Combine calculation repeats using inverse-variance weighting

[jthorton]

Currently, calculation repeats (i.e multiple runs of the solvent leg of an RBFE calculation) are combined by taking the average with the uncertainty calculated by propagating the standard deviation of each legs estimates, which does not use the free energy estimator (MBAR) error at all. Instead, would it not be better to combine the independent estimates using an uncertainty-aware method like inverse-variance weighting which becomes the average when the variances are equal. In this case the repeats which have the lowest variance will contribute more to the estimate.