用于数据驱动材料设计的联合自动化综合模拟存储库(JARVIS)
材料科学
2021-07-13 v2 计算物理
摘要
联合自动化综合模拟存储库(Joint Automated Repository for Various Integrated Simulations, JARVIS)是一个利用密度泛函理论(DFT)、经典力场(FF)和机器学习(ML)技术加速材料发现与设计的一体化基础设施。JARVIS受材料基因组计划(MGI)原则的推动,该计划旨在开发开放获取数据库与工具,以降低材料发现、优化和部署的成本与开发时间。JARVIS的主要组成部分为:JARVIS-DFT、JARVIS-FF、JARVIS-ML和JARVIS-Tools。迄今为止,JARVIS-DFT包含40,000种材料和1百万个计算性质,JARVIS-FF包含1,500种材料和110个力场,JARVIS-ML包含25个用于材料性质预测的ML模型,所有这些均在持续扩展。JARVIS-Tools提供用于运行和分析各类模拟的脚本与工作流。我们在适用处将计算数据与实验或高保真计算方法进行比较,以评估预测中的误差/不确定性。除现有工作流外,该基础设施还能支持作为数据驱动材料设计范式一部分的多种其他重要技术应用。JARVIS数据集与工具公开于网站:https://jarvis.nist.gov 。
引用
@article{arxiv.2007.01831,
title = {The Joint Automated Repository for Various Integrated Simulations (JARVIS) for data-driven materials design},
author = {Kamal Choudhary and Kevin F. Garrity and Andrew C. E. Reid and Brian DeCost and Adam J. Biacchi and Angela R. Hight Walker and Zachary Trautt and Jason Hattrick-Simpers and A. Gilad Kusne and Andrea Centrone and Albert Davydov and Jie Jiang and Ruth Pachter and Gowoon Cheon and Evan Reed and Ankit Agrawal and Xiaofeng Qian and Vinit Sharma and Houlong Zhuang and Sergei V. Kalinin and Bobby G. Sumpter and Ghanshyam Pilania and Pinar Acar and Subhasish Mandal and Kristjan Haule and David Vanderbilt and Karin Rabe and Francesca Tavazza},
journal= {arXiv preprint arXiv:2007.01831},
year = {2021}
}