Weak Lensing Simulations for the SKA Prina Patel∗a,b, Ian Harrisonc, Sphesihle Makhathinid,b, Filipe Abdallae,d, David Baconf, Michael L. Brownc, Ian Heywoodg,d, Matt Jarvish,a, Oleg Smirnovd,b 5 aUniversityoftheWesternCape,bSKASouthAfrica,cJodrellBank,dRhodesUniversity, 1 eUniversityCollegeLondon, fICG,gCSIRO,hUniversityofOxford, 0 E-mail: [email protected] 2 n Weakgravitationallensingisaverypromisingprobeforcosmology. Measurementsaretradition- a J allymadeatopticalwavelengthswheremanyhighlyresolvedgalaxyimagesarereadilyavailable. 6 However, the Square Kilometre Array (SKA) holds great promise for this type of measurement 1 at radio wavelengths owing to its greatly increased sensitivity and resolution over typical radio ] O surveys. Thekeytosuccessfulweaklensingexperimentsisinmeasuringtheshapesofdetected C sourcestohighaccuracy.Inthisdocumentwedescribeasimulationpipelinedesignedtosimulate . radioimagesofthequalityrequiredforweaklensing, andwillbetypicalofSKAobservations. h p We provide as input, images with realistic galaxy shapes which are then simulated to produce - o images as they would have been observed with a given radio interferometer. We exploit this r pipelinetoinvestigatevariousstagesofaweaklensingexperimentinordertobetterunderstand t s a theeffectsthatmayimpactshapemeasurement. WefirstshowhowtheproposedSKA1-Midar- [ rayconfigurationsperformwhenwecomparethe(known)inputandoutputellipticities. Wethen 1 investigate how making small changes to these array configurations impact on this input-outut v 2 ellipticity comparison. We also demonstrate how alternative configurations for SKA1-Mid that 9 aresmallerinextent,andwithafastersurveyspeedsproducesimilarperformancetothoseorig- 8 3 inally proposed. We then show how a notional SKA configuration performs in the same shape 0 measurement challenge. Finally, we describe ongoing efforts to utilise our simulation pipeline . 1 to address questions relating to how applicable current (mostly originating from optical data) 0 shapemeasurementtechniquesaretofutureradiosurveys. Asanalternativetosuchimageplane 5 1 techniques, we lastly discuss a shape measurement technique based on the shapelets formalism : v thatreconstructsthesourceshapesdirectlyfromthevisibilitydata. Weendwithadiscussionof i X extensionstotheoutcurrentsimulationsandconcludingremarks. r a AdvancingAstrophysicswiththeSquareKilometreArray June8-13,2014 GiardiniNaxos,Italy ∗Speaker. (cid:13)c Copyrightownedbytheauthor(s)underthetermsoftheCreativeCommonsAttribution-NonCommercial-ShareAlikeLicence. http://pos.sissa.it/ SKAWeakLensingSimulations PrinaPatel 1. Introduction Weakgravitationallensingisapromisingprobeforcosmology. Lightraysfromdistantsources are bent by the gravitational potential of objects on the path to an observer, leading to a coherent ellipticity or shear on images of galaxies near each other on the sky. We can measure statistics of this shear for galaxies in the Universe, at different redshifts and angular separations. These statistics are sensitive to the growth history of density fluctuations in the Universe (and therefore thematterpowerspectrum), andtotheexpansionhistoryoftheUniverse(andhenceforinstance, darkenergyparameters). Weak shear measurements are already maturing at optical frequencies (e.g. Kilbinger et al., 2013), and a range of future optical experiments are planned to provide tight constraints on cos- mologicalparametersusingthisprobe(forinstancethegroundbasedLargeSynopticSurveyTele- scope (LSST), (LSST Science Collaboration et al., 2009), and Euclid space telescope, (Laureijs et al., 2012). See also Bacon et al. (2014) and Kitching et al. (2014)). In addition, shear can be measuredatradiofrequencies(e.g.Changetal.,2004;Pateletal.,2010), andasshowninBrown etal.