,v_*,I,I_*) \\_\times \varphi_\alpha(r) \, _(1-R) R^{1/2} \psi_\alpha(R) \, \mathrm{d} \sigma\,_\mathrm{d}r \, \mathrm{d}R \, \mathrm{d}I_* \mathrm{d} v_*_\mathrm{d}I \mathrm{d}v, \\ = \frac{1}{2} {C_{\gamma,\alpha}^{{ub}}}\int_{\Delta^2_ } \left( \int_{S^2_\times [0,1]^2_} \left( \langle v', I' \rangle^{{_2k }}+ \langle_v'_*, I'_*_\rangle^{{ 2k }}_ \right) b(\hat{u}\cdot\sigma) \, {d_{\gamma}^{{ub}}(r)}\, \varphi_\alpha(r) _\, \

an example, for $N = 2$, repeating the above procedure will yield us $y_1 = 0.5, y_2 = 0.75, y_3 = 0.25, y_4 = 0.875, \cdots$. Note that the above values are in fact the thresholds reported in Table \[tab:osa\_two\_users\]. 3. In Remark \[rem:osa\_2\], for the $N = 2$ case, we noticed that the feedback from the base station corresponding to a threshold can be viewed as the binary representation of the threshold itself. For general $N$, the feedback from the base station can still be viewed as the binary representation of the threshold, however, with the weights corresponding to a position computed from the thresholds $\{ y_k \}$ obtained from the pseudo-code. For example, the weight of the first position will be equal to $y_1$. ![Plot of the average delay performance of OSA and MPA as a function of the number of users $N$. We have also plotted the entropy of the threshold random variable corresponding to MPA.[]{data-label="fig:osa_vs_mpa"}](osa_vs_mpa.pdf) In Figure \[fig:osa\_vs\_mpa\], we plot the average delay performance of OSA (as described in Section \[sec:osa\]) and the maximal probability allocation code. As expected, the performance of OSA and MPA are similar and in fact, MPA performs better than OSA as it identifies the optimal thresholds without any approximations (see Remark \[rem:osa\]). We have also plotted in the Figure \[fig:osa\_vs\_mpa\], the entropy of the maximal probability allocation code in bits. As expected, the entropy of the random variable reflects the average delay performance of the contention resolution algorithm as a function of $N$ very well. Entropy Minimization -------------------- Entropy is a concave function of the distribution. The maximal probability allocation code identifies a local minima in the space of probability distributions. From limited numerical work (not reported in this paper), we conjecture that the maximal probability allocation code is a globally entropy minimizing strategy as well. The following theorem proves the optimality of MPA for the $N = 2$ case. MPA is a delay minimizing strategy and an entropy minimizing strategy for $N = 2$ case. Let $(x, 1]$ be the contention

an example, for $ N = 2 $, repeating the above procedure will yield us $ y_1 = 0.5, y_2 = 0.75, y_3 = 0.25, y_4 = 0.875, \cdots$. notice that the above value are in fact the thresholds reported in Table   \[tab: osa\_two\_users\ ]. 3. In Remark   \[rem: osa\_2\ ], for the $ N = 2 $ subject, we notice that the feedback from the base station represent to a doorsill can be viewed as the binary representation of the threshold itself. For general $ N$, the feedback from the nucleotide station can still be view as the binary theatrical performance of the threshold, however, with the weights correspond to a position computed from the threshold $ \ { y_k \}$ obtained from the pseudo - code. For example, the weight unit of the first position will be equal to $ y_1$. ! [ Plot of the average delay operation of OSA and MPA as a function of the issue of users $ N$. We have also plotted the information of the threshold random variable corresponding to MPA.[]{data - label="fig: osa_vs_mpa"}](osa_vs_mpa.pdf) In Figure   \[fig: osa\_vs\_mpa\ ], we plot the average delay performance of OSA (as described in Section   \[sec: osa\ ]) and the maximal probability allocation code. As expected, the performance of OSA and MPA are similar and in fact, MPA performs better than OSA as it identifies the optimal thresholds without any approximations (attend Remark   \[rem: osa\ ]). We have also plotted in the Figure   \[fig: osa\_vs\_mpa\ ], the randomness of the maximal probability allotment code in bits. As expected, the entropy of the random variable star reflects the modal delay performance of the contention settlement algorithm as a function of $ N$ very well. Entropy Minimization -------------------- Entropy is a concave function of the distribution. The maximal probability allocation code identifies a local minima in the space of probability distributions. From circumscribed numerical work (not reported in this paper), we conjecture that the maximal probability allocation code is a globally entropy minimize strategy as well. The following theorem proves the optimality of MPA for the $ N = 2 $ case. MPA is a delay minimize strategy and an entropy minimizing scheme for $ N = 2 $ case. get $ (x, 1]$ be the contention

an example, for $N = 2$, repeatikg the above proewdure xill yisld us $y_1 = 0.5, y_2 = 0.75, y_3 = 0.25, y_4 = 0.875, \cdots$. Note thet tye abive values are in fact the thredholds rwporued in Table \[tab:ose\_fwo\_users\]. 3. In Dcmark \[xen:osa\_2\], for the $N = 2$ case, we noticed that dhd yeedback from the base station correfpondinb ho a threshold can fe vjvwtd as the binary representation or the tireshold itself. For general $N$, the feedbacn frlm the base statioj can still be duewed as the binary rekrzsentation kf the threshold, however, with tfe wenghts correwpindltg to a posmtion bomputed from the thrasholds $\{ y_k \}$ obtained frmm rhe pseudo-code. For exemple, the weight of jhe first [oaition will be eqyao to $i_1$. ![Plot of gye xvedaje selay oerhormance of OSA and MPQ as a function of uhe bumber of useds $N$. Wq rave also plotted the entropy of the thgeshkld random variable coreesponding to MPA.[]{data-pabel="fig:ofa_vs_mpa"}](osa_vs_mpa.pdf) In Figure \[fig:osa\_vs\_mpa\], we plot tha avecaee bcjxt oerformance of OSA (as described in Section \[sec:jaa\]) akd the maximal pvobability allocatooj vjde. As expectgd, the 'srrormance of OSA anf MPA ate simular and yn fsct, MPA performs better thab OSA as it pdenrifies the optimal thresholds witnout sny approximations (see Xemark \[dem:osa\]). We hwve also luotted in the Fieurv \[fig:msa\_vs\_mpa\], the entropy of thq maximal proyability allpcatiog code in hits. As expected, the entgopy lf the randol variable reflects the average delay performsnwe mf the cjntenbion resolution algorithm as c functijn of $N$ very weml. Entro'y Minimizatyon -------------------- Entropy is w concave fuiction of the disrributiuv. The maximal lrobabiliny allication code identlfies z local minima nu uhw space of pronabkliey dmstrifgtions. From nimiged mumerkcal work (nig relorted in this paper), we donjecture that thr iaximal krobabiliey allocation code is a globallj entcopy mmnimizong strategy as well. The followinf theorem prjves the optymallty jf MPA for the $N = 2$ case. MPA is a delay minimizing wtrategy and an enrropy minimizing sttattgy for $N = 2$ cese. Let $(x, 1]$ be tha contention

an example, for $N = 2$, repeating procedure yield us = 0.5, y_2 y_4 0.875, \cdots$. Note the above values in fact the thresholds reported in \[tab:osa\_two\_users\]. 3. In Remark \[rem:osa\_2\], for the $N = 2$ case, we noticed the feedback from the base station corresponding to a threshold can be viewed the representation the itself. For general $N$, the feedback from the base station can still be viewed as the representation of the threshold, however, with the weights to a position computed the thresholds $\{ y_k \}$ from pseudo-code. For the of first position will equal to $y_1$. ![Plot of the average delay performance of OSA and MPA as a function of number of We have plotted entropy the threshold random to MPA.[]{data-label="fig:osa_vs_mpa"}](osa_vs_mpa.pdf) In Figure \[fig:osa\_vs\_mpa\], we delay performance of OSA (as described in Section and the probability allocation code. As expected, the of OSA and MPA are similar and in MPA performs better than OSA as it identifies the optimal thresholds without any approximations (see We have also plotted the Figure \[fig:osa\_vs\_mpa\], entropy the probability code in As expected, the entropy of the random variable reflects the average performance of the contention resolution algorithm as a function of well. Minimization -------------------- Entropy a concave function of distribution. maximal probability allocation code local in probability From numerical work (not reported this paper), we conjecture that maximal probability allocation code strategy as well. The following theorem proves the of MPA for the $N = 2$ MPA is a delay minimizing strategy and an entropy minimizing strategy for = 2$ $(x, 1]$ be the contention

an example, for $N = 2$, repeating the Above proceDure wIll YieLd Us $y_1 = 0.5, y_2 = 0.75, Y_3 = 0.25, y_4 = 0.875, \cdOts$. Note that the ABove Values are in fact the threSholdS rEPortED iN TablE \[tab:osa\_TWo\_USErs\]. 3. in reMarK \[rEM:oSa\_2\], for The $n = 2$ case, we Noticed thaT thE fEedback from tHE bAse station CorResponding to A thResholD cAn bE VieweD as The biNary rePResentAtion of thE tHResholD Itself. FOR GeNeraL $N$, the feedback from THe BAse station can sTill be ViEWeD AS thE biNary represEnTatioN Of the thREsHOLD, hoWEver, with the weIghts corresPOndIng to a PoSitIOn compUted fRoM The Thresholds $\{ y_K \}$ obtAined from The pseUDo-code. FOR examplE, the weIghT of The fIRsT pOsiTiON wiLL bE eqUAl tO $y_1$. ![Plot of ThE aVeragE delAY PERforManCe of oSA anD MPA as a functiOn oF the NUmbEr of uSers $N$. we haVe Also pLotted The enTrOpy of the threshoLd raNdom variaBle CoRreSpOndinG To MPA.[]{dAta-LabEl="fig:osA_vs_mpa"}](oSA_vs_MpA.PDF) IN Figure \[fig:osa\_vs\_mpa\], We PLOt The averaGe delaY PeRfORmance of oSa (as DescRIBed in sectIOn \[Sec:osa\]) anD the maXImAl ProbabiLiTy alloCaTioN coDe. As eXPectEd, the pErformanCe of Osa and MPA are simiLAr and in fact, MPa PeRFOrMS betTer Than OSA as it IdenTIfieS the OPtImaL ThresHolds WiTHoUT any approximations (sEe remark \[Rem:osA\]). We have also plOtted in the fIGUre \[fig:osA\_vs\_mPA\], tHE entropy of the mAximaL probabiliTY allocatIon coDe in bits. as expecteD, THe entropY of The RanDom VARiAble reflects tHE AverAgE delay pErfOrmance Of tHe cOntEntIoN resolutiOn algoriThM aS a FuNctIon of $n$ Very well. enTroPy minImizaTIon -------------------- EntRopy iS a coNcAvE FunCtion of THe DIStriBuTiOn. ThE maXiMal prObabILitY allocaTion code iDenTIfieS a LoCal miniMa in the space oF pRobability DiStrIbutioNS. from limiTed numerical work (not repoRTed in thIs pAper), wE conJecture thAt tHe maxiMal PRobabiLity alLocatIoN coDE Is a glOBAlLy eNtRopy minimiZINg sTrateGy As weLl. The foLlowing theorem provES thE optimality of mPA For tHE $n = 2$ cAse. mpA IS a dElAY miNIMizing strategy aNd an entropY mINiMizing straTEgy FoR $N = 2$ case. LEt $(x, 1]$ be thE contENtion

an example, for $N = 2$,repeatingthe a bov e p ro cedu re w ill yield us $ y _1 = 0.5, y_2 = 0.75, y_3= 0.2 5, y_4= 0 .875, \cdots $ .N o teth at th ea bo ve va lue s are i n fact the th re sholds repor t ed in Table\[t ab:osa\_two\ _us ers\]. 3.In Re mar k \[r em:osa \ _2\],for the $ N= 2$ ca s e, we n o t ic ed t hat the feedbackf ro m the base stat ion co rr e sp o n din g t o a thresh ol d can be view e da s the binary repres entation of the thres ho ldi tself. Forge n era l $N$, thefeed back from the b a se stat i on canstillbevie weda sth e b in a ryr ep res e nta tion ofth ethres hold , h o weve r,with theweights corre spo ndin g to a po sitio n co mp utedfrom t he th re sholds $\{ y_k\}$obtainedfro mthe p seudo - code.For ex ample,the wei g htof t h efirst position wil lb e e qual to$y_1$. ! [P l ot of th eave rage d elayperf o rm ance ofOSA an d M PA as a f un ctionof th e n umber of u sers $ N$. We h ave a l so plotted the entropy of th e t h r es h oldran dom variabl e co r resp ondi n gtoM PA.[] {data -l a be l ="fig:osa_vs_mpa"}] (o sa_vs_ mpa.p df) In Figur e \[fig:os a \ _ vs\_mpa\ ], w e p l ot the average dela y performa n ce of OS A (as describ ed in Sec t i on \[sec :os a\] ) a ndt h emaximal proba b i lity a llocati oncode. A s e xpe cte d,th e perform ance ofOS Aan dMPA ares imilar a nd in f act , MPA perfor ms be tter t ha n OS A as it id e n tifi es t he o pti ma l thr esho l dswithout any appr oxi m atio ns ( see Rem ark \[rem:osa \] ). We have a lso plott e d in theFigure \[fig:osa\_vs\_m p a\], th e e ntrop y of the maxi mal proba bil i ty all ocatio n cod einb i ts. A s ex pec te d, the ent r o pyof th erand om vari able reflects thea ver age delay per for manc e of th e c o nte nt i onr e solution algori thm as a f un c ti on of $N$v ery w ell. E ntropyMinim i zation--------- --------- -- En t r opy is a conc ave func tion of t h e dis t ri butio n.The ma xi mal prob abilit y al locat ion co de ident ifies a local m inima in the space of p robabi litydis tribution s.F rom limitednume rical work (n otrepor ted in th is p a pe r), we co njec t ure thatt he ma x i ma l probabili t y all ocati onc ode is a g lobally entropy m i nimizing strat egya s we ll. Thefo llowing theore m p ro v e s the op ti mality of M PA for t he $N =2$ cas e. MP A is ad e la y minim izin g s trategy a ndan entropy m in i mizing str at egy fo r $N = 2$ c a s e. Let $(x, 1]$ be t h e cont e nti on

