|STEP2 telecon : 25th April 2005|
|STEP home||Simulations||Data||Mailing List|
Joel Berge, Sarah Bridle, Håkon Dahle, Marco Hetterscheidt, Catherine Heymans, Henk Hoekstra, Mike Jarvis, Konrad Kuijken, Rachel Mandelbaum, Richard Massey, Reiko Nakajima, Alexandre Refregier, Jason Rhodes, Tim Schrabback, Elisabetta Semboloni, Ludovic Van Waerbeke. Apologies from Gary Bernstein. Apologies to Mischa Schirmer, whose phone didn't work with the telecon system.
Many thanks to all those involved and to Richard Ellis for funding the telecon.
A draft STEP2 paper is taking shape headed by Richard Massey. There are many new issues identified in STEP2 (see discussion below) and the analysis is genuinely useful. The rotated images pairs were particularly effective at reducing noise in the measurements. We should be able to quantify some of the issues, and highlight other areas for future investigation. To fully understand some of the issues may require special simulations targeted to specific problems (STEP4!).
Authors will be given 3 more weeks to make any final modifications to STEP2 pipelines, provided they are not simple calibration correction factors or magnitude/size selection cuts. Input shears and galaxy catalogues will then be made public on 15th May.
Catherine summarised the STEP2 results. An enormous set of plots are available.
Sarah - Why do we measure different m and c sometimes for the rotated and
Catherine - This is only true for some of the analyses. Some are understood (eg MS and S2) where there are some individual images with bad measurements which skew the linear fit to shear in vs shear out. A good example of where this isn't the case is for Reiko PSFC where m2 ~ -0.1 for the original image and m2~0.0 for the rotated image.
Sarah - Suggests making plots of m and c as a function of angle between galaxy major axis and shear direction. Catherine to investigate.
Everyone says yes.
Reiko wants to include some small modifications.
Elisabetta is making improvements and is confident she can do something in 3 weeks.
Final deadline for any additional analyses set to 15th May. Input catalogues will be made public then.
Comparing the results of PSF A (shapelet galaxies) and PSF B (exponential galaxies) tells you about the impact of galaxy morphology
Catherine - In general it looks like shapelet morphology does not effect the results
Konrad - Thats true for galaxies but not true for the more complex PSFs
Discussion - Some people found the STEP1 PSFs more complicated that the STEP2 and vice versa
Rachel - In the m,c as a function of galaxy mag and size plots, there is a difference between PSFA and B - is this morphology?
Ludo - We should look at m and c as a function of galaxy morphology.
Catherine - Encourages people to investigate this when the galaxy input catalogues are made public. The fact that galaxy morphology doesn't have a dramatic effect on the results is enough for the general paper.
Rachel (later) - One difference is that there is no variation of m2 with galaxy size in RM (or m1 in KK) results for exponent galaxies, but there is for shapelets.
Comparing the results of PSF A (0.6 arcsec seeing) and PSF C (0.8 arcsec seeing) tells you about the impact of seeing
Catherine - Most people have equally good results on PSF A and C, or are better with PSF A.
Mike - Strangely, MJ and RM do slightly better on PSF C than PSF A.
Rachel (by e-mail later) - For most methods, cuts are performed on galaxy size relative to PSF size, might be important to compare the different criteria for galaxy size selection with these two PSFs.
Henk - The relative size of PSF to pixels is the relevant quantity.
Ludo - Both 0.6 and 0.8 arcsec seeing PSFs are well sampled with these simulations so wouldn't expect to see a dramatic difference.
Alex - STEP1 simulations had the same pixel scale but worse seeing, so the PSFs are even better sampled there. Perhaps Nyquist is not enough?.
Sarah - Can we look at the error measured on different galaxies binned by size relative to PSF size? Catherine to investigate.
Generally the e1 components of shear are measured to a higher accuracy than the e2 component - is this because of pixelisation?
Catherine - generally m2 is more underestimated than m1
Richard - PSF F is perfectly circular so, for at least that simulation, it can't be caused by anything to do with PSF anisotropy.
Henk - Need simulations with 0.4 arcsec seeing to investigate pixelisation, so that the (isotropic) PSF isn't well sampled.
Tim - Tests with Gaussian profiles also find poorer accuracy with the e2 component. Both Psm and Psh underestimated; roughly cancel out.
Catherine - Why do we see this difference? Is it for the e1 we're interpolating the light intensity over 1 pixel but for the e2 we're interpolating over sqroot(2) pixels?
Konrad - We can speculate but really need different simulations.
Richard - Will High has made those simulations, with a changing pixel scale. In these, the calibration factor disappears at infinite pixel resolution, and gets worse with the sampling. Also need different calibration corrections for the e1 and e2 components.
Catherine (later) - SpaceSTEP should also be able to help in this respect.
PSF D is pure e1. PSF E is pure e2. Everyone has -e2 residuals for PSF D and e1 residuals for PSF E - Why?