(2014)SKAwillbeabletoprovidecompetitivegravitationallensingmeasurements. In order to demonstrate this, it is necessary to simulate realistic images which could be ob- tainedusingparticularSKAmeasurementsets,includingtheeffectsofrealisticgravitationallens- ing and telescope effects. Shear measurement techniques should be carried out on these images, and we should verify that we can obtain faithful estimates of the true shear. In this chapter, we describe our efforts to make such simulations and confirm that SKA configurations which are be- ingconsideredbythecommunityaresuitableformakingtheSKAapowerfulgravitationallensing telescope. 2. SimulationPipeline Thesimulationsweutilisearebasedonapipelinebuiltintwopartswhichwebrieflydescribe hereandrefertoPateletal.(2013)forfurtherdetails. Thepipelineworksbytakinginputimages thatcontainrealisticgalaxyshapesandrunningthemthroughasimulator. Withinthesimulatorwe candefinetheinterferometertouseandotherobservationdetailssuchasthefrequency,bandwidth, integrationtimeetc. Thesimulatorthenpredictsthevisibilitiesforthegivenimageandobservation. Thelastpartofthesimulationthentakesthesevisibilitiesandimagesthemaswouldbedonewith real observed data and a restored image is produced. Once the simulation has produced a final restoredimagewethenre-analysetheseimageswithourchosenshapemeasurementtechniqueand comparetheinputandoutellipticities. 2.1 Simulation Wefirstlydescribetheinputimageswehavecreatedthatareusedthroughoutthiswork. The input images are based on a shapelet (see §4.2) based method that is described in detail in Re- fregier(2003)andRefregier&Bacon(2003). Briefly, theshapeletsmethoddecomposesagalaxy imageintoaseriesoflocalisedbasisfunctionscalledshapelets. Theshapeletsareacompleteand orthonormal set of basis functions consisting of weighted Hermite polynomials, corresponding to perturbationsaroundacircularGaussian. Theprocessofgeneratingtheseinputimagesisdescribed 2 SKAWeakLensingSimulations PrinaPatel indetailin(Roweetal.,2013),resultinginshapeletmodelsthatrepresentsimulated,butrealistic, (known)galaxyshapesaswouldhavebeenobservedwiththeHubbleSpaceTelescope(HST).Al- though these are simulated optical galaxies we make use of them as there exists no large enough sample of highly resolved galaxy images in the radio. In Patel et al. (2010) it was found that on acase-by-casebasistheintrinsicshapesofradioandopticalsourceswereonlyweaklycorrelated, but that the overall distribution of ellipticities were very similar at the two wavelengths. To keep computation time low we have created 100 such images that are 0.85×0.85 arcminutes2 with a pixelscaleof0.05arcseconds,thateachcontain(cid:39)100sourceseach. Thisgivesaresultinginput numberdensityofn(cid:39)140arcminutes−2,whichisfarlargerthananycurrentlensingsurvey. These images are then fed into the simulator which in combination with the chosen interfer- ometer and observation particulars, predicts the visibilities. If desired we are then able to include affects that would effect the visibilities in a variety of ways using the Radio Interferometer Mea- surement Equation (RIME) formalism (e.g. Smirnov, 2011), which relates the propagation of the signal from the source to detector via various observational effects. We are also able to include Gaussian measurement noise on the visibilities. Note in our simulations we do not currently em- ployanyobservationeffectsbutwedoincludeGaussianmeasurementnoise. Thesevisibilitiesare thenimagedusingstandardtechniques(e.g. CLEAN)thatweareagainabletocontrol,generating theoutputrestoredimage. 