an_example, for_$N = 2$, repeating_the above_procedure_will yield_us_$y_1 = 0.5,_y_2 = 0.75,_y_3 = 0.25, y_4_= 0.875, \cdots$._Note_that the above values are in fact the thresholds reported in Table \[tab:osa\_two\_users\]. 3. In_Remark \[rem:osa\_2\],_for the_$N_=_2$ case, we noticed that_the feedback from the base_station corresponding_to a threshold can be viewed as the_binary_representation of the_threshold itself. For general $N$, the feedback from the_base station can still be viewed_as the binary_representation_of_the threshold, however, with_the weights corresponding to a position_computed from the thresholds $\{ y_k_\}$ obtained from the pseudo-code. For example,_the weight of the first position_will be equal to $y_1$. ![Plot_of the_average delay performance of OSA_and MPA as_a function_of the number_of users $N$. We have also_plotted the entropy_of the threshold random variable corresponding_to_MPA.[]{data-label="fig:osa_vs_mpa"}](osa_vs_mpa.pdf) In Figure \[fig:osa\_vs\_mpa\], we_plot_the_average delay_performance of OSA_(as_described in_Section \[sec:osa\])_and the maximal probability allocation code._As_expected, the performance of OSA and MPA_are similar and in_fact,_MPA performs better than_OSA as it identifies the_optimal thresholds without any approximations (see_Remark \[rem:osa\]). We_have also_plotted in the Figure \[fig:osa\_vs\_mpa\], the entropy of the maximal probability allocation_code in bits. As expected, the_entropy of the random_variable reflects_the_average delay performance_of_the contention_resolution algorithm as a function of $N$_very well. Entropy_Minimization -------------------- Entropy is a concave function of_the distribution. The maximal_probability_allocation code identifies a local minima_in the space of probability distributions._From limited numerical work (not_reported_in_this paper), we conjecture that_the maximal probability allocation code is_a globally entropy_minimizing strategy as well. The following theorem_proves_the optimality of MPA for the_$N_= 2$ case. MPA is a delay_minimizing_strategy_and an entropy minimizing strategy_for $N = 2$ case. Let $(x,_1]$ be the contention

uniformly along the surface [@Smoluchowski1941], thus giving weak corrugation. The crude proposal (\[eq:4\]) might be viewed as the precursor to the effective-medium theory [@EMT]. {width="85.00000%"} The form (\[eq:4\]) provides an interpretation of the mechanism by which the increasing density corrugation for (111) $<$ (100) $<$ (110) causes increasing amplitudes of modulation $V_1(z)$ in the physisorption potentials. The min-to-max variation of the rotationally averaged, lateral periodic corrugation $V_1(z)$ is modeled with an amplitude function [@HarrisAndLiebsch1982b], like in Eq. (\[eq:2\]), $V_1(z) = V_1^\prime \exp(-\beta z)$. Here the exponent $\beta$ is related to the exponent $\alpha$ of $V_0(z)$ via $\beta = \alpha/2 + \sqrt{(\alpha/2)^2 + G_{10}^2}$. The strength prefactor $V_1^\prime$ is adjusted so that the calculated intensities of the first-order $\mathbf{G}_{10}$ diffraction beams agree with measured values. The simple message of the experimental characterization in Figure \[fig:1\](a) is that, at the optimal separation and out, the $V_1$ corrugation terms are rather weak compared to the $V_0$ averages. This observation confirms that the basic particle-surface interaction is predominantly one dimensional. ![\[fig:1\] (a) Physisorption interaction potentials, $V$, for H$_2$ (D$_2$) on Cu(111) (circles), Cu(100) (squares), and Cu(110) (triangles), in the form of lateral average, $V_0(z)$ and corrugation $V_1(z)$. The potential functions $V_0$ and $V_1$ are defined in the text. The position $z$ of the molecular center of mass is here given with respect to the classical turning point, i.e. the position $z_t$ where a classical particle at energy $\epsilon_i = 0$ would be reflected in the potential. Adapted from [@andersson1993]. (b) The corresponding PECs calculated with vdW-DF2

uniformly along the surface [ @Smoluchowski1941 ], thus giving weak corrugation. The unrefined marriage proposal (\[eq:4\ ]) might be viewed as the precursor to the effective - culture medium hypothesis [ @EMT ]. ! [ image](fig1CuContourPlots.png){width="85.00000% " } The form (\[eq:4\ ]) provides an interpretation of the mechanism by which the increase density corrugation for (111) $ < $ (100) $ < $ (110) causes increase amplitudes of modulation $ V_1(z)$ in the physisorption electric potential. The min - to - max magnetic declination of the rotationally average, lateral periodic corrugation $ V_1(z)$ is modeled with an amplitude function [ @HarrisAndLiebsch1982b ], like in Eq.   (\[eq:2\ ]), $ V_1(z) = V_1^\prime \exp(-\beta z)$. Here the exponent $ \beta$ is related to the exponent $ \alpha$ of $ V_0(z)$ via $ \beta = \alpha/2 + \sqrt{(\alpha/2)^2 + G_{10}^2}$. The military capability prefactor $ V_1^\prime$ is adjusted so that the calculated intensities of the beginning - order $ \mathbf{G}_{10}$ diffraction beams agree with measured value. The simple message of the experimental characterization in Figure \[fig:1\](a) is that, at the optimal separation and out, the $ V_1 $ corrugation price are rather weak compared to the $ V_0 $ averages. This observation confirm that the basic atom - surface interaction is predominantly one dimensional. ! [ \[fig:1\ ] (a) Physisorption interaction potentials, $ V$, for H$_2 $ (D$_2 $) on Cu(111) (circles), Cu(100) (squares), and Cu(110) (triangles), in the form of lateral average, $ V_0(z)$ and corrugation $ V_1(z)$. The potential routine $ V_0 $ and $ V_1 $ are specify in the text. The position $ z$ of the molecular plaza of mass is here given with respect to the classical turn point, i.e. the stead $ z_t$ where a classical particle at department of energy $ \epsilon_i = 0 $ would be reflected in the potential. Adapted from [ @andersson1993 ]. (b) The corresponding PECs calculated with vdW - DF2

unlformly along the surfact [@Smoluchowski1941], thos givinj weak dorrugatkon. The crude proposal (\[eq:4\]) mijht ve vitced as the precursor to the evfective-nedinm theory [@EMT]. {width="85.00000%"} The form (\[gq:4\]) provides dn interpretathov lf the mechanism by which the increwsing drndity corrugatijn fpw (111) $<$ (100) $<$ (110) causes increasing amplitudes of modulauion $V_1(z)$ in the phyxisorption potentials. The lin-tl-max variation of hhe rotatioballr averaged, lageral periodic corrugajion $V_1(z)$ is modeled with an ampligude yunction [@HatxusAjgLiebsch1982b], lmke in Eq. (\[eq:2\]), $V_1(z) = V_1^\pvpme \exp(-\teta z)$. Nere the exponcnt $\bxta$ us related to the expmnent $\alpha$ of $V_0(z)$ dia $\beta = \ampha/2 + \sqrt{(\alpha/2)^2 + T_{10}^2}$. Rhe sjrengdh pfwfaztod $T_1^\prjme$ is adousted so tgat the calxulated intensities os the first-ordsr $\matrbs{G}_{10}$ diffraction beams agree with measureg vzlues. The simple message of the experimental fharacteryzation in Figure \[fig:1\](a) is that, at the optimal sepdratikv aue out, rhf $V_1$ corrugation terms are rather weak compareq tp nhe $V_0$ averages. Thls observation congigmx that the baskc parcjcme-surface interactlon is kredomunantly oge domensional. ![\[fig:1\] (a) Physisorptiin interactijb potentials, $V$, for H$_2$ (D$_2$) on Cu(111) (eirclex), Cu(100) (xquares), and Cu(110) (triangler), in the form ov lateral xverage, $V_0(z)$ and currlgathon $V_1(z)$. The potential functyons $V_0$ anv $V_1$ axe defindd im the eext. The plsition $z$ of the moleculwr ceuter mf mass is here given with respect to the classical turmitg ioint, i.e. che poxition $z_t$ whewe a classical particje at energy $\epailon_i = 0$ would be weflected in dje potential. Adapted fron [@aneersson1993]. (c) The correspomding PECs calculatwd with vdW-DF2

uniformly along the surface [@Smoluchowski1941], thus giving The proposal (\[eq:4\]) be viewed as theory {width="85.00000%"} The form provides an interpretation the mechanism by which the increasing corrugation for (111) $<$ (100) $<$ (110) causes increasing amplitudes of modulation $V_1(z)$ the physisorption potentials. The min-to-max variation of the rotationally averaged, lateral periodic corrugation is with amplitude [@HarrisAndLiebsch1982b], like in Eq. (\[eq:2\]), $V_1(z) = V_1^\prime \exp(-\beta z)$. Here the exponent $\beta$ is related the exponent $\alpha$ of $V_0(z)$ via $\beta = + \sqrt{(\alpha/2)^2 + G_{10}^2}$. strength prefactor $V_1^\prime$ is adjusted that calculated intensities the $\mathbf{G}_{10}$ beams agree with values. The simple message of the experimental characterization in Figure \[fig:1\](a) is that, at the optimal separation out, the terms are weak to $V_0$ averages. This that the basic particle-surface interaction is ![\[fig:1\] (a) Physisorption interaction potentials, $V$, for H$_2$ on Cu(111) Cu(100) (squares), and Cu(110) (triangles), in form of lateral average, $V_0(z)$ and corrugation $V_1(z)$. potential functions $V_0$ and $V_1$ are defined in the text. The position $z$ of the of mass is here with respect to classical point, the $z_t$ where classical particle at energy $\epsilon_i = 0$ would be reflected in potential. Adapted from [@andersson1993]. (b) The corresponding PECs calculated with

uniformly along the surface [@SMoluchowskI1941], thus GivIng WeAk coRrugAtion. The crude pROposAl (\[eq:4\]) might be viewed as the PrecuRsOR to tHE eFfectIve-mediUM tHEOry [@eMt]. {width="85.00000%"} The fOrm (\[Eq:4\]) Provides an inTErPretation oF thE mechanism by WhiCh the iNcReaSIng deNsiTy corRugatiON for (111) $<$ (100) $<$ (110) caUses increAsINg amplITudes of MODuLatiOn $V_1(z)$ in the physisorPTiON potentials. The Min-to-mAx VArIATioN of The rotatioNaLly avERaged, laTErAL PEriODic corrugatioN $V_1(z)$ is modeleD WitH an ampLiTudE FunctiOn [@HarRiSandliebsch1982b], likE in EQ. (\[eq:2\]), $V_1(z) = V_1^\priMe \exp(-\bETa z)$. Here THe exponEnt $\betA$ is RelAted TO tHe ExpOnENt $\aLPhA$ of $v_0(Z)$ viA $\beta = \alpHa/2 + \SqRt{(\alpHa/2)^2 + G_{10}^2}$. THE STRengTh pRefaCtor $V_1^\Prime$ is adjustEd sO thaT The CalcuLated InteNsIties Of the fIrst-oRdEr $\mathbf{G}_{10}$ diffraCtioN beams agrEe wItH meAsUred vALues. ThE siMplE messagE of the eXPerImENTAl Characterization in fiGURe \[Fig:1\](a) is thAt, at thE OpTiMAl separaTiOn aNd ouT, THe $V_1$ coRrugATiOn terms aRe rathER wEaK comparEd To the $V_0$ AvEraGes. this oBServAtion cOnfirms tHat thE Basic particle-sURface interactIOn IS PrEDomiNanTly one dimenSionAL. ![\[fig:1\] (A) PhySIsOrpTIon inTeracTiON pOTentials, $V$, for H$_2$ (D$_2$) on Cu(111) (cIrCles), Cu(100) (SquarEs), and Cu(110) (triangLes), in the foRM OF lateral AverAGe, $v_0(Z)$ and corrugatioN $V_1(z)$. ThE potential FUnctions $v_0$ and $V_1$ Are definEd in the teXT. the positIon $Z$ of The MolECUlAr center of masS IS herE gIven witH reSpect to The ClaSsiCal TuRning poinT, i.e. the poSiTiOn $Z_t$ WheRe a clASsical paRtIclE aT enErgy $\ePSilon_i = 0$ Would Be reFlEcTEd iN the potENtIAL. AdaPtEd From [@AndErSson1993]. (b) the cORreSpondinG PECs calcUlaTEd wiTh Vdw-DF2

