CosmoMC: optical depth convergence error

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Graeme Addison
Posts: 34
Joined: July 17 2014
Affiliation: Johns Hopkins University

CosmoMC: optical depth convergence error

Post by Graeme Addison » March 03 2017

I'm getting this error a few thousand steps into a CosmoMC run and am not sure what it means.

Code: Select all

Reionization_zreFromOptDepth: Did not converge to optical depth
 tau =  0.103192158035086      optical_depth =   0.116118419386404
   50.0000000000000        49.9984741210938
 (If running a chain, have you put a constraint on tau?)
 MpiStop:            5
application called MPI_Abort(MPI_COMM_WORLD, 0) - process 5
I have these lines in my .ini file:

Code: Select all

param[tau] = 0.07 0.01 0.15 0.02 0.02
prior[tau] = 0.07 0.02
In my fit I don't have any primary CMB data (no TT/TE/EE measurements), however I do have the Planck 2015 CMB lensing phi-phi power spectrum, and am allowing some freedom in tau because I noticed it made some difference to this likelihood. I see tau is fixed in batch2/lensonly.ini, so I will rerun with fixed tau to see if that helps. But what's the reason this error comes up?

Antony Lewis
Posts: 1936
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: optical depth convergence error

Post by Antony Lewis » March 04 2017

I'm not sure why you are seeing this with those parameters - usually this error means that tau was accidentally unconstrained. Or possibly some other parameter is in some crazy part of parameter space.

Graeme Addison
Posts: 34
Joined: July 17 2014
Affiliation: Johns Hopkins University

CosmoMC: optical depth convergence error

Post by Graeme Addison » March 04 2017

Hmm, in my fit the baryon density was not being well constrained. Maybe that was the problem.

Do you know why the Planck lensing likelihood depends on tau (i.e. the chi2 changes if I hold other params fixed but shift tau)? Is it because there's a bias subtraction that depends on the primary CMB fluctuations? I haven't touched any of the reionization assumptions.

Antony Lewis
Posts: 1936
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC: optical depth convergence error

Post by Antony Lewis » March 05 2017

Sounds likely.

The N1 bias and normalization of the estimator are corrected for deviations from the assumed fiducial model, so there is some dependence on all parameters. I'd have thought it should be a rather weak dependence on tau, at least within the usual posterior range.

Post Reply