RENORMALIZED QUANTUM TOMOGRAPHY GIACOMOMAUROD’ARIANOANDMASSIMILIANOFEDERICOSACCHI 9 0 0 ABSTRACT. Thecore ofquantum tomography is thepossibility ofwriting agenerally unbounded complexoperator informofanexpansion overoperators thataregenerally 2 nonlinearfunctionsofagenerallycontinuoussetofspectraldensities—-theso-calledquo- n rum of observables. The expansion is generally non unique, the non unicity allowing a furtheroptimizationforgivencriteria. Themathematicalproblemoftomographyisthus J theclassificationofallsuchoperatorexpansionsforgiven(suitablyclosed)linearspaces 9 ofunboundedoperators—e.g.Banachspacesofoperatorswithanappropriatenorm.Such 1 problemisadifficultone,andremainsstillopen,involvingthetheoryofgeneralbasisin Banachspaces,astillunfinishedchapterofanalysis. Inthispaperwepresentnewnon- ] trivialoperatorexpansionsforthequorumofquadraturesoftheharmonicoscillator, and h introduceafirstverypreliminarygeneralframeworktogeneratenewexpansionsbasedon p theKolmogorovconstruction. Thematerialpresentedinthispaperisintendedtobehelp- - fulforthesolutionofthegeneralproblemofquantumtomographyininfinitedimensions, t n whichcorrespondstoprovideacoherentmathematicalframeworkforoperatorexpansions a overfunctionsofacontinuoussetofspectraldensities. u q [ 2000MathematicsSubjectClassification.47N30,47N40,47N50.PACS:03.65.Wj,02.30.Tb, 1 42.50.-pKeywordsandphrases.Quantumtomography,operatorexpansions. v 6 6 8 2 . 1 0 CONTENTS 9 1. Introduction 2 0 2. Quantumtomographyandquorumofobservables 3 : v 2.1. Unbiasingnoise 6 i 3. Thecaseofhomodynetomography 6 X 4. Calculuswithnullfunctions 8 r a 5. TheKolmogorovconstruction 9 5.1. Homodynetomography 10 5.2. Spintomography 11 5.3. Alternateexpansions 12 6. Expansionofunboundedoperatorsoverthequadratures 12 7. Canonicaldualforhomodynetomography 14 7.1. Alternatedualframes 16 8. Framesofnormal-orderedmoments 16 9. Frameofmomentsversusquadraturedistribution 17 10. Generatingnewframes 18 11. Otherframes 20 Appendix 20 11.1. ProofofLemma4 20 11.2. Alternativederivationofidentity(87) 21 References 23 2 GIACOMOMAUROD’ARIANOANDMASSIMILIANOFEDERICOSACCHI 1. INTRODUCTION Thestateofaphysicalsystemisthemathematicaldescriptionthatprovidesacompleteinforma- tiononthesystem.Itsknowledgeisequivalenttoknowtheresultofanypossiblemeasurementonthe system.Inclassicalmechanicsitisalwayspossible,atleastinprinciple,todeviseaproceduremade ofmultiplemeasurementswhichfullyrecoversthestateofasinglesystem. Inquantummechanics, onthecontrary,thereisnoway,noteveninprinciple,toinferthequantumstateofasinglesystem withouthavingsomepriorknowledgeonit[1]. Itishoweverpossibletoestimatethequantumstate ofasystemwhenmanyidenticalcopiesareavailablepreparedinthesamestate,sothatadifferent measurementcanbeperformedoneachcopy. Suchaprocedureiscalledquantumtomography. Theproblemoffindingastrategyfordeterminingthestateofasystemfrommultiplecopiesdates backto1957byFano[2],whocalledquorumasetofobservablessufficientforacompletedetermi- nationofthedensitymatrix.However,quantumtomographyenteredtherealmofexperimentsmore