Add perturbed optics module

[cwwalter]

This may belong as a GalSim issue (does one already exist?) @aaronroodman was thinking about this problem.

[cwwalter]

@aaronroodman Do you want help with this or is this something you or Josh can do as a followup to your summer student's work?

[djperrefort]

There is code in place that takes a given array of optical deviations and generates a set of Zernike coefficients. Currently, the input deviations are generated randomly using a uniform distribution. The range of this distribution extends from an arbitrary minimum to a maximum value determined using results from a single AOS simulation provided by Bo. A much better approach would be to pull deviations from a distribution describing the size of each deviation over multiple AOS simulations. @bxin do you have any data on the optical deviations taken over multiple AOS simulations?

[bxin]

Unfortunately I have not done many AOS simulation runs. I may have one or two, in addition to the one I sent over email already. Our focus has not been to accumulate large statistics, but to test the control strategies of the AOS. Meanwhile, what are you doing with the gravitational and thermal impacts on the system?

[djperrefort]

How computationally expensive are the simulations? I have a few computers here at PITT and I could run them over a few nights to generate the distribution myself.

Currently I'm using the sensitivity matrix to map a set of simulated deviations in the optical system to deviations in the Zernike wavefront. The resulting values are intended to counter the gravitational and thermal impacts that were uncorrected by the look-up table. It was my understanding that these Zernike deviations are what is required when adding a new layer to results from Josh's (@jmeyers314) atmospheric simulations. If I was mistaken, then perhaps a conversation is warranted to clear up exactly what is needed - otherwise the AOS simulation results are all thats left.

[cwwalter]

@bxin, @djperrefort is looking after the lookup table so he needs to understand the size of the input deviations needed to produce the set of the abberated Zernikes.

We should try to close the loop on this:

@djperrefort what actual values are you using now?

@bxin Do you think Daniel should run a set of the simulations himself? Or do you think given the one that you sent him we already have enough information to set the size of the residuals?

[aaronroodman]

My understanding from emails from Bo is that the exact sort of simulations that would be needed have not been performed, and I suspect doing this is not trivial. Since the purpose for GalSim is to generate ‘typical’ deviations in the pattern of aberrations, I suggest that Daniel proceed with an uncorrelated set of deviations, with guesses for the RMS size of each term.

I can make reasonable guesses for these if needed.

Aaron

On Mar 4, 2018, at 7:28 PM, Chris Walter notifications@github.com<mailto:notifications@github.com> wrote:

@bxinhttps://github.com/bxin, @djperreforthttps://github.com/djperrefort is looking after the lookup table so he needs to understand the size of the input deviations needed to produce the set of the abberated Zernikes.

We should try to close the loop on this:

@djperreforthttps://github.com/djperrefort what actual values are you using now?

@bxinhttps://github.com/bxin Do you think Daniel should run a set of the simulations himself? Or do you think given the one that you sent him we already have enough information to set the size of the residuals?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/LSSTDESC/imSim/issues/9#issuecomment-370299995, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AEEa86neGKK24pmkZ0IRjiT91NNbwPsJks5tbLDBgaJpZM4KaUJG.

---

Prof. Aaron Roodman Chair Dept. of Particle Physics & Astrophysics Kavli Institute for Particle Astrophysics & Cosmology SLAC National Accelerator Laboratory Stanford University

SLAC National Accelerator Laboratory E-mail: roodman@slac.stanford.edumailto:roodman@slac.stanford.edu 2575 Sand Hill Rd. Phone: 650-926-2705 MS 29 Menlo Park, CA 94025 URL: http://www.slac.stanford.edu/~roodman

---

[djperrefort]

@cwwalter I'm currently using peak values from the results of a single AOS simulation as the maximum deviation magnitude and then setting the minimum deviation magnitude as 20% lower. The actual value in the simulation is then found using a uniform distribution in that range.

Depending on how expensive the AOS simulations are, we have a few computers here at Pitt that I could run overnight (or potentially longer if necessary), but from what Aaron has said the simulations sound like they are fairly intensive.

As an alternative, or temporary stopgap, I can use @aaronroodman's suggestion and create a Gaussian distribution using the RMS of a single simulation. However, since I've only seen results from a single run, I'm not sure how consistent the results of each simulation are.

[cwwalter]

I think for right now you should use @aaronroodman's suggestion and make a uncorrelated set of guesses with Aaron guiding you on the values. Write this in such a way that we could later replace it with a function based on more AOS runs if necessary. This way we can push through to the end and upgrade it later if necessary.

[djperrefort]

@aaronroodman, When refering to an "uncorrelated set of deviations", I'm interpreting this as meaning a set of separate distributions for each optical degree of freedom. If this is correct, could you please provide an idea of the average size and standard deviation for each of the optical deviations? If you want, I could calculate this from the AOS run I have, but again, I'm unsure how consistent the results of each simulation are.