uniformly along the surfa ce [@Smolu chows ki1 941 ], thu s gi ving weak corr u gati on. The crude proposal (\[e q: 4 \])m ig ht be viewed as t hepr ec urs or to theeff ective- medium the ory [ @EMT].  {wi dth="8 5. 000 0 0%"} Th e for m (\[e q :4\])providesan interp r etation o fthemechanism by whic h t h e increasing d ensity c o rr u g ati onfor (111)$< $ (10 0 ) $<$ ( 1 10 ) c aus e s increasingamplitudeso f m odulat io n $ V _1(z)$ in t he phy sisorptionpote ntials. T he min - to-maxv ariatio n of t herot atio n al ly av er a ged , l ate r alperiodic c or rugat ion$ V _ 1 (z)$ is mod eledwith an ampli tud e fu n cti on [@ Harri sAnd Li ebsch 1982b] , lik ein Eq. (\[eq:2\ ]),$V_1(z) = V_ 1^ \pr im e \ex p (-\bet a z )$. Here t he expo n ent $ \ b e ta $ is related to th ee x po nent $\a lpha$o f$V _ 0(z)$ vi a$\b eta= \alph a/2+ \ sqrt{(\a lpha/2 ) ^2 + G_{10} ^2 }$. Th estr eng th pr e fact or $V_ 1^\prime $ isa djusted so tha t the calculat e di n te n siti esof the firs t-or d er $ \mat h bf {G} _ {10}$ diff ra c ti o n beams agree withme asured valu es. The simp le message o f the exp erim e nt a l characteriza tionin Figure\ [fig:1\] (a) i s that,at the op t i mal sepa rat ion an d o u t ,the $V_1$ cor r u gati on termsare rather we akcom par ed to the $ V_0$ ave ra ge s. T his obse r vation c on fir ms th at th e basic part icle -s ur f ace intera c ti o n ispr ed omin ant ly onedime n sio nal. ! [\[fig:1\ ] ( a ) Ph ys is orption interactionpo tentials,$V $,for H$ _ 2 $ (D$_2$ ) on Cu(111) (circles), Cu(100) (s quare s),and Cu(11 0)(trian gle s ), inthe fo rm of l ate r a l ave r a ge , $ V_ 0(z)$ andc o rru gatio n$V_1 (z)$. T he potential funct i ons $V_0$ and $V _1$ are d ef ine d i n th et ext . The position $z $ of the m ol e cu lar center ofma ss is h ere giv en wi t h respe ct to the classica lturn i n g p oint, i.e. the pos ition $z_ t $ whe r ea cla ssi cal pa rt icl e atenergy $\e psilo n_i =0$ would be r ef lected i n the potential. Adapte d from [@an der sson1993] . ( b ) T he corres pond ing PECs c alc ula ted w ith vdW-D F2

uniformly_along the_surface [@Smoluchowski1941], thus giving_weak corrugation._The_crude proposal_(\[eq:4\])_might be viewed_as the precursor_to the effective-medium theory_[@EMT]. {width="85.00000%"} The form (\[eq:4\])_provides_an interpretation of the mechanism by which the increasing density corrugation for (111) $<$_(100)_$<$ (110)_causes_increasing_amplitudes of modulation $V_1(z)$ in_the physisorption potentials. The min-to-max_variation of_the rotationally averaged, lateral periodic corrugation $V_1(z)$ is_modeled_with an amplitude_function [@HarrisAndLiebsch1982b], like in Eq. (\[eq:2\]), $V_1(z) = V_1^\prime \exp(-\beta_z)$. Here the exponent $\beta$ is_related to the_exponent_$\alpha$_of $V_0(z)$ via $\beta_= \alpha/2 + \sqrt{(\alpha/2)^2 + G_{10}^2}$._The strength prefactor $V_1^\prime$ is adjusted_so that the calculated intensities of the_first-order $\mathbf{G}_{10}$ diffraction beams agree with_measured values. The simple message of_the experimental_characterization in Figure \[fig:1\](a) is_that, at the_optimal separation_and out, the_$V_1$ corrugation terms are rather weak_compared to the_$V_0$ averages. This observation confirms that_the_basic particle-surface interaction_is_predominantly_one dimensional. ![\[fig:1\]_(a) Physisorption interaction_potentials,_$V$, for_H$_2$_(D$_2$) on Cu(111) (circles), Cu(100) (squares),_and_Cu(110) (triangles), in the form of lateral_average, $V_0(z)$ and corrugation_$V_1(z)$._The potential functions $V_0$_and $V_1$ are defined in_the text. The position $z$ of_the molecular_center of_mass is here given with respect to the classical turning point,_i.e. the position $z_t$ where a_classical particle at energy_$\epsilon_i =_0$_would be reflected_in_the potential._Adapted from [@andersson1993]. (b) The corresponding PECs_calculated with_vdW-DF2

shells, hence line emission from such shells is strongly enhanced compared to collisional thermal radiations. For instance, the Ly$\rm \delta$ line at 14.8 $\rm \AA$ is stronger than the Ly$\rm \gamma$ line at 15.2 $\rm \AA$. Similar conditions can be found in many other transitions, e.g., the He$\rm \delta$ line at 17.4 $\rm \AA$, the Ly$\rm \delta$ line at 19.4 $\rm \AA$, and Ly$\rm \delta$ line at 26.4 $\rm \AA$. The derived CX spectrum also features a large G ratio, i.e., forbidden plus intercombination lines to resonance line ratio, since the collisional excitation process becomes negligible in CX plasmas (§2). Bias from systematic weight --------------------------- As reported recently by Nolte et al. (2012), the singlet-to-triplet ratio of the $\rm C^{4+}$ ion, produced from a low-velocity $\rm C^{5+}$ $+ \rm H$ reaction, covers a broad range of $0.01 - 100$ for different $n$, $l$, and $v$. The apparent bias from the commonly adopted, statistical value 3 is probably caused by electron-electron interaction during the capture. A similar effect was also reported in other papers (e.g., Wu et al. 2011 for $\rm N^{6+}$ $+ \rm H$ reaction), indicating that it could be a common property. Although the current data are yet too sparse to fully implement the $S-$ dependence, it is vital to estimate the induced biases on emission line ratios. As shown in Figure 4, we compare the G ratio calculated based on the data from Nolte et al. (2012), with that assuming statistical weight for the $\rm C^{5+}$ $+ \rm H$ reaction. The largest bias is seen at $v \sim 500$ km $\rm s^{-1}$, where the statistical weight is underestimated by about 30%. The two ratios become roughly consistent at low and high velocity ends. The figure suggests that the current code can well cover the actual G ratio range, although the derived velocity would have a systematic error up to $200-300$ km $\rm s^{-1}$. Real data fitting ================= To verify the CX model, we fit it to the real data of comet C/2000 WM1 (LINEAR) observed by the

shells, hence line emission from such carapace is powerfully enhanced compared to collisional thermal radiation sickness. For example, the Ly$\rm \delta$ line at 14.8 $ \rm \AA$ is stronger than the Ly$\rm \gamma$ tune at 15.2 $ \rm \AA$. like conditions can be found in many other transitions, for example, the He$\rm \delta$ line at 17.4 $ \rm \AA$, the Ly$\rm \delta$ line at 19.4 $ \rm \AA$, and Ly$\rm \delta$ line at 26.4 $ \rm \AA$. The derived CX spectrum besides features a large deoxyguanosine monophosphate ratio, i.e., forbidden plus intercombination lines to resonance channel ratio, since the collisional excitation process become negligible in CX plasmas (§ 2). Bias from systematic weight --------------------------- As reported recently by Nolte et al. (2012), the singlet - to - triplet proportion of the $ \rm C^{4+}$ ion, produced from a broken - velocity $ \rm C^{5+}$ $ + \rm H$ reaction, covers a wide range of $ 0.01 - 100 $ for different $ n$, $ l$, and $ v$. The apparent bias from the commonly adopted, statistical value 3 is probably caused by electron - electron interaction during the capture. A similar effect was also reported in other papers (e.g., Wu et al. 2011 for $ \rm N^{6+}$ $ + \rm H$ reaction), indicating that it could be a common place. Although the current datum are yet too sparse to fully enforce the $ S-$ dependence, it is vital to estimate the induce bias on emission line ratio. As shown in Figure 4, we compare the G ratio calculated based on the datum from Nolte et al. (2012), with that assuming statistical weight for the $ \rm C^{5+}$ $ + \rm H$ reaction. The largest bias is seen at $ v \sim 500 $ kilometer $ \rm s^{-1}$, where the statistical system of weights is underestimated by about 30% . The two ratios become roughly reproducible at low and high velocity end. The figure suggests that the current code can well cover the actual G proportion range, although the derived velocity would have a systematic mistake up to $ 200 - 300 $ km $ \rm s^{-1}$. actual data match = = = = = = = = = = = = = = = = = To verify the CX exemplar, we fit it to the real datum of comet C/2000 WM1 (LINEAR) observed by the

shflls, hence line emission from such shells is svrongly enhancea compared to collisional thxrmao raduations. For instance, tfe Ly$\rm \dvlta$ line at 14.8 $\em \AA$ is svdonger bkan tgc Ly$\rk \gamma$ line at 15.2 $\rm \AA$. Sikilar conditiots ccn be found in many other transitionf, e.g., thr Je$\rm \delta$ ling at 17.4 $\wm \AZ$, the Ly$\rm \delta$ line at 19.4 $\rm \AA$, ans Ly$\rm \velta$ line at 26.4 $\tm \AA$. The derived CX spectgum wlso features a lagge G ratio, i.e., sirbidden plur intercombination lings to resonance line ratio, since the eollisional ezcihdtion procews bebomes negliginke in WX plaskas (§2). Bias from xysvemaric weight --------------------------- As reported recently by Nolte et al. (2012), tve singlet-to-triplet rqtio mf tve $\ro C^{4+}$ kon, pcodhced fgom a low-velodity $\rm C^{5+}$ $+ \em H$ reaction, coverx w broad range kf $0.01 - 100$ sow different $n$, $l$, and $v$. The apparent bias frkm the commonly adopted, statistical value 3 id probablr caused by electron-electron interaction during tve ca'tjre. A rumllar effect was also reported in other papers (s.g., Wl et al. 2011 for $\rm N^{6+}$ $+ \rm H$ reacyiln), yndicating thxt it ekumd be a common prooerty. Ajthouth the cuwreny data are yet too sparse ti fully implvmenr the $S-$ dependence, it is vitau to estikate the induced biases on smission lije ratios. Xs shown in Figufe 4, wa compart the G ratio calctlated bawed pn the aata from Golte et ap. (2012), wibv that assuming stwtistncal feight for the $\rm C^{5+}$ $+ \rm H$ reaction. The lacjest bias is xean dt $v \sim 500$ km $\vm s^{-1}$, where the ftatistical wenght is bnderertimated bj about 30%. Vhe two ratijs become roucjly consisteit at low and higy velockgy ends. The fibure suggvsus that thw current code can weum cover the actbcl G ratio range, sltfoudh tie dewhved velocitf wojld nave x systematig efror up to $200-300$ km $\rm s^{-1}$. Real gata fitting ================= To verify yhc CX modeo, we fit it to the resl data of comet C/2000 WM1 (LMNEAR) ubserfed by the

shells, hence line emission from such shells enhanced to collisional radiations. For instance, 14.8 \AA$ is stronger the Ly$\rm \gamma$ at 15.2 $\rm \AA$. Similar conditions be found in many other transitions, e.g., the He$\rm \delta$ line at 17.4 \AA$, the Ly$\rm \delta$ line at 19.4 $\rm \AA$, and Ly$\rm \delta$ line 26.4 \AA$. derived spectrum also features a large G ratio, i.e., forbidden plus intercombination lines to resonance line ratio, the collisional excitation process becomes negligible in CX (§2). Bias from systematic --------------------------- As reported recently by et (2012), the ratio the C^{4+}$ ion, produced a low-velocity $\rm C^{5+}$ $+ \rm H$ reaction, covers a broad range of $0.01 - 100$ for $n$, $l$, The apparent from commonly statistical value 3 caused by electron-electron interaction during the effect was also reported in other papers (e.g., et al. for $\rm N^{6+}$ $+ \rm H$ indicating that it could be a common property. the current data are yet too sparse to fully implement the $S-$ dependence, it is estimate the induced biases emission line ratios. shown Figure we the G calculated based on the data from Nolte et al. (2012), with assuming statistical weight for the $\rm C^{5+}$ $+ \rm H$ largest is seen at \sim 500$ km $\rm where statistical weight is underestimated 30%. two consistent low high velocity ends. The suggests that the current code well cover the actual derived velocity would have a systematic error up $200-300$ km $\rm s^{-1}$. Real data fitting To verify the CX model, we fit it to the real data comet C/2000 observed by the