2.2 ShapeMeasurement Wetaketherestoredimagesproducedbythesimulationandthenanalysethemusingtheimage basedshapeletmethoddescribedin§4.2andcomparetheinputandoutputellipticities. Note,that in producing the restored image using CLEAN, a convolution is performed between the model imageandthemainlobeofthesynthesisedbeam(PSF).WesimulatethisPSFaswellandperform a deconvolution within the shapelet analysis, further details on deconvolution with shapelets can be found in Rowe et al. (2013), and further details about the image plane shapelet analysis can be found in Patel et al. (2013). We are then left with shapelet models (i.e. shapelet coefficients f )bothbeforesimulationandpost,fromwhichweestimatea2-component(complex)ellipticity n,m accordingto √ 2f(cid:48) 2,2 ε = , (2.1) (cid:104)f − f (cid:105) 0,0 4,0 (Masseyetal.,2007). ThisellipticityestimatoristheGaussianweightedquadrupolemomentcast intoshapeletspace. WethenfollowHeymansetal.(2006)andfitalinearmodeltoourdatapoints ε −εt =mεt+c, (2.2) i i i i i whereεt isthetrueinputellipticityandε isthemeasuredellipticity. Inallthefollowstherelative i i meritofeachexperimentcanbecomparedthroughthecalculatedvaluesofm andc. Foraperfect i i experimentwherewefullyrecoverexactlyalltheshapesintheimagesm =0,c =0. Anon-zero i i m isindicativeofacalibrationbiasresultingfrompoorcorrectionoffactorsthatcirculariseimages, i poorPSFcorrectionforexample. c (cid:54)=0suggestsasystematicwhereevencircularobjectsappear i elliptical. Wehavemadenoattempttooptimiseanyofouranalysestoreducethebiasesinanyway, neither have we looked for the origin of these biases. As such, the bias values presented here are 3 SKAWeakLensingSimulations PrinaPatel onlymeaningfulinacomparativesenseandshouldnotbetakentorepresentthefinalperformance thatanysuchexperimentmightachieve. Inafullanalysisonewouldhopetounderstandthenature and origin of such biases to high precision in order to correct for them. One clear use of our pipelineistoassesthelevelsofbiasthatmaybeintroducedbyobservationeffects(e.g. Direction DependantEffects(DDEs)). The key to making a weak lensing measurement requires accurate measurements of many galaxy shapes. Our simulation pipeline offers a way in which all parts of the data processing pipeline,fromrawvisibilitiestorestoredimages,canbeexplored. Intherestofthisdocumentwe presentinvestigationsthathavebeencarriedouttoaddresssomeofthekeyquestionsmostrelevant toweaklensingstudieswiththeSKA. 3. SKABaselinesConfigurations SincetheconfigurationforSKA1-Midisyettobecompletelyfinalisedweexplorehowchang- ing the array configuration as proposed by a small amount would effect the ability to accurately measuretheshapesofgalaxies. WegeneratedmanydifferentSKA1-Midarrayconfigurationsand ran them through our pipeline. In this section we describe what impact minor changes had on the calculatedcalibrationandadditivebiases. 3.1 BaselinesChangesandImpactonWeakLensing We initially calculated the bias values for the two proposed SKA1-Mid configurations. The first is that proposed by the SKAO (referred to as SKAM) and the other, a short time after, by Robert Braun (SKAM12), both are shown in Figure 1. Both these arrays contain 254 dishes with 197ofthemwithina4kmcore,andtheother57dividedinto3logarithmicallyspacedspiralarms extendingoutto100km. Forallsimulationsruninthissectionwehaveadoptedan8hourobser- vation at 700MHz and 10 50 MHz channels pointing at declination δ =−40◦. We add Gaussian measurementnoisetothevisibilitiesresultinginthesourcesintheoutputimageshavingasignal- to-noise of (cid:39) 10. We have also adopted a uniform weighting scheme through this work. Also shown in Figure 1 are the recovered ellipticity distributions derived for both configurations. Un- surprisingly, both these two configurations produce similar calibration values, with m (cid:39)−0.261 i and very small additive bias. We use these recovered values of m and c as our base values to i i which we can compare the values derived from modified SKA1-Mid configurations. Note that all thecalculatedbiasvaluesforallconsideredarraysaregiveninTable1. Thechangesweexploredwere: changingthespacinginthearms,takingdishesfromthecore and redistributing them into the 3 spiral arms and adding new dishes to the spiral arms. In the