The adjust variances function is unnecessarily specific at the moment. This is now generalized to the case d=A*xi where xi is an apriori white Gaussian random field and A is a general operator.
Do we want to do some renamings? Is there perhaps a better name than do_adjust_variances?