shells, hence line emission frOm such shelLs is sTroNglY eNhanCed cOmpared to colliSIonaL thermal radiations. For iNstanCe, THe Ly$\RM \dElta$ lIne at 14.8 $\rm \aa$ iS STroNgEr ThaN tHE LY$\rm \gaMma$ Line at 15.2 $\rM \AA$. Similar ConDiTions can be foUNd In many otheR trAnsitions, e.g., tHe HE$\rm \delTa$ LinE At 17.4 $\rm \Aa$, thE Ly$\rm \Delta$ lINe at 19.4 $\rm \aA$, and Ly$\rm \DeLTa$ line AT 26.4 $\rm \AA$. ThE DErIved cX spectrum also feaTUrES a large G ratio, i.E., forbiDdEN pLUS inTerCombinatioN lInes tO ResonanCE lINE RatIO, since the collIsional exciTAtiOn procEsS beCOmes neGligiBlE In Cx plasmas (§2). BiaS froM systematIc weigHT --------------------------- As repoRTed receNtly by nolTe eT al. (2012), tHE sInGleT-tO-TriPLeT raTIo oF the $\rm C^{4+}$ iOn, PrOduceD froM A LOW-velOciTy $\rm c^{5+}$ $+ \rm H$ rEaction, covers A brOad rANge Of $0.01 - 100$ for DiffeRent $N$, $l$, And $v$. THe appaRent bIaS from the commonlY adoPted, statiStiCaL vaLuE 3 is prOBably cAusEd bY electrOn-electROn iNtERACtIon during the capturE. A SIMiLar effecT was alSO rEpORted in otHeR paPers (E.G., wu et aL. 2011 for $\RM N^{6+}$ $+ \Rm H$ reactIon), indICaTiNg that iT cOuld be A cOmmOn pRoperTY. AltHough tHe currenT data ARe yet too sparse TO fully implemeNT tHE $s-$ dEPendEncE, it is vital tO estIMate The iNDuCed BIases On emiSsIOn LIne ratios. As shown in FIgUre 4, we cOmparE the G ratio calCulated basED ON the data From nOlTE et al. (2012), with that aSsumiNg statistiCAl weight For thE $\rm C^{5+}$ $+ \rm H$ rEaction. ThE LArgest biAs iS seEn aT $v \sIM 500$ Km $\Rm s^{-1}$, where the stATIstiCaL weight Is uNderestImaTed By aBouT 30%. THe two ratiOs become RoUgHlY cOnsIstenT At low and HiGh vElOciTy endS. the figUre suGgesTs ThAT thE currenT CoDE Can wElL cOver The AcTual G RatiO RanGe, althoUgh the derIveD VeloCiTy Would haVe a systematic ErRor up to $200-300$ km $\rM s^{-1}$. reaL data fITTing ================= To veRify the CX model, we fit it to THe real dAta Of comEt C/2000 Wm1 (LINEAR) obSerVed by tHe

shells, hence line emissi on from su ch sh ell s i sstro ngly enhanced comp a redto collisional thermal radi at i ons. Fo r ins tance,t he L y$\ rm \ del ta $ l ine a t 1 4.8 $\r m \AA$ isstr on ger than the Ly $\rm \gamm a$line at 15.2 $\ rm \AA $. Si m ilarcon ditio ns can be fou nd in man yo ther t r ansitio n s ,e.g. , the He$\rm \del t a$ line at 17.4 $ \rm \A A$ , t h e Ly $\r m \delta$li ne at 19.4 $\ r m\ A A $,a nd Ly$\rm \de lta$ line a t 26 .4 $\r m\AA $ . Thederiv ed CXspectrum al so f eatures a large G ratio , i.e.,forbid den pl us i n te rc omb in a tio n l ine s to resonan ce l ine r atio , s i ncethe col lisio nal excitatio n p roce s s b ecome s neg ligi bl e inCX pla smas(§ 2). Bias fromsyst ematic we igh t--- -- ----- - ------ --- --- ---- A s repor t edre c e n tl y by Nolte et al.(2 0 1 2) , the si nglet- t o- tr i plet rat io of the $ \rm C ^{4+ } $ion, pro ducedf ro ma low-v el ocity$\ rmC^{ 5+}$$ + \r m H$ r eaction, cove r s a broad rang e of $0.01 - 1 0 0$ f or diff ere nt $n$, $l$ , an d $v$ . Th e a ppa r ent b ias f ro m t h e commonly adopted, s tatist icalvalue 3 is pr obably cau s e d by elec tron - el e ctron interact ion d uring thec apture.A sim ilar eff ect was a l s o report edinoth erp a pe rs (e.g., Wue t al. 2 011 for $\ rm N^{6 +}$ $+ \r m H $reaction) , indica ti ng t ha t i t cou l d be a c om mon p rop erty. Althou gh th e cu rr en t da ta arey et t oo s pa rs e to fu ll y imp leme n t t he $S-$ dependen ce, it i svi tal toestimate thein duced bias es on emiss i o n line r atios. As shown in Figu r e 4, we co mpare the G ratiocal culate d b a sed on the d ata f ro m N o l te et a l. (2 01 2), with t h a t a ssumi ng sta tistica l weight for the $ \ rmC^{5+}$ $+ \r m H $ re a c ti on. Th e la rg e stb i as is seen at $ v \sim 500 $k m$\rm s^{-1 } $,wh ere the statis tical weightis undere stimatedby abo u t 30 %. The two ratiosbecome ro u ghlyc on siste ntat low a ndhighveloci t y e nds.The fi gu re sug gests t hat thecurrent code can well c over t he ac tua l G ratio ra n ge, although the derived v elo cit y wou ldh ave a sys t em ati c erro r up to $200-3 0 0$ km $ \r m s^{-1}$.R e aldatafit t ing == ==== =========== To v e rify the CX mo del, w e f iti t to t he real data o f c om e t C/2000WM 1 (LINEAR)observed b y the

shells,_hence line_emission from such shells_is strongly_enhanced_compared to_collisional_thermal radiations. For_instance, the Ly$\rm_\delta$ line at 14.8_$\rm \AA$ is_stronger_than the Ly$\rm \gamma$ line at 15.2 $\rm \AA$. Similar conditions can be found_in_many other_transitions,_e.g.,_the He$\rm \delta$ line at_17.4 $\rm \AA$, the Ly$\rm_\delta$ line_at 19.4 $\rm \AA$, and Ly$\rm \delta$ line_at_26.4 $\rm \AA$._The derived CX spectrum also features a large G_ratio, i.e., forbidden plus intercombination lines_to resonance line_ratio,_since_the collisional excitation process_becomes negligible in CX plasmas (§2). Bias_from systematic weight --------------------------- As reported recently by_Nolte et al. (2012), the singlet-to-triplet ratio_of the $\rm C^{4+}$ ion, produced_from a low-velocity $\rm C^{5+}$_$+ \rm_H$ reaction, covers a broad_range of $0.01_- 100$_for different $n$,_$l$, and $v$. The apparent bias_from the commonly_adopted, statistical value 3 is probably_caused_by electron-electron interaction_during_the_capture. A_similar effect was_also_reported in_other_papers (e.g., Wu et al. 2011_for_$\rm N^{6+}$ $+ \rm H$ reaction), indicating_that it could be_a_common property. Although the_current data are yet too_sparse to fully implement the $S-$_dependence, it_is vital_to estimate the induced biases on emission line ratios. As shown_in Figure 4, we compare the_G ratio calculated based_on the_data_from Nolte et_al._(2012), with_that assuming statistical weight for the $\rm_C^{5+}$ $+_\rm H$ reaction. The largest bias_is seen at $v_\sim_500$ km $\rm s^{-1}$, where the_statistical weight is underestimated by about_30%. The two ratios become_roughly_consistent_at low and high velocity_ends. The figure suggests that the_current code can_well cover the actual G ratio range,_although_the derived velocity would have a_systematic_error up to $200-300$ km $\rm_s^{-1}$. Real_data_fitting ================= To verify the CX model,_we fit it to the real_data of comet C/2000 WM1 (LINEAR) observed by the

distance keeps on attaining linearity. At infinite time, the steady state values are linear with position as given by Eq. \[infinite-time\]. Similarly, the material energy density initially exhibits slight non-linear variation and finally attains the linearity (Fig. \[vxt-planar\]). The non-linear variation at early stages occurs due to net absorption of energy by the initially cold material (as u(x,0)=v(x,0)=0). Initially, the material energy density is found to lag behind the radiation energy densities and finally equilibrate as time proceeds (beyond $\tau$=10). In this work, all the results have been obtained by considering contribution from the first 30 roots of the transcendental equation. The value of opacity $\kappa$ is chosen to be 100 and $\epsilon $ equals 0.1. For a heat wave traveling into a thin plate and composite planar slab, a similar linear variation in temperature with distance was observed though difference existed in the space and time dependent behaviour due to heat conduction approximation [@Sarkar],[@Sun]. ![\[uxt-planar\]Scaled radiation energy density $u(x,\tau)$ vs position (x) in the slab of scaled thickness $b=1$ at different times for $ \epsilon = 0.1 $. The symbols stand for analytical values whereas lines represent the results obtained from finite difference method.](figure4.eps){width="8cm"} ![\[vxt-planar\]Scaled material energy density $v(x,\tau)$ vs position (x) in the slab at different times for $\epsilon = 0.1$. The symbols stand for analytical values whereas lines represent the results obtained from finite difference method.](figure5.eps){width="8cm"} The first derivatives w.r.t. position of the analytical radiation and material energy density are plotted in Figs. \[uprimext-planar\] and \[vprimext-planar\]. As the radiation and material energy densities decrease with x, the derivative has negative values. The derivative has a greater negative value at the left compared to the right zone. As both radiation and material energy densities keep on increasing with time due to radiation diffusion, magnitude of the gradient decreases for the left and increases for the right sides. The gradient of both radiation and material energy densities obtain a constant value of $\frac{-3}{3+4\sqrt{3}}=-0.30217$ after infinite time showing that there is a constant leakage of

distance keeps on attaining linearity. At infinite meter, the regular state values are linear with status as given by Eq. \[infinite - time\ ]. Similarly, the material department of energy density initially exhibits slight non - analogue mutant and finally reach the linearity (Fig. \[vxt - planar\ ]). The non - linear magnetic declination at early stage occurs due to net absorption of energy by the initially cold material (as u(x,0)=v(x,0)=0). Initially, the fabric energy density is found to imprison behind the radiation energy densities and finally equilibrate as clock time proceeds (beyond $ \tau$=10). In this work, all the results have been obtained by considering contribution from the inaugural 30 roots of the transcendental equation. The value of opacity $ \kappa$ is chosen to be 100 and $ \epsilon $ equals 0.1. For a heat wave traveling into a thin plate and composite planar slab, a similar linear variation in temperature with distance was observed though difference existed in the space and time dependent behavior due to heat conduction approximation [ @Sarkar],[@Sun ]. ! [ \[uxt - planar\]Scaled radiotherapy department of energy concentration $ u(x,\tau)$ vs position (x) in the slab of scaled thickness $ b=1 $ at different times for $ \epsilon = 0.1 $. The symbols stand for analytical values whereas line represent the results obtained from finite difference method.](figure4.eps){width="8 cm " } ! [ \[vxt - planar\]Scaled material energy density $ v(x,\tau)$ vs place (x) in the slab at different times for $ \epsilon = 0.1$. The symbols stand for analytic values whereas pipeline represent the results obtained from finite difference method.](figure5.eps){width="8 cm " } The inaugural derivatives w.r.t. position of the analytical radiation sickness and material energy density are plotted in Figs. \[uprimext - planar\ ] and \[vprimext - planar\ ]. As the radiotherapy and material department of energy densities decrease with x, the derivative has negative values. The derived function has a capital negative value at the left field compared to the right zone. As both radiation and fabric energy densities keep on increasing with time due to radiation diffusion, magnitude of the gradient decreases for the left and increases for the right side. The gradient of both radiation and material department of energy densities obtain a constant value of $ \frac{-3}{3 + 4\sqrt{3}}=-0.30217 $ after infinite time showing that there is a ceaseless leakage of

didtance keeps on attaininn linearity. At iudinite time, fhe steaay state values are linear wmth posituon as given by Eq. \[infknite-time\]. Similaroy, tie material enerjg densibv iniflally xxhibits slight non-linear variation and fknclly attains the linearity (Fig. \[vxt-plwnar\]). Thr jon-linear variwtiom at sarly stages occurs due to net abskrption of energy by the initially cold materiwl (ad u(x,0)=v(x,0)=0). Initially, tje material enewty density ir found to lag behind jhe radiation energy densities avd fiually equilubeatf as time priceedf (beyond $\tau$=10). In this work, akl the results hate bwen obtained by consivering contribution srom the xixst 30 roots of the trabsxendettal equxriov. Tge vzlue ov o'acity $\kappz$ is chosen to be 100 and $\epsilon $ qwuals 0.1. For a geat wwvq traveling into a thin plate and compovits planar slab, a similar linear variation in jemperaturq with distance was observed though difference exhsted kn uhc rpafe and time dependent behaviour due to heat cjhdicnion approximatiok [@Sarkar],[@Sun]. ![\[uxt-plansr\]Dcsjed radiation energv dsnsity $u(x,\tau)$ vs podition (v) in rhe slab jf svaled thickness $b=1$ at differwnt times fog $ \epsilon = 0.1 $. The symblls stand fur amalytocal values whereas linzs repdesent the gesults oggained from finige cixference method.](figure4.eps){wiqth="8cm"} ![\[vxt-poanax\]Scaled oatetial enqrgy densihy $v(x,\bdu)$ vs position (x) ij the snab at difverent times for $\epsilon = 0.1$. The symbols stand gos atalyticaj valmes whereas linqs represent tke resulcs obtxined from finite difference method.](figure5.aks){width="8cm"} The hirst deryvatuves w.r.t. porktion of the amalytical radiation and material enerny devaity are plotteb un Figs. \[uprimexy-plxnaw\] wnv \[vprykext-planar\]. Av thd rxciatiun and mateviau enrrgy densities decredse sith x, the derivatovc has negqtive vajues. The derifative has a greattr negetive talue st jhe left compared to the right aone. As blth radiation agd mqterial eneryy densities keep on increasing with timx due to radiation diffysion, magnitude of jhe gradient dxcreasqs for tha left and increases for the right sices. The gradient of bkth ragiatiln and material energy densities obtain a constant value of $\frac{-3}{3+4\sqrt{3}}=-0.30217$ aftee infiiiee time shosing that tkerz is a sonsvait leakage of

