ImpactofInstrumental SystematicContaminationonthe Lensing MassReconstruction using the CMB Polarization Meng Su1, Amit P.S. Yadav1, and Matias Zaldarriaga1,2 ∗ 1Harvard-Smithsonian CenterforAstrophysics, 60Garden St., Cambridge, MA02138, USAand 2Jefferson Laboratory of Physics, Harvard University, Cambridge, MA 02138, USA Inthispaper, westudytheeffectsofinstrumentalsystematicsonthereconstructionofthedeflectionangle power spectrum from weak lensing of Cosmic Microwave Background (CMB) temperature and polarization observations. Weconsider7typesofeffectswhicharerelatedtoknowninstrumentalsystematics: calibration, rotation, pointing, spin-flip, monopole leakage, dipoleleakageandquadrupole leakage. Theseeffectscanbe characterizedby11distortionfields. Eachofthesesystematiceffectscanmimictheeffectiveprojectedmatter 9 power spectrumandhence contaminatethelensingreconstruction. Todemonstratetheeffect oftheseinstru- 0 mentalsystematicsonCMBlensingmeasurements,weconsidertwotypesofexperiments,onewithadetector 0 noiselevelforpolarizationof9.6µK-arcminandFWHMof8.0,typicalofupcominggroundandballoon-based ′ 2 CMBexperiments,andaCMBPol-likeinstrumentwithadetectornoiselevelforpolarizationof2.0µK-arcmin n andFWHMof4.0′,typicaloffuturespace-basedCMBexperiments. Foreachsystematicsfield,weconsider a various choices of coherence scale α , starting from α = 10 to the maximum of α = 120. Among all s s ′ s ′ J the11systematicparameters,rotationω,andmonopoleleakageγ &γ placethemoststringentrequirements, a b 5 whilequadrupole leakageq, pointingerror pa and pb,andcalibrationaareamongtheleastdemanding. The requirements from lensing extraction are about 1-2 orders of magnitude less stringent than the requirements ] tomeasure the primordial B-modes withinflationary energy scale of 1.0 1016 GeV. On the other hand the O × requirementsforlensingreconstructionarecomparableorevenmorestringentforsomesystematicparameters C thantherequirementstodetectprimordialB-modeswithinflationaryscaleE =3.0 1016GeV. i × . h PACSnumbers: p - o r t s I. INTRODUCTION a [ ObservationsofthetemperatureanisotropiesoftheCosmicMicrowaveBackground(CMB) havebeena majortoolto con- 1 v strain cosmological parameters. The polarization data of the CMB can help us to extract additional information beyond the 5 temperatureinformation[1,2]. ThenextgenerationofCMBobservationswillfocusontheprecisemeasurementofpolarization 8 oftheCMB,especiallythesocalledB-modepolarization,whichisatleasttwoordersofmagnitudesmallerthanthetempera- 2 tureanisotropysignal. IncontrasttoE-modepolarization,whichcanbegeneratedbyscalarortensorperturbationsintheearly 0 universe,theprimordialB-modesaregeneratedonlybytensorperturbations[3,4]. Howeverevenintheabsenceofprimordial . 1 B-modes,subsequentgravitationallensingbythelargescalestructureoftheUniverseconvertsE-modepolarizationtoB-mode 0 polarization[5,6,7,8]. Althoughtheamplitudeoftheprimordialgravitationalwavesignalisuncertainbymanyordersmagni- 9 tudeandmightnotbedetectablebythenextgenerationofpolarizationexperiments,thelensingB-modesignalisaguaranteed 0 predictionofthecurrentcosmologicalmodel. Inaddition,theB-modelensingsignalwillhelptobreakdegeneraciesbetween : v cosmologicalparameters[9,10,11,12,13,14]. i The weak gravitational lensing of CMB anisotropies provides a unique opportunity to map the matter distribution of the X universe. Thenon-GaussianhigherordercorrelationsintheCMBgeneratedbytheweaklensingcanbeusedtoreconstructthe r a massdistributionoftheinterveninglarge-scalestructure. Theprincipleiseasytounderstand. TheCMBphotonsareremapped by gravitationallensing which introducescorrelationsbetween differentangularmoments. One way of extractingthe lensing information is to use a quadratic combination of the CMB multipoles to define an estimator for the projected gravitational potential[15,16,17,18,19,20,21,22]. AlthoughtheB-modepolarizationobservationsarecurrentlystillnoisedominated,thenextgenerationofCMBpolarization instruments has the sensitivity to make first detections, at least of the lensing induced B-mode signal. However, there are severalchallengesforCMBlensingdetection,mainlycomingfromastrophysicalforegroundsandinstrumentalsystematics. It is important to estimate and control those spurious signals as well