Lifting the Curse of Dimensionality on Enhanced Sampling of Reaction Networks with Parallel Bias Metadynamics

A common
challenge to applying metadynamics to the study of complex
systems is selecting the proper collective variables to bias. The
advent of generic collective variables, specifically social permutation
invariant (SPRINT) coordinates, has helped to address this challenge
by reducing the level of a priori knowledge required to just basic
chemical fundamentals. However, the efficiency of biasing SPRINT coordinates
can be severely handicapped by the high dimensionality of the bias
potential. Here, we circumvent this deficiency by biasing SPRINT coordinates
using the parallel bias metadynamics framework. We demonstrate the
efficacy of this method to efficiently explore a complex system, without
any prior knowledge about transition pathways, by applying it to study
the decomposition of γ-ketohydroperoxide and generating a comprehensive
reaction network of relevant pathways. The reduction in both computational
cost and chemical intuition makes this method a promising option for
studying complex reacting systems.