CosmoMC-June2016: Question about how to write the prior of parameters

Use of Healpix, camb, CLASS, cosmomc, compilers, etc.
Post Reply
Ningqiang Song
Posts: 3
Joined: August 06 2016
Affiliation: Stony Brook University

CosmoMC-June2016: Question about how to write the prior of p

Post by Ningqiang Song » March 30 2017

Dear All,

When running cosmomc by adding additional parameters other than base LambdaCDM model, I constantly run into the error:

MCMC.f90: Couldn't start after 1000 tries - check starting ranges

For example, I run 10 chains with

param[mnu] = 0.06 0 5 0.2 0.03
param[w] = -1 -3.5 -0.5 1.0 0.1
param[nnu] = 3.046 0.05 10.0 2.0 0.15


and PlanckTTTEEE+lowTEB+lensing+BAO data without propose matrix it reports this error. This happens even if I add parameter nnu only. The range above is not much different from the paper of Planck Collaboration (e.g. Table 1 in 1303.5076). I noticed the webpage of cosmomc reads
The propose_width should be of the order of the conditional posterior width (i.e. the expected error in that parameter when other parameters are fixed).
and tried to adjust both the start_width and propose_width but it didn't work. Any advice on how to choose the width or how this error can be solved?

Thank you,
Ningqiang

Antony Lewis
Posts: 1394
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC-June2016: Question about how to write the prior

Post by Antony Lewis » March 30 2017

Your start_width and propose_width should both be small (2 for nnu is huge compared to the posterior; even 0.15 is large)

Ningqiang Song
Posts: 3
Joined: August 06 2016
Affiliation: Stony Brook University

CosmoMC-June2016: Question about how to write the prior of p

Post by Ningqiang Song » March 31 2017

Thank you, that solved my problem.

Ningqiang

Post Reply