数据与计算发展前沿 ›› 2023, Vol. 5 ›› Issue (2): 37-49.

CSTR: 32002.14.jfdc.CN10-1649/TP.2023.02.003

doi: 10.11871/jfdc.issn.2096-742X.2023.02.003

• 专刊：“人工智能&大数据”科研范式变革专刊（上） • 上一篇    下一篇

基于深度学习方法研究分子/固体界面量子化质子耦合的电荷转移过程

涂又友(),郑奇靖*(),赵瑾*()   

  1. 中国科学技术大学物理系，国际功能材料量子设计中心/合肥微尺度物质科学国家研究中心，安徽 合肥 230026
  • 收稿日期:2023-02-15 出版日期:2023-04-20 发布日期:2023-04-24
  • 通讯作者: 郑奇靖,赵瑾
  • 作者简介:涂又友，2021年本科毕业于中国科学技术大学物理系，现为中国科学技术大学物理系研究生，在赵瑾教授课题组下进行研究工作。
    本文主要承担计算测试和文章整理工作。
    TU Youyou, who graduated from the Department of Physics, University of Science and Technology of China in 2021, is currently a graduate student of the Department of Physics, University of Science and Technology of China, conducting research work in Professor Jin Zhao’s research group.
    In this paper, he is mainly responsible for computational testing and article arrangement.
    E-mail: tyy2017@mail.ustc.edu.cn|郑奇靖，2009年毕业于中国科学技术大学物理系，2016年于中国科学技术大学物理系获得博士学位。2016年6月至2018年9月在中国科学技术大学微尺度物质科学国家研究中心从事博士后研究。2018年10月至2022年3月在中国科学技术大学物理系任特任副研究员，2022年3月至今转为中国科学技术大学物理系副教授岗位，主要致力于第一性原理计算研究凝聚态体系中激发态动力学的方法发展及应用。
    本文主要负责计算结果分析和文章撰写指导。
    ZHENG Qijing, who graduated from the Department of Physics, University of Science and Technology of China in 2009, and was awarded his Ph.D. from the Department of Physics, University of Science and Technology of China in 2016. From June 2016 to September 2018, he was engaged in postdoctoral research at the Department of Physics of the University of Science and Technology of China and Hefei National Laboratory for Physical Sciences at Microscale. From October 2018 to March 2022, he served as a special associate researcher in the Department of Physics, University of Science and Technology of China. Since March 2022, he has become an associate professor in the Department of Physics, University of Science and Technology of China. He has been focusing on the first principles computational method development and application of excited state dynamics.
    In this paper, he is mainly responsible for the analysis of calculation results and the guidance of article writing.
    E-mail: zqj@ustc.edu.cn|赵瑾，1998年毕业于中国科学技术大学物理系，2003年于中国科学技术大学理化科学中心获得博士学位（指导教师：侯建国院士, 杨金龙院士）。2004年3月起在美国匹兹堡大学Hrvoje Petek教授组内工作。2008年8月起成为匹兹堡大学物理系研究助理教授，2010年初成为中国科学技术大学物理系及合肥微尺度国家实验室教授。赵瑾教授的研究小组关注于利用第一性原理计算激发态动力学，发展了具有独立知识产权、自主可控的激发态动力学第一性原理计算软件Hefei-NAMD，初步构建了可以同时从时间、空间、动量、能量、自旋等多个维度研究凝聚态体系激发态动力学的理论和程序框架，率先实现了自旋分辨的real-time GW+BSE (GW+rtBSE) 激子动力学。共发表SCI文章152多篇，作为第一/理论第一/通讯作者共发表Science 3篇，Chem. Rev. 2篇，Nat. Photo. 1篇，Sci. Adv. 3篇，Nat. Commun. 1篇，Phys. Rev. Lett. 5篇，JACS 4篇等, 全部论文他引次数6700余次，H因子45。
    本文负责深度学习框架的构建和指导，以及文章修改审阅工作。
    ZHAO Jin, who graduated from the Department of Physics, University of Science and Technology of China in 1998, and was awarded her Ph.D. Since March 2004, she has been working in the group of Professor Hrvoje Petek at the Univer-sity of Pittsburgh, USA. Since August 2008, she was a research assistant professor in the Department of Physics of the University of Pittsburgh. In early 2010, she became a professor in the Department of Physics of the University of Science and Technology of China and Hefei National Laboratory for Physical Sciences at Microscale. Professor Jin Zhao’s research group focuses on calculation of excited state dynamics using first principles, and has developed an independent controllable first-principles calculation software for excited state dynamics with independent intellectual property rights named Hefei-NAMD. Her group has constructed the basic theory and pro-gram framework for studying the excited state dynamics of con-densed matter systems in multiple dimensions such as time, space, momentum, energy, and spin, and is the first group to realize spin-resolved real-time GW+BSE (GW+rtBSE) exciton dynamics. She published more than 152 SCI articles, as the first/theory first/corresponding author, published 3 Science articles, 2 Chem. Rev. articles, 1 Nat. Photo. article, 3 Sci. Adv. articles, 1 Nat. Commun. article, 5 Phys. Rev. Lett. Articles and 4 JACS articles, etc. They have been cited more than 6,700 times, with an H-index of 45.
    In this paper, she is responsible for the construction and gui-dance of the deep learning framework, as well as the revision and review of the article.
    E-mail: zhaojin@ustc.edu.cn
  • 基金资助:
    国家自然科学基金面上项目(GG2030020150);中国科学院网信专项资助(CAS-WX2021SF-0105)

