Overfill Protection and Hyperdynamics in Adaptively Biased Simulations

Abstract

Two problems associated with adaptively biased simulations are considered: overfilling and time scale estimation. First, a simple and computationally efficient procedure for limiting the bias fill-depth of any adaptive biasing potential is introduced. The resulting bias potential floods only up to a specified level and avoids bias accumulation in higher regions of free energy. Second, hyperdynamics can be invoked in combination with this depth-limited bias potential to estimate time scales. We argue that allowing the bias to equilibrate to a set depth is one crux in efficient hyperdynamics application. In both simple (alanine dipeptide) and complex (ligand residence) test cases, the useable boost factors are 6-8 times larger when the bias fill depth is directly controlled, as opposed to controlling the update frequency. Update frequency has emerged as a proxy for controlling the fill height of adaptive biases, with infrequent metadynamics being an example. The Kolmogorov-Smirnov test is shown to be insufficient for determining the trustworthiness of a hyperdynamics simulation, and a more robust strategy is described. Overfill limiting and hyperdynamics have been added to the fABMACS project.