recently,withthepioneeringexperimentsbyRaymer’sgroup[3]inthedomainofquantumoptics.In quantumoptics,infact,usingabalancedhomodynedetectoronehastheuniqueopportunityofmea- suringallpossiblelinearcombinationsofpositionandmomentum—theso-calledquadratures—of theharmonicoscillatorrepresentingasinglemodeoftheradiationfield. Thefirsttechniquetoreconstructthedensitymatrixfromhomodynemeasurements—socalled homodynetomography—originatedfromtheobservationbyVogelandRisken[4]thatthecollec- tionofprobabilitydistributionsachievedbyhomodynedetectionisjusttheRadontransformofthe WignerfunctionW. Therefore, similarlytoclassicalimaging, onecanobtainW byinvertingthe Radontransform,andthenfromW onecanrecoverthematrixelementsofthedensityoperator.This originalmethod,however,workswellonlyinasemi-classicalregime,whereasgenerallyforsmall photonnumbersitisaffectedbyanunknownbiascausedbythesmoothingprocedureneededforthe analyticalinversionoftheRadontransform. Thesolutiontotheproblemistobypasstheevaluation oftheWignerfunction,andtoevaluatethematrixelementsofthedensityoperatorbysimplyaverag- ingsuitablefunctionsofhomodynedata:thisisthebasisofthefirstunbiasedtopographictechnique presentedinRef. [5]. Clearlythestateisperfectlyrecoveredinprincipleonlyinthelimitofinfin- itelymanymeasurements: however,forfinitelymanymeasurementsonecanestimatethestatistical error affecting each matrix element. For infinite dimensions there is the further problem that the propagationofstatisticalerrorsofthedensitymatrixelementsmakethemuselessforestimatingthe ensembleaverageofsomeoperators(e. g. unbounded), andamethodforestimatingtheensemble averageisneeded,whichbypassestheevaluationofthedensitymatrixitself,aswasfirstsuggested in Ref. [6]. For a brief historical excursus on quantum tomography, along with a review on the generalizationtoanynumber ofradiationmodes, arbitraryquantumsystems, noisedeconvolution, adaptivemethods,andmaximum-likelihoodstrategiesthereaderisaddressedtoRef.[7]. Themostcomprehensivetheoreticalapproachtoquantumtomographyusestheconceptofframe of observables, i. e. a set of observables spanning the linear space of operators, from which one derivescontextuallythequorumofobservablesandtheestimationrule. Theensembleaverage H ofanyarbitraryoperatorH onaHilbertspaceHisestimatedusingmeasurementoutcomesofhthei quorum O upon expanding H over a set of functions f (O) of the observables O . What l n l l { } { } makesthegeneraltheorynontrivialininfinitedimensionsisthecrucialroleofthenonlinearfunc- tionsf (O)inmakingtheinfiniteexpansionconvergent.Let’sdenotebyP :=f (O),j =(n,l), n l j n l suchcompletesetofoperators.OnceyouhavetheP ,thentheproblemisreducedtothelinearprob- j lemofexpandinganoperatorasH = jhQ†j,HiPj,forasuitable“dual”setofoperators{Qj}. (Noticethat generally theindex l iscontinuous, whence also theoperator expansion.) Thescalar P productintheexpansionisgenerallynotsimplytheFrobenius’swhenweneedtoexpandoperators thatarenotHilbert-Schmidt. Themathematicaltheoryofframes[8,9,10,11]istheperfecttoolfor establishingcompletenessof P andforfindingdualsets Q . Inmostpracticalsituationsthe j j { } { } set P isover-complete, andtherearemanyalternatedualsets Q ,thenonunicityproviding j j { } { } roomforoptimization.AgeneraltheoryshouldalsoclassifytheoperatorsHhavingboundedscalar productwith{Ql}andexpansionH = jhQ†j,HiPj converginginaverageforagivenclassof quantumstates.Ininfinitedimensions—e.g.forhomodynetomography—theeasyknownapproach P RENORMALIZEDQUANTUMTOMOGRAPHY 3 worksonlyforHilbert-Schmidtoperators,whenceincludingtrace-classones—e.g. forestimating thematrix elements of thedensity operator over an orthonormal basis. For unbounded operators, however,suchoperatorexpansionbecomesaninfinitesumofunboundedterms. Ontheotherhand, converging(andevenfinite)alternateexpansionsareknowntoexistforvariousunboundedoperators [12].Aswewillseeinthispaper,themechanismallowing”renormalization”oftheexpansionrelies ontheexistenceofnull-estimators—namelyoperator-valuedfunctionsthathavezeromeanoverthe quorum—theirexistencebeingrelatedtoagroupofsymmetriesofthequorum. Thenotionofnull- estimatorwasfirstintroducedinRef. [13], inthecontext ofofhomodyne tomography, wherethe quorumismadeofthequadraturesofthefieldmode—thequantumanalogueoftheRadontransform. HerethesymmetrygroupofthequorumisthegroupU(1)ofrotationsofthequadraturephase.The existenceofnull-estimatorsleadstoinfinitelymanyalternateexpansionsofthesameoperatorover thequorum, allowingcancellationsoftheinfinitiesintheexpansion—a kindof”renormalization” procedure. Theproblemofclassifyingalloperatorexpansionsforagivenquorumininfinitedimensionsfor givenspacesofunboundedoperatorsisadifficultone,andremainsstillopen. Itinvolvesthetheory offramesor evenmoregeneral notionsofbasisinBanach spaces[8,9,10,11], astillunfinished chapter of analysis. In this paper we present new nontrivial operator expansions for the quorum of quadratures of the harmonic oscillator, and introduce a first preliminary general framework to generate and classify new expansions, based on the Kolmogorov construction. We hope that the materialpresented herewill open theway tothesolution of theproblem of quantum tomography ininfinitedimensions, leading toageneral mathematical framework for operator expansions over functionsofasetofspectraldensities. 2. QUANTUMTOMOGRAPHYANDQUORUMOFOBSERVABLES Thegeneral ideaof quantum tomography isthatthereisaset of observables Xξ ξ X onthe HilbertspaceofthesystemH—called”quorum”—bywhichonecanestimateanyd{esir}ed∈ensemble averageby measuring the observables of thequorum, each at thetime, in ascheme of arepeated measurements.Theobservablesofthequorummustbeindependenteachother,namelytheyarenot commuting: [Xξ′,Xξ′′]=0 ⇐⇒ ξ′=ξ′′. (1) GenerallythesetXparameterizingthequorumisinfinite,andmostcommonly,isacontinuum. In thesecases,sinceclearlyonecanmeasureonlyafinitenumberofobservables,thesearerandomly pickedoutaccordingtoagivenprobabilitymeasureonX, which,therefore, mustbeaprobability space. Inthefollowing,forsimplicity,wewillalsoassumeaprobabilitydensityoverXanddenote itwiththesymboldµ(ξ). Itfollowsthattheensembleaverageofa(generallycomplex)operatoris