Catherine - Triple checked that there hasn't been a label error or image naming error, and everything looks fine. From TS stellar catalogues the PSFD has been measured to be D(~0.1,0.02) and E(~-0.02,0.1) so it appears that it is the PSF measurement that is wrong and in correcting for the PSF e2/1 signal is added.
Reiko (by email) - Finds D(~0.15,0.05) E(~-0.05,0.15) (note different definitions of e in the BJ02 and KSB methods).
Richard - The m=2 shapelet coefficients were chosen so e2=0 in PSF D and e1=0 in PSFE.
Mike - Ellipticity could also arise from the m=6 and m=10 shapelet coefficients, depending on the definition of ellipticity.
Richard - Off-diagonal components of shear susceptibility?
Sarah - Shear measurement bias involving angle (dot product) between directions of shear and PSF ellipticity?
Marco - A significant portion of the spurious ellipticity is in the stars, rather than the galaxies, and enters the shear estimator.
Discussion - General confusion. Oh dear.
For every method bright galaxies are overestimated and faint galaxies are underestimated. Typically small galaxies are underestimated and large galaxies overestimated, although this is not always the case for the BJ02-style methods. The PSF correction is always the worst for the small galaxies.
Rachel - Can we bin as a function of magnitude and size? Catherine to investigate.
Konrad - To compare properly we need to all start from the same catalogue of sources.
Richard - It's probably too late to do this now, but should be considered for future STEP analyses.
Sarah/Rachel - The mean shear of a population is always good...
Richard - ...but that can arise either by calibration or by construction.
Catherine - Are the trends in the KK analysis to do with the (1-e^2) responsivity correction from ellipticity to shear? i.e do galaxy properties change as we go to fainter magnitudes?
Richard - The shape properties are taken from COSMOS so they are a good representation of real data. e^2 likely to change as a function of galaxy magnitude. This points to needing calibration corrections as a function of magnitude and size (a la KSB).
Discussion - Sounds reasonable. Konrad and Catherine to investgate.
Catherine - How do we best present this delicate subject in a positive light for lensing?
Alex? - The majority of galaxies are small and faint; we really only need to measure these correctly.
Alex? - Highlight that the problem is with the faint galaxies where we knew we'd have problems?
Catherine - But there are problems with the bright galaxies too. Should we re-calculate m,c including faint/bright big/small galaxy cuts?
Discussion - General disagreement with this suggestion.
Richard - How should we physically present the results? There too many plots!
Rachel - Display only a representative sample.
Sarah - Fit m(mag,size) and list fit parameters.
Sarah/Catherine - Fits will be useful for theorists to include in their parameter forecasts.
Richard - Unfortunately, the galaxies don't have assigned redshifts, so this can only be done as a function of size and magnitude. Will have to live with these.
Catherine - Fitting impractical for some results (eg MJ) but definetly possible for others. Catherine to experiment.
Any investigation into this is still fundamentally limited by shape noise (combining rotated and original images removes the shape information so you can't investigate selection effects in this way)
Sarah (earlier) - Shear bias due to angle between shear and PSF could explain difference between individual and paired analysis for MS and SP.
Catherine - Hasn't properly investigated the weights yet - will do. There are no notable selection effects as in STEP1, as proven by the observed mean ellipticity of galaxies.
Richard - Was also thoroughly investigated in STEP1. Will therefore basically omit from this paper. No objections.
Some of the problems we have seen could be due to biased measurements around an incorrect centre.
Catherine - Centroiding was seen to be very accurate in STEP1. Similar precision here. Catherine to check quickly for PSFs D and E.
Henk - All(?) of the methods perform iteration after SExtractor/hfindpeaks/etc. to satisfy some condition and thereby obtain good centroiding.
Richard - This section will also be very short in the STEP2 paper, or incorporated into other sections.
The biggest complaint about STEP1 was that there were no firm statements about the impact for WL measurements.
Catherine - That was on purpose.
Catherine - CFHTLS is the only relevant new measurement since STEP1. Elisabetta to investigate her method.
Alex - How about including and error on sigma_8 from the m and c parameters measured?
Ludo - The calibration bias as a function of mag and size means the results are dependent on the depth of the survey etc.
Alex - Could include something just from the m,c results that we have - the simulations are quite deep.
Ludo/Tim - mag/size dependent shear calibration errors are very important and so we can't say what the impact on WL is without knowing the mag distribution of a given survey.
Sarah - That's why the fits of m(size,mag) are useful.
Jason - SpaceSTEP deadline moved to 1st June.
Catherine to arrange analysis pipeline in a state where other people can do the spaceSTEP analysis as she will be on maternity leave from early June till Jan 07.
Richard to write a fairly complete draft paper by mid-May, provided that we're not expecting any significant changes in the results.
25th April 2006.
Minutes by Catherine Heymans and Richard Massey. STEP pages maintained by Catherine Heymans: heymans[at]phas.ubc.ca