map-making, self-calibration, stellar rotation

In my stars group meeting at CCA, Ruth Angus (Columbia) told us about her work to replace standard methods for determining stellar rotation with a probabilistic model, based on Gaussian Processes with a quasi-periodic kernel. This seems to work extremely well! There are some pesky outlier stars, even in simulated data, in which all methods (including Angus's best) seem to get the wrong answer for the stellar rotation; these are interesting for further investigation.

In my cosmology group meeting, I got fully taken to school. First Brian Keating (UCSD) told us about self-calibration of CMB polarimetry. It turns out that you can't self-calibrate for some of the most important science. That is, your signal might be in exactly the modes to which the self-calibration is orthogonal! That's bad. And an aspect of self-calibration that I haven't thought about before. He discussed many crazy and creative ways to do absolute calibration of polarimetry devices; none of them look good enough (or cheap enough) at this point.

Then Colin Hill (Columbia) told us about map-making things he is working on in CMB data. I got all crazy because he is only considering linear combinations of observed data to (say) produce the thermal S-Z map (from, say, Planck data). But then he pointed out (correctly) that all least-squares methods return linear combinations of the data! Oh duh! All L2-like methods return linear combinations of the data. So then we went on to think about combined L1 and L2 methods that could permit him to open up his model space without enormously over-fitting. At the end of the discussion I had a job to do: Write down the largest class of convex map-making methods I can, given what I know about L1 and L2.

In between group meetings Cameron Hummels (Caltech) talked about open-source codes he is building that take simulation outputs from cosmological hydro simulations and predict observables, especially those that relate to the inter-galactic medium and circum-galactic medium. We talked a lot about the differences between resolution effects and sub-grid physics choices, which are confusingly inter-related.

No comments:

Post a Comment