Research on Quantum Proton Coupled Charge Transfer Process Based on Deep Neural Network

TU Youyou(),ZHENG Qijing*(),ZHAO Jin*()   

  1. Department of Physics and ICQD/Hefei National Research Center for Physical Sciences at the Microscale, University of Science and Technology of China, Hefei, Anhui 230026, China
  • Received:2023-02-15 Online:2023-04-20 Published:2023-04-24
  • Contact: ZHENG Qijing,ZHAO Jin

RichHTML

29

PDF

190

摘要：

【目的】理解分子与固体界面发生的超快电荷转移物理机制，对于提升太阳能电池以及光催化反应的能量转化效率起着至关重要的作用。【文献范围】在许多含有氢元素的分子/固体界面，分子会以一定结构吸附，分子中的氢会在固体表面形成氢键网络。由于氢元素具有很轻的质量，同时，它在分子中常常以质子的形式存在，因此人们常常可以在分子/固体界面观察到质子电子耦合在一起的电荷转移过程，而这类过程中质子的核量子效应（Nuclear quantum effects, NQEs）无法被忽略。最近，我们发展了结合路径积分分子动力学（Path-integral molecular dynamics, PIMD）与非绝热分子动力学（Nonadiabatic molecular dynamics, NAMD）的计算方法RP-NAMD，可以有效地研究质子电子耦合的电荷转移过程。然而，其中PIMD/RPMD的巨大计算量限制了这类方法在真实体系中的应用。【方法】在本文中，我们利用深度学习方法，大幅度减小了PIMD/RPMD的计算量，使得RP-NAMD可以应用于更大的体系，并可以用来研究更长时间尺度的动力学过程。为了验证新的方法，我们选择了甲醇/二氧化钛界面作为原型体系，并研究了核量子效应在界面质子耦合的电荷转移过程的重要作用。【结论】总之，本文基于深度学习，为大家提供了一种实用的、可以用来研究核量子效应与载流子动力学耦合效应的计算方法。

关键词: 深度学习, 路径积分动力学, 质子耦合电荷转移

Abstract:

[Objective] Understanding the physical mechanism of ultrafast charge transfer at the molecule/solid interfaces plays a crucial role in improving the energy conversion efficiency of solar cells and photocatalytic reactions. [Scope of the literature] At many molecules/solid interfaces containing hydrogen, molecules adsorption forms a certain structure, then molecular hydrogen ions can form the hydrogen bond (H-bond) network on the surface. Because of the very light mass of hydrogen, and its frequent existence in the form of proton in molecules, one can often observe the charge transfer process in which protons couple with electrons at the molecule/solid interface, and the nuclear quantum effects (NQEs) of proton cannot be ignored in these processes. Recently, we developed a computational method RP-NAMD combining path-integral molecular dynamics (PIMD) and nonadiabatic molecular dynamics (NAMD), which can effectively study the proton-charge coupled charge transfer process. However, the huge amount of computation required by PIMD/RPMD limits the application of these methods in real systems. [Methods] In this paper, we use the deep learning method to greatly reduce the computational cost of PIMD/RPMD, so that RP-NAMD can be applied to larger systems and can be used to study dynamical processes at a longer time scale. To validate this new method, we chose the CH3OH/anatase TiO2 interface as a practical system and investigated the important role of nuclear quantum effects in the interfacial proton-coupled charge transfer process. [Conclusions] In short, based on deep learning, this paper provides a practical calculation method that can be used to study the coupling between nuclear quantum effects and carrier dynamics.

Key words: deep learning, path-integral molecular dynamics, proton-coupled charge transfer