STEP 4: simple simulations
STEP home Simulations Data Mailing List

The aim of STEP 4 is to separate the effects of varying galaxy and PSF properties to understand and remove the dependence of shear measurement accuracy on; galaxy size, galaxy profile, signal-to-noise or magnitude, PSF type, PSF ellipticity and PSF size.

In this first stage, Konrad has developed a suite of simulations that can be downloaded from You will need 57GB of disk space to download the full set. To see what is proposed for future stages, read the STEP 4 discussion wiki

Details of the first stage STEP4 simulations.

These simulations test the accuracy of shear measurement using very well-controlled simulations. The galaxies are at (actually near) well-known positions, do not overlap, and intrinsically average to exactly random orientation. In these simulations all galaxies are either perfectly exponential or de Vaucouleur. PSFs have constant-ellipticity isophotes and are the same for all galaxies in a simulation. PSF FWHM is 3.5 pixels, galaxy FWHM is 7 pixels after PSF smearing.

There are six simulations, each consisting of a number of shearsets of fits images with 3600 galaxies each:

galtype SN=15 SN=50 SN=200
exp 32 sets of 10 32 sets of 5 32 sets of 2 images
dev 32 sets of 10 32 sets of 5 32 sets of 2 images

Within each shearset the PSF anisotropy and shear are constant, and the galaxies in the images have a uniform spread in intrinsic position angle, at a discrete set of intrinsic axis ratios drawn from realistic distributions. All images in a set need to be analysed to derive a shear.

The filenames are of the form

indicating S/N (sn15, sn50, sn200)
galaxy type (exp or dev)
shearset (shear1 to shear32)
image number (im1 to im10 for SN=15, fewer for other S/N as per table above)

Each image contains 3600 galaxies and 240 stars. The stars are 10x brighter than the galaxies. The galaxies are placed on a rectangular grid of spacing 60, near x and y coordinates 90, 150, ..., 3630. The actual centroid of each galaxy is randomly offset from the pixel grid. The stars are arranged around the edge, near columns x=30 and 3690, and lines y=30 and 3690.

Each of the 32 shearsets has a different shear value and PSF anisotropy. These values differ from simulation to simulation, so eg set 10 of the exponential, SN=50 simulation has a different shear from set 10 in the devauc, SN=15 simulation. The PSF anisotropies will also be different.

In total there are 3600x32x10 ~ 1E6 galaxies at S/N 15 (33Gb), for exp and dev each, 3600x32x5 ~ 5E5 galaxies at S/N 50 (17Gb) and 3600x32x2 ~ 2E5 at S/N 200 (7Gb).

The input shear and PSF anisotropy of sets 1-8 is public, and IS the same for each simulation: these values are

set id g1 g2 e1psf e2psf
1 0.00 0.02 0.02 0.00
2 0.00 0.02 0.00 0.02
3 0.02 0.00 0.02 0.00
4 0.02 0.00 0.00 0.02
5 0.00 0.02 0.00 0.00
6 0.02 0.00 0.00 0.00
7 0.00 0.00 0.02 0.00
8 0.00 0.00 0.00 0.02

Feel free to contact Konrad with further questions: (kuijken[at]

Rules for the analysis

In contrast to previous STEPs, we encourage you to use information about the simulations to aid your analysis. This might involve the following;

  • Use your knowledge about the location of object centroids to remove any false detections from your source catalogues and separate stars from galaxies.
  • Make a PSF model for each image from the average properties of all 240 stars.
  • Correct for the PSF as if it is constant across the image.
  • Modify and improve your shear measurement pipeline until it can achieve <1% accuracy on the first 8 image sets for which the input shear is public (as listed above).

Once you can demonstrate an accuracy of <1%, i.e m<0.01 on the first 8 image sets for a particular SN, you can send your results for the full SN simulation set to Konrad in the following format

  • Filename: STEP4_SN***_$$$_##.dat, where *** is 200, 50 or 15, $$$ is exp or dev and ## is and ID for you.
  • File format: setid, g1, g1_error, g2, g2_error

A piece of code will be made available shortly to compute these values from your shear catalogues. Konrad will return STEP m and c results to you. Repeat analyses are permitted, but please limit yourself to as few as possible. In the long term Tim Schrabback has offered to create a web-tool to perform this analysis automatically. Note that the input shears for STEP4 simulations will remain blind for the foreseeable future.

Last modified 4th April.
STEP pages maintained by Catherine Heymans: heymans[at]
Questions regarding the STEP4 simulations should be directed to Konrad Kuijken (kuijken[at]