as possible when analyzing upcoming CMB data. These challengeswillhavetobeovercomeinordertoprobethephysicsoftheearlyuniversethroughB-modepolarizationortoinfer theprojectedlargescalematterdistributionfromE/Bpolarization. ∗Electronicaddress:[email protected] 2 Lensing studies can be considered as secondary science for an experimentdevoted to B-mode detection. Impact of instru- mental systematics on the projected matter power spectrum is helpfulto both instrumentdesign and futuredata forecast. On theotherhand,lensinginducedB-modepolarizationisacosmologicalcontaminantforthedetectionofprimordialB-modes,a systemicstudyofinstrumentalsystematicsforlensingreconstructionmaywellberequiredofsuchanexperimenttodelensethe observedCMBfields[23,24,25,26,27,28,29,30,31]. Intheliterature,instrumentalsystematicshavebeendiscussedextensively [32,33,34,35,36,37,38,39]. Themaingoalof thispaperistoillustratestheeffectsofinstrumentalsystematicsandsystematicallystudytheimpactonthemassreconstruction process for upcomingCMB experiments[24, 25, 26, 27, 28, 29, 30]. To calculate the effects of instrumentalsystematics on theprojectedmatterpowerspectrum,wemakeuseofthequadraticestimatortoreconstructtheprojectedgravitationalpotential [15],andclassifyandparameterizethesystematicsfollowing[32]. Wedividepolarizationcontaminationsintotwocategories: those which are associated with a transfer between the polarization state of the incoming radiation (from detection system), includingcalibrationandrotation, spin-flipcouplingandmonopoleleakageerrors, andthosewhich areassociated with CMB anisotropyinducedbythefiniteresolutionorbeamofthetelescope. Forthepolarizationtransfersystematics,weonlyconcern ourselves with polarization transfer in a single, perfectly known, direction on the sky. However, in reality, every experiment necessarilyhasfiniteresolutionandthisthereforeisanadditionalclassofcontaminationassociatedwiththeresolutionorbeam oftheexperiment. We referthereaderto[32]fora detaileddiscussionoftheparametrizationofthe systematicerrorsweuse inthispaper. We calculatetheunlensedandlensedCMB powerspectrumusingCMBFAST [40]. Inthecalculation,wehave assumed a flat ΛCDM cosmology with following cosmological parameter values: Ω = 0.045,Ω = 0.23,H = 70.5,n = b c 0 s 0.96,n =0.0,τ=0.08. t Thispaperisorganizedasfollows: InSec.IIwereviewthebasiclensingformalism,defineournotation,andintroducethe quadraticestimator of the lensing potentialreconstructionwhich we later use to explorethe effects of instrumentalsystemat- ics. InSec.III,wefirstconsiderarelativelysimplecaseofthecalibrationsystematicsonthetemperatureestimatoroflensing reconstruction. Thenwe considerEB estimatoras an example, studying7 typesof instrumentalsystematic effects onlensing reconstruction,andcomparetothesystematiccontaminationoftheB-modepowerspectrumdetection.Weconsidertwoinstru- mental configurations, one with noise sensitivity for polarization of 9.6 µK-arcmin and FWHM of 8.0 (Exp1from here on), ′ and another CMBPol like instrument with noise sensitivity for polarization of 2.0 µK-arcmin and FWHM of 4.0 (reference ′ experimentfrom here on). In Sec. IV and V, we describe our results and conclude with a discussion of the implications for experimentsdedicatedtomeasureprimordialB-modesorthesecondarylensingsignal. Weleavediscussionsoftheotherthree lensingpotentialestimators(EE,TE,TB)toAppendix. II. LENSINGFORMALISM GravitationallensingdeflectsthepathofCMBphotonsfromthelastscatteringsurfaceresultinginaremappingoftheCMB temperature/polarizationpattern on the sky. In this section, we review the basic lensing calculation for both temperatureand polarizationfieldsasthestartingpointofourdiscussion.WeformulateCMBlensingusingtheflat-skyapproximation[22]. The flat-sky approachsimplifies the derivationby replacingsummationsoverWigner symbolsof sphericalharmonicmomentsby integralsinvolvingmodecouplingangles[7]. MoredetailsofCMBlensingcanbefoundinthenicereviewpaper[41]. Atacertainpositionnˆ onthesky,theobservedCMBfieldX˜(nˆ)islensedfromanotherdirectionintheprimordialCMBsky X(nˆ +d(nˆ))atz=1090.Theremappingprocesscanbedescribedas T˜(nˆ) = T(nˆ +d(nˆ)), (1) [Q˜ iU˜](nˆ) = [Q iU](nˆ +d(nˆ)), ± ± whereT˜(nˆ)(T(nˆ))representsthelensed(unlensed)temperaturefluctuationfield, Q˜(nˆ)(Q(nˆ))andU˜(nˆ)(U(nˆ))arelensed(un- lensed)polarizationStokesparameters,andd(nˆ)isthedeflectionanglewhichisrelatedtoφ(nˆ),thelensinggravitationalpoten- tial,byd(nˆ)= φ(nˆ). Hereandthroughoutthispaper,weuseboldfacequantitiestoidentifyvectors,andX˜ (X)standsforlensed ∇ (unlensed)temperatureandpolarizationfields. Thelensingpotentialφ(nˆ)isgivenby φ(nˆ) = 2 r0dr dA(r0−r) Φ(r,rnˆ), (2) − Z0 dA(r)dA(r0) whered isthecomovingdistancealongthelineofsight;r isthecomovingdistancetothesurfaceoflastscattering,andΦis A 0 gravitationalpotential. ThelensingremappingprocessconservesthesurfacebrightnessdistributionoftheCMB,thusdoesnot changetheone-pointstatistics. The observed temperature and polarization fluctuations also include secondary effects, such as Sunyaev-Zel’dovich (SZ) effect[42] and IntegratedSachs Wolfe (ISW) effect[43], which comefromthe first orderdensityor potentialfluctuationand 3 thusalsocorrelatewiththelensingdeflectionangle.WedenotethesephysicalcontaminationtolensingreconstructionbyXsec(nˆ). WedenotethenoisecomponentbyXn(nˆ). ThetotalobservedCMBanisotropythereforeincludesthelensedprimarysignal,any secondaryeffects, andnoise, i.e. Xt(nˆ) = X˜(nˆ)+Xsec(nˆ)+Xn(nˆ). Inthe nextsection, we will introduceanothercontribution to Xt(nˆ), which comes from instrumentalsystematics Xsys(nˆ). We define the observedCMB field Xobs(nˆ) = Xt(nˆ)+Xsys(nˆ). Here we write the secondary contributionas an independentcomponentfrom the lensed CMB. However, in reality it is hard toseparateXsec(nˆ)fromXt(nˆ)becausesecondariesarealsolensedbygravitationalpotentialswithdeflectionanglesdepending ontheirredshifts. Inthispaper,wesimplydropthecontributioneffectsXsec(nˆ),asthistopicisbeyondthefocusofthispaper. We refer the readers to [22] for a treatment of the secondary anisotropy as a physical contamination to the lensing potential reconstructionanalysis1. It is convenientto work in Fourier space. If one considersa small enough patch of sky, spherical harmonic modes can be replaced by Fourier modes. Generalization from the flat-sky to the full sky is straightforward. The Fourier transform of the TaylorexpendedlensedCMBtemperatureandpolarizationfieldis d2l T˜(l) = dnˆT˜(nˆ)e ilnˆ =T(l) ′ T(l)L(l,l), (3) − · ′ ′ Z −Z (2π)2 d2l E˜ ±iB˜ (l) = Z dnˆ[Q˜(nˆ)±iU˜(nˆ)]e∓2iϕle−iˆl·nˆ =[E(l)±iB(l)]−Z (2π)′2[E(l′)±iB(l′)]Lp(l,l′), h i φ(l) = dnˆφ(nˆ)e ilnˆ, (4) − · Z where 1 d2l L(l,l) φ(l l) (l l) l + ′′ φ(l ) φ(l l l )(l l) (l +l l) l +... , (5) ′ ′ ′ ′ ′′ ′ ′′ ′′ ′ ′′ ′ ′ ≡ − − · 2Z (2π)2 × − − · − · (cid:2) (cid:3) (cid:2) (cid:3) 1 d2l LP(l,l′) ≡ e±2i(ϕl′−ϕl)φ(l−l′) (l−l′)·l′ + 2Z (2π)′′2e±2i(ϕl′−ϕl)φ(l′′)×φ(l−l′−l′′)(l′′·l′) (l′′+l′−l)·l′ +... . (cid:2) (cid:3) (cid:2) (cid:3) We can immediately see that lensing induces remapping of CMB fields by lensing potential gradients, hence in Fourier space lensing acts as a convolutionwhichcouplesdifferentharmonicmodes. Fouriermoments, powerspectrum, bispectrum, trispectrumandsoonoftheCMBfieldsandthelensingpotentialcanbedefinedintheusualmanner: Xi(l )X j(l ) (2π)2δ (l +l )Cij (l ), D 1 ′ 2 E ≡ D 1 2 XiX′j 1 Xi(l )X j(l )X k(l ) (2π)2δ (l +l +l )Bijk (l ,l ,l ), 1 ′ 2 ′′ 3 c ≡ D 1 2 3 XX′X′′ 1 2 3 D E Xi(l )X j(l )X k(l )X m(l ) (2π)2δ (l +l +l +l )Tijkm (l ,l ,l ,l ), 1 ′ 2 ′′ 3 ′′′ 4 c ≡ D 1 2 3 4 XX′X′′X′′′ 1 2 3 4 D E ... (6) wheretheanglebracketsrepresentensembleaveragesoverrealizationsoftheprimordialCMBfields,thelarge-scalestructure between observers and the last scattering surface, and the experimental noise. The connected part of the n-point function is denotedbythesubscriptc. Thefields X, X , X , X areamong T(l), E(l), B(l),φ(l) . Thesuperscriptsi, j,k,mrepresentthe ′ ′′ ′′′ unlensedfieldX,thelensedfield X˜,theinstrumentalnoiseXn,the{ instrumentalsystem}aticsXsys,CMBsecondarycontribution Xsec,thetotalsignalXt,ortheobservedfieldXobs. Wenotethatthebispectrumandhigherorderodd-correlationsvanishifone ignoresthesecondaryeffects. ThisisbecauseoddmomentscontainsampleaveragesovertheoddprimordialCMBfieldswhich weassumetobeGaussian. Wemaketheassumptionthatfluctuationsinthelarge-scalestructurebetweentheobserverandthelastscatteringsurfaceare Gaussianandhencecanbefullydescribedbyapowerspectrum. Weusethelensingpotentialpowerspectrumcalculatedfrom