Page 1 of 1

Cosmomc: fast estimation of covariance matrix

Posted: October 27 2007
by gongbo zhao
Dear all,

Is there any fast method to estimate the covariance matrix among cosmological parameters using real data by cosmomc? Cosmomc allows to leave the proposal matrix blank and estimate the proposal matrix by inverse Hessian. Is it accurate? Has anybody tried to do something by this?

Another related question, when letting cosmomc to find best fit point, seems it can only find local minima close to the number one sets as the central value in the ini file. Is there any good method to find the global best fit point quickly?

Thanks very much.

Cheers
Gongbo

Cosmomc: fast estimation of covariance matrix

Posted: November 21 2007
by Brian Powell
My understanding is that the covariance estimation routine is a bit shaky. It often doesn't work reliably. The best thing to do is to turn on MPI learning and run a few chains for a while. Then, generate a covariance matrix from these chains for use with an actual run.