CosmoCoffee

Hello,

I am running some chains on CosmoMC and I would like them to have the convergence of R-1~0.005, which I specify at MPI_Converge_Stop. However, for the cases when I am combining Planck with multiple datasets, the final chains seem to have finished running before this convergence is achieved (at ~0.02)

I am aware that getdist measures convergence differently than it is calculated during the run, but even if only use the second half of the chains with no derived parameters, the Gelman-Rubin parameter of final chains is larger than what I specified (~0.04)

Has anyone faced this issue before and have any idea how I could work around this?

Thanks a lot for any insight!

The run didn't time out on the server?

No, that was what I also suspected initially...

Have a look at the chains' .converge_stat file and see if it is consistent, and check the .inputparams actually has what you expect.

The .converge_stat file shows a value that is ~an order of magnitude higher than what is specified in MPI_Converge_Stop in .inputparams...

I might just stop the chains manually rather than relying on MPI_Converge_Stop if there is no obvious solution to this behaviour. But I really appreciate you took the time to reply Antony!