A high-throughput multiobjective genetic-algorithm workflow for in situ training of reactive molecular-dynamics force fields

The reactive molecular dynamics (RMD) method has enabled large-scale simulations of chemical reactions involving multimillion atoms, but its reliability is severely limited by the quality of reactive force fields (ReaxFF). For accurate RMD simulations, we have proposed a dynamic approach, where ReaxFF is trained by directly fitting RMD trajectories against a quantum molecular dynamics (QMD) trajectory on the fly, instead of a conventional approach of fitting a quantum-mechanical database of static quantities. To do so, multiobjective genetic algorithms (MOGA) were previously implemented using file-based communications between RMD, QMD and genetic-algorithm computations. However, this file-based approach is not scalable for a highthroughput workflow involving hundreds of concurrent RMD simulations. Here, we present a scalable in situ MOGA (iMOGA) workflow that eliminates the file I/O bottleneck using interprocess communications but with minimal modification of the original parallel RMD code. For a population of 120 RMD simulations, the new iMOGA workflow has achieved a speed-up of factor 2.15 over the original MOGA workflow. Furthermore, iMOGA exhibits a weak-scaling parallel efficiency of 0.848 on 120 processors, which is much higher than 0.720 of MOGA.