CAMBFAST.TheinstrumentalnoiseXnisalsoassumedtobeGaussian. Wenotethatprimordialnon-Gaussianitycanpossibly contributepercentleveluncertaintytoouranalysiswhichisconsideredin[45]. We are now in the position to calculate any order (cross) correlation functions of CMB fields and the lensing potential in Fourierspace. WewillassumeuniformGaussiannoisewiththepowerspectrumCXXngivenby l ClXXn =w−X1el2σ2b. (7) 1 Asanote,thermalSZeffectcaninprincipalbeseparatedfromtheprimaryfluctuationsbyitsspectraldependence. ForthekineticSZeffect,itwasclaimed thatbyusingaspeciallydesignedestimator[44],itispossibletoseparateitoutfromreallensingsignal. However,someimportantsecondarycontributions suchastheISWeffectcannotbeseparatedeasilyandwillleadtoadditionalnoisecontributionsduetocorrelationswiththelensingpotentials[22]. 4 XX 1W (l ,l ) 2W (l ,l ) ′ XX′ 1 2 XX′ 1 2 TT (L l ) (L l ) · 1 · 2 TE cos2(ϕ ϕ )(L l ) (L l ) l1− l2 · 1 · 2 TB sin2(ϕ ϕ )(L l ) 0 l1 − l2 · 1 EE cos2(ϕ ϕ )(L l ) cos2(ϕ ϕ )(L l ) l1− l2 · 1 l1− l2 · 2 EB sin2(ϕ ϕ )(L l ) sin2(ϕ ϕ )(L l ) l1 − l2 · 1 l1− l2 · 2 BB cos2(ϕ ϕ )(L l ) cos2(ϕ ϕ )(L l ) l1− l2 · 1 l1− l2 · 2 TABLEI:WindowfunctionswhichappearinEq.(11)andEq.(12);hereL=l +l 1 2 w 1isthedetectornoisevariancepersteradianareafortemperature(X=T)orpolarization(X=EorB),andσ =θ /√8ln2 −X b fwhm istheeffectivebeamwidthoftheinstrumentcalculatedfromitsfull-widthhalf-maximumresolutionθ . Wewillassumefully fwhm polarizeddetector,forwhich2w =w =w . T E B Quadratic combinations of CMB fields can be used as estimators of the lensing potential field and hence the intervening projected mass between us and the last scattering surface. Furthermore, a CMB-field-squared map appropriately filtered in Fourier space can serve as an optimal estimator for the deflection field. Optimal filters for quadratic estimators have been designed[15] A (L) d2l d (L) XX′ 1 Xt(l )Xt(l )F (l ,l ), (8) XX′ ≡ L Z (2π)2 1 ′ 2 XX′ 1 2 whereXandX canbeT,E,andB.ThenormalizationA ischosensuchthat d (L) =d(L) Lφ ′ XX XX CMB ′ h ′ i ≡ A (L) L2 d2l1 f (l ,l )F (l ,l ) −1 , (9) XX′ ≡ "Z (2π)2 XX′ 1 2 XX′ 1 2 # where F (l ,l ) = ClX1′X′tClX2XtfXX′(l1,l2)−ClX1X′tClX2X′tfXX′(l2,l1), (10) XX′ 1 2 CXXtCX′X′tCX′X′tCXXt (CXX′tCXX′t)2 l1 l2 l1 l2 − l1 l2 whereforXX =TT,EE,BB,andTE, ′ fXX′(l1,l2)=ClX1X′1WXX′(l1,l2)+ClX2X′2WXX′(l1,l2), (11) andforX = T,E ,X = B, ′ { } fXX′(l1,l2)=ClX1E1WXX′(l1,l2)+ClX2′X′2WXX′(l1,l2). (12) ThewindowfunctionsW aregiveninTable I. InFig. 1, we showthe inputpowerspectrumof thelensed CMB fields, the XX ′ reconstructed deflection field, and the corresponding Gaussian noise as obtained by using the estimators defined in Eq. (8). Notethattheaverage denotesanensembleaveragerestrictedonlytodifferentGaussianrealizationsoftheprimordial CMB h i CMB and instrumentnoise but assuming a fixed realization of the large-scale structure. The unmarkedaverage, , means h i theaverageovertheprimordialCMBfieldand thelarge-scalestructurerealizationsasdefinedinEq.(6). Forthepurposesof estimatingthelarge-scalestructureintherealobservableuniverse,itisessentialtoensurethattheestimatorsafterappropriate averagingoverrealizationsaretrulyunbiasedforatypicalrealizationoftheprimordialGaussianCMBfield. Aswewillseein thenextsectionthatinthepresenceofnon-zerosystematiccontamination,theestimatorsarebiased. III. INSTRUMENTALSYSTEMATICSEFFECTONTHELENSINGPOTENTIALPOWERSPECTRUM Inthissection,weusethequadraticestimatorstoderivetheeffectsofinstrumentalsystematicsonthereconstructedlensing potentialpowerspectrum. Weshowthatinstrumentalsystematicscanintroducenon-GaussiancorrelationsofCMBfields. The instrumental systematics-induced CMB trispectrum gives an extra contamination to the reconstructed deflection angle power spectrum. In subsection IIIA, we consider the simple case of the TT estimator to explain how systematics contaminate the lensingreconstructionprocess.InsubsectionIIIB,wetaketheEBestimatorasanexampletoshowhowinstrumentalsystematics in CMB polarization measurementsaffect the lensing reconstruction. In order to compare the experimentalrequirementsfor primordialB-modedetectionandlensingpotentialreconstruction,wealso calculatetheeffectsofinstrumentalsystematicson B-modedetectionforagiveninflationaryenergyscale. 