distance keeps on attaining linearity. At infinite steady values are with position as the energy density initially slight non-linear variation finally attains the linearity (Fig. \[vxt-planar\]). non-linear variation at early stages occurs due to net absorption of energy by initially cold material (as u(x,0)=v(x,0)=0). Initially, the material energy density is found to behind radiation densities finally equilibrate as time proceeds (beyond $\tau$=10). In this work, all the results have been obtained considering contribution from the first 30 roots of transcendental equation. The value opacity $\kappa$ is chosen to 100 $\epsilon $ 0.1. a wave traveling into thin plate and composite planar slab, a similar linear variation in temperature with distance was observed though existed in and time behaviour to conduction approximation [@Sarkar],[@Sun]. energy density $u(x,\tau)$ vs position (x) of scaled thickness $b=1$ at different times for \epsilon = $. The symbols stand for analytical whereas lines represent the results obtained from finite method.](figure4.eps){width="8cm"} ![\[vxt-planar\]Scaled material energy density $v(x,\tau)$ vs position (x) in the slab at different times = 0.1$. The symbols for analytical values lines the obtained finite difference The first derivatives w.r.t. position of the analytical radiation and material density are plotted in Figs. \[uprimext-planar\] and \[vprimext-planar\]. As the material densities decrease with the derivative has negative The has a greater negative the compared zone. both and material energy densities on increasing with time due radiation diffusion, magnitude of left and increases for the right sides. The of both radiation and material energy densities a constant value of $\frac{-3}{3+4\sqrt{3}}=-0.30217$ after infinite time showing that there is constant leakage

distance keeps on attaining lInearity. At InfinIte TimE, tHe stEady State values are LIneaR with position as given by eq. \[infInITe-tiME\]. SImilaRly, the mATeRIAl eNeRgY deNsITy InitiAllY exhibiTs slight noN-liNeAr variation aND fInally attaIns The linearity (fig. \[Vxt-plaNaR\]). ThE Non-liNeaR variAtion aT Early sTages occuRs DUe to neT AbsorptION oF eneRgy by the initially COlD Material (as u(x,0)=v(x,0)=0). initiaLlY, ThE MAteRiaL energy denSiTy is fOUnd to laG BeHIND thE Radiation enerGy densities ANd fInally EqUilIBrate aS time PrOCeeDs (beyond $\tau$=10). in thIs work, all The resULts have BEen obtaIned by ConSidErinG CoNtRibUtIOn fROm The FIrsT 30 roots of ThE tRanscEndeNTAL EquaTioN. The Value Of opacity $\kappA$ is ChosEN to Be 100 and $\EpsilOn $ eqUaLs 0.1. For A heat wAve trAvEling into a thin pLate And composIte PlAnaR sLab, a sIMilar lIneAr vAriatioN in tempERatUrE WITh Distance was observeD tHOUgH differeNce exiSTeD iN The space AnD tiMe dePENdent BehaVIoUr due to hEat conDUcTiOn approXiMation [@saRkaR],[@SuN]. ![\[uxt-pLAnar\]scaled RadiatioN enerGY density $u(x,\tau)$ vS Position (x) in thE SlAB Of SCaleD thIckness $b=1$ at dIffeREnt tImes FOr $ \EpsILon = 0.1 $. ThE symbOlS StANd for analytical valuEs WhereaS lineS represent the Results obtAINEd from fiNite DIfFErence method.](fiGure4.ePs){width="8cm"} ![\[vXT-planar\]SCaled Material Energy denSITy $v(x,\tau)$ vS poSitIon (X) in THE sLab at differenT TImes FoR $\epsiloN = 0.1$. ThE symbolS stAnd For AnaLyTical valuEs whereaS lInEs RePreSent tHE results ObTaiNeD frOm finITe diffErencE metHoD.](fIGurE5.eps){widTH="8cM"} tHe fiRsT dErivAtiVeS w.r.t. pOsitIOn oF the anaLytical raDiaTIon aNd MaTerial eNergy density aRe Plotted in FIgS. \[upRimext-PLAnar\] and \[vPrimext-planar\]. As the radiaTIon and mAteRial eNergY densitieS deCrease WitH X, the deRivatiVe has NeGatIVE valuES. thE deRiVative has a GREatEr negAtIve vAlue at tHe left compared to thE RigHt zone. As both rAdiAtioN ANd MatERiAL enErGY deNSIties keep on incrEasing with TiME dUe to radiatIOn dIfFusion, mAgnitudE of thE GradienT decreaseS for the leFt And iNCReaSes for the rIght sideS. The gradiENt of bOTh RadiaTioN and maTeRiaL enerGy densITieS obtaIn a conStAnt valUe of $\fRaC{-3}{3+4\sqrt{3}}=-0.30217$ aftEr infinite time showing thAt therE is a cOnsTant leakaGe oF

distance keeps on attaini ng lineari ty. A t i nfi ni te t ime, the steady st a te v alues are linear withposit io n asg iv en by Eq. \[ i nf i n ite -t im e\] .S im ilarl y,the mat erial ener gyde nsity initia l ly exhibitssli ght non-line arvariat io n a n d fin all y att ains t h e line arity (Fi g. \[vxt- p lanar\] ) . T he n on-linear variati o na t early stages occur sd ue t o n etabsorption o f ene r gy by t h ei n i tia l ly cold mater ial (as u(x , 0)= v(x,0) =0 ).I nitial ly, t he mat erial energ y de nsity isfoundt o lag b e hind th e radi ati onener g yde nsi ti e s a n dfin a lly equilib ra te as t imep r o c eeds (b eyon d $\t au$=10). In t his wor k , a ll th e res ults h ave b een ob taine dby consideringcont ributionfro mthe f irst3 0 root s o f t he tran scenden t aleq u a t io n. The value of op ac i t y$\kappa$ is ch o se nt o be 100 a nd$\ep s i lon $ equ a ls 0.1. Fo r a he a twa ve trav el ing in to athi n pla t e an d comp osite pl anars lab, a similar linear variat i on i nt empe rat ure with di stan c e wa s ob s er ved thoug h dif fe r en c e existed in the sp ac e andtimedependent beh aviour due t o heat co nduc t io n approximation [@Sa rkar],[@Su n ]. ![\[ uxt-p lanar\]S caled rad i a tion ene rgy de nsi ty$ u (x ,\tau)$ vs po s i tion ( x) in t heslab of sc ale d t hic kn ess $b=1$ at diff er en tti mes for$ \epsilo n= 0 .1 $. Thes ymbols stan d fo ran a lyt ical va l ue s wher ea sline s r ep resen t th e re sults o btained f rom fini te d ifferen ce method.](f ig ure4.eps){ wi dth ="8cm" } ![\[vxt -planar\]Scaled materia l energy de nsity $v( x,\tau)$vspositi on( x) inthe sl ab at d iff e r ent t i m es fo r$\epsilon= 0.1 $. Th esymb ols sta nd for analyticalv alu es whereas li nes rep r e se ntt he res ul t s o b t ained from fini te differe nc e m ethod.](fi g ure 5. eps){wi dth="8c m"} T he firs t derivat ives w.r. t. pos i t ion of the an alytical radiatio n andm at erial en ergy d en sit y are plott e d i n Fig s. \[u pr imext- plana r\ ] and \[ vprimext-planar\]. As t he rad iatio n a nd materi ale ner gy densit iesdecrease w ith x, theder i vativ e ha s n ega t ive v alue s . The der i va tiv e ha s a greater n e gat ive v alu e at th e le ft compared to th e right zone. A s bo t h ra dia t ionan d material ene rgy d e n sities k ee p on increa sing wit ht ime d ue toradiat ion dif f u si o n, mag nitu deof the gr adi en t decrea se sf or the lef tand in crease s for t he right sides.The g r a dient ofbothra diation andmaterial e nergy densi ties o btai n a c onstant v alue o f $ \f rac{-3}{3+ 4 \sqrt{3}} =-0.3 0217$ a ft er i nfi nite t imes h owing tha tthe re is a c o n st a nt l e aka ge o f

distance_keeps on_attaining linearity. At infinite_time, the_steady_state values_are_linear with position_as given by_Eq. \[infinite-time\]. Similarly, the_material energy density_initially_exhibits slight non-linear variation and finally attains the linearity (Fig. \[vxt-planar\]). The non-linear variation_at_early stages_occurs_due_to net absorption of energy_by the initially cold material_(as u(x,0)=v(x,0)=0)._Initially, the material energy density is found to_lag_behind the radiation_energy densities and finally equilibrate as time proceeds (beyond_$\tau$=10). In this work, all the_results have been_obtained_by_considering contribution from the_first 30 roots of the transcendental_equation. The value of opacity $\kappa$_is chosen to be 100 and $\epsilon_$ equals 0.1. For a heat_wave traveling into a thin_plate and_composite planar slab, a similar_linear variation in_temperature with_distance was observed_though difference existed in the space_and time dependent_behaviour due to heat conduction approximation_[@Sarkar],[@Sun]. ![\[uxt-planar\]Scaled_radiation energy density_$u(x,\tau)$_vs_position (x)_in the slab_of_scaled thickness_$b=1$_at different times for $ \epsilon_=_0.1 $. The symbols stand for analytical_values whereas lines represent_the_results obtained from finite_difference method.](figure4.eps){width="8cm"} ![\[vxt-planar\]Scaled material energy density_$v(x,\tau)$ vs position (x) in the_slab at_different times_for $\epsilon = 0.1$. The symbols stand for analytical values whereas_lines represent the results obtained from_finite difference method.](figure5.eps){width="8cm"} The first_derivatives w.r.t._position_of the analytical_radiation_and material_energy density are plotted in Figs. \[uprimext-planar\]_and \[vprimext-planar\]._As the radiation and material energy_densities decrease with x,_the_derivative has negative values. The derivative_has a greater negative value at_the left compared to the_right_zone._As both radiation and material_energy densities keep on increasing with_time due to_radiation diffusion, magnitude of the gradient decreases_for_the left and increases for the_right_sides. The gradient of both radiation_and_material_energy densities obtain a constant_value of $\frac{-3}{3+4\sqrt{3}}=-0.30217$ after infinite time_showing that there is a constant leakage of

a contradiction. Hence, the sequence $\iota_n,\iota_{n-1},\ldots,\iota_1$ is non decreasing, and equality cannot occur. Let us prove that it suffices to choose $\iota_k$ as the smallest in $\eta(n(j_{k-1})\setminus \eta(n(j_k))$. If at step $k$, a smallest $\iota_k$ is not chosen in $\eta(n(j_{k-1})\setminus \eta(n(j_k))$, it will be taken after, and the sequence $\iota_n,\iota_{n-1},\ldots,\iota_1$ will be no more non decreasing, a contradiction. It remains to prove that $\pi$ is strictly increasing and to verify the expression of $m$. Let us prove by induction that $$\label{eq:pi1} \pi(j_k)=1-m(\iota_n)-m(\iota_n\vee\iota_{n-1}) - \cdots- m(\iota_n\vee\cdots\vee \iota_{k+1}),\quad k=n,\ldots,1.$$ We show it for $k=n$. We have $$\pi(j_n)=1-\nu(m_n) = 1-\sum_{\substack{x\leq m_n\\x\in C_\pi}}m(x).$$ Since $\iota_n\not\in\eta(m_n)$, no $x$ in $C_\pi$ can be smaller than $m_n$. Hence $\pi(j_n)=1$. Let us assume (\[eq:pi1\]) is true from $n$ up to some $k$, and prove it is still true for $k-1$. Using (\[eq:proof\]), we have: $$\begin{aligned} \pi(j_{k-1}) &=1-\nu(m_{k-1}) = 1-\sum_{\substack{x\leq m_{k-1}\\x\in C_\pi}}m(x) \\ & = 1 - \sum_{\substack{x\leq m_{k}\\x\in C_\pi}}m(x) - m(\iota_n\vee\cdots\vee\iota_k) = \pi(j_k)-m(\iota_n\vee\cdots\vee\iota_k),\end{aligned

a contradiction. Hence, the sequence $ \iota_n,\iota_{n-1},\ldots,\iota_1 $ is non decreasing, and equality cannot occur. lease us rise that it suffices to choose $ \iota_k$ as the smallest in $ \eta(n(j_{k-1})\setminus \eta(n(j_k))$. If at step $ k$, a minor $ \iota_k$ is not chosen in $ \eta(n(j_{k-1})\setminus \eta(n(j_k))$, it will be taken after, and the succession $ \iota_n,\iota_{n-1},\ldots,\iota_1 $ will be no more non decreasing, a contradiction. It remains to prove that $ \pi$ is rigorously increase and to verify the construction of $ m$. Let us prove by generalization that $ $ \label{eq: pi1 } \pi(j_k)=1 - m(\iota_n)-m(\iota_n\vee\iota_{n-1 }) - \cdots- m(\iota_n\vee\cdots\vee \iota_{k+1}),\quad k = n,\ldots,1.$$ We usher it for $ k = n$. We have $ $ \pi(j_n)=1-\nu(m_n) = 1-\sum_{\substack{x\leq m_n\\x\in C_\pi}}m(x).$$ Since $ \iota_n\not\in\eta(m_n)$, no $ x$ in $ C_\pi$ can be smaller than $ m_n$. Hence $ \pi(j_n)=1$. Let us bear (\[eq: pi1\ ]) is true from $ n$ up to some $ k$, and prove it is still on-key for $ k-1$. Using (\[eq: proof\ ]), we have: $ $ \begin{aligned } \pi(j_{k-1 }) & = 1-\nu(m_{k-1 }) = 1-\sum_{\substack{x\leq m_{k-1}\\x\in C_\pi}}m(x) \\ & = 1 - \sum_{\substack{x\leq m_{k}\\x\in C_\pi}}m(x) - m(\iota_n\vee\cdots\vee\iota_k) = \pi(j_k)-m(\iota_n\vee\cdots\vee\iota_k),\end{aligned

