run on new PSF simulations

[esheldon]

We have created some PSF simulations to test PIFF and PSFEx. I can give more details as needed, but in brief these are using galsim with

• realistic star flux and S/N distributions based on DES, with a cutoff at low flux and high flux
• DES optical paramaters, with spatial variation
• wcs based on DES
• Kolmogorov atmosphere
• randomly located stars, and thus blending

Still TODO

• add tree rings and edge distortions
• galaxies (for testing s/g separation and blending further, and ultimately shear recovery)
• full field of view simulation, with common optical and atmospheric models for all CCD images

Logical steps in processing

• run SExtractor or other reduction - volunteer needed
• run PSFEx - volunteer needed
• run PIFF - volunteer needed

The first step in analysis is running SExtractor or another reduction such as LSST DM stack. This is important since the stars are intentionally placed randomly and thus blended.

I think the next logical step is running PSFEx. This is an interesting test in itself since PSFEx is widely used, including in DES. We will learn something about PSFEx from this, and probably iron out some bugs in the sims.

Last is running PIFF. I don't know the current status, whether this is feasible at a large scale.

[rmjarvis]

I've finally managed to get a reasonable result from Piff on the same simulation. It took a while to track down some subtle errors in the I/O and draw routine when drawing from a file that's been written to disk. It's not quite as good as PSFEx yet, but at least it's in the same ball park now. (Early versions of this plot were off the top of the range for most of them.)

Here is the Piff config file I used for this. (Although all the 000000 values are changed on the command line when I run it to go through all the different images.) I'll try to few adjustments to see what parameters make things better or worse and report back.

[rmjarvis]

I iterated on this for a while and here is the best result I found: This uses the following piff configuration:

input:
    image_dir: '/gpfs/mnt/gpfs01/astro/workarea/esheldon/lensing/des-lensing/psfsim/v009/output'
    images: 'psfsim-stars-v009-000000.fits'
    image_hdu: 0
    cat_dir: '/astro/u/mjarvis/work/sims/psfsim-stars-v009/v01'
    cats: 'psfsim-stars-v009-000000_psfcat.fits'
    cat_hdu: 2

    x_col: XWIN_IMAGE
    y_col: YWIN_IMAGE
    gain: 4.0
    noise: 469.  # Variance in ADU per pixel.
    sky: 231.    # Sky level in ADU per pixel.

    stamp_size: 31

output:
    dir: '/astro/u/mjarvis/work/sims/psfsim-stars-v009/v04'
    file_name: 'psfsim-stars-v009-000000_piff.fits'

psf:

    model:
        type: PixelGrid
        scale: 0.20
        size: 17
        start_sigma: 0.4

    interp:
        type: BasisPolynomial
        order: 3

    outliers:
        type: Chisq
        nsigma: 4
        max_remove: 3

verbose: 2

The file names and directories are updated on the command line, so those are just the first file in the list. This was a useful exercise, so thanks for making these sims, @esheldon.

I've now moved on to trying this on real data (DES Y1A1), where it seems to have some new problems for me to figure out, so I'm going to close this issue.