Iterative steady-state restarting of weighted ensemble simulations

Although the weighted ensemble (WE) algorithm provides an efficient, unbiased framework for rare-event sampling in molecular dynamics (MD), WE convergence timescales are often still limiting for very slow systems. This is because unbiased estimates of observables generally are computed from simulations which have converged to steady state. Recent work has shown that history-augmented Markov models (haMSMs) can provide estimates of steady state from transient, unconverged WE data; additionally, a new WE simulation can be initialized using structures from the initial simulation, weighted according to steady state. We demonstrate how this process improves performance, as well as a new iterative pipeline of repeated restarts based on haMSM steady-state estimates.