As noted in #6 , our first test for Bayesian sampling in parameter space will be bond parameter moves for simulations of isolated molecules. Reweighting of existing simulations will be unlikely to take us very far for these, @mrshirts remarks, because bond potentials are so stiff. However, key will be to know when reweighting stops working so that new simulations are required. Presumably to do so, we need realistic uncertainty estimates, but @mrshirts raised concerns that these uncertainty estimates may be unreliable due to the extremely small number of snapshots that may contribute significantly once the bond length is varied significantly.
@mrshirts and @bmanubay are going to test this out and report back so we can decide what action will need to be taken.
As noted in #6 , our first test for Bayesian sampling in parameter space will be bond parameter moves for simulations of isolated molecules. Reweighting of existing simulations will be unlikely to take us very far for these, @mrshirts remarks, because bond potentials are so stiff. However, key will be to know when reweighting stops working so that new simulations are required. Presumably to do so, we need realistic uncertainty estimates, but @mrshirts raised concerns that these uncertainty estimates may be unreliable due to the extremely small number of snapshots that may contribute significantly once the bond length is varied significantly.
@mrshirts and @bmanubay are going to test this out and report back so we can decide what action will need to be taken.