5 4 10 Cldd Cldd K)1023 DetecTtorT noise for reference experiment 10-5 Exp1 NN((00)) ffoorr TTTB eessttiimmaattoorr RExepfeerreinmceen t NNN(((000))) fffooorrr TTETBE eeessstttiiimmmaaatttooorrr l(l+1)C(/21l 1110000-011 DetecTtorE noise for Exp1 BB dd ()+ll1C/2 11l00--76 NN((00)) ffoorr EEBE eessttiimmaattoorr TETEBETB dd ()+ll1C/21l0-7 ETTETB N(0) for EB estimator -2 EE -8 10 10 EB -3 -8 10 10 10 100 1000 10 100 1000 10 100 1000 l l l FIG.1: Leftpanel: CMBpowerspectrumfortheTT,TE,EEandlensedBBpowerspectrum. Thelower(upper)dashedblacklineshowsthe temperature(polarization)noiseforExp1. Thelower(upper)dottedblacklineshowsthetemperature(polarization)noiseforCMBPol-like referenceexperiment. Centerpanel: GaussiannoisefordifferentquadraticestimatorsfortheExp1. Rightpanel: Sameascentralpanelbut forCMBPol-likereferenceexperiment. Notethat theEBestimatorhasthelowestGaussian noise, andmaythusbeconsidered asthebest estimatoramongallthepossiblequadraticestimators. A. Asimpleexampleoftemperaturesystematics Weintroducethecalibrationparameter(gainfluctuationofreceivers)a(nˆ)fortemperaturemeasurement,definedas: T˜obs(nˆ)=[1+a(nˆ)]T˜t(nˆ). (13) Ifweassumethatthereisnocorrelationbetweenthelensingpotentialφandtheinstrumentalsystematica(nˆ),thepowerspec- trumofthelensedCMBtemperaturewithsystematicscorrectiontermcanbefoundas(moredetailsaregiveninAppendixA): d2l d2l d2l C˜TT = 1 1 Cφφ(l l)2 CTT + 1 CTT Cφφ[(l l ) l ]2 + 1 Caa CTT. (14) l " −Z (2π)2 l1 1· # l Z (2π)2 |l−l1| l1 − 1 · 1 Z (2π)2 |l−l1| l1 This result is given to linear order in the lensing-potential power spectrum Cφφ and the gain fluctuation systematics power l spectrumCaa. Thelasttermrepresentsthebiasintroducedbythecalibrationsystematics. Intheabsenceofsystematiceffects, l itiseasytoprovethatthedeflectionangleestimatoris d (L) = Lφ(L) d (L),asdesired. Butoncethecontribution TT CMB TT h i ≡ fromthecalibrationparametera(nˆ)isconsidered,onefinds: A (L) d2l d (L) =d (L)+ TT 1 F (l ,l )a(L)(CTT +CTT), (15) h TT iCMB TT L Z (2π)2 TT 1 2 l1 l2 i.e. in the presenceof a(L)the estimator d (L) is a biased estimatorfor thedeflectionfield in Fourierspace. Consequently, TT the deflection angle power spectrum d (L)d (L) would be biased due to systematic contamination, and is given by (see TT TT h i AppendixAfordetails) 6 A (L)A (L) d (L) d (L) = TT TT ′ TT TT ′ CMB h · i LSS SYS L L DD E E ′ d2l1 d2l′1 (2π)2F (l ,l )F (l ,l ) ×Z (2π)2 Z (2π)2 TT 1 2 TT 1′ 2′ Cφφf (l ,l )f (l ,l )δ (L+L) ( L TT 1 2 TT 1′ 2′ D ′ +(2π)2CTTtCTTt δ (l +l )δ (l +l )+δ (l +l )δ (l +l ) l1 l2 D 1′ 1 D 2′ 2 D 2′ 1 D 1′ 2 h i + Cφφ f (l ,l )f (l ,l )+Cφφ f (l ,l )f (l ,l ) δ (L+L) h |l1+l1′| TT 1 1′ TT 2 2′ |l1+l2′| TT 1 2′ TT 2 1′ i D ′ + Caaf (l ,l )f (l ,l )+Caa f (l ,l )f (l ,l ) h L aa 1 2 aa 1′ 2′ |l1+l1′| aa 1 1′ aa 2 2′ +Caa f (l ,l )f (l ,l ) δ (L+L) |l1+l2′| aa 1 2′ aa 2 1′ i D ′ ) = (2π)2δ (L+L) Cdd+N(0) (L)+N(1) (L)+N(S) (L)+... , (16) D ′ " L TT,TT TT,TT TT,TT # where we define f (l ,l ) = CTT +CTT. In the last line, the first term in the square bracketis the deflectionangle power aa 1 2 l1 l2 spectrumCdd. Thesecondterm is the so calledGaussian noise N(0) (L) whichgivesthe dominantnoise contributionto the L TT,TT variance of the deflection power spectrum. The third term N(1) (L) is the the leading order non-Gaussian noise which is TT,TT first order in Cφφ and gives correction to the dominant Gaussian noise N(0) (L). The forth term, N(S) (L), is the leading L TT,TT TT,TT orderinstrumentalsystematiccontributiontothevarianceandisfirstorderinCSS. TheGaussiannoiseN(0) (L)andthefirst L TT,TT ordernon-Gaussiannoise N(1) (L)havebeenpreviouslycalculatedin[15]and[21],respectively. Thesystematicnoiseterm TT,TT N(S) (L)isanewcontributiontolensingpowerspectrum. Inprinciple,oneshouldincludenoisetermswhicharehigherorder TT,TT inCφφ andCSS,howeversincebothofthemaresmall,wetruncateatthefirstorderandexpectthathigherordercontributions L L aremuchsmaller. The quadraticestimator given in Eq. (8) is optimizedin the presence of N(0) (L), and assuming no contributionfrom the TT,TT firstordernon-GaussiannoiseN(1) (L)andinstrumentalsystematicnoiseN(S) (L). Hencetheestimatorisoptimal2 aslong TT,TT TT,TT asN(1) (L) N(0) (L),andN(S) (L) N(0) (L). Ithasbeenshownthatthenon-GaussiannoiseN(1) (L)isaboutone TT,TT ≪ TT,TT TT,TT ≪ TT,TT TT,TT orderof