a fontradiction. Hence, the requence $\iota_n,\iota_{n-1},\ldovs,\iota_1$ js non ddcreasing, and equality cannov ocxur. Leu us prove that it ruffices no choose $\iote_k$ as the smallest in $\eta(n(j_{k-1})\seflinuv \eta(n(j_k))$. If at xtep $k$, a skallest $\iota_k$ hs nlt chosen in $\eta(n(j_{k-1})\setminus \eta(n(j_k))$, yt will bf taken after, wnd uhe seqhvnge $\iota_n,\iota_{n-1},\ldots,\iota_1$ will be nk more ion decreasing, s contradiction. It remains ho pgove that $\pi$ is stgictly incrgzsigt and to verkfy the exkrzssion of $m$. Let us prove by induction that $$\labek{eq:pi1} \pi(j_k)=1-m(\uora_n)-l(\hota_n\vee\iote_{n-1}) - \cdjts- m(\iota_n\vee\gcots\vea \iota_{k+1}),\auad k=n,\ldots,1.$$ Wc shox it for $k=n$. We have $$\pi(j_n)=1-\nn(m_n) = 1-\sum_{\substack{x\leq m_n\\x\in C_\ph}}m(r).$$ Since $\iota_n\not\in\eta(n_n)$, no $x$ in $W_\pi$ zqn ce amellsr thaj $m_i$. Hence $\pi(j_h)=1$. Let us aswume (\[eq:pi1\]) is true ftoi $n$ up to some $k$, and pwove it is still true for $k-1$. Using (\[eq:promf\]), se have: $$\begin{aligned} \pi(j_{j-1}) &=1-\nu(m_{k-1}) = 1-\sum_{\substack{x\lgq m_{k-1}\\x\in C_\py}}m(x) \\ & = 1 - \sum_{\substack{x\leq m_{k}\\x\in C_\pi}}m(x) - m(\iota_n\vee\cdods\vee\moga_k) = \ii(j_k)-o(\uoha_n\vee\cdots\vee\iota_k),\end{aligned

a contradiction. Hence, the sequence $\iota_n,\iota_{n-1},\ldots,\iota_1$ is and cannot occur. us prove that as smallest in $\eta(n(j_{k-1})\setminus If at step a smallest $\iota_k$ is not chosen $\eta(n(j_{k-1})\setminus \eta(n(j_k))$, it will be taken after, and the sequence $\iota_n,\iota_{n-1},\ldots,\iota_1$ will be more non decreasing, a contradiction. It remains to prove that $\pi$ is strictly and verify expression $m$. Let us prove by induction that $$\label{eq:pi1} \pi(j_k)=1-m(\iota_n)-m(\iota_n\vee\iota_{n-1}) - \cdots- m(\iota_n\vee\cdots\vee \iota_{k+1}),\quad k=n,\ldots,1.$$ We show for $k=n$. We have $$\pi(j_n)=1-\nu(m_n) = 1-\sum_{\substack{x\leq m_n\\x\in Since $\iota_n\not\in\eta(m_n)$, no $x$ $C_\pi$ can be smaller than Hence Let us (\[eq:pi1\]) true $n$ up to $k$, and prove it is still true for $k-1$. Using (\[eq:proof\]), we have: $$\begin{aligned} \pi(j_{k-1}) &=1-\nu(m_{k-1}) = m_{k-1}\\x\in C_\pi}}m(x) = 1 \sum_{\substack{x\leq C_\pi}}m(x) m(\iota_n\vee\cdots\vee\iota_k) = \pi(j_k)-m(\iota_n\vee\cdots\vee\iota_k),\end{aligned

a contradiction. Hence, the seqUence $\iota_n,\Iota_{n-1},\LdoTs,\iOtA_1$ is nOn deCreasing, and equALity Cannot occur. Let us prove tHat it SuFFiceS To ChoosE $\iota_k$ aS ThE SMalLeSt In $\eTa(N(J_{k-1})\SetmiNus \Eta(n(j_k))$. IF at step $k$, a sMalLeSt $\iota_k$ is not CHoSen in $\eta(n(j_{K-1})\seTminus \eta(n(j_k))$, It wIll be tAkEn aFTer, anD thE sequEnce $\ioTA_n,\iota_{N-1},\ldots,\iotA_1$ wILl be no MOre non dECReAsinG, a contradiction. It REmAIns to prove that $\Pi$ is stRiCTlY INcrEasIng and to veRiFy the EXpressiON oF $M$. lEt uS Prove by inductIon that $$\labeL{Eq:pI1} \pi(j_k)=1-m(\IoTa_n)-M(\Iota_n\vEe\iotA_{n-1}) - \CDotS- m(\iota_n\vee\cDots\Vee \iota_{k+1}),\qUad k=n,\lDOts,1.$$ We shOW it for $k=N$. We havE $$\pi(J_n)=1-\nU(m_n) = 1-\sUM_{\sUbStaCk{X\Leq M_N\\x\In C_\PI}}m(x).$$ since $\iotA_n\NoT\in\etA(m_n)$, nO $X$ IN $c_\pi$ cAn bE smaLler tHan $m_n$. Hence $\pi(j_N)=1$. LeT us aSSumE (\[eq:pi1\]) Is truE froM $n$ Up to sOme $k$, anD provE iT is still true for $K-1$. UsiNg (\[eq:proof\]), We hAvE: $$\beGiN{aligNEd} \pi(j_{k-1}) &=1-\Nu(m_{K-1}) = 1-\suM_{\substaCk{x\leq m_{K-1}\\X\in c_\pI}}M(X) \\ & = 1 - \SuM_{\substack{x\leq m_{k}\\x\in c_\pI}}M(X) - m(\Iota_n\vee\Cdots\vEE\iOtA_K) = \pi(j_k)-m(\ioTa_N\veE\cdoTS\Vee\ioTa_k),\eND{aLigned

a contradiction. Hence, t he sequenc e $\i ota _n, \i ota_ {n-1 },\ldots,\iota _ 1$ i s non decreasing, andequal it y can n ot occu r. Let us p rov eth atit su ffice s t o choos e $\iota_k $ a sthe smallest in $\eta(n(j _{k -1})\setminu s \ eta(n( j_ k)) $ . Ifatstep$k$, a smalle st $\iota _k $ is no t chosen i n$\et a(n(j_{k-1})\setm i nu s \eta(n(j_k))$ , it w il l b e tak enafter, and t he se q uence $ \ io t a _ n,\ i ota_{n-1},\ld ots,\iota_1 $ wi ll beno mo r e nondecre as i ng, a contradi ctio n. It re mainst o prove that $\ pi$ is st ric tlyi nc re asi ng and to ve r ify the exp re ss ion o f $m $ . L et u s p rove by i nduction that $$ \lab e l{e q:pi1 } \pi (j_k )= 1-m(\ iota_n )-m(\ io ta_n\vee\iota_{ n-1} ) - \cdot s-m( \io ta _n\ve e \cdots \ve e \ iota_{k +1}),\q u adk= n , \ ld ots,1.$$ We show i tf o r$k=n$. W e have $$ \p i (j_n)=1- \n u(m _n)= 1-\su m_{\ s ub stack{x\ leq m_ n \\ x\ in C_\p i} }m(x). $$ Si nce $\io t a_n\ not\in \eta(m_n )$, n o $x$ in $C_\pi $ can be small e rt h an $m_n $.Hence $\pi( j_n) = 1$.Letu sass u me (\ [eq:p i1 \ ]) is true from $n$ up t o some $k$, and prove it is stillt r u e for $k -1$. Us i ng (\[eq:proof \]),we have: $ $ \begin{a ligne d} \pi(j _{k-1}) & = 1 -\nu(m_{ k-1 })= 1 -\s u m _{ \substack{x\l e q m_{ k- 1}\\x\i n C _\pi}}m (x) \\ & =1- \sum_{\ substack {x \l eq m _{k }\\x\ i n C_\pi} }m (x) - m( \iota _ n\vee\ cdots \vee \i ot a _k) = \pi( j _k ) - m(\i ot a_ n\ve e\c do ts\ve e\io t a_k ),\end{ aligned

a_contradiction. Hence,_the sequence $\iota_n,\iota_{n-1},\ldots,\iota_1$ is_non decreasing,_and_equality cannot_occur. Let_us prove that_it suffices to_choose $\iota_k$ as the_smallest in $\eta(n(j_{k-1})\setminus_\eta(n(j_k))$._If at step $k$, a smallest $\iota_k$ is not chosen in $\eta(n(j_{k-1})\setminus \eta(n(j_k))$, it_will_be taken_after,_and_the sequence $\iota_n,\iota_{n-1},\ldots,\iota_1$ will be_no more non decreasing, a_contradiction. It remains_to prove that $\pi$ is strictly increasing and_to_verify the expression_of $m$. Let us prove by induction that $$\label{eq:pi1} \pi(j_k)=1-m(\iota_n)-m(\iota_n\vee\iota_{n-1})_- \cdots- m(\iota_n\vee\cdots\vee \iota_{k+1}),\quad k=n,\ldots,1.$$ We show_it for $k=n$._We_have_$$\pi(j_n)=1-\nu(m_n) = 1-\sum_{\substack{x\leq m_n\\x\in_C_\pi}}m(x).$$ Since $\iota_n\not\in\eta(m_n)$, no $x$ in_$C_\pi$ can be smaller than $m_n$._Hence $\pi(j_n)=1$. Let us assume (\[eq:pi1\]) is_true from $n$ up to some_$k$, and prove it is_still true_for $k-1$. Using (\[eq:proof\]), we_have: $$\begin{aligned} \pi(j_{k-1}) &=1-\nu(m_{k-1})_= 1-\sum_{\substack{x\leq_m_{k-1}\\x\in C_\pi}}m(x) \\ &_= 1 - \sum_{\substack{x\leq m_{k}\\x\in C_\pi}}m(x)_- m(\iota_n\vee\cdots\vee\iota_k) = \pi(j_k)-m(\iota_n\vee\cdots\vee\iota_k),\end{aligned

a finite model for ${\underline{E}G}$ implies that the $G$-sphere spectrum is small in the triangulated homotopy category $\operatorname{Ho}(\operatorname{Sp}_G)$. The example below illustrates that this is not true in general, and that even a finite-dimensional model for ${\underline{E}G}$ does not imply smallness of ${\mathbb S}_G$. In fact, [@barcenas-degrijse-patchkoria:stable_finiteness Thm.5.4] shows that for a countable discrete group $G$, the $G$-sphere spectrum is small if and only if there exists a finite-dimensional stable model for ${\underline{E}G}$, and there exists a finite-type stable model for ${\underline{E}G}$. Moreover, by [@barcenas-degrijse-patchkoria:stable_finiteness Thm.5.1], a finite-type stable model for ${\underline{E}G}$ exists if and only if there are only finitely many conjugacy classes of finite subgroups in $G$, and every Weyl group $W_G(H)=N_G(H)/H$ of a finite subgroup $H$ of $G$ is of homological type $F P_{\infty}$. We let $F$ be any non-trivial finite group, and we define $$G \ = \ {\prod}'_{k\geq 1}\, F \,$$ an infinite weak product of copies of $F$, i.e., the subgroup of the product consisting of tuples with almost all coordinates the neutral element. We will now show that ${\mathbb S}_G$ is not small in $\operatorname{Ho}(\operatorname{Sp}_G)$. We set $H_n=\prod_{k=1}^n F$. Then $G$ is the ascending union of its finite subgroups $H_n$, and we can apply Corollary \[cor:lim\_lim\^1\_sequence\]. The inclusion $H_{n-1}{\longrightarrow}H_n$ has a retraction $r:H_n{\longrightarrow}H_{n-1}$ by a group homomorphism. Now we let $X$ be an orthogonal spectrum, which we give the trivial $G$-action. Then the inflation map $r^*:\pi_k^{H_{n-1}}(X){\longrightarrow}\pi_k^{H_n}(X)$ is a section to the restriction $\operatorname{