writtenintheformofdoubleexpectation X = dµ(ξ) f (X X) , (2) ξ ξ h i h | i ZX wherethegenerallynonlinearfunctionf (xX)ofthevariablexhasananalyticformwhichdepends ξ | ontheparticularoperatorX. Wewillcallthefunctionf (xX)thetomographicestimatorforX ξ | withquorum Xξ ξ X. Ifwewanttoachievetheestimationof X = Tr[ρX]—theexpectation { } ∈ h i beingsupposedly bounded onthestateρ—byaveragingtheestimator f (xX)over bothquorum ξ | andmeasurementoutcomeswithaboundedvariance,weneedtohavethefunctionf (xX)square- ξ | summableoverxandξ,moreprecisely dµ(ξ) dE (x) f (xX)2 < , (3) ξ ξ h i| | | ∞ ZX ZXξ where denotes the spectrum of X , and dE (x) its spectral measure. In the following, for ξ ξ ξ X simplicity, wewill consider the spectrum independent on ξ. Clearly, the above square- ξ X ≡ X summabilitywilldependagainonthestateρandontheoperatorX. Wefirstwanttonoticetwomainfeaturesofestimators: 4 GIACOMOMAUROD’ARIANOANDMASSIMILIANOFEDERICOSACCHI (1) Theestimatorf (xX)isgenerallynotunique,namelytherecanbemanydifferentestima- ξ | torsforthesameoperatorX.Thisisequivalenttotheexistenceofnullestimators,namely functionsn (x)suchthat ξ dµ(ξ) dE (x)n (x)=0. (4) ξ ξ ZX ZX Accordingly,theestimatorsaregroupedintoequivalenceclasses,eachclasscorresponding toanoperatorX.Forsuchequivalencewewillusethenotation ,i.e.wewillwritef g ≃ ≃ orf g 0todenotethatthetwoestimatorsareequivalent,namelytheydifferbyanull − ≃ estimator. (2) Forfixedxandξtheestimatorf (xX)mustbealinearfunctionalofX,namely ξ | fξ(xaX+bY)=afξ(xX)+bfξ(xY), fξ(xX†)=fξ(xX)∗. (5) | | | | | Example1(Homodynetomography). [7]Thequorumisgivenby X withX denotingthe . { φ}[0,π) φ quadrature Xφ = 12(a†eiφ +ae−iφ), a,a† being the annihilation and creation operators of the harmonicoscillatorwithcommutator[a,a†] = 1. Estimatorsforthedyads n m madewiththe | ih | orthonormalbasis n ,n=0,..., are {| i} ∞ fφ(x||nihm|) = +∞ dk4|k|e18−ηηk2−ikxhn+d|eikXφ|ni Z−∞ = eid(φ+π2)s(n+n!d)! +∞dk|k|e1−2η2ηk2−i2kxkdLdn(k2), (6) Z−∞ whereLd(x)denotes thegeneralized Laguerre polynomials. For theunbounded operators aand n a†aonecancheckthatthefollowingareunbiasedestimators f (xa) = 2eiφx, φ | fφ(x|a†a) = 2x2− 21. (7) Theproblemofquantumtomographyistoestablishthegeneralruleforestimation,namely Definition1(Estimationrule). Giventhequorum Xξ ξ X findthecorrespondence: { } ∈ f (xX) X, foreveryoperatorXonH, (8) ξ | ⇐⇒ wherewepossiblymeantofindthewholeequivalenceclassofestimatorsf (X X). ξ ξ | Beforesolvingthistask,firstoneneedstoknowthatthesetofobservables Xξ ξ X isactually { } ∈ aquorum. Theeasiest thingtodo, however, istoderiveboth thequorum and theestimationrule contextually,startingfromaspanning-setofobservables—shortlyobservablespanning-set—namely asetofobservables Fω ω O intermsofwhichwecanlinearlyexpandoperatorsasfollows { } ∈ X = dωc (X)F . (9) ω ω ZO Noticethatthenotionofoperatorspanning-setusedheregeneralizesthenotionofframesforBanach spacestounboundedoperators(seealsothefollowing),andisgenerallynotstrictlyaframeaccord- ingtothedefinitionofRefs. [8,9]. Inthefollowingthroughout thepaper wewillalwaysassume probability