​催化反应能垒计算：机器学习力场

计算模拟在理解原子尺度上的多相催化作用中起着核心作用。结合实验观察，模拟可应用于发现详细的反应机制，合理化催化趋势，并指导催化材料的设计。受从头算方法计算成本的限制，密度泛函理论通常会忽略复杂的效应，如不同吸附物之间的相互作用、催化剂中毒、有限温度和复杂的表面形貌等等，这可能会严重限制计算研究与实验观察的相关性。

另一种选择是使用经验参数化的力场，其速度要快几个数量级。然而，由于存在反作用力场，它的参数需要针对每个新系统进行调整，且由于其有限的表现力，它往往与真实的势能面显著偏离。机器学习力场（ MLFFs ）提供了一种弥合这一问题的方法。

Fig. 2 Training protocol. 

来自英国剑桥大学工程实验室的 Lars L. Schaaf 等人，提出了一种机器学习力场的自动训练方法，能够准确地捕捉给定异质反应路径的能量。利用已被广泛研究的具有单个氧空位的氧化铟上将二氧化碳转化为甲醇，作者对方法进行了验证。

作者发现 ， MLFFs 为常规计算催化任务提供的不仅仅是计算成本的降低，他们还能成为深入机理催化研究的重要工具。通过对每个反应进行多次 NEB 计算，作者发现了速率限制步骤的首选最低能量途径（ MEP ），其能垒降低了 40% 。

准确描述单个反应的真实 MEP 会显著影响计算研究与实验的相关性。作者预计，主动学习 方法 将有助于在更广泛的催化中采用 MLFF ，从而能够对催化循环进行更全面的机制探索。 该文近期发布 于 npj Computational Materials 9: 180 (2023).

Computational modeling plays a central role in understanding heterogeneous catalysis at an atomic scale. By complementing experimental observations, simulations are used to discover detailed reaction mechanisms, rationalize catalytic trends, and guide the design of catalytic materials. Limited by the computational cost of ab-initio methods, Density Functional Theory (DFT) usually neglects more complex effects, such as interactions between different adsorbates, poisoning, finite temperature, and complex surface morphology, which can severely limit the relevance of computational studies to experimental observations. An alternative is to use empirically parameterized force fields, which are orders of magnitude faster. 

However, while there exist reactive force fields, their parameters need to be adjusted for every novel system and often deviate significantly from the true PES due to their limited expressivity. Machine learning force fields (MLFFs) offer a way to bridge this gap.

Here, Lars L. Schaaf et al. from the Engineering Laboratory, University of Cambridge, introduced an automatic training protocol for machine learning force fields capable of accurately capturing the energetics of a given heterogeneous reaction path. The authors validated the approach on the extensively explored conversion of CO ₂ to methanol on indium oxide with a single oxygen vacancy. The authors showed that MLFFs offer more than just computational cost reduction for routine computational catalysis tasks; they emerge as an essential tool for in-depth mechanistic catalytic investigations. By running multiple nudged elastic band calculations for each reaction, they found a preferred minimum energy path (MEP) for the rate-limiting step, with a 40% lower energy barrier. Accurately describing the true MEP of individual reactions significantly influences the relevance of computational studies to experiment. They anticipated that active learning protocols will facilitate the adoption of MLFFs in the wider catalysis community, enabling more comprehensive mechanistic explorations of catalytic cycles. This article was recently published in npj Computational Materials 9: 180 (2023) .

原文Abstract及其翻译

Accurate energy barriers for catalytic reaction pathways: An automatic training protocol for machine learning force fields ( 催化反应路径的精确能垒：一种机器学习力场的自动训练方法 )

Lars L. Schaaf ,  Edvin Fako ,  Sandip De ,  Ansgar Schäfer  &  Gábor Csányi  

Abstract  We introduce a training protocol for developing machine learning force fields (MLFFs), capable of accurately determining energy barriers in catalytic reaction pathways. The protocol is validated on the extensively explored hydrogenation of carbon dioxide to methanol over indium oxide. With the help of active learning, the final force field obtains energy barriers within 0.05 eV of Density Functional Theory. Thanks to the computational speedup, not only do we reduce the cost of routine in-silico catalytic tasks, but also find an alternative path for the previously established rate-limiting step, with a 40% reduction in activation energy. Furthermore, we illustrate the importance of finite temperature effects and compute free energy barriers. The transferability of the protocol is demonstrated on the experimentally relevant, yet unexplored, top-layer reduced indium oxide surface. The ability of MLFFs to enhance our understanding of extensively studied catalysts underscores the need for fast and accurate alternatives to direct ab-initio simulations.

摘要 我们引入了一种开发机器学习力场（ MLFFs ）的训练方法，能够准确确定催化反应途径中的能垒。该方法在已被广泛研究的氧化铟上二氧化碳氢化制甲醇中得到了验证。在主动学习的帮助下，最终力场得到的能垒在密度泛函理论的 0.05eV 内。得益于更快的计算速度，我们不仅降低了常规硅催化任务的成本，而且为之前建立的限速步骤找到了一条替代途径，使活化能降低了 40% 。此外，我们还证明了有限温度效应的重要性，并计算了自由能垒。该方法的可移植性在实验相关、但尚未探索的顶层还原氧化铟表面上得到了证实。机器学习力场能够增强我们对已被广泛研究的催化剂的理解，强调了对于开发更快更准的从头算模拟替代方案的需求。