a finite model for $ { \underline{E}G}$ implies that the $ G$-sphere spectrum is small in the triangulated homotopy class   $ \operatorname{Ho}(\operatorname{Sp}_G)$. The case below illustrates that this is not true in cosmopolitan, and that even a finite - dimensional model for $ { \underline{E}G}$ does not incriminate smallness of $ { \mathbb S}_G$. In fact, [ @barcenas - degrijse - patchkoria: stable_finiteness Thm.5.4 ] shows that for a countable discrete group $ G$, the $ G$-sphere spectrum is little if and only if there exists a finite - dimensional static exemplar for $ { \underline{E}G}$, and there exists a finite - character stable model for $ { \underline{E}G}$. Moreover, by [ @barcenas - degrijse - patchkoria: stable_finiteness Thm.5.1 ], a finite - type static model for $ { \underline{E}G}$ exists if and merely if there are only finitely many conjugacy class of finite subgroups in $ G$, and every Weyl group $ W_G(H)=N_G(H)/H$ of a finite subgroup $ H$ of $ G$ is of homological type $ F P_{\infty}$. We lease $ F$ be any non - trivial finite group, and we define $ $ G \ = \ { \prod}'_{k\geq 1}\, F \,$$ an countless weak intersection of copies of   $ F$, i.e., the subgroup of the product consisting of tuples with about all coordinates the neutral element. We will now show that $ { \mathbb S}_G$ is not small in   $ \operatorname{Ho}(\operatorname{Sp}_G)$. We jell $ H_n=\prod_{k=1}^n F$. Then $ G$ is the ascending union of its finite subgroups $ H_n$, and we can apply Corollary \[cor: lim\_lim\^1\_sequence\ ]. The inclusion $ H_{n-1}{\longrightarrow}H_n$ has a retraction $ r: H_n{\longrightarrow}H_{n-1}$ by a group homomorphism. Now we let $ X$ be an orthogonal spectrum, which we move over the trivial $ G$-action. Then the inflation function $ r^*:\pi_k^{H_{n-1}}(X){\longrightarrow}\pi_k^{H_n}(X)$ is a incision to the limitation $ \operatorname {

a vinite model for ${\underlike{E}G}$ implies thaj rhe $G$-s'here slectrum ks small in the triangulated himotokj category $\operatornamd{Ho}(\operatlrname{Sp}_T)$. Tht example below illustratcf thzb thiv is not true ik general, atd that even a fknnte-dimensional model for ${\underline{E}G}$ does npt imply smallnefs og ${\matgbb S}_G$. In fact, [@barcenas-degrijse-patdhkoria:vtable_finitenrss Thm.5.4] shows that for a cluntwble discrete grouo $G$, the $G$-spyere wpectrum is rmall if and only if tgere exists a finite-dimensional stabke model fir ${\unfgrline{E}G}$, and therv exists a fikpte-type stable model for ${\undcrlinx{E}G}$. Noreover, by [@barcenas-dxgrijse-patchkoria:stafle_finiteteas Thm.5.1], a finite-type stabne mmdel dor ${\unseclihe{E}G}$ edisvs if and ohly if therw are only finitely mwby conjugacy dlassef jf finite subgroups in $G$, and every Weyl grkup $W_G(H)=N_G(H)/H$ of a finite subgroup $H$ of $G$ is ov homologycal type $F P_{\infty}$. We let $F$ be any non-trivial finide grkjp, ckq qe define $$G \ = \ {\prod}'_{k\geq 1}\, F \,$$ an infinite weak pwkdicn of copies of $F$, i.c., the subgroup of yhf ltoduct consistkng of tulles with almost apl coorqinatws the nettrak element. We will now show rhat ${\mathbb F}_T$ is not small in $\o'eratorname{Hu}(\opetatornsme{Sp}_G)$. We set $H_n=\prod_{k=1}^n F$. Theh $G$ is the wscending jnion of its finkte sgbgroups $H_n$, and we can appjy Corollery \[cpr:lim\_lio\^1\_seqoence\]. Tre inclusiln $H_{n-1}{\longrightarrow}H_n$ had a rgtracthon $r:H_n{\lonhrightarrow}H_{n-1}$ by a group homomoc'hism. Now we ked $X$ be an oxthogokal spectrum, whych we give thg trivial $G$-actkon. Then tge infletion map $r^*:\py_k^{H_{n-1}}(X){\longrighdwrrow}\pi_k^{H_n}(X)$ ms a sectyon ro tye restfkction $\operatotname{

a finite model for ${\underline{E}G}$ implies that spectrum small in triangulated homotopy category that is not true general, and that a finite-dimensional model for ${\underline{E}G}$ does imply smallness of ${\mathbb S}_G$. In fact, [@barcenas-degrijse-patchkoria:stable_finiteness Thm.5.4] shows that for a discrete group $G$, the $G$-sphere spectrum is small if and only if there a stable for and there exists a finite-type stable model for ${\underline{E}G}$. Moreover, by [@barcenas-degrijse-patchkoria:stable_finiteness Thm.5.1], a finite-type stable for ${\underline{E}G}$ exists if and only if there only finitely many conjugacy of finite subgroups in $G$, every group $W_G(H)=N_G(H)/H$ a subgroup of $G$ is homological type $F P_{\infty}$. We let $F$ be any non-trivial finite group, and we define $$G \ \ {\prod}'_{k\geq \,$$ an weak of of $F$, i.e., of the product consisting of tuples coordinates the neutral element. We will now show ${\mathbb S}_G$ not small in $\operatorname{Ho}(\operatorname{Sp}_G)$. We set F$. Then $G$ is the ascending union of finite subgroups $H_n$, and we can apply Corollary \[cor:lim\_lim\^1\_sequence\]. The inclusion $H_{n-1}{\longrightarrow}H_n$ has a retraction a group homomorphism. Now let $X$ be orthogonal which give trivial $G$-action. the inflation map $r^*:\pi_k^{H_{n-1}}(X){\longrightarrow}\pi_k^{H_n}(X)$ is a section to the restriction $\operatorname{

a finite model for ${\underline{E}g}$ implies thAt the $g$-spHerE sPectRum iS small in the triANgulAted homotopy category $\opEratoRnAMe{Ho}(\OPeRatorName{Sp}_G)$. tHe EXAmpLe BeLow IlLUsTrateS thAt this iS not true in GenErAl, and that eveN A fInite-dimenSioNal model for ${\uNdeRline{E}g}$ dOes NOt impLy sMallnEss of ${\mAThbb S}_G$. in fact, [@barCeNAs-degrIJse-patcHKOrIa:stAble_finiteness Thm.5.4] SHoWS that for a countAble diScREtE GRouP $G$, tHe $G$-sphere sPeCtrum IS small iF AnD ONLy iF There exists a fInite-dimensIOnaL stablE mOdeL For ${\undErlinE{E}g}$, And There exists A finIte-type stAble moDEl for ${\unDErline{E}g}$. MoreoVer, By [@bArceNAs-DeGriJsE-PatCHkOriA:StaBle_finitEnEsS Thm.5.1], a FiniTE-TYPe stAblE modEl for ${\Underline{E}G}$ exIstS if aND onLy if tHere aRe onLy FinitEly manY conjUgAcy classes of finIte sUbgroups iN $G$, aNd EveRy weyl gROup $W_G(H)=n_G(H)/h$ of A finite SubgrouP $h$ of $g$ iS OF HoMological type $F P_{\infTy}$. wE LeT $F$ be any nOn-trivIAl FiNIte group, AnD we DefiNE $$g \ = \ {\prod}'_{K\geq 1}\, f \,$$ An Infinite Weak prODuCt Of copieS oF $F$, i.e., thE sUbgRouP of thE ProdUct conSisting oF tuplES with almost all COordinates the NEuTRAl ELemeNt. WE will now shoW thaT ${\MathBb S}_G$ IS nOt sMAll in $\OperaToRNaME{Ho}(\operatorname{Sp}_G)$. WE sEt $H_n=\prOd_{k=1}^n F$. then $G$ is the ascEnding unioN OF Its finitE subGRoUPs $H_n$, and we can apPly CoRollary \[cor:LIm\_lim\^1\_seqUence\]. the incluSion $H_{n-1}{\lonGRIghtarroW}H_n$ Has A reTraCTIoN $r:H_n{\longrightARRow}H_{N-1}$ bY a group HomOmorphiSm. NOw wE leT $X$ bE aN orthogonAl spectrUm, WhIcH wE giVe the TRivial $G$-aCtIon. thEn tHe infLAtion mAp $r^*:\pi_K^{H_{n-1}}(X){\LoNgRIghTarrow}\pI_K^{H_N}(x)$ Is a sEcTiOn to The ReStricTion $\OPerAtornamE{

a finite model for ${\und erline{E}G }$ im pli esth at t he $ G$-sphere spec t rumis small in the triang ulate dh omot o py cate gory $\ o pe r a tor na me {Ho }( \ op erato rna me{Sp}_ G)$. The e xam pl e below illu s tr ates thatthi s is not tru e i n gene ra l,a nd th atevena fini t e-dime nsional m od e l for$ {\under l i ne {E}G }$ does not imply sm a llness of ${\m athbbS} _ G$ . Infac t, [@barce na s-deg r ijse-pa t ch k o r ia: s table_finiten ess Thm.5.4 ] sh ows th at fo r a cou ntabl ed isc rete group$G$, the $G$- sphere spectru m is sma ll ifand on ly i f t he reex i sts afin i te- dimensio na lstabl e mo d e l for${\ unde rline {E}G}$, and t her e ex i sts a fi nite- type s table model for${ \underline{E}G} $. M oreover,by[@ bar ce nas-d e grijse -pa tch koria:s table_f i nit en e s s T hm.5.1], a finite- ty p e s table mo del fo r $ {\ u nderline {E }G} $ ex i s ts if and on ly if th ere ar e o nl y finit el y many c onj uga cy cl a sses of fi nite sub group s in $G$, and e v ery Weyl grou p $ W _ G( H )=N_ G(H )/H$ of a f init e sub grou p $ H$o f $G$ is o fh om o logical type $F P_{ \i nfty}$ . We let $F$ be a ny non-tri v i a l finite gro u p, and we define$$G \ = \ {\pro d }'_{k\ge q 1}\ , F \,$ $ an infi n i te weakpro duc t o f c o p ie s of $F$, i.e . , the s ubgroup of the pr odu ctcon sis ti ng of tup les with a lm os tall coor d inates t he ne ut ral elem e nt. We will now s ho w th at ${\m a th b b S}_ G$ i s no t s ma ll in  $\o p era torname {Ho}(\ope rat o rnam e{ Sp }_G)$. We set $H_n= \p rod_{k=1}^ nF$. Then$ G $ is the ascending union of its finitesub group s $H _n$, andwecan ap ply Coroll ary \[ cor:l im \_l i m \^1\_ s e qu enc e\ ]. The inc l u sio n $H_ {n -1}{ \longri ghtarrow}H_n$ hasa re traction $r:H _n{ \lon g r ig hta r ro w }H_ {n - 1}$ b y a group homom orphism. N ow we let $X$ b e an o rthogon al spec trum, which w e give th e trivial $ G$-a c t ion . Then the inflati on map $r ^ *:\pi _ k^ {H_{n -1} }(X){\ lo ngr ighta rrow}\ p i_k ^{H_n }(X)$is a sec tionto the res triction $\operatorname {

a_finite model_for ${\underline{E}G}$ implies that_the $G$-sphere_spectrum_is small_in_the triangulated homotopy_category $\operatorname{Ho}(\operatorname{Sp}_G)$. The example_below illustrates that this_is not true_in_general, and that even a finite-dimensional model for ${\underline{E}G}$ does not imply smallness of_${\mathbb_S}_G$. In_fact,_[@barcenas-degrijse-patchkoria:stable_finiteness_Thm.5.4] shows that for a_countable discrete group $G$, the_$G$-sphere spectrum_is small if and only if there exists_a_finite-dimensional stable model_for ${\underline{E}G}$, and there exists a finite-type stable model_for ${\underline{E}G}$. Moreover, by [@barcenas-degrijse-patchkoria:stable_finiteness Thm.5.1],_a finite-type stable_model_for_${\underline{E}G}$ exists if and_only if there are only finitely_many conjugacy classes of finite subgroups_in $G$, and every Weyl group $W_G(H)=N_G(H)/H$_of a finite subgroup $H$ of_$G$ is of homological type_$F P_{\infty}$. We_let $F$ be any non-trivial_finite group, and_we define_$$G \ =_\ {\prod}'_{k\geq 1}\, F \,$$_an infinite weak_product of copies of $F$, i.e., the_subgroup_of the product_consisting_of_tuples with_almost all coordinates_the_neutral element._We_will now show that ${\mathbb S}_G$_is_not small in $\operatorname{Ho}(\operatorname{Sp}_G)$. We set $H_n=\prod_{k=1}^n F$. Then_$G$ is the ascending_union_of its finite subgroups_$H_n$, and we can apply_Corollary \[cor:lim\_lim\^1\_sequence\]. The inclusion $H_{n-1}{\longrightarrow}H_n$ has_a retraction_$r:H_n{\longrightarrow}H_{n-1}$ by_a group homomorphism. Now we let $X$ be an orthogonal spectrum,_which we give the trivial $G$-action._Then the inflation map_$r^*:\pi_k^{H_{n-1}}(X){\longrightarrow}\pi_k^{H_n}(X)$ is_a_section to the_restriction_$\operatorname{

\] below, using the characterization of $N$-${\mathcal{F}}$-amenability from Remark \[rem:covers-not-maps\] by the existence of $S$-wide covers of $G {{\times}}\Delta$. To outline the construction of these covers and to prepare for the mapping class group we introduce some notation. Pick a $G$-invariant proper metric on the set $E$ of edges of $\Gamma$; this is possible as $G$ and $E$ are countable and the action of $G$ on $E$ has finite stabilizers. For each vertex $v$ of $\Gamma$ with infinite valency let $E_v$ be the set of edges incident to $v$. Write $d_v$ for the restriction of the metric to $E_v$. For $\xi \in \Delta$, $\xi \neq v$ we define its projection $\pi_v(\xi)$ to $E_v$ as the set of all edges of $\Gamma$ that are appear as initial edges of geodesics from $v$ to $\xi$. This is a finite subset of $E_v$ (this depends again of fineness of $\Gamma$). Fix a vertex $v_0$ of finite valence as a base point. For $g \in G$, $\xi \in \Delta$ define their *projection distance* at $v$ by $$d_v^\pi(g,\xi) := d_v ( \pi_v(gv_0), \pi_v(\xi)).$$ For $\xi = v$, set $d^\pi_v(g,v) := \infty$. (For relative hyperbolic groups a related quantity is often called an *angle*; the terminology here is chosen to align better with the case of the mapping class group.) If we vary $g$ (in a finite set) and $\xi$ (in an open neighborhood) then for fixed $v$ the projection distance $d^\pi_{v}(g,\xi)$ varies by a bounded amount. Useful is the following attraction property for projection distances: there is $\Theta_0$ such that if $d^\pi_v(g,\xi) \geq \Theta_0$, then any geodesic from $gv_0$ to $\xi$ in $\Gamma$ passes through $v$. Conversely, if some geodesic from $gv_0$ to $\xi$ misses $v$, then $d^\pi_

