Adaptive accelerated reactive molecular dynamics driven by parallel collective variables overcoming dimensionality explosion.

Abstract

ReaxFF reactive molecular dynamics has significantly advanced the exploration of chemical reaction mechanisms in complex systems. However, it faces several challenges: (1) the prevalent use of excessively high temperatures (>2000K), (2) a time scale considerably shorter than the experimental timeframes (nanoseconds vs seconds), and (3) the constraining impact of dimensionality growth due to collective variables on the expansiveness of research systems. To overcome these issues, we introduced Parallel Collective Variable-Driven Adaptive Accelerated Reaction Molecular Dynamics (PCVR), which integrates metadynamics with ReaxFF. This method incorporates bond distortion based on each bond type for customized Collective Variable (CV) parameterization, facilitating independent parallel acceleration. Simultaneously, the sampling was confined to fixed cutoff ranges for distinct bond distortions, effectively overcoming the challenge of the CV dimensionality explosion. This extension enhances the applicability of ReaxFF to non-strongly coupled systems with numerous reaction energy barriers and mitigates the system size limitations. Using accelerated reactive molecular dynamics, the oxidation of ester-based oil was simulated with 31 808 atoms at 500K for 64s. This achieved 61% efficiency compared to the original ReaxFF and was ∼37 times faster than previous methods. Unlike ReaxFF's high-temperature constraints, PCVR accurately reveals the pivotal role of oxygen in ester oxidation at industrial temperatures, producing polymers consistent with the sludge formation observed in ester degradation experiments. This method promises to advance reactive molecular dynamics by enabling simulations at lower temperatures, extending to second-level timescales, and accommodating systems with millions of atoms.