Open dkirkby opened 4 years ago
Hi David,
For 2-pt correlation functions like this the current behaviour is the standard thing to do, and Tilman has recently (in another context) drawn my attention to a paper showing that it's actually the correct thing to do, since otherwise you include artificial information that's not actually valid: https://arxiv.org/abs/1204.4724 .
Cheers, Joe
And we checked this explicitly for non-Gaussian fields in LSST here: https://arxiv.org/abs/1811.11584
It looks like the (Omega_c, sigma8) and (w0, wa) FoM values are calculated from a Fisher matrix that is assumed to have constant covariance, i.e., the
tr[Cinv . grad(C, i) . Cinv . grad(C, j)]/2
term is not included.Has someone already estimated the impact of this approximation on the results? How feasible is it to use the full Fisher matrix?