CosmoMC (w/ latest update) issues: 1 chain not running & GSL interp.c error

Use of Healpix, camb, CLASS, cosmomc, compilers, etc.
Post Reply
Luke Hart
Posts: 40
Joined: July 13 2015
Affiliation: University of Manchester

CosmoMC (w/ latest update) issues: 1 chain not running &

Post by Luke Hart » March 28 2017

Hi Antony and everyone else,

I've been having a problem with CosmoMC with new run parameters using CosmoRec instead of Recfast. There are two runs, each with a different problem:

1) Chain not running and error: Error in Matrix_Diagonalize using GetDist


Running CosmoMC with CosmoRec in CAMB and with another parameter, I have noticed that one of my chains hasn't even started whilst the others are all stacking up. They are at the order of several thousand now and this one chain hasn't started. I don't know if this is related, but I also don't have a .converge_stat file showing in my chains/ directory either. When I try to run getdist on this chainset (even though I know its probably not wise since the converge_stat file hasn't popped up), it is giving this error

Error in Matrix_Diagonalize

Looking at previous threads it seems to be an error in reading the column size but why is this happening with my parameters? Is there something missing?

2) GSL interp.c error

In one of my other runs with another parameter, as I try to vary this parameter, after a chain length ~500, the program aborts with the error

gsl: interp.c:150: ERROR: interpolation error
Default GSL error handler invoked.


I tracked the error to the gsl library and found that it seems to be some valuable x<interp->xmin || x>interp->xmax. So it seems to tried to extrapolate at some point but I don't understand where or why. I'm not sure whether GSL even targets the CosmoMC source code, however, it is used in CosmoRec. When I ran this over CosmoRec for the widest settings on my priors, it worked. Similarly so did CAMB. So my parameter range is okay for CosmoRec and CAMB but not for CosmoMC.

These errors are quite picky and seem to be slightly debugging errors so if anyone has an idea why they're happening or how to fix that would be great!

Thanks

Antony Lewis
Posts: 1379
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: CosmoMC (w/ latest update) issues: 1 chain not running &

Post by Antony Lewis » March 28 2017

I would write out in cosmorec the parameters being passed from CosmoMC to check everything is being passed around correctly after your modifications. (you can also use action=4 to test-calculate a specific model).

Luke Hart
Posts: 40
Joined: July 13 2015
Affiliation: University of Manchester

CosmoMC (w/ latest update) issues: 1 chain not running &

Post by Luke Hart » March 28 2017

Okay, I should have thought of that! Thanks Antony :)

Luke Hart
Posts: 40
Joined: July 13 2015
Affiliation: University of Manchester

CosmoMC (w/ latest update) issues: 1 chain not running &

Post by Luke Hart » April 01 2017

Update: It's all communicated properly its become apparent that my priors were too wide and breaking interpolation within CosmoRec I believe :)

Thanks for your help Antony

Post Reply