distributions admitting densities. Generally, the set O isunbounded, and themeasure dωisnotnormalizable,whence,assuch,theexpansion(9)cannotbeusedforquantumtomography. However,generallythisfeatureisrelatedtotheredundancyoftheobservablespanning-set,whichin- cludesmanyobservablesF thatarejustdifferentfunctionsofthesameobservable.Then,collecting ω theobservablesofthespanning-setintofunctionalequivalenceclassesK ,eachcorrespondingtoan ξ . observableofthequorum Xξ ξ X,onecanrelabeltheobservablespanning-setasFκ,ξ =fκ(Xξ) withκ K ,writing { } ∈ ξ ∈ X = dµ(ξ) dν(κ)c (X)f (X )= dµ(ξ)f (X X), (10) κ,ξ κ ξ ξ ξ | ZX ZKξ ZX RENORMALIZEDQUANTUMTOMOGRAPHY 5 wherethefunctionf (xX)istheintegralovertheobservablesequivalenttoX ,namely ξ ξ | . f (X X)= dν(κ)c (X)f (X ). (11) ξ ξ κ,ξ κ ξ | ZKξ NoticethatintermsofthespectraldecompositionofX wecanwrite ξ X = dµ(ξ) dE (x)f (xX), (12) ξ ξ | ZX ZXξ andsincethisexpansionislinearinthespectralmeasuredE (x),thelattercanberegardeditselfas ξ . anobservablespanning-set. Indeed,uponintroducingthespectraldensitydE (x)=Z dx,and . . ξ ξ,x thedensitydµ(ξ)=m(ξ)dξ,andrenamingζ =(ξ,x)andZ= (ξ,x),x ,ξ X Eq.(12) ξ { ∈X ∈ } canberewritteninthesameformofEq.(9),namely X = dζc′ζ(X)Zζ, (13) ZZ wherethenewexpansioncoefficientsarenowgivenby c′ξ,x(X)=m(ξ)fξ(x|X)=m(ξ) dν(κ)cκ,ξ(X)fκ(x). (14) ZKξ ForhomodynetomographytheabovequantitiesareexplicitlygiveninTable1. For F an operator frame, the coefficients of the expansion (9) can be rewritten in form of a ω pairing( )withadualframeG asfollows ω ·|· c (X)=(G X), (15) ω ω | intermsofwhichEq.(14)rewritesinthepairingform c′ξ,x(X)=m(ξ)(Wξ,x|X), (16) withdualframe Wξ,x = dν(κ)Gξ,xfκ∗(x). (17) ZKξ Fromthelastequationitfollowsthattheestimatoritselfcanbewrittenusingthepairing f (xX)=(W X), (18) ξ ξ,x | | or,intermsoftheoriginalobservableframe f (xX)= dν(κ)(G X)f (x). (19) ξ ξ,x κ | | ZKξ general homodyne general homodyne general homodyne Xξ Xφ cω(X) Tr[D†(α)X] dµ(ξ) dπφ X [0,π) ω α Zξ,x x φφ x dµξ(ξ) dφφ dνK(ξk) 1dRkk XZξ [0|,iπR) h R| ω απ cξ,κ(X) Tr[4e−ikX|φ|X] m(ξ) π1× Fω D(α) fκ(Xξ|X) eikXφ Wξ,x −41π(x−PXφ)2 TABLE1. Tableofcorrespondenceforhomodynetomography. 6 GIACOMOMAUROD’ARIANOANDMASSIMILIANOFEDERICOSACCHI 2.1. Unbiasingnoise. Itispossibletoestimatetheidealensembleaverage X bymeasuringthe quoruminthepresenceofinstrumental noise, whenthenoisemapN isinvhertiible, or, moregen- erally,ifthereexiststherightinverseofN . Intermsofobservableframesthisjustcorrespondsto usingadifferentdualframe.Moreprecisely,onehas: X =N N −1(X)= dµ(ξ)N [fξ(Xξ N −1(X))] | ZX (20) = dµ(ξ) N (dEξ(x))fξ(xN −1(X)). | ZX ZXξ NoticethatEq.(20)meansthat X = dµ(ξ) dEξ(x) Nfξ(xN −1(X)), (21) h i h i | ZX ZXξ . . where = Tr[ρ ] denotes the ideal ensemble average, whereas N = Tr[N τ(ρ) ] de- notes thhe·iexperiment·al ensemble average, N being the predual mapho·fiN (Schroedinger v·ersus Heisenberg picture). This also means that fo∗r left invertible map N the noisy spectral measures N (dE (x))arestillaquorum.IntermsofthepairinginEq.