Fast way of computing the MBAR analytical error

[xiki-tempula]

Hi, I'm trying to optimise some order parameter and use the MBAR analytical error as a scoring function.
My current implementation involves running full MBAR calculation until convergence and get the analytical error, which is a bit slow as I need to run millions of MBAR calculations.
I know that the analytical error should converge much faster than the MBAR convergence. I wonder if I could get some advice on how to get the analytical error faster? Thanks.