CosmoNest: How long should it run to obtain the evidence?

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Jie Ren
Posts: 28
Joined: May 13 2006
Affiliation: Theoretical Physics Division, Chern Institute of Mathematics, Nankai University, Tianjin, China

CosmoNest: How long should it run to obtain the evidence?

Post by Jie Ren » September 28 2006

In my computer, CosmoMC with CosmoNest run for a long time and it didn't stop. In an ordinary computer (512M, 3GHz), how long should the program run to obtain the Baysian evidence of the six-parameter model provided by CosmoMC? Thank you!

David Parkinson
Posts: 23
Joined: January 19 2005
Affiliation: KASI
Contact:

CosmoNest: How long should it run to obtain the evidence?

Post by David Parkinson » September 29 2006

It depends on how big your prior is and how many live points you are using. The more live points you use, the more finely the code will sample the prior space. The larger the prior width on the parameters, the longer the code takes to work through them to the likelihood peak.

In his paper Skilling states that it will take NH steps to reach the peak, where N is the number of live points and H is the information content. For the standard set up we were using, we had N=300 and we found H=12 (for our prior). So it took about 3600 steps to reach the peak (plus a few extra hundred steps to sample it). This means, assuming CAMB plus the WMAP likelihood code takes about 4 seconds, and COSMONEST can achieve about a 10% replecament efficiency, I would expect it to run in about 3600*10*4=1440000 seconds, or about a day and a half. This is approximatley how long it took us.

David Parkinson
Posts: 23
Joined: January 19 2005
Affiliation: KASI
Contact:

CosmoNest: How long should it run to obtain the evidence?

Post by David Parkinson » September 29 2006

Remember, if you are having problems with running CosmoNest, you can email [Log in to view email]

Post Reply