(18)unbiasingthenoiseisequivalent ξ tousethenewdualframeN −1†(Wξ,x). WhenN isnotright-invertibleonecanstillestimatethe ensembleaverageofoperatorsintherangeofthemap. Moreover, ininfinitedimension, whenthe noiseCPmapN iscompactitsinversemapisunbounded,andonegenerallycannotunbiasthenoise withoutrestrictingthespaceofreconstructedoperators. Otherwise,onehasaHadamardill-posed problem,forwhichtherearebiasedcompromises,suchasputtingacutoffonthevanishingsingular valuesofN . Example2(PaulitomographyinaPaulichannel). [7]Theoperators[σ /√2]makeanobservable α orthonormalbasisforC⊗2.WeconsidernowthenoisedescribedbythedepolarizingPaulichannel p N =(1 p)I + T, (22) − 2 . where I denotes the identity map and T the trace map T(X) = ITr(X). This noise can be simplyunbiasedvianoise-mapinversion: 1 p N −1† = I T. (23) 1 p − 2(1 p) − − Example3(Homodynetomographywithη <1). [7]ThesetofdisplacementsoperatorsD(α):= eαa†−α∗a, α∈CmakeanobservableDirac-orthonormalframeforT1/2,where Ts = X T, X =:f(a,a†):, s.t. lim f(α,α¯)es|α|2 =0 (24) { ∈ α } →∞ where: :denotesnormalordering. Inthepresenceofnoisefromnonunitquantumefficiencyη,the unbiased reconstruction ispossible foroperators inTs ifη (2s)−1. Infact, one usesthenew ≥ dual: D(α) N −1†(D(α))=ηD(η1/2α)e1−2η|α|2. (25) → Example4 (Homodyne tomography in Gaussian noise). [7] As forquantum efficiency, Gaussian noisecanbeunbiasedformeanthermalphotonnumbern¯ s 1.Onehasthenewdual: ≤ − 2 D(α) N −1†(D(α))=D(α)en¯|α|2. (26) → 3. THECASEOFHOMODYNETOMOGRAPHY Before addressing the general problem of deriving a general tomographic rule for unbounded operators,inthissectionwewillre-derivetheknownpatternfunctionofhomodynetomographyin ordertoillustratethegeneralconceptsintroducedintheprevioussection. RENORMALIZEDQUANTUMTOMOGRAPHY 7 Thestartingpointistheobservableframe D(α) α C,intermsofwhichthedecomposition(9) { } ∈ forHilbert-Schmidtoperatorsrewritesasfollows d2α X = Tr[D†(α)X]D(α). (27) π ZC Bychangingtopolarvariablesα=( i/2)keiφ,Eq.(27)becomes − X = π dφ +∞ dk|k| Tr[X eikXφ]e−ikXφ. (28) π 4 Z0 Z−∞ Eq.(28)canbeusedonlyforHilbert-Schmidtoperators,forwhichthetraceundertheintegralinEq. (28)exists.Intermsofthequadraturespectralmeasure,onehas X = π dπφ +∞ dk4|k| Tr[X eikXφ]e−ikXφ = π dπφ +∞dEφ(x)Tr[XWφ,x], (29) Z0 Z−∞ Z0 Z−∞ where Wφ,x =e−iφa†aD(x)W0,0D†(x)eiφa†a, W0,0 =−12PX12, (30) 0 PdenotingtheCauchyprincipalvalue. Ontheotherhand,inthenextsectionwewillshowthatfor unboundedoperatorswealsohavetheexpansion π dφ ∞ X = dtTr[G†(t,φ)X]F(t,φ), π Z0 Z−∞ (31) F(t,φ)= 1 e−21(t−i2Xφ)2, G(t,φ)= d tet22 1dθi(1 θ)teiφ iθteiφ. √2π dt | − ih− | Z0 whichintermsofthethequadraturespectralmeasurereads π dφ ∞ X = π dEφ(x)Tr[XWφ′,x], (32) Z0 Z−∞ wherenow Wφ′,x= ∞ dt√12πe−21(t−i2x)2ddttet22 1dθ|i(1−θ)teiφih−iθteiφ|. (33) Z−∞ Z0 Alternatively,asshowninSec. (9)bymeansoftheframeofnormal-orderedmoments,onehas theexpansion X = ∞ a†namTr[gn†,mX] n,m=0 X (34) 1 =Z0π dπφZ−+∞∞dEφ(x)Tr"Xn,Xm∞=0gn†,m n+nm!− Hn√+m2n(+√m2x)# where min(n,m) ( 1)j gn,m = − n j m j . (35) j! (n j)!(m j)!