克服即时机器学习分子动力学模拟中的化学复杂性瓶颈
计算物理
2024-06-13 v2 材料科学
化学物理
摘要
我们开发了一个基于多极特征化方案的即时机器学习力场分子动力学模拟框架，克服了化学元素数量带来的瓶颈。针对包含多达 6 种元素的体相系统，我们证明了密度泛函理论调用的次数与化学元素数量近似无关，这与原子位置平滑重叠方案中调用次数增加的现象形成对比。
引用
@article{arxiv.2404.07961,
title = {Overcoming the chemical complexity bottleneck in on-the-fly machine learned molecular dynamics simulations},
author = {Lucas R. Timmerman and Shashikant Kumar and Phanish Suryanarayana and Andrew J. Medford},
journal= {arXiv preprint arXiv:2404.07961},
year = {2024}
}
备注
15 total pages including references and SI, 4 main figures, 1 supplemental, 6 supplemental tables. Submitted to ACS Journal of Chemical Theory and Computation. Written and reviewed by all