[CosmoMC] Propose-widths

Use of Healpix, camb, CLASS, cosmomc, compilers, etc.
Post Reply
Luke Hart
Posts: 58
Joined: July 13 2015
Affiliation: University of Manchester

[CosmoMC] Propose-widths

Post by Luke Hart » September 03 2019

Dear all,

This has been discussed several times but I'm just curious about some of the width settings in CosmoMC. I have tried a much larger and a smaller propose-width. The smaller one actually led to an incredibly small posterior distribution in one of the cases, however the larger one then just wouldn't converge. Do you know why this is occurring? I've currently now gone back to a finer width settings but with a previous successful run (of the same parameters and same data types) proposal matrix as that seems to be recommended heavily.

Any other ideas would be gratefully received :)

Luke

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

Re: [CosmoMC] Propose-widths

Post by Antony Lewis » September 03 2019

Too large will never move, so better to start too small if needed as it should then move and be able to learn the correct proposal (assuming running with default settings in MPI). But usually best to use the closest existing .covmat in which case those settings are ignored.

Post Reply