机器学习：分子动力学模拟探测原子力的新方案

分子动力学模拟是探索原子尺度物理和化学过程随时间演变的有力工具。要进行这种模拟，需要定义初始原子配置，并为每个时间步长输入原子力。虽然计算原子力场可用基于量子力学的各种方法，但这些方法在长时程上应用于大型系统，很是不便。新兴的机器学习（ML）技术为研究各种物理和化学问题提供了强大而新颖的工具。来自美国康涅狄格大学的Rampi Ramprasad等，提出了创建基于ML的原子力场的通用的4步方案，可执行高保真分子动力学模拟，扩大其模拟范围，为纳米级数纳秒系统的分子动力学模拟提供一条有效途径。他们按这个方案已为六种元素的块体（包括Al、Cu、Ti、W、Si和C）创建了ML力场，均能达到各自的化学精确度（chemical accuracy）。该方案是一般性和通用性的，有为任何材料生成ML力场的潜力，任何材料都可使用相同的统一工作流程，几乎无需人为修正。还可通过逐步添加新的训练数据来代表前所未遇的原子环境，从而可以系统地改进力场。该文近期发表于npj Computational Materials 3: 37 (2017)；doi:10.1038/s41524-017-0042-y (原文链接：https://www.nature.com/articles/s41524-017-0042-y)，点击文末阅读原文可以获取。 英文标题与摘要如下：

Tran Doan Huan, Rohit Batra, James Chapman, Sridevi Krishnan, Lihua Chen & Rampi Ramprasad

Emerging machine learning (ML)-based approaches provide powerful and novel tools to study a variety of physical andchemical problems. In this contribution, we outline a universal strategy tocreate ML-based atomistic force fields, which can be used to perform high-fidelity molecular dynamics simulations. This scheme involves (1) preparing a big reference dataset of atomic environments and forces with sufficiently low noise, e.g., using density functional theory or higher-levelmethods, (2) utilizing a generalizable class of structural fingerprints for representing atomic environments, (3) optimally selecting diverse and non-redundant training datasets from the reference data, and (4) proposing various learning approaches to predict atomic forces directly (and rapidly) from atomic configurations. From the atomistic forces, accurate potential energies can then be obtained by appropriate integration along a reaction coordinate or along a molecular dynamics trajectory. Based on this strategy, we have created model ML force fields for six elemental bulk solids, including Al,Cu, Ti, W, Si, and C, and show that all of them can reach chemical accuracy. The proposed procedure is general and universal, in that it can potentially be used to generate ML force fields for any material using the same unified workflow with little human intervention. Moreover, the force fields can be systematically improved by adding new training data progressively to represent atomic environments not encountered previously.