Reducing the Cost of Neural Network Potential Generation for Reactive Molecular Systems

• DOI: 10.1021/acs.jctc.3c00391
• Publication Date: 2023-09-25T18:08:55Z
• Authors: Krystof Brezina; Hubert Beck; Ondrej Marsalek
• Keywords: Speedup

Although machine learning potentials have recently had a substantial impact on molecular simulations, the construction of robust training set can still become limiting factor, especially due to requirement reference ab initio simulation that covers all relevant geometries system. Recognizing this be prohibitive for certain systems, we develop method transition tube sampling mitigates computational cost and model generation. In approach, generate classical or quantum thermal around path describing conformational change chemical reaction using only sparse local normal mode expansions along select from these by an active protocol. This yields with characterize whole without need costly trajectory. The performance is evaluated different systems complexity potential energy landscape increasing single minimum double proton-transfer high barriers. Our results show leads sets give rise models applicable in integral simulations alike are par those based directly calculations while providing speedup come expect potentials.