Simulations to test PSF models and shear recovery about piff

[esheldon]

We might want to launch a new git repository to host the config files and scripts for this sub-project.

[esheldon]

This might be a good start.

https://github.com/rmjarvis/DESWL/blob/master/sims/v005.yaml

It only outputs galaxies, so we need to add some ability to output the star field.

[esheldon]

Need advice from @mjarvis on which branch of galsim supports the new configuration files. Will master work?

[esheldon]

the config I linked doesn't seem to work. I tried on master and #774, so I think I just need to get the right branch from someone

Traceback (most recent call last):
  File "/astro/u/esheldon/envs/psfsim/bin/galsim", line 288, in <module>
    main()
  File "/astro/u/esheldon/envs/psfsim/bin/galsim", line 275, in main
    galsim.config.Process(config, logger, njobs=args.njobs, job=args.job, new_params=new_params)
  File "/astro/u/esheldon/envs/psfsim/lib/python2.7/site-packages/galsim/config/process.py", line 610, in Process
    ImportModules(config)
  File "/astro/u/esheldon/envs/psfsim/lib/python2.7/site-packages/galsim/config/process.py", line 445, in ImportModules
    exec('import '+module)
  File "<string>", line 1, in <module>
  File "./des_wcs.py", line 170, in <module>
    galsim.config.RegisterInputType('des_wcs', DES_FullFieldWCS, ['DES_LocalWCS'])
TypeError: RegisterInputType() takes exactly 2 arguments (3 given)

[rmjarvis]

That v005 config file was a strawman that isn't expected to work. In particular, there is currently no implementation of the epoch stuff yet, which will require some development (probably connected to issue #699).

However, the error you ran into is probably because you are using an old version of des_wcs.py. You should get the new one in the master branch of GalSim (which will soon be an official release as v1.4).

[esheldon]

I didn't realize this des_wcs.py was part of galsim

[esheldon]

I tried the v004.yaml, with the newer des_wcs.py, and it runs fine, thanks for the pointers.

Now for a star field

[esheldon]

I've put some config files here: https://github.com/esheldon/psfsim_config

The star field is in psfsim-stars-v001.yaml

I could use some help making this suitable for our tests; I've marked TODO some things in the file, but also

• is this producing a variable psf across the field?
• what is a good distribution function for flux?
• are are the right DES image parameters?

[rmjarvis]

is this producing a variable psf across the field?

No. The psf.index_key = file_num means that this will keep a single PSF model for each output file. And most of the prescriptions for picking parameters are not appropriate for how to construct a reasonable PSF variation across a chip.

For useful tests of PSF interpolation, we'll want the various PSF parameters be functions of (x,y) with the meta-parameters fixed in each file. e.g.

astig1: 
    type: Eval
    str: "mean + a * image_pos.x + b * image_pos.y + c * image_pos.x**2 + d * image_pos.y**2 + e * image_pos.x*image_pos.y"
    fmean: { type: RandomGaussian, sigma: 0.13, index_key: file_num }
    fa: { type: RandomGaussian, sigma: 2.e-3, index_key: file_num }
    fb: { type: RandomGaussian, sigma: 2.e-3, index_key: file_num }
    fc: { type: RandomGaussian, sigma: 1.e-6, index_key: file_num }
    fd: { type: RandomGaussian, sigma: 1.e-6, index_key: file_num }
    fe: { type: RandomGaussian, sigma: 1.e-6, index_key: file_num }

Something like this at least. I don't really know what the right sigmas are here, but I think something around this order of magnitude. And if we do this for all the optical terms, then they will couple non-linearly, so a polynomial interpolation won't be trivially correct, so it will be an interesting test.

[rmjarvis]

gain: 2.1 # e- / ADU TODO: match to DES

I think this was the SV value. But Y1 is about 4. And Y3 will be 1.

[rmjarvis]

read_noise: 3.4 # Variance in ADU^2 TODO: match to DES

Again, I think I took this from the header of the FITS file for the SV image that we used in the GalSIm example, but I suspect it is different for Y1, and in fact probably varies somewhat from image to image. We should check what the right distribution for this is.

[esheldon]

Those values were from the lsst.yaml file

[rmjarvis]

Those values were from the lsst.yaml file

Oh. OK. Here are the values for one random Y1 image:

GAINA   =                4.039 / [electrons/adu] Gain for amp A         
RDNOISEA=                6.139 / [electrons] Read noise for amp A
GAINB   =                3.781 / [electrons/adu] Gain for amp B
RDNOISEB=                5.982 / [electrons] Read noise for amp B

I don't know how stable these values are...

[esheldon]

are those sigmas fractional or absolute?

[esheldon]

Framework for running these sims is here https://github.com/esheldon/psfsim

[rmjarvis]

are those sigmas fractional or absolute?

They're the normal definition of sigma for a Gaussian random deviate. N(mu, sigma). You can give a mean too if you like; the default is 0. cf.

https://github.com/GalSim-developers/GalSim/wiki/Config-Values#float_value

[esheldon]

What I meant, is, is sigma=0.13 a 13% scatter or is that the total scatter around the mean. Since it is fmean I thought maybe that was fractional.

[rmjarvis]

The f means float. The first letter for eval variables tells the config processing what type to use. cf. https://github.com/GalSim-developers/GalSim/wiki/Config-Values#user-defined-variables

So fmean defines a float variable called mean that is used in the eval string as just mean. Likewise fa, fb, etc. show up in the eval string as a, b, etc.

[esheldon]

I think we can close this issue, since the effort has begun. Let's continue over at https://github.com/esheldon/psfsim_config