Cavity pole frequency uncertainty fix

Historically, the uncertainty budget did not include any cavity pole frequency from the MCMC, only the TDCF value uncertainty. This MR now calculates the sampled cavity pole frequency so that it includes the uncertainty from both MCMC and TDCF contributions.

The sample_response() method creates response function curves by sampling the MCMC and TDCF values for sensing and actuation function parameters. The nominal_response() method creates a response function curve assuming the mean MCMC and mean TDCF values are creating the nominal response.

Closes #59 (closed)

changed milestone to %pyDARM 0.1.1

requested review from @hsiang-yu.huang, @ethan.payne, @jeffrey-kissel, and @ling.sun

assigned to @evan-goetz

added 1 commit

• 61656f40 - Cavity pole frequency uncertainty fix

Compare with previous version

approved this merge request

approved this merge request

approved this merge request

approved this merge request

@evan-goetz together with @hsiang-yu.huang and @ethan.payne came up with the following fix:

                 f_cc^sample(t)
f_cc^sample(t) = -------------- * f_cc^sample(t=0)
                    f_cc^MAP

This accounts for sampling from the TDCF as well as the MCMC, to shift the original reference MAP value to a new central value and then include the sample over both TDCF and MCMC.

@jeffrey-kissel had the initial idea that one takes the ratio of transfer functions in order to include both MCMC and TDCF:

       1          1 + if/f_cc^sample(t=0)
--------------- * -----------------------
1 + if/f_cc^MAP    1 + if/f_cc^sample(t)

We don't think the original proposal is wrong, but we also can't see anything wrong with Jeff's proposal either.

I'll make some plots to show the this fix (implementing the former) satisfies the original request to include MCMC sampling to get an updated f_cc value.

By looking at this again, Jeff's initial idea seems to make more sense to me, because for each draw of the fcc, the effect is stacked through the transfer function ratios. But I could not find an apparent flaw in the original fix either. Maybe in the end they are similar (don't think they are equivalent). I'd be interested to see the plots!

@jeffrey-kissel @ling.sun @ethan.payne @hsiang-yu.huang Here is a simulation comparison showing the cavity pole sampling via both methods There is certainly a difference in the ratio of the two sample draws, but by eye the two distributions are the same overall.

Here is a really quick-n-dirty code to generate the figure test_f_cc_samp.py

Next simulation would be to implement this in pydarm to compare in the real thing. My hunch is that there would be something similar, but the overall distribution of response correction curves will not be distinguishable.

See also as a reference for the above discussion in the DCC: https://docs.google.com/presentation/d/1HMqNIIMmAJXPv39zMDcFnd4ukIujR1OXIUsu0uNGqCg/edit?usp=sharing

resolved all threads

approved this merge request

merged

mentioned in commit 5175ff47