Yes, the minimizer is not super robust. That's why by default (and run with MPI) it runs several minimizations at once and checks for the lowest one; usually this is OK.
Of course if there are big parameter degeneracies, the minimum in the degeneracy direction may well not be well define wrt. numerical noise in the theory and likelihood calculations.
CosmoMC action=2 robustness
-
- Posts: 9
- Joined: January 21 2015
- Affiliation: Johns Hopkins University