Page 1 of 1

MCMC: How to improve the plots?

Posted: August 04 2005
by Jun-Qing Xia
Dear all,

I use MPI to run CosmoMC and the program now stops automatically. But in the 1D marginalized plots the dotted lines distinctly separate from the solid lines. The CosmoMC Readme says there is a much better fit in a small region of parameter space.
How to improve the result to fit much better? Which parameter should be modified?

Thanks!

Jun-Qing

Re: MCMC: How to improve the plots?

Posted: August 04 2005
by Antony Lewis
Dotted lines are mean likelihoods, solid lines are marginalized probabilities. As discussed in the cosmomc paper (astro-ph/0205436) for non-Gaussian distributions these are different. i.e. your result is probably correct as the posterior distribution is in general non-Gaussian.

The distribution may be more Gaussian if you add more data - e.g. HST or BBN prior, SDSS, etc.