I'm running MCMC for a dark energy model and got a problem. In the program, we have a mpi convergence function to end the program, if the convergence reaches the requirement, the program ends normally, if the convergence doesn't reach the requirement, then the program will be ended by the number of runs. Now I run 4 mpi chains, one chain may reach the maximum number first, and that chain got end, but now the problem
is that other chains can't continue and the program just hang over there. Does anyone get the same problem like this?
cosmomc-mpi problem
-
- Posts: 1944
- Joined: September 23 2004
- Affiliation: University of Sussex
- Contact:
Re: cosmomc-mpi problem
The maximum number of samples isn't intended to be used with MPI convergence tests: set it to a very high number and wait for the chain to converge, or turn off the convergence stopping.