CosmoMC: Accuracy of post-process

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
gongbo zhao
Posts: 73
Joined: January 04 2005
Affiliation: NAOC
Contact:

CosmoMC: Accuracy of post-process

Post by gongbo zhao » August 12 2006

Dear all,
Has anyone test the accuracy of post-process? If some of the parameter space is not sampled by the original run but allowed by the latter constraint (pointed out by Hiranya Peiris), is the result still correct?
For example, are the following two results both correct?
I) Run MCMC with CMB+LSS+SN to generate *.data and then post process to give constrains of CMB+SN.
II) Run MCMC with CMB+SN to generate *.data and then post process to give constrains of CMB+SN+LSS.
I guess II) should be more accurate.
Am I right?
Thanks.

Cheers
Gong-Bo Zhao

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: Accuracy of post-process

Post by Antony Lewis » August 12 2006

Post-processing is importance sampling. So the distributions need to be similar, or the new one a subset of the old. Hence only II is likely to be good (as long as the extra constraint doesn't shrink the parameter space volume too much).

gongbo zhao
Posts: 73
Joined: January 04 2005
Affiliation: NAOC
Contact:

CosmoMC: Accuracy of post-process

Post by gongbo zhao » August 13 2006

Hi,Antony,
So you mean it is better to use the least dataset to generate *.data. Importance sampling does a better job to "add" dataset rather than "substract". Am I Right?
Thanks again.


Best,
Gong-Bo Zhao

Antony Lewis
Posts: 1941
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: Accuracy of post-process

Post by Antony Lewis » August 13 2006

Yes, though the less data you use originally the worse the importance sampling will be (less distribution overlap).

Of course if the datasets are not consistent (overlapping well) then it won't work at all. This is usually obvious by looking at the 1D plots.

Post Reply