CosmoMC: Problems when using params_H and simulated data

Use of Healpix, camb, CLASS, cosmomc, compilers, etc.
Post Reply
Ana Vasile
Posts: 25
Joined: March 26 2006
Affiliation: Institute for Space Sciences
Contact:

CosmoMC: Problems when using params_H and simulated data

Post by Ana Vasile » March 10 2008

Hello!

Can someone please help me with the following problem:

I'm trying to use data in all_l_exact format with the non-default parametrization which uses H_0 instead of \theta and I never seem to get convergence.

I've tested my data and with the default parametrization and everythings works properly.

Antony Lewis
Posts: 1501
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: Problems when using params_H and simulated data

Post by Antony Lewis » March 11 2008

If you are not using a .covmat you need to chose your propose widths very carefully with high-quality data. The widths should be of the order of the conditional parameter posterior - typically very small for things like A_s (much smaller than the marginalized posterior width). Make sure the chains also start in a sensible place.

Best thing is to use GetDist to generate a .covmat from chains from a default run, where you've replaced old parameter values by the values of the parameters you want to switch to. Then you can use this new .covmat for your run in the other parameterization.

Or do a provisional run where you greatly increase the temperature.

Ana Vasile
Posts: 25
Joined: March 26 2006
Affiliation: Institute for Space Sciences
Contact:

Re: CosmoMC: Problems when using params_H and simulated data

Post by Ana Vasile » March 12 2008

Antony Lewis wrote:If you are not using a .covmat you need to chose your propose widths very carefully with high-quality data. The widths should be of the order of the conditional parameter posterior - typically very small for things like A_s (much smaller than the marginalized posterior width). Make sure the chains also start in a sensible place.

Best thing is to use GetDist to generate a .covmat from chains from a default run, where you've replaced old parameter values by the values of the parameters you want to switch to. Then you can use this new .covmat for your run in the other parameterization.

Or do a provisional run where you greatly increase the temperature.
Thank you for your answer Antony!

I was using the basic6_cmb.covmat. Maybe this was the problem.

Post Reply