exoSAMSI day three, likelihood function, wavelets

We generated a scope for the detrending working group at SAMSI, which is to concentrate on small planets on long periods around quiet G-type stars. That is, the signals that are most likely to make us famous. We put deliverables for each Friday for the next three Fridays. We began the discussion about what kinds of instrumental and astrophysical effects we would need to consider and also how to test methods that we want to consider.

In the afternoon, Baines (Davis) and a bunch of other statisticians convinced me that wavelets are a great tool for modeling non-parametric, stochastic processes. I was surprised at first but then came around when I realized that because they are always linearly related to the time-domain data, they preserve Gaussianity; in many cases of interest you can design a wavelet basis in which the noise—or some component of the noise—is independent (diagonal in the covariance matrix).

Important piece of context: We agreed that there is a deep sense in which detrending is just writing down a better likelihood function. It would please me to no end if one of the outputs of this project was a new likelihood function for the Kepler mission.

No comments:

Post a Comment