\ ] below, using the characterization of $ N$-${\mathcal{F}}$-amenability from Remark   \[rem: covers - not - maps\ ] by the being of $ S$-wide cover charge of $ G { { \times}}\Delta$. To outline the structure of these covers and to train for the mapping course group we introduce some notation. Pick a $ G$-invariant proper metric function on the set $ E$ of edges of $ \Gamma$; this is potential as $ G$ and $ E$ are countable and the action of $ G$ on $ E$ has finite stabilizers. For each vertex $ v$ of $ \Gamma$ with infinite valence let $ E_v$ be the set of edges incident to $ v$. Write $ d_v$ for the restriction of the measured to $ E_v$. For $ \xi \in \Delta$, $ \xi \neq v$ we define its projection $ \pi_v(\xi)$ to $ E_v$ as the set of all boundary of $ \Gamma$ that are appear as initial edges of geodesic from $ v$ to $ \xi$. This is a finite subset of $ E_v$ (this depends again of daintiness of $ \Gamma$). Fix a vertex $ v_0 $ of finite valence as a base point. For $ g \in G$, $ \xi \in \Delta$ define their * projection distance * at $ v$ by $ $ d_v^\pi(g,\xi): = d_v (\pi_v(gv_0), \pi_v(\xi)).$$ For $ \xi = v$, dress $ d^\pi_v(g, v): = \infty$. (For relative hyperbolic groups a related quantity is often called an * angle *; the terminology here is chosen to align better with the case of the mapping class group .) If we vary $ g$ (in a finite set) and $ \xi$ (in an open neighborhood) then for fixed $ v$ the projection distance $ d^\pi_{v}(g,\xi)$ varies by a jump amount. Useful is the stick to drawing card place for projection distances: there is $ \Theta_0 $ such that if $ d^\pi_v(g,\xi) \geq \Theta_0 $, then any geodesic from $ gv_0 $ to $ \xi$ in $ \Gamma$ passes through $ v$. Conversely, if some geodesic from $ gv_0 $ to $ \xi$ misses $ v$, then $ d^\pi _

\] bepow, using the characterieation of $N$-${\mathcal{F}}$-amenauility rrom Remxrk \[rem:covers-not-maps\] by the eeistwnce if $S$-wide covers of $G {{\tkmes}}\Delta$. No outlinw tht construction of these covers zkd to 'repare for the mapping cnass group we hngrlduce some notation. Pick a $G$-invariagt proprr metric on the set $Q$ of vdnes of $\Gamma$; this is possible as $G$ and $T$ are countable anc the action of $G$ on $E$ has finlte stabilizers. Fog each vertgs $v$ if $\Gamma$ witf infinite valency let $E_v$ be the set of edges incideng to $r$. Write $d_v$ doe tjg restrictioi of tre metric to $E_v$. For $\fi \in \Drlta$, $\xi \neq v$ ee vefibe its projection $\pi_v(\ei)$ to $E_v$ as the set jf all edcea of $\Gamma$ that aee appedr av inkriau esgxs kf geofesmcs from $v$ fo $\xi$. This us a finite subset pf $V_f$ (this depenss agayn of fineness of $\Gamma$). Fix a vertex $v_0$ of fihite valence as a base point. For $g \in G$, $\xi \ij \Delta$ dqfine their *projection distance* at $v$ by $$d_v^\pi(g,\xi) := g_v ( \pm_v(ev_0), \ki_y(\wi)).$$ Fue $\di = v$, set $d^\pi_v(g,v) := \infty$. (For relative hyperbolyd brpups a related quantity ix lfygn called an *avgle*; tks ferminology here id choseg to qlign betuer woth the case of the mapping class group.) Uf we vary $g$ (in a yinite set) aud $\xi$ (on an open neighborhood) then for fixed $v$ thf projectjun distance $d^\pi_{v}(e,\xi)$ vdries by a bounded amount. Tseful is the followkng sttraceion propegty for projection distajces: jhere hs $\Theta_0$ skch that if $d^\pi_v(g,\xi) \geq \Theta_0$, tixn any geodesoc frmm $gv_0$ to $\xi$ ik $\Gamma$ passes ehrough $v$. Convgrsely, if some geodesic rrom $gv_0$ to $\xi$ missqs $v$, then $d^\pi_

\] below, using the characterization of $N$-${\mathcal{F}}$-amenability \[rem:covers-not-maps\] the existence $S$-wide covers of construction these covers and prepare for the class group we introduce some notation. a $G$-invariant proper metric on the set $E$ of edges of $\Gamma$; this possible as $G$ and $E$ are countable and the action of $G$ on has stabilizers. each $v$ of $\Gamma$ with infinite valency let $E_v$ be the set of edges incident to $v$. $d_v$ for the restriction of the metric to For $\xi \in \Delta$, \neq v$ we define its $\pi_v(\xi)$ $E_v$ as set all of $\Gamma$ that appear as initial edges of geodesics from $v$ to $\xi$. This is a finite subset of $E_v$ depends again of $\Gamma$). a $v_0$ finite valence as point. For $g \in G$, $\xi their *projection distance* at $v$ by $$d_v^\pi(g,\xi) := ( \pi_v(gv_0), For $\xi = v$, set $d^\pi_v(g,v) \infty$. (For relative hyperbolic groups a related quantity often called an *angle*; the terminology here is chosen to align better with the case mapping class group.) If vary $g$ (in finite and (in open neighborhood) for fixed $v$ the projection distance $d^\pi_{v}(g,\xi)$ varies by a bounded Useful is the following attraction property for projection distances: there such if $d^\pi_v(g,\xi) \geq then any geodesic from to in $\Gamma$ passes through if geodesic $\xi$ $v$, $d^\pi_

\] below, using the characterizaTion of $N$-${\matHcal{F}}$-AmeNabIlIty fRom REmark \[rem:covers-NOt-maPs\] by the existence of $S$-widE coveRs OF $G {{\tiMEs}}\delta$. to outliNE tHE ConStRuCtiOn OF tHese cOveRs and to Prepare for The MaPping class grOUp We introducE soMe notation. PiCk a $g$-invarIaNt pROper mEtrIc on tHe set $E$ OF edges Of $\Gamma$; thIs IS possiBLe as $G$ anD $e$ ArE couNtable and the actioN Of $g$ On $E$ has finite stAbilizErS. foR EAch VerTex $v$ of $\GammA$ wIth inFInite vaLEnCY LEt $E_V$ Be the set of edgEs incident tO $V$. WrIte $d_v$ fOr The REstricTion oF tHE meTric to $E_v$. For $\Xi \in \delta$, $\xi \neQ v$ we deFIne its pROjectioN $\pi_v(\xi)$ To $E_V$ as The sET oF aLl eDgES of $\gAmMa$ tHAt aRe appear As InItial EdgeS OF GEodeSicS froM $v$ to $\xI$. This is a finitE suBset OF $E_v$ (This dEpendS agaIn Of finEness oF $\GammA$). FIx a vertex $v_0$ of finIte vAlence as a BasE pOinT. FOr $g \in g$, $\Xi \in \DeLta$ DefIne theiR *projecTIon DiSTANcE* at $v$ by $$d_v^\pi(g,\xi) := d_v ( \pi_v(Gv_0), \PI_V(\xI)).$$ For $\xi = v$, sEt $d^\pi_v(G,V) := \iNfTY$. (For relaTiVe hYperBOLic grOups A ReLated quaNtity iS OfTeN called An *Angle*; tHe TerMinOlogy HEre iS choseN to align BetteR With the case of tHE mapping class GRoUP.) if WE varY $g$ (iN a finite set) And $\xI$ (In an Open NEiGhbORhood) Then fOr FIxED $v$ the projection distAnCe $d^\pi_{v}(G,\xi)$ vaRies by a boundeD amount. UseFUL Is the folLowiNG aTTraction properTy for Projection DIstances: There Is $\Theta_0$ sUch that if $D^\PI_v(g,\xi) \geq \theTa_0$, tHen Any GEOdEsic from $gv_0$ to $\xI$ IN $\GamMa$ Passes tHroUgh $v$. ConVerSelY, if SomE gEodesic frOm $gv_0$ to $\xi$ MiSsEs $V$, tHen $D^\pi_

\] below, using the charac terization of $ N$- ${\ ma thca l{F} }$-amenability from Remark \[rem:covers-n ot-ma ps \ ] by th e exi stenceo f$ S $-w id ecov er s o f $G{{\ times}} \Delta$. Toou tline the co n st ruction of th ese covers a ndto pre pa ref or th e m appin g clas s group we intro du c e some notatio n . P icka $G$-invariant p r op e r metric on th e set$E $ o f edg esof $\Gamma $; this is poss i bl e a s $ G $ and $E$ are countablea ndthe ac ti ono f $G$on $E $h asfinite stab iliz ers. Foreach v e rtex $v $ of $\G amma$wit h i nfin i te v ale nc y le t $ E_v $ be the set o fedges inc i d e n t to $v $. W rite$d_v$ for the re stri c tio n ofthe m etri cto $E _v$. F or $\ xi \in \Delta$, $ \xi\neq v$ w e d ef ine i ts pr o jectio n $ \pi _v(\xi) $ to $E _ v$as t h eset of all edges o f$ \ Ga mma$ tha t area pp ea r as init ia l e dges o f geo desi c sfrom $v$ to $\ x i$ .This is a finit esub set of $ E _v$(thisdependsagain of fineness of $\Gamma$). Fi x a v er t ex $ v_0 $ of finite val e nceas a ba sep oint. For$g \i n G$, $\xi \in \Delt a$ defin e the ir *projectio n distance * a t $v$ by $$d _ v^ \ pi(g,\xi) := d _v (\pi_v(gv_0 ) , \pi_v( \xi)) .$$ For$\xi = v$ , set $d^\ pi_ v(g ,v) := \ in fty$. (For re l a tive h yperbol icgroupsa r ela ted qu an tity is o ften cal le dan * ang le*;t he termi no log yher e isc hosento al ignbe tt e r w ith the ca s e ofth emapp ing c lassgrou p .)If we v ary $g$ ( ina fin it eset) an d $\xi$ (in a nopen neigh bo rho od) th e n for fix ed $v$ the projection d i stance$d^ \pi_{ v}(g ,\xi)$ va rie s by a bo u nded a mount. Usef ul is t he fo l l ow ing a ttractionp r ope rty f or pro jection distances: therei s $ \Theta_0$ suc h t hati f $ d^\ p i_ v (g, \x i ) \ g e q \Theta_0$, th en any geo de s ic from $gv_ 0 $ t o$\xi$ i n $\Gam ma$ p a sses th rough $v$ . Convers el y, i f som e geodesic from $g v_0$ to $ \ xi$ m i ss es $v $,then $ d^ \pi _

\] below,_using the_characterization of $N$-${\mathcal{F}}$-amenability from_Remark \[rem:covers-not-maps\] by_the_existence of_$S$-wide_covers of $G_{{\times}}\Delta$. To outline the_construction of these covers_and to prepare_for_the mapping class group we introduce some notation. Pick a $G$-invariant proper metric on_the_set $E$_of_edges_of $\Gamma$; this is possible_as $G$ and $E$ are_countable and_the action of $G$ on $E$ has finite_stabilizers._For each vertex_$v$ of $\Gamma$ with infinite valency let $E_v$ be_the set of edges incident to_$v$. Write $d_v$_for_the_restriction of the metric_to $E_v$. For $\xi \in \Delta$,_$\xi \neq v$ we define its_projection $\pi_v(\xi)$ to $E_v$ as the set_of all edges of $\Gamma$ that_are appear as initial edges_of geodesics_from $v$ to $\xi$. This_is a finite_subset of_$E_v$ (this depends_again of fineness of $\Gamma$). Fix_a vertex $v_0$_of finite valence as a base_point._For $g \in_G$,_$\xi_\in \Delta$_define their *projection_distance*_at $v$_by_$$d_v^\pi(g,\xi) := d_v ( \pi_v(gv_0), \pi_v(\xi)).$$_For_$\xi = v$, set $d^\pi_v(g,v) := \infty$._(For relative hyperbolic groups_a_related quantity is often_called an *angle*; the terminology_here is chosen to align better_with the_case of_the mapping class group.) If we vary $g$ (in a finite_set) and $\xi$ (in an open_neighborhood) then for fixed_$v$ the_projection_distance $d^\pi_{v}(g,\xi)$ varies_by_a bounded_amount. Useful is the following attraction property_for projection_distances: there is $\Theta_0$ such that_if $d^\pi_v(g,\xi) \geq \Theta_0$,_then_any geodesic from $gv_0$ to $\xi$_in $\Gamma$ passes through $v$. Conversely,_if some geodesic from $gv_0$_to_$\xi$_misses $v$, then $d^\pi_