| − ih − | Xj=0 − − The above expansions in Eqs. (32) apnd (34) are just examples of alternate expansions which are equivalent for the estimation of the expectation values of (even unbounded) observables, but can beverydifferentasregardsthestatisticalnoiseaffectingsuchestimation. Asamatteroffact, the problemofclassifyingallpossibleexpansionshasbeenneversolved,and,hopefully, theresultsof thepresentpapermaysuggestaunifyingapproachtothesolutionofsuchadifficultproblem.Aswe willseeinthenextsection,theexistenceofmanyalternateexpansionsisduetothesymmetryofthe quorumofquadratureoperators,andtheresultingpropertiesofnullestimatorfunctions. 8 GIACOMOMAUROD’ARIANOANDMASSIMILIANOFEDERICOSACCHI 4. CALCULUSWITHNULLFUNCTIONS Wefirstnoticethattoanullestimatorfunctionn 0itcorrespondsanullexpansionoverthe ξ ≃ quorum,namely dµ(ξ) dE (x)n (x)=0 dµ(ξ)n (X )=0. (36) ξ ξ ξ ξ ⇐⇒ ZX ZX ZX Letusrecalltheorderingrelation[14] k,l { } k!l! s r j :a†kal:s = j!(k j)!(l j)! −2 :a†k−jal−j:r , (37) Xj=0 − − “ ” where k,l := min(k,l), and s = 1,0, and1 correspond to normal, symmetrical, and anti- { } − normalordering,respectively.WewillalsowritethesymmetricalorderingasS a†kal :=:a†kal:0, { } andthenormalorderingas:a†kal::=:a†kal: 1. − Thenwehave: Lemma1(Mainequivalencerelation). [13]Thefollowingequivalencerelationholds xke±i(k+2n+2)φ 0, k,n 0. (38) ≃ ∀ ≥ Proof.SinceXφk = 21k kl=0 kl S{a†lak−l}eiφ(2l−k),onehas P π`dφ´ π e±i(k+2+2n)φXφk =0, ∀k,n≥0, (39) Z0 whichisequivalentto(38). Stateddifferently: Lemma2. Thefollowingequivalencerelationholds Hk(√2x)e±i(k+2n+2)φ 0, k,n 0, (40) ≃ ∀ ≥ whereH (x)denotethek-thHermitepolynomial. k Proof.FromthedefinitionofHermitepolynomialsonehas √12nHn(√2Xφ)= √12n ∂∂tnn e−t2+√2t(a†eiφ+ae−iφ) ˛t=0 = √12n ∂∂tnn˛˛t=0e√2ta†e˛˛˛iφe√2tae−iφ =Xk=n0 nk!a†kan−keiφ(2k−n) =2n:Xφn:. (41) Then,itfollowsthat ˛˛ πdφ π e±i(k+2+2n)φHk(√2Xφ)=0, ∀k,n≥0, (42) Z0 whichisequivalentto(40). Moreover,wealsohave Lemma 3 (Equivalence of truncated Hermite polynomials). The following equivalence relations hold: e±inφH2(ll)+n(κx)≃e±inφH2l+n(κx), (43) whereweintroducedthetruncatedHermitepolynomial H(l)(z)= l (−)mn!(2z)n−2m. n 2l. (44) n m!(n 2m)! ≥ m=0 − X RENORMALIZEDQUANTUMTOMOGRAPHY 9 Proof. Thetwoidentitiesarejustthecomplexconjugatedofeachother. Therefore,itissufficient toprovetheidentitywiththeplussign. Thelatterisasimpleconsequenceofidentity(38),namely xke±i(k+2n+2)φ 0.Byusingtheun-truncatedHermitepolynomial,wehave ≃ einφH2l+n(κx)=einφl+[[n/2]](−)mm(2!(l2+l+n)n!(2κ2)2ml+)n!−2m (45) m=0 − X x2l+n−2mei[n+2(l−m)+2(m−l)]φ, × fromwhichitfollowsthatalltermswithm>lareequivalenttozero. Finally,onecanshowthePoissonidentities(whoseproofcanbefoundintheAppendix) Lemma4(Poissonidentities). Thefollowingidentitieshold f(x2)δ (φ) 1 f(x2e2iφ) + f(x2e−2iφ) , π ≃ π 1 e 2iφ 1 e2iφ » − − − – (46) xf(x2)δ (φ) 1 xeiφf(x2e2iφ) + xe−iφf(x2e−2iφ) , π ≃ π 1 e 2iφ 1 e2iφ » − − − – Inparticular,wehavetheidentity 1 δ (φ) . (47) π ≃ π 5. THEKOLMOGOROVCONSTRUCTION In thisSectionwe present theKolmogorov construction, and itsrelationwiththe fundamental identityofquantumtomography. Inthefollowing, byl wedenotetheHilbertspaceofsquaresummablesequencesofcomplex 2 numbers,andbyL (X)wedenotetheHilbertspaceofsquaresummablefunctionsoverthespaceX. 