CAREER: Accurate Electrochemical Barriers Accelerated by Machine-learning

摘要(Peterson；1553365)这项研究将促进在机器学习工具的推动下对电催化的理论理解的进步。由此达成的谅解将有助于太阳能燃料装置、电池、燃料电池和电解槽等相关技术的发展--所有这些都与可再生能源和可持续技术的商业化有关。这项研究还将为美国和肯尼亚农村地区的学生提供不同水平的教育机会，并将通过主要研究人员公开可用的代码“AMP”，促进在广泛的研究社区中引入高保真、加速原子计算。近年来，电子结构理论彻底改变了多相催化剂的设计，但由于计算过渡态能垒(通常决定催化性能)的困难，它在电催化方面面临着更大的挑战。这项研究将首次系统地研究一系列催化材料和吸附反应物之间的这种电位依赖的电催化势垒，从而促进新材料和电催化过程的发现。过渡态计算将通过开发独特的、以原子为中心的机器学习工具来实现，这些工具可以极大地加速原子计算，同时与计算密集(通常非常耗时)的“从头计算”计算的精确度相匹配。PI的AMP软件将以原子为中心的机器学习模块化，并将用于加速搜索势能面以寻找局部极小值、过渡态和全局极小值。此外，机器学习提供的加速也将有助于引入与电化学环境相关的复杂现象，如溶剂效应和典型材料的大单元电池。更广泛地说，该项目将提供的信息将被PI用作教学工具，向从当地高中和布朗大学本科生和研究生到肯尼亚农村一所新科技大学(JOOUST)的学生传达反应可视化。此外，AMP的持续开发，以及新的共享应用，将在与能源和环境相关的广泛应用中加速材料发现。

Abstract (Peterson;1553365)The study will promote advances in theoretical understanding of electrocatalysis as accelerated by machine-learning tools. The resulting understanding will aid the development of related technologies such as solar-fuel devices, batteries, fuel cells and electrolyzers - all of relevance to renewable energy and the commercialization of sustainable technologies. The research will also provide educational opportunities to students at various levels in both the U.S. and in rural Kenya, and will facilitate introduction of high-fidelity, accelerated atomistic calculations across a broad research community via the principal investigator's publicly available code, "Amp".Electronic structure theory has revolutionized heterogeneous catalyst design in recent years, however it has had greater challenges in electrocatalysis due to the difficulty of calculating transition state energy barriers (which often dictate catalytic performance). This study will provide the first systematic study of such potential-dependent electrocatalytic barriers across a range of catalytic materials and adsorbed reactants, thus facilitating the discovery of new materials and electrocatalytic processes. The transition-state calculations will be enabled by the development of unique, atom-centered, machine-learning tools that dramatically accelerate atomistic calculations while matching the accuracy of computationally-intensive (and often exceedingly time-consuming) "ab initio" calculations. The PI's Amp software modularizes atom-centered machine learning, and will be used to accelerate the search of potential energy surfaces for local minima, transition states, and global minima. Moreover, the acceleration provided by machine learning will also facilitate the introduction of complicated phenomena associated with the electrochemical environment such as solvent effects and large unit cells of typical materials. More broadly, the project will provide information that will be used by the PI as teaching tools to convey reaction visualization to students ranging from local high school and Brown undergraduate and graduate students to students at a new science and technology university (JOOUST) in rural Kenya. Moreover, continued development of Amp, with new shared applications, will accelerate materials discovery across a broad range of applications related to energy and the environment.