formercase,welookedatchangingthearmdistributionfromlogarithmictoequidistantandlinear. Inthecaseofthelattertwo,9,21,30,39,51,and60disheswereaddedtothearmswhilekeeping themaximumextentofthearmsthesame. InTable1weshowtheellipticityrecoveryperformance oftheseotherSKA1-Midconfigurationsdescribedabove. Thefirsttwoentriesinblueandmagenta correspondtotheSKAOandRobertBraunconfigurationsrespectively. Thecyanrowscorrespond to the configurations with equidistant and linear arms spacing. Entries in green and yellow are those where either dishes were redistributed from the core (green), or new dishes (yellow) were addedtothespiralarms. 4 SKAWeakLensingSimulations PrinaPatel Figure 1: Left panel: Array configurations for SKA1-Mid as proposed by the SKAO (April 2013) and RobertBraun(September2013). RightPanel: Recoveredellipticity(onlycomponent1oftheellipticityis shown)distributionsfortheseSKA1-Midarrays. Figure2: Cross-sectionsthroughthePSFsobtainedfromconfigurationsdescribedin§3. Since these are only modest changes to the configuration (i.e. ∼20% movement/addition of dishes) we see no significant improvements in performance of the recovered ellipticity distribu- tions. Since the deconvolution of the PSF in known to be one of the major causes of systematic error in shear measurement, in Figure 2 we show cross-sections of the PSFs for all the consid- ered configurations. As can be seen, the change in the PSF is small and so reaffirms the result of consistentcalibrationvaluesintheabsenceofanyotherpotentialcausesofnoise. 3.2 AlternativeConfigurations For weak lensing the main aspect of the baselines configuration is concerned with achieving high sensitivity at scales where we can measure the shapes of sources in the continuum. This 5 SKAWeakLensingSimulations PrinaPatel Table 1: SKA1-Mid ellipticity recovery performance results. The first two entries in blue and magenta correspond to the SKAO and Robert Braun configurations respectively. The cyan rows correspond to the configurationswithequidistantandlineararmsspacing. Entriesingreenandyellowarethosewhereeither dishes were redistributed from the core (green), or new dishes (yellow) were added to the spiral arms. aquotedrelativetotheSKAOandRobertBraunconfigurations. Name Ntotal Ncore Narms Arm aSensitivityfor NumberDensity mi ci Dishes Dishes Dishes Spacing SNR=10 narcminute−2 −0.278±0.021 0.001±0.005 SKAM(SKAO) 254 197 19 Logarithmic 1.0 21.97 −0.258±0.020 0.017±0.005 −0.227±0.015 −5×10−4±0.004 SKAM12(RobertBraun) 254 197 19 Logarithmic 1.0 30.43 −0.280±0.016 0.001±0.004 −0.319±0.015 −0.006±0.004 SKAM12EQ 254 197 19 Equidistant 1.25 19.32 −0.301±0.015 0.010±0.004 −0.297±0.015 −0.006±0.004 SKAM12LIN 254 197 19 Linear 1.0 30.69 −0.282±0.016 0.004±0.004 −0.292±0.019 −0.006±0.005 SKAM9C 254 188 22 Logarithmic 1.0 32.20 −0.264±0.017 0.002±0.005 −0.307±0.018 −0.002±0.005 SKAM21C 254 176 26 Logarithmic 1.0 32.33 −0.278±0.019 0.001±0.005 −0.308±0.015 −0.006±0.004 SKAM30C 254 167 29 Logarithmic 1.0 30.67 −0.286±0.017 0.001±0.005 −0.333±0.014 −0.006±0.004 SKAM39C 254 158 32 Logarithmic 1.0 25.41 −0.293±0.015 0.011±0.004 −0.314±0.016 −0.005±0.005 SKAM51C 254 146 36 Logarithmic 1.13 31.21 −0.285±0.015 0.008±0.004 −0.318±0.015 −0.005±0.004 SKAM60C 254 137 39 Logarithmic 1.13 25.72 −0.297±0.015 0.005±0.004 −0.234±0.016 −0.007±0.004 SKAM263 263 197 22 Logarithmic 1.13 33.52 −0.264±0.016 −0.003±0.004 −0.256±0.017 −0.004±0.005 SKAM275 275 197 26 Logarithmic 1.13 29.61 −0.256±0.018 −0.009±0.005 −0.256±0.017 −0.002±0.005 SKAM284 284 197 29 Logarithmic 1.13 29.79 −0.268±0.017 0.002±0.005 −0.260±0.018 −0.003±0.005 SKAM293 293 197 32 Logarithmic 1.25 29.87 −0.269±0.016 −0.001±0.004 −0.272±0.016 −0.005±0.004 SKAM305 305 197 36 Logarithmic 1.25 30.09 −0.281±0.016 0.002±0.005 −0.322±0.015 −0.006±0.004 SKAMPLUS 314 197 39 Logarithmic 1.38 31.42 −0.289±0.017 0.005±0.005 translatesroughlytohavingsignificantsensitivityatscalescorrespondingto0.5-1arcseconds. For