Abstract: In this talk, I will introduce some machine learning based materials simulation methods developed or co-developed in my group, including the machine learning and graph theory assisted crystal structure prediction method (MAGUS) [1], machine learning molecular dynamics engine GPUMD [2], and the high order tensor massage passing interatomic potential (HotPP) [3]. In addition, I will show some of our recent progress in the applications of these methods to predict new compounds of different research fields, including planetary minerals [4-8] and functional materials, including superhard, high energy density, and superconducting materials, etc.

MAGUS code registration：

https://www.wjx.top/vm/m5eWS0X.aspx

REFERENCE

1. Junjie Wang et al., “MAGUS: machine learning and graph theory assisted universal structure searcher”, Natl. Sci. Rev., nwad128 (2023).

2. Zheyong Fan et al., “GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations”, J. Chem. Phys. 157, 114801 (2022).

3. Junjie Wang et al., “E(n)-Equivariant Cartesian Tensor Passing Potential”, arXiv:2402.15286 (2024).

4. Shuning Pan et al., “Magnesium oxide-water compounds at megabar pressure and implications on planetary interiors”, Nat. Commun. 14, 1165 (2023).

5. Hao Gao et al., “Superionic Silica-Water and Silica-Hydrogen Compounds in the Deep Interiors of Uranus and Neptune”, Phys. Rev. Lett. 128, 035702 (2022).

                                        主讲人简介：

  孙建，南京大学物理学院和固体微结构物理国家重点实验室教授，国家杰出青年科学基金获得者（2021）。研究方向为：计算凝聚态与高压物理、材料设计、行星深部物质等。发展了若干计算模拟新方法，包括晶体结构预测新方法（MAGUS）、机器学习力场（HotPP）和机器学习分子动力学软件（GPUMD）；预言了多个新材料，若干被实验证实；预言了若干系统在高温高压下的超离子态、塑晶态等新奇物态。已发表学术论文120余篇，以第一或通讯作者在重要期刊（Nat. Phys./Nat. Commun. /PRL/PRX/PNAS/JACS）发表论文20余篇。曾获2011年中国国家自然科学二等奖（第五完成人）、2013年“国家海外高层次青年人才”、2014年国际高压领域青年科学家奖（Valkenburg奖）。担任高压物理、计算材料学、高压化学等专委会委员、《物理学进展》副主编、MRE和《高压物理学报》编委。