magnitudesmaller than the Gaussian noise contribution[21]. In IIIB, we calculate the systematic noise term NS(L) contributiontothelensingpotentialreconstructionfortheEBestimator. B. Generalanalysisonpolarizationsystematics WeparametrizethefieldsofinstrumentalsystematicsforCMBpolarizationmeasurementsfollowing[32]. Thepolarization contaminations fall into two categories, one associated with the detector system which distorts the polarization state of the incomingpolarizedsignal(TypeIhereafter),andanotherassociatedwithdistortionoftheCMBsignalduetothebeamanisotropy (TypeIIhereafter). Thisparametrizationcanbe generalizedto differentpolarimeters. The instrumentalresponsetoincoming CMBradiationisusuallydescribedbytheJonestransfermatrix. BiasinducedinthematrixdeterminationwillmixtheStokes parametersdeterminedfromit. Tofirstorder,theeffectofTypeIsystematicsontheStokesparameterscanbewrittenas[32] δ[Q iU](nˆ)=[a i2ω](nˆ)[Q iU](nˆ)+[f if ](nˆ)[Q iU](nˆ)+[γ iγ ](nˆ)T(nˆ). (17) 1 2 1 2 ± ± ± ± ∓ ± a is a scalar field which describesthe miscalibrationof the polarizationmeasurements(recallthatin last subsection, we used a to denote the miscalibration of temperature measurements), ω is also a scalar field that describes the rotation angle of the instrument,(f if )arespin 4fieldsthatdescribethecouplingbetweentwospinstates(spin-flip),and(γ iγ )arespin 2 1 2 1 2 ± ± ± ± fieldsthatdescribemonopoleleakagefromthetemperaturetopolarization. 2IfthesystematiccontributionsarecomparabletotheGaussiannoisecontributionthenweneedtodesignnewoptimallensingreconstructionestimatorsto takeintoaccounttheinstrumentalsystematicseffect. 7 SimilartotheTypeIsystematics,theeffectofTypeIIsystematicsontheStokesparameterscanbewrittenas[32] δ[Q iU](nˆ;σ)=σp(nˆ) [Q iU](nˆ;σ)+σ[d id ](nˆ)[∂ i∂ ]T(nˆ;σ)+σ2q(nˆ)[∂ i∂ ]2T(nˆ;σ) (18) 1 2 1 2 1 2 ± ·∇ ± ± ± ± the systematic fields are smoothed over the average beam σ of the experiment. Therefore the type II systematic fields are sensitivetotheimperfectionofthebeamonthescaleσ. (p ip )arespin 1fieldsthatdescribepointingerrors,(d id )are 1 2 1 2 ± ± ± alsospin 1fieldsthatdescribedipoleleakagefromtemperaturetopolarization,andqisascalarfieldthatdescribesquadrupole ± leakage[32]. As a simple model, we willassume thatthe contaminationfields, as definedin (17) and (18), are statistically isotropicand Gaussian(althoughsomeofthesystematicsfieldsneednotbeso),thustheirstatisticalpropertiescanbefullydescribedbytheir powerspectra, S(l)S(l) =(2π)2δ(l+l)CSS , (19) ′ ′ l (cid:10) (cid:11) whereS standsforanyofthe11systematicfields. Thesystematicfieldscanbemodeledwiththepowerspectraoftheform CSS =C exp( l(l+1)α2), (20) l 0 − S i.e.whitenoiseabovecertaincoherencescaleα ,whichisakeyquantitytoaffectthelevelofcontaminationofeachsystematics S effects. ThenormalizationfactorC canbedeterminedby 0 d2l −1 C2 =A2 exp( l(l+1)α2) , (21) 0 S"Z (2π)2 − S # whereA characterizesthermsofthecontaminationfieldS. S TheinstrumentalsystematicsinducedistortionsontheCMBfileds. ThecontaminationstotheBBandEEpowerspectradue todifferentmeasurementsystematicstaketheform d2l d2l δClBB =XSS Z (2π)′2C|Sl−Sl′′|C|El′E| (σ)[WBS(l,l′)]2 +XSS Z (2π)′2C|Sl−Sl′′|C|Tl′T| (σ)[WBS(l,l′)]2, (22) ′ ′ d2l d2l δClEE =XSS Z (2π)′2C|Sl−Sl′′|C|El′E| (σ)[WES(l,l′)]2 +XSS Z (2π)′2C|Sl−Sl′′|C|Tl′T| (σ)[WES(l,l′)]2. (23) ′ ′ The explicit forms of WS(l ,l ) and WS(l ,l ) are given in Table II, which are the window functionsof each systematic S B 1 2 E 1 2 forB-modeandE-modeharmonics,respectively. The summationsofthe first termonthe RHS of Eq. (22) andEq. (23) run overcalibrationa,rotationω,spinflip f and f ,andpointingerrorγ andγ . Thesummationsofthesecondtermrunoverthe a b a b restofthesystematicsparameterswhichdescribethetemperatureleakagegiveninTableII. CEE(σ)andCTT(σ)arethebeam l l smoothedtemperatureandE-modepolarizationpowerspectra. CEE(σ)=CEEexp( l(l+1)σ), CTT(σ)=CTTexp( l(l+1)σ). (24) l l − l l − WeuseEq. (22)andEq. (23)tocalculatethesystematicrequirementsforB-modedetectioninordertocomparetherequire- mentsforlensingreconstruction.WeshowtheresultsinTablesIIIandIV. Now we move on to calculate the systematic contamination on the lensing power spectrum. The polarization fields are essentiallyuncorrelatedwiththelensingpotential,soifwedonotconsidersecondaryeffects,then-pointfunctionswithnodd