Bridging the accuracy-speed divide in reactive molecular dynamics with QuantaMind MD

Abstract

Simulating chemical reactivity remains a central challenge in molecular dynamics, historically constrained by the trade-off between ab initio accuracy and classical efficiency. We present QuantaMind, a machine learning force field (MLFF) framework that attains density functional theory (DFT)–level accuracy, enabling fully reactive simulations of complex molecular systems. QuantaMind achieves exceptional numerical stability over tens-of-nanosecond trajectories, maintaining accuracy throughout long-time reactive dynamics. It reproduces spontaneous bond formation and cleavage in key chemical processes—including proton transfer, acid-base neutralization, and phosphate buffering—and captures biologically essential phenomena such as histidine titration under constant pH. As a demonstration of the framework’s applicability to enzyme catalysis, QuantaMind recapitulates the complete catalytic cycle of the PETase-catalyzed hydrolysis reaction. By uniting quantum-level accuracy with computational efficiency comparable to state-of-the-art MLFFs, QuantaMind establishes a paradigm for long-timescale, fully reactive molecular simulation, opening avenues for rational catalyst design, reactive materials discovery, and predictive modeling of biochemical function.