Minimize bouncing on extended diffraction spikes

[Borlaff]

Hi,

I am trying to generate a very extended PSF for Roman WFI with GalSim. However, I can't find a way to minimize the "reflected" (FFT-generated) diffraction spikes in the image array. Here is an example code and the product.

`import galsim
import galsim.roman as galsim_roman

import numpy as np
from astropy.io import fits
import matplotlib.pyplot as plt

SCA = 1
bandpass = "W146"
gsparams = galsim.GSParams(maximum_fft_size=30000, folding_threshold=1.e-20)

psf = galsim_roman.getPSF(SCA, bandpass, SCA_pos=None, pupil_bin=1, wcs=None,
n_waves=None, extra_aberrations=None, wavelength=1460.,
gsparams=gsparams, logger=None, high_accuracy=None, approximate_struts=None)

img = psf.drawImage()
img.write('test_psf.fits')

psf_fits = fits.open("test_psf.fits")
plt.imshow(np.log10(psf_fits[0].data), vmin=-13, vmax=-9)
plt.colorbar()`

Following the instructions here (https://galsim-developers.github.io/GalSim/_build/html/roman.html#galsim.roman.getPSF) I reduced the folding_threshold to 1.e-20, no difference.

Is there any possibility to minimize that effect? Thank you,
Alex

[rmjarvis]

You're talking about features that are 10^-9 smaller than the flux in the central pixel. I think this is the best we can do with the existing pupil plane images. These features would normally be hidden by any reasonable noise level in the image.

A hint about the underlying problem is given in the GalSim output when you run this script:

GalSimWarning: Input pupil plane image may not be sampled well enough!
Consider increasing sampling by a factor 29794976971454552.000000, and/or check PhaseScreenPSF outputs for signs of folding in real space.

Obviously that kind of increase in sampling is just silly, but it's because you tried to crank the folding_threshold way down (1.e-20). Once it's gone below the resolution of the pupil plane image, it can't really do any better, so lowering it beyond that just makes it complain harder about how you're likely to get folding in real space. GalSim doesn't have any ability to magically make the image higher resolution.

[Borlaff]

Dear Mike,

Thanks for the insights. I was not aware of the inner workings of the folding_threshold. I thought that this was the product of a misconfiguration on the gsparameters on my side, not a hardcore limitation. We will need to find a workaround then.

Thanks,
Alex

[Borlaff]

Dear Mike,

One further question that might help this task. By using the configuration described in the first comment, the size limit is (4098, 4098) pixels. Is there any way to make the area even larger? Ideally we would like it to span multiple detectors, keeping the same resolution, but with wider area.

Best,
Alex

[rmjarvis]

You can, but it won't look good. The pixel scale of the pupil plane image already sets the maximum extent in real space of the observed PSF. So it will still have the bounces at that point and outside that will look like crap.

If you want to see, you can use the size options for drawImage: E.g. psf.drawImage(nx=8000,ny=8000) You can also draw onto an existing image, which can be whatever size you want. Also, it would be good practice to set the pixel scale of the image somewhere, rather than just use the GalSim default (the profile's Nyquist scale). E.g. psf.drawImage(nx=4000, ny=4000, scale=0.11)