arezero. Thenextnon-zeroorderisthe trispectrum. Thecalculationfortheconnectedpartofthetrispectrumof polarization in the presence of instrumental systematics is similar to the connected temperature trispectrum presented in the last section. Here we give the results for the trispectrum related to lensing reconstructionusing EB estimator, and refer the readersto the AppendixCfortheexplicitcalculationforotherquadraticestimators. Atleadingorderwehave 8 TypeofS WS(l ,l ) WS(l ,l ) B 1 2 E 1 2 Calibrationa sin[2(ϕ φ )] cos[2(ϕ ϕ )] l2− L l2− L Rotationω 2cos[2(ϕ ϕ )] 2sin[2(ϕ ϕ )] l2− L − l2− L Pointingp σ(l ˆl ) zˆsin[2(ϕ ϕ )] σ(l ˆl )sin[2(ϕ ϕ)] a 2× 1 · l2− L 2· 1 l2− l Pointingp σ(l ˆl )sin[2(ϕ ϕ)] σ(l ˆl ) zˆsin[2(ϕ ϕ )] b 2· 1 l2− l − 2× 1 · l2− L Flip f sin[2(2ϕ ϕ ϕ )] cos[2(2ϕ ϕ ϕ )] a l1 − l2− L l1− l2− L Flip f cos[2(2ϕ ϕ ϕ )] sin[2(2ϕ ϕ ϕ )] b l1 − l2− L − l1− l2− L Monopoleγ sin[2(ϕ ϕ)] cos[2(φ ϕ)] a l1− l l1 − l Monopoleγ cos[2(ϕ ϕ)] sin[2(ϕ φ) b l1− l − l1− l Dipoled (l σ)cos[ϕ +ϕ 2ϕ] (l σ)sin[ϕ +φ 2ϕ] a − 2 l1 l2− l 2 l1 l2 − l Dipoled (l σ)sin[ϕ +φ 2ϕ] (l σ)cos[ϕ +ϕ 2ϕ] b 2 l1 l2− l 2 l1 l2− l Quadrupoleq (l σ)2sin[2(ϕ ϕ)] (l σ)2cos[2(ϕ ϕ)] − 2 l2− l − 2 l2− l TABLEII:Windowfunctionsforallthe11systematicparameters. Firstcolumnindicatesthetypeofsystematicparametersinconsideration. SecondandthirdcolumnsshowwindowfunctionsforsystematicsinducedB-modeWS(l ,l ),andforE-modeWS(l ,l )respectively. These B 1 2 E 1 2 windowfunctionsareneededtocalculatesystematiccontaminationonprimordialgravitationalwavedetectionorthedeflectionanglepower spectrumreconstruction.Wenotethatl =l ˆl , l =L l ,andl =l ˆl . 1 1 1 2 − 1 2 22 E˜(l )obsB˜(l )obsE˜(l )obsB˜(l )obs = (2π)2δ (l +l +l +l ) 1 2 ′1 ′2 c D 1 2 ′1 ′2 × D E CEECEE Cφφ W (l , l )W (l , l )+Cφφ W (l , l )W (l , l ) ( l1 l′1 " |l1+l2| B 2 −1 B ′2 −′1 |l1+l′2| B 2 −′1 B ′2 −1 P distortion P distortion − − + CSS WS(l , l )WS(l , l ) + CSS WS(l , l )WS(l , l ) XS |l1+l2| B 2 −1 B ′2 −′1 XS |l1+l′2| B 2 −′1 B ′2 −1 i T leakage − +CTECTE CSS WS(l , l )WS(l , l ) + l1 l′1 h XS |l1+l2| B 2 −1 B ′2 −′1 T leakage − CSS WS(l , l )WS(l , l ) , (25) XS |l1+l′2| B 2 −′1 B ′2 −1 #) wherewe definedthelensingB-modewindowfunctionW (l,l) l (l l)sin2(ϕ ϕ )andWS(l , l )isthesystematics B ′ ≡ ′ · − ′ l− l′ B 1 −′1 windowfunctionforanyofthe11systematicsparameters. TheformulaforeachofthesystematicwindowfunctionsWS(l,l) B ′ canbefoundinTableII. Differenttrispectrumcanbeconstructedfromcombinationsofthetemperatureandpolarizationfields. Wediscussothercases intheAppendixCinordertocalculateEE,TEandTBestimatorsincludingcontributionsfromsystematicscontamination.The formulasshownherearereadilygeneralizedtothefullsky. Foradiscussionofthesphericalgeneralizationofthepolarization trispectra, see [47]. We have shown that E and B-modes are mixed not only by weak lensing, but also by instrumental sys- tematics. Evenifthereisnolensinginducedcorrelation,certainkindsofsystematicscangivea non-zerocontributionsto the trispectrum. Now we move on to construct quadratic lensing estimators from E/B polarization modes, and we quantitatively showhowinstrumentalsystematicscontaminationaffectsthe lensingreconstructionprocess. Againwe take the EBestimator as an example, and leave the discussions of other estimators to the Appendix C. The variance of the deflection angle power spectrumincludingsystematiceffectscanbewrittenas d (L) d (L) = AEB(L)AEB(L′) d2l1 d2l′1 F (l ,l )F (l ,l ) h EB · EB ′ iCMB LSS SYS L L Z (2π)2 Z (2π)2 EB 1 2 EB 1′ 2′ DD E E ′ E˜(l )obsB˜(l )obsE˜(l )obsB˜(l )obs × 1 2 ′1 ′2 D E = (2π)2δ (L+L) Cdd(L)+N(0) (L)+N(1) (L)+N(S) (L)+... , (26) D ′ " EB,EB EB,EB EB,EB # whereL = l +l , andCdd(L)isthedeflectionanglepowerspectrum. ThetermsN(0) (L), N(1) (L),and N(S) (L)arethe 1 2 EB,EB EB,EB EB,EB Gaussian noise, first order non-Gaussiannoise, and the first order systematics noise. The Gaussian noise contributioncomes 9 Cldd s=20’ Cldd s=20’ 10-6 a Exp1 NN((01)) ss==6300’’ a Reference NN((01)) ss==6300’’ d/2 -7 ss==1105’’ s=120’ experiment ss==1105’’ s=120’ d 10 Cl 1) + -8 (l 10 l -9 10 -10 10 CNl(d0d) ss==2300’’ Cldd s=20’ -6 Exp1 N(1) s=60’ Reference N(0) s=30’ 10 s=10’ s=120’ experiment N(1) s=60’ 2 s=15’ s=10’ s=120’ dd/ s=15’ Cl -7 1) 10 + (l l -8 10 -9 10 10 100 1000 10 100 1000 l l FIG.2: Upperpanels: ContaminationfromthecalibrationsystematicsatothedeflectionanglepowerspectrumusingtheEBestimator. The rmsfluctuation, A isassumedtobe10%. TheleftandrightpanelareforExp1andreferenceexperimentrespectively. Inboththepanels, S the solid black, dashed blue, and dot red curves show deflection angle power spectrumCdd(L), Gaussian noise N(0)(L), and the firstorder non-GaussiannoiseN(1)(L). Theremainingcurvesshowtheabsolutevalueofthesystematicbiasforvariouschoicesofcoherencelengthα , s startingfromα =10 toα =120.Lowerpanels:Sameastheupperpanelsbutforrotationsystematicsω. s ′ s ′ fromthedisconnectedpartofthefour-pointfunction,whileboththefirst-ordernon-Gaussiannoise N(1) (L)andsystematics EB,EB noise N(S) (L) contributioncomes from the connected part. In derivingEq. (26) we have used filters F which are given EB,EB EB inEq. (10),andthetrispectrumwhichisgiveninEq. (25). Theellipsesstandsfortermsbeyondfirstorderinthesystematics power spectra or the deflection angle power spectrum. We note that the Gaussian noise term also includessystematic effects implicitlysinceinstrumentalsystematicsbiasthemeasuredpowerspectrumaswehaveshown.TheGaussianandnon-Gaussian noisetermsin theabsenceof systematiccontributionhavebeenpreviouslyreportedin [15]and [21]. Thesystematicsnoise termN(S) (L)isnew,forwhichtheexplicitformisgivenas EB,EB N(S) (L) = AEB(L)AEB(L′) d2l1 d2l′1 F (l ,l )F (l ,l ) EB,EB L L Z (2π)2 Z (2π)2 EB 1 2 EB 1′ 2′ ′ P distortion P distortion CEECEE − CSS WS(l , l )WS(l , l ) + − CSS WS(l , l )WS(l , l ) ( l1 l′1 " XS |l1+l2| B 2 −1 B ′2 −′1 XS |l1+l′2| B 2 −′1 B ′2 −1 # T leakage T leakage − − +CTECTE CSS WS(l , l )WS(l , l ) + CSS WS(l , l )WS(l , l ) . (27) l1 l′1 " XS |l1+l2| B 2 −1 B ′2 −′1 XS |l1+l′2| B 2 −′1 B ′2 −1 #) Eq. (27)isourmainresultforthesystematicscontaminationonlensingpowerreconstructionusingtheEBestimator. Wewill usethisequationtonumericallycomputethesystematic-inducedbiasforthe11systematicparameters. Theresultsareshown 10 -6 10 Cldd s=20’ Cldd s=20’ fa Exp1 NN(s(0=1)) 1 0 ’ ss==63s=001’’ 20’ fa Rexepfeerriemnceen t NN((01)) ss==6300’’ s=15’ s=10’ s=120’ d/210-7 s=15’ d Cl ) 1 + (l -8 l10 -9 10-6 10 Cldd s=20’ Cldd s=20’ fb Exp1 NN(s(0=1)) 1 0 ’ ss==63s=001’’ 20’ fb Rexepfeerriemnceen t NN((01)) ss==6300’’ s=15’ s=10’ s=120’ -7 210 s=15’ d/ d Cl ) 1 + (l10-8 l -9 10 10 100 1000 10 100 1000 l l FIG.3:Upperpanels:Contaminationfromthespin-flipsystematics f tothedeflectionanglepowerspectrumusingtheEBestimator.Therms a fluctuation,A isassumedtobe10%.TheleftandrightpanelareforExp1andreferenceexperimentrespectively.Inboththepanels,thesolid S black,dashedblue,anddotredcurvesshowdeflectionanglepowerspectrumCdd(L),GaussiannoiseN(0)(L),andthefirstordernon-Gaussian noiseN(1)(L). Theremainingcurvesshowtheabsolutevalueofthesystematicbiasforvariouschoicesofcoherencelengthα ,startingfrom s α =10 toα =120.Lowerpanels:Sameastheupperpanelsbutforspin-flipsystematics f . s ′ s ′ b inFigures(2)to(7)andarediscussedinnextsection3. IV. RESULTS Figures2to7andTablesIII&IVsummarizeourmainfindings.Wehavefocusedonthesystematics-inducedbiasfortheEB estimator(resultsfortheotherestimatorsareprovidedintheAppendixC)becauseithasthehighestGaussiansignal-to-noise ratioforreconstructingtheprojectedmatterpowerspectrum(SeeFig.1or[15]).Figures2-7showthecontaminationintroduced bydifferentsystematiceffects(thetermN(S) (L)inEq. (27))inthedeflectionanglepowerspectrumreconstruction. Wehave EB,EB assumedthermsfluctuationofthesystematicsfieldstobe10%,andvariedcoherencelengthstartingfromminimumα = 10 s ′ tothemaximumα =120. Forcomparison,wealsoshowthelevelofGaussiannoiseandfirstordernon-Gaussiannoiseofthe s ′ EB estimator, the terms N(0) (L) and N(1) (L) respectivelyin Eq. (27). Systematics-inducedbiasgenerallyincreaseswith EB,EB EB,EB 3Asanote,foreachinstrumentalsystematicsparameterS,forboththepolarizationdistortionandtemperatureleakage,therearetwotermswhichcontribute tothefinalresults. ThetermsproportionalCSS isgenerallymuchsmallerthanthetermsproportionaltoCSS butfornon-GaussiannoiseN(1)[21],the termsproportionaltoCφφ areaboutanord|el1r+ol2f|magnitudelargerthanothertwotermsproportionaltoCφφ |l1+al′2n|dCφφ . |l1+l2| |l1+l1′| |l1+l2′|