Discussing the brighter-fatter correction Augustin Guyonnet
Pixel level correction It seems to me that : We have done many progress in our understanding, and several attempt to correct for the effect on real data have been made. In this context, there are two questions that I am asking myself : - What can be expected from the electrostatic simulation? - How to assess the correctness of the correction?
Highlight of the principal steps of the correction (ADU) Model fitting
First step : PTC and pixels spatial correlation second degree polynomial fit on the PTC second degree polynomial fit on the PTC LSST, e2v, 50ke 0.014 0.012 0.01 0.008 0.006 0.004 0.002 0 0 1 2 3 4 5 5 Y direction (pix.) 4 3 2 1 X direction (pix.) 0
Dynamical pixel boundaries and pixel effective size The evolution of the electric field within pixels due to collection of charges. 20 15 10 5 0 0 5 10 15 20 µm
Second step : relating covariances to dynamical pixel size. 2 choices : covariances measurement from flat fields + Model fitting, or covariances measurement + Electrostatic simulation. A. Rasmussen Correlations A. Rasmussen I^2 + J^2
Third step : Redistribution of charges Q 0 0,0 = Q 0,0 + X X X i,j 1 (Q 0,0 + Q X ) 2 ax i,jq i,j 2 (Q 0,0 + Q X ) 2 Q i,j
Fourth step : assessing the quality of the correction (ADU) Model fitting evaluated by looking at moments
The quality of the correction following path (1), evaluated on second moments. (pixel) / 0 for 100 ke (pixel) 2.11 2.10 2.09 2.08 2.07 2.06 2.05 2.04 x Your content (1) goes here. y (2) (3) (4) 900 nm spots (4+LC) 2.03 0 20 40 60 80 100 120 140 flux pixel max. (ke ) 0.020 0.015 0.010 0.005 0.000 Correction VS extention of the solution (900-nm spots) 0 1 2 3 4 5 extention (pix.) x y (Q 00 + Q X ) 2 PSF(x 00 + x X ) => The approximation underestimate the ef by about 4% @ IQ =1.6 pix 2% @ IQ =2 pix. The correction has a : 5% relative precision 5% positive residual
This brings me to the 2 questions I would like to discuss with you : - The assessment of the quality of the correction, and - the expectation we can have when moving to a simulation-based correction.
The error budget is dominated by the measurements of the correlations Correlations measurements at 0.0001 level (ADU) Model fitting Turns into a 5% uncertainty on bf local charge density approximation (few % bias) m = 0.027 for 1% bf bf =0.0005+/-0.0005 shear requires m < 0.003
The electrostatic simulation will help understand the physics refine our understanding of the physics (ADU) Model fitting
One exemple : what happens beyond the first order model? residuals from quadratic fit on PTCs mean sign in flatfield pairs (adu) Simulation shall help understand the physics that could lead to higher order effects
What can be expected from following path 2? (ADU) Model fitting an assumption on the relative decrease of the shifts scaled either using 1/gain or variance regarding 2nd moments, the improvement may be limited
But, maybe we have to extend our evaluation of the correction : -What is the impact of our intervention at the pixel level in an actual image analysis pipeline? - What observables are needed to assess the quality of the correction for shear purpose? Here are two examples to illustrate these questions in connection with a shear measurement pipeline : Linearity? psf deconvolution ellipticities measurement
The correction of the non linearities of the sensors are entangled with the brighter-fatter correction Decam CCDs, slopes of the brighter-fatter effect, before (dash line indicates center of the distribution) and after correction of the non-linearities (red) and compared to the model (blue). CCD 12 10 X hmes.i =0.0249 hmod.i =0.0222 8 6 4 2 0 0.005 0.010 0.015 0.020 0.025 0.030 0.035 12 Y B-F slopes (pixel / 100ke ) 10 hmes.i =0.0252 hmod.i =0.0222 8 CCD 6 4 2 0 0.005 0.010 0.015 0.020 0.025 0.030 0.035 B-F slopes (pixel / 100ke )
Pixel level correction, ellipticities measurement and deconvolution of the PSF = gmxx gm yy gm xy +2i gm xx + gm yy gm xx + gm yy - Mxx is also sensible to the redistribution of charges. - It is noisy on laboratory data. gmxy -0.070-0.075-0.080-0.085-0.090-0.095-0.100-0.105 raw corrected 20 40 60 80 100 120 flux pixel max. (ke ) Depending on the shear method, the deconvolution also uses higher order moments. I may be stating the obvious, but : I think that we should also look at them when we evaluate the correction. In practice, we want to be sure that we rightly put the charges back to the pixel where they were generated. How do we do that?
How should we evaluate the performance of the correction? = gmxx gm yy gm xx + gm yy gm xy +2i gm xx + gm yy 0.001 e 1 raw e 1 corrected -0.030 e 2 raw e 2 corrected 0.000-0.035 ellipticities -0.001-0.002-0.003 ellipticities -0.040-0.045-0.004-0.005-0.050 20 40 60 80 100 120 flux pixel max. (ke ) 20 40 60 80 100 120 140 flux pixel max. (ke )
annex : SuprimeCam, r-band, ellipticities of stars as a function of flux. -0.020 e1 e1 corr -0.025 0.035 e2 e2 corr 0.040-0.030 0.045-0.035 0.050 0.055 0 10 20 30 40 flux in pixel max (adu) 0 10 20 30 40 flux in pixel max (adu)
Annex : linearity of correlation offset for R10 0.04 0.03 LSST E2V candidate [zoom] R(0,1), Amp1-8 R(1,0), Amp1-8 R(1,1), Amp1-8 E2V 250, all BSS, all segments correlation (frac.) 0.02 0.01 y intercept 0.00 clocking voltage (V) 0 20 40 60 80 100 120 flux (ke )
annex : Fe55 vs PTC, e2v250 Your content goes here. Illustration
Annex : psf photometry PSF Photometry is defined by a least square function : sum over pixel sky-subtracted image Solving for flux estimator : d 2 d 2 = X p =0 Assuming that the faint object has an actual PSF smaller than the one of the model: I p = ˆ = (I p PSF) w p flux estimator gives the ˆ PSF An error in the PSF translates in an error on the flux the following way : = P p (PSF P p PSF2 pixel weights P p (w pi p PSF) P p (w p PSF 2 ) ˆ PSF) Overestimation of integrated flux of faint objects PACCD Meeting (2014)
Annex : the size of objects depend on their flux. Decam @ CTIO, SuprimeCam @ Subaru 0.040 0.08 SC2 - Cluster180 0.035 0.030 x/ 0 x y/ 0 y 0.06 x y / 0 (relative) 0.025 0.020 0.015 - [pixel] 0.04 0.02 0.010 0.00 0.005 0.000 0 20 40 60 80 100 120 140 160 flux pixel max. (ke ) -0.02-10 0 10 20 30 40 50 60 70 flux pixel max. [kadu]
SuprimeCam@Subaru Stars second moments before/after correction ( )/ [%] 0.03 0.02 0.01 Subaru Camera - Stars second moments xraw yraw x corr y corr gmxx corr. gmyy corr. gmxx obs. gmyy obs. 0.00-0.01 0 10 20 30 40 50 flux pixel max. [kadu]
Consequences : It biases psf deconvolution and psf photometry Observation «true» PSF brighter-fatter «true» galaxy Reconstructed ellipticity Usual parametrization of shear bias ˆ =(1+m) + c Impact of a 1% «brighter-fatter» on the LSST m 0.027 Meanwhile, the requirement is m req 0.003 J.Mayers, PACCD Meeting (2014)