this reason increasing the the number of antennas in the spiral arms out to 70-80 km is beneficial forweaklensingwhilethelackofthesebaselinesmakessuchasurveyunfeasible. To accommodate the other 3 cosmology science cases (cosmology with continuum and HI galaxysurveysandintensitymapping)aswellasweaklensingalternativeconfigurationshavebeen proposedsuchthattheuvcoverageisasfullaspossibleouttobaselinesof70-80km. Inorderto achieve a smooth transition between the three sections of the array (inner core, outer core and the spiral arms), the core is ‘puffed’ up slightly while the total number of dishes in the (inner+outer) coreispreserved. Thetwoproposedconfigurationsarereferredtousingthefollowingconvention SKA1Wi-jAkBl, where i refers to the number of dishes moved from the outer core to the spiral arms,jisthenumberofnewdishesaddedtothearms,kisthemaximumextentofthespiralarms andlthemaximumbaseline. Thelattertwoarefixedinbothcasestok=72kmandl=120km. We shallrefertothetwoconfigurations,SKA1W9-0A72B120andSKA1W9-12A72B120,asAandB respectively. 6 SKAWeakLensingSimulations PrinaPatel Figure3: SKA1-MidconfigurationsproposedbyRobertBraun(blue)andalsoonethatissmallerinextent andtakesintoconsiderationthesitegeography(red). Resultsforboththeseconfigurationsarecomparedin Table2. To assess the capabilities of these alternative configurations we run a new set of simulations at600MHz,800MHzand1000MHzwith150MHzchannelobservingatdeclinationδ =−30◦. To compare we also run the same simulation using the Robert Braun SKA1-Mid configuration discussedabove,theresultsarepresentedinTable2. We notice immediately that as we increase the frequency of the observation, the calibration valuesdecrease. Thisisbecausetheresolutionis∝λ/D,whereλ isthewavelengthofobservation and D the maximum baseline. In this case, our PSF is effectively becoming smaller, and so the higherthefrequency,themoreaccurateourshapemeasurement. Also,asdescribedinMakhathini etal.(inpreparation)thesearrayshaveoptimisedthedistributionofdishesinthespiralarmssuch that the sensitivity at angular scales of 0.4−1 arc second at 650 MHz can be enhances without significantly compromising the larger scales, so we expect them to perform best at the lower part ofthefrequencyspaceexploredhere. Also shown in Table 2 is the same simulation ran for the so-called SKA1V8 configuration. This is a slightly altered configuration of SKA1-Mid that also has a smaller extent than the two originally proposed configurations (maximum baseline of (cid:39)150km opposed to (cid:39)170km), it also takes into account the geography of the site. This configuration is plotted (along with SKAM12 from above) in Figure 3. Encouragingly, even with site topology incorporated this configuration producessimilarcalibrationvaluestotheoneoriginallyproposedthatdidnottakethisintoaccount, whilealsobringingdownthemaximumextentofthespiralarms. 3.3 SKACapabilities InthisSectionwecomputetheperformanceofSKAandcompareittoSKA1-Mid. Inorderto dothiswehavecreatedaSKAconfigurationthatconsistsof5spiralarmsextendingouttoo150km (eachspiralarmhas50disheslogarithmicallyspaced),butwehaveneglectedallthedisheswithin the1kmcore. ThisisduetotheverylargenumberofbaselinesinvolvedinafullSKAsimulation (and hence massive computation time). Instead, since weak lensing is primarily concerned with 7 SKAWeakLensingSimulations PrinaPatel Table2: WeaklensingsimulationsresultsforSKAM12proposedbyRobertBraunandalso2alternatives (SKA1W9-0A72B120andSKA1W9-12A72B120)thatgiveafulleruvcoveragetobaselinesbetween70- 80km. SKA1V8istheSKA1-Midconfigurationwithasmallerextentandwithsitegeographyconsidered. ArrayConfiguration 600MHz 800MHz 1000MHz mi ci mi ci mi ci −0.560±0.039 0.028±0.011 −0.508±0.031 0.052±0.008 −0.424±0.021 0.021±0.005 SKA1REF2(RobertBraun) −0.491±0.040 −0.004±0.010 −0.434±0.033 −0.007±0.008 −0.400±0.021 −0.001±0.005 −0.655±0.027 0.033±0.007 −0.604±0.020 0.034±0.005 −0.533±0.020 0.016±0.005 SKA1W9-0A72B120 −0.639±0.025 −0.006±0.006 −0.557±0.021 −0.006±0.005 −0.530±0.021 0.003±0.005 −0.582±0.043 0.038±0.011 −0.563±0.020 0.008±0.005 −0.530±0.018 0.002±0.004 SKA1W9-12A72B120 −0.596±0.040 −0.006±0.010 −0.532±0.020 −0.007±0.005 −0.530±0.017 −0.008±0.004 −0.545±0.042 0.026±0.011 −0.519±0.037 0.071±0.010 −0.458±0.023 0.052±0.006 SKA1V8 −0.506±0.046 0.016±0.011 −0.480±0.036 −0.003±0.009 −0.428±0.024 0.004±0.020 thelongerbaselinesanyway,andalsoatthisstageweareonlyinterestinggainingsomeideaasto whatSKAmightbeabletoachieve,weadoptthissimplification. TheresultantSKAconfiguration isshowninthelefthandpanelofFigure4. TocompareappropriatelytotheSKA1-Midsimulationsdiscussedin§3weagainruntheSKA simulationwithan8hourobservationat700MHzand1050MHzchannelspointingatdeclination δ =−40◦. Wehavealsoagaincorruptedthevisibilitiesbyanamountthatresultsinsourcesbeing at a SNR (cid:39)10. The received ellipticity distribution is shown in the right hand panel of Figure 4, theresultingcalibrationvaluesare: m = −0.357±0.005 1 c = 3×10−4±0.001 1 m = −0.354±0.005 2 c = −0.002±0.001. 2 (3.1) We see that the multiplicative bias values recovered from SKA seem to be worse than for SKA1-Mid. We note at this stage that this is only a notional SKA configuration that we have simulated and so can not be completely relied upon when comparing to the more sophisticated SKA1-Mid configurations. Since our SKA configuration has some dishes missing, we are invari- ably missing many short and intermediate length baselines, that also carry shape information of scalesthatarerelevant. Wealsoseethaterrorbarsonm areafactorof10smallerthanforSKA1- i Mid. This is due to the many more sources that reach our final catalogue as SNR ∼10 sources, meaningthisisamoreprecisemeasurement. 3.4 CalibrationRequirementsforSKA To provide some context for the obtained calibration values, we calculate the requirements on m and c for stages of SKA1-Mid and SKA and also for comparison, current and future optical surveyssuchastheDarkEnergySurvey(TheDarkEnergySurveyCollaboration,2005),Euclidand LSST.WeadopttherequirementsascomputedinAmara&Réfrégier(2008),whicharebasedupon theparameters: skyarea,galaxymedianredshiftandgalaxynumberdensity. Therequirementsare setsuchthatthestatisticalerrorisequaltothesystematicerrorandthusprovidesanupperlimiton thelevelofbiasallowed. 8 SKAWeakLensingSimulations PrinaPatel Figure 4: Left panel: A mock SKA configuration with 5 spiral arms extending to 150 km, but no dishes withintheinner1kmofthecore. Rightpanel: RecoveredellipticitydistributionforSKA. Inaddition,wefollowtheconventionofconvertingthemultiplicativeandadditivebiasesinto asinglequalityfactorQ,computedhereasinVoigt&Bridle(2010)withanassumedrmscosmic shear of σ = 0.03. In Table 3 we show the requirements for notional surveys conducted with γ an early phase of SKA1-Mid (SKA1-Mid early), SKA1-Mid and SKA, along with corresponding numbersforDES-likeandEuclid/LSST-likesurveysforcomparison. Thevaluesquotedforthenumberdensityandmedianredshiftsarederivedforanenvisaged2 year (net) continuum survey over 3 possible survey areas. The specification used are those given in Braun (2013), which in turn use the SKA1-Mid baseline design and the SKADS simulations of Wilman et al. (2008). The sensitivity levels have been chosen appropriately for weak lensing angularscalesof0.5arcsecondsatBand2andthegalaxynumberdensitiescorrespondto>10σ detections. SKA1-Midearlyisdefinedtobesuchthatithas50%ofthesensitivityofSKA1-Mid. We note how these requirements are orders of magnitude smaller than those derived in the preceding section. In our simulations we have not attempted to optimise any of the parameters (either in the simulation or the shape measurement analysis) to seek out the smallest calibration values,e.g. wehavemadenoattempttooptimisetheimagingofthesimulateddatabyinvestigating otherimagingmethodsotherthanCLEAN.Therequirementsquotedrepresentthelevelsthatneed to be achieved in order for the error budget to be equal between the systematics and statistics. Wehopethatwecanutiliseourpipelinefurthertounderstandthevarioussystematicsandexplore differentimagingtechniquesetc. toprovidemorerobustvaluesofforthecalibrationbiases. 