Force-Field Optimization by End-to-End Differentiable Atomistic Simulation.

IF 5.7 1区 化学 Q2 CHEMISTRY, PHYSICAL
Abhijeet Sadashiv Gangan, Ekin Dogus Cubuk, Samuel S Schoenholz, Mathieu Bauchy, N M Anoop Krishnan
{"title":"Force-Field Optimization by End-to-End Differentiable Atomistic Simulation.","authors":"Abhijeet Sadashiv Gangan, Ekin Dogus Cubuk, Samuel S Schoenholz, Mathieu Bauchy, N M Anoop Krishnan","doi":"10.1021/acs.jctc.4c01784","DOIUrl":null,"url":null,"abstract":"<p><p>The accuracy of atomistic simulations depends on the precision of the force fields. Traditional numerical methods often struggle to optimize the empirical force-field parameters for reproducing the target properties. Recent approaches rely on training these force fields based on forces and energies from first-principle simulations. However, it is unclear whether these approaches will enable the capture of complex material responses such as vibrational or elastic properties. To this extent, we introduce a framework employing inner loop simulations and outer loop optimization that exploits automatic differentiation for both property prediction and force-field optimization by computing gradients of the simulation analytically. We demonstrate the approach by optimizing classical potentials such as Stillinger-Weber and EDIP for silicon and BKS for SiO<sub>2</sub> to reproduce properties like the elastic constants, vibrational density of states, and phonon dispersion. We also demonstrate how a machine-learned potential can be fine-tuned using automatic differentiation to reproduce any target property such as radial distribution functions. Interestingly, the resulting force field exhibits improved accuracy and generalizability to unseen temperatures compared to those fine-tuned on energies and forces. Finally, we demonstrate the extension of the approach to optimize the force fields toward multiple target properties. Altogether, differentiable simulations, through the analytical computation of their gradients, offer a powerful tool for both theoretical exploration and practical applications toward understanding physical systems and materials.</p>","PeriodicalId":45,"journal":{"name":"Journal of Chemical Theory and Computation","volume":" ","pages":""},"PeriodicalIF":5.7000,"publicationDate":"2025-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Chemical Theory and Computation","FirstCategoryId":"92","ListUrlMain":"https://doi.org/10.1021/acs.jctc.4c01784","RegionNum":1,"RegionCategory":"化学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"CHEMISTRY, PHYSICAL","Score":null,"Total":0}
引用次数: 0

Abstract

The accuracy of atomistic simulations depends on the precision of the force fields. Traditional numerical methods often struggle to optimize the empirical force-field parameters for reproducing the target properties. Recent approaches rely on training these force fields based on forces and energies from first-principle simulations. However, it is unclear whether these approaches will enable the capture of complex material responses such as vibrational or elastic properties. To this extent, we introduce a framework employing inner loop simulations and outer loop optimization that exploits automatic differentiation for both property prediction and force-field optimization by computing gradients of the simulation analytically. We demonstrate the approach by optimizing classical potentials such as Stillinger-Weber and EDIP for silicon and BKS for SiO2 to reproduce properties like the elastic constants, vibrational density of states, and phonon dispersion. We also demonstrate how a machine-learned potential can be fine-tuned using automatic differentiation to reproduce any target property such as radial distribution functions. Interestingly, the resulting force field exhibits improved accuracy and generalizability to unseen temperatures compared to those fine-tuned on energies and forces. Finally, we demonstrate the extension of the approach to optimize the force fields toward multiple target properties. Altogether, differentiable simulations, through the analytical computation of their gradients, offer a powerful tool for both theoretical exploration and practical applications toward understanding physical systems and materials.

基于端到端可微原子模拟的力场优化。
原子模拟的精度取决于力场的精度。传统的数值方法往往难以优化经验力场参数以再现目标特性。最近的方法依赖于基于第一性原理模拟的力和能量来训练这些力场。然而,目前尚不清楚这些方法是否能够捕获复杂的材料响应,如振动或弹性特性。在这种程度上,我们引入了一个采用内环模拟和外环优化的框架,该框架通过解析计算模拟的梯度来利用自动微分进行属性预测和力场优化。我们通过优化经典势(如硅的Stillinger-Weber和EDIP)和SiO2的BKS来重现弹性常数、态的振动密度和声子色散等特性,从而证明了这种方法。我们还演示了如何使用自动微分对机器学习潜力进行微调,以重现任何目标属性,如径向分布函数。有趣的是,与那些对能量和力进行微调的力场相比,由此产生的力场表现出更高的准确性和对看不见的温度的通用性。最后,我们展示了该方法的扩展,以优化多目标属性的力场。总之,可微模拟,通过对其梯度的分析计算,为理解物理系统和材料的理论探索和实际应用提供了一个强大的工具。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
Journal of Chemical Theory and Computation
Journal of Chemical Theory and Computation 化学-物理:原子、分子和化学物理
CiteScore
9.90
自引率
16.40%
发文量
568
审稿时长
1 months
期刊介绍: The Journal of Chemical Theory and Computation invites new and original contributions with the understanding that, if accepted, they will not be published elsewhere. Papers reporting new theories, methodology, and/or important applications in quantum electronic structure, molecular dynamics, and statistical mechanics are appropriate for submission to this Journal. Specific topics include advances in or applications of ab initio quantum mechanics, density functional theory, design and properties of new materials, surface science, Monte Carlo simulations, solvation models, QM/MM calculations, biomolecular structure prediction, and molecular dynamics in the broadest sense including gas-phase dynamics, ab initio dynamics, biomolecular dynamics, and protein folding. The Journal does not consider papers that are straightforward applications of known methods including DFT and molecular dynamics. The Journal favors submissions that include advances in theory or methodology with applications to compelling problems.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信