 |
CosmoCoffee
|
A CosmoMC convergence issue
|
View previous topic :: View next topic |
Author |
Message |
Erick Almaraz
Joined: 10 Jun 2012 Posts: 25 Affiliation: University of Mexico
|
Posted: June 16 2016 |
|
|
Hi everybody.
I have been running a CosmoMC analysis to find the constraints for a model of dynamical dark energy. After three weeks (the number of samples has surpassed 200 K - two chains), the R−1 values stored in the log files have been wandering around 0.01< R−1< 0.02, the plots retrieved from GetDist GUI show barely noticeable changes from time to time and still the program is running because it doesn't fulfill the R−1 < 0.01 stopping criterion. On the other hand, if I use the GetDist tools to assess the convergence (particularly, "getConvergeTests" ), I get
var(mean)/mean(var), remaining chains, worst e-value: R−1 = 0.00824
According to the "CosmoMC readme", there are differences between the GetDist and MPI run-time statistics, so I suppose that this discrepancy is normal. However, I don't know, given the amount of samples obtained and the results from GetDist, if it is safety to stop the program and take the results as reliable constraints to the model I have been working with.
Any idea? |
|
Back to top |
|
 |
Antony Lewis
Joined: 23 Sep 2004 Posts: 1332 Affiliation: University of Sussex
|
Posted: June 16 2016 |
|
|
You probably have a broad tailed distribution, and/or weak multi-modality, which can make convergence slow. R−1 =0.01 is very good convergence, so 0.02 is probably fine, though look at 2d plots to understand if the shape looks reasonable. |
|
Back to top |
|
 |
Erick Almaraz
Joined: 10 Jun 2012 Posts: 25 Affiliation: University of Mexico
|
Posted: June 16 2016 |
|
|
Thanks for the quick reply! |
|
Back to top |
|
 |
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
|