4. ShearMeasurementTechniques As discussed above, the signal in weak gravitational lensing is the small shearing of galaxy imagesbyforegroundmatter. Thesmallnessofthisshearing(typicallyoforder1%)anditssensi- tivitytochangeincosmologicalparameters(typicallyoforder0.01%fora1%changeinthedark energyequationofstatew)meansthatanyeffectwhichiscapableofbiasingresultsmustbecare- fullycontrolled. Oneplaceinwhichsuchabiasmayenterisinthetranslationfromreal,noisydata 9 SKAWeakLensingSimulations PrinaPatel Table3: RequirementsonmultiplicativeandadditivebiasesonellipticitymeasurementforproposedSKA weaklensingsurveystobedominatedbystatisticalratherthansystematicsuncertainties,andforDES-like andEuclid/LSST-likeforcomparison. QiscalculatedfrommandcasinVoigt&Bridle(2010). Experiment A n z m< c< Q> sky gal m DES-like 5000 12 0.8 0.004 0.0006 260 Euclid/LSST-like 20000 35 0.9 0.001 0.0005 990 SKA1-Midearly 1000 3.0 1.0 0.014 0.0012 62 SKA1-Midearly 5000 1.2 0.8 0.012 0.0011 79 SKA1-Midearly 30940 0.35 0.5 0.011 0.0011 80 SKA1-Mid 1000 6.1 1.2 0.0090 0.00095 103 SKA1-Mid 5000 2.7 1.0 0.0067 0.00082 140 SKA1-Mid 30940 0.9 0.7 0.0058 0.00076 164 SKA 1000 37 1.6 0.0031 0.00055 318 SKA 5000 23 1.4 0.0019 0.00043 523 SKA 30940 10 1.3 0.0012 0.00035 825 to a map of shear measurements across the field-of-view. Typically this is done by measuring the ellipticityofgalaxiesidentifiedinthedata,whichischangedbyshear. Thepreponderanceofopti- caldataofthequalitynecessaryforweaklensinghasledtothedevelopmentofalargenumberof differenttechniquesforperformingthisshapemeasurementprocesswhichtakeimageplanedataas theirinputs. Amongthefirstderived(andsubsequentlymostwidelyused)aremethodswhichuse weightedquadropolemomentofcombinedgalaxy-PSFimagestomeasureellipticitiesdirectlyina non-parametric way (KSB Kaiser et al. (1995); KSB+ Hoekstra et al. (1998); Re-Gaussianization Hirata&Seljak(2003)). Anotherpopularapproachistoassertthatthegalaxyimagesmaybemod- elledwithsomeanalyticbrightnessdistribution(suchasaGaussianorSersicprofile)andfindthe best fitting parameters, including ellipticity parameters, for each source (IM3SHAPE Zuntz et al. (2013),lensfitMilleretal.(2007)). Intheradioregimetheapproachwhichhasfoundmostapplicationisthatofshapelets,which reconstructsthedatausinganorthonormalsetofbasisfunctions. Howthesebasisfunctionstrans- form with shear is known, meaning the best-fitting coefficients for an image may be used to form anunbiasedestimatorfortheshearingithasundergonethroughcomparisonwithcoefficentsfrom some‘unlensed’sample. Shapeletsalsohavetheadvantageofhavingsimilarlysimpleandanalytic Fourier transformations which also remain localised, facilitating their use in directly modelling visibilitydataratherthanreconstructedimages. Asdemonstratedabove,usingsimulationswithknownellipticitydistributionsprovidesaway of probing different aspects of the weak lensing pipeline. Most notably in the optical community simulationshavebeenusedfortestingshapemeasurementtechniques. Overtheprevious10years, theShearTEstingProgramme(STEP)andGRavitationallEnsingAccuracyTesting(GREAT)(see Mandelbaum et al. (2014) and references therein for a brief history) initiatives have simulated largeopticalweaklensingdatasetsandinvitedparticipantsto(blindly)measuretheshearinthose images. This has allowed relative proficiency of different shear measurement methods and how they react to changes in data parameters, such as source size, S/N and simulated galaxy model 10