Issue with n_dims when running MontePython with MultiNest

[dchooper]

Hi, When running MontePython with MultiNest, there seems to be a compiler-dependent issue with the handling of the n_dims variable when it is passed between pyMultiNest and MultiNest. The value of n_dims is correctly initialised, but becomes random when it is passed later as an argument of the functions prior() and loglike().

As a temporary fix for this, one can hard-code the value of n_dims in the functions prior() and loglike() in the nested_sampling.py file, but this needs to be done for every run. Here is an example of our hard-coded fix assuming 6 parameters (what matters here is the dimension of the free parameter space, cosmological+nuisance; derived parameters do not count): between

def prior(cube, ndim, *args):

and the loop

    for i, name in zip(range(ndim), NS_param_names):
        cube[i] = data.mcmc_parameters[name]['prior']\
            .map_from_unit_interval(cube[i])

we added the line

   ndim=6

and the same just after

def loglike(cube, ndim, *args):

An issue has been opened about this in the pyMultiNest git as well (https://github.com/JohannesBuchner/PyMultiNest/issues/72).