2 Forexample,X=R,andL (X)theHilbertspaceofsquaresummablefunctionsontherealaxis,or 2 X=S1,andL (X)istheHardyspaceofsquare-summablecomplexfunctionsonthecircle. Inthe 2 followingwewillfocusattentiononlytothecaseX R.Considernowacompleteorthonormalset offunctions[υ (x)]forL (X).Thecompletenessof≡thesetcorrespondstothefollowingdistribution n 2 identity υn(x)∗υn(y)=δ(x y), (48) − n whereδ denotestheusualDirac-dXelta. Considernowa(infinite-dimensional)HilbertspaceHand denoteby[w ]anorthonormalbasisforit.Thefollowingvector n υ(x) = υ (x)w , (49) n n | i | i n X isDirac-normalizable,inthesensethat υ(y)υ(x) =δ(x y). (50) h | i − Considernowanother(infinite-dimensional)HilbertspaceK H.Totheorthonormalbasis[υ (x)] n forL2(X)and [zn]forKweassociate amap fromtheobse≃rvables OX withspectrumX onHto operatorsinB(H,H K)givenby ⊗ υ(X)= υ (X) z , (51) n n ⊗| i n X where,asusual,wedefinetheoperatorsυ (X)intermsofthespectralresolutionofXasfollows n υ (X)= dE (x)υ (x), (52) n X n ZX wheredE (x)denotesthespectralmeasureofX. X ForX,Y OX formallywewrite ∈ υ(X)†υ(Y)= υn(X)†υn(Y), (53) n X 10 GIACOMOMAUROD’ARIANOANDMASSIMILIANOFEDERICOSACCHI ConsidertheintegralkernelK(x,y),x,y XcorrespondingtothepositiveoperatorK B(H) ∈ ∈ K(x,y)= υ(x)K υ(y) . (54) h | | i Foranytwoself-adjointoperatorsX,Y onL (R),theexpressionK(X,Y)iswelldefinedinthe 2 followingsense K(X,Y)=. υ(X)†(I K˜)υ(Y), (55) ⊗ whereK˜ B(K)isgivenby ∈ K˜ = z w K w z , (56) n n m m | ih | | ih | n,m X namelywecanalsowrite K(X,Y)= υn(X)† wnK wm υm(Y)= dEX(x) dEX(y)K(x,y). (57) h | | i n,m ZX ZX X Thisisalsoequivalent tosaythatfor anyexpansion of K(x,y)inseriesof products offunctions ofsinglevariable,K(X,Y)isdefinedasthesameexpansion, orderedwiththefunctionsofX on theleftandthefunctionsofY ontheright. ForcommutingX,Y,thenK(X,Y)simplyrepresents thesameanalyticexpressionofK(x,y),nowsubstitutingtheoperatorsinplaceofthevariables.As anexample,theidentityoperatorK = I correspondstotheDirac-deltakernel,andforcommuting X,Y OX wehave ∈ υ(X)†υ(Y)=δ(X Y). (58) − ByreplacingnowH H⊗2,evenfornoncommutingXandY,onehas → (υ(X)† I)(I υ(Y))=δ(X I I Y). (59) ⊗ ⊗ ⊗ − ⊗ Moreover,similarlytoEq.(57),onehas K(X I,I Y) =. (υ(X)† I)(IH⊗2 K˜)(I υ(Y)) ⊗ ⊗ ⊗ ⊗ ⊗ = υn(X)† υm(Y) wn K wm ⊗ h | | i n,m X = dE (x) dE (y)K(x,y). (60) X X ⊗ ZX ZX Thefundamentalidentitiesofquantumtomographyareobtainedasanexpansionoftheswapoperator Eoverthequorum,sinceforanystateρandobservableAonehasTr[ρA]=Tr[(ρ A)E],where ⊗ E ψ φ = φ ψ . | i⊗| i | i⊗| i 5.1. Homodynetomography. FromEq.(28),itiscleartheswapoperatorcanbewrittenas E = π dφ +∞ dk|k|e−ikXφ eikXφ. (61) π 4 ⊗ Z0 Z−∞ Then,theusualhomodynetomographicformulacanbeobtainedbytheKolmogorovconstructionin writingtheswapoperatorasfollows π dφ E = π (υ(Xφ)†⊗I)(IH⊗2⊗K˜)I⊗υ(Xφ), (62) Z0 correspondingtothepositivekernel K(x,x′)= π x Y x′ = dk keik(x′−x) = P 1 , (63) 2h || || i ZR 4 | | −2 (x−x′)2 whereY isthequadratureconjugatedtoX,with[X,Y] = i. Thekernelisclearlypositive,since 2 onehas 2 dk K(xi,xj)ξi∗ξj = R 4 |k|˛ eixjξj˛ (64) Xi,j Z ˛Xj ˛ ˛ ˛ ˛ ˛ ˛ ˛