Memory Dependent Langevin Dynamics to Study Nonadiabatic Dynamics at Metal Surfaces

金属表面的化学反应是从催化到电化学和表面科学的各种工业的支柱。然而，这些反应并不像看起来那么简单。这些反应的动力学经常受到电子激发的影响，使它们成为一个有待解开的谜团。近年来，朗之万动力学已被用作解开金属表面化学反应动力学秘密的钥匙。然而，在朗之万动力学中常用的马尔可夫近似[1]被发现是缺乏的，未能捕捉到这些反应的重要物理。最近的研究表明，马尔可夫近似没有考虑记忆效应，而记忆效应在描述金属表面化学反应动力学中是至关重要的。研究目的为了揭示金属表面化学反应的隐藏动力学，我们的目标是发展完全依赖于位置和频率的Langevin动力学[4]。这将涉及开发依赖于内存的摩擦内核Langevin集成算法，并改进机器学习代理模型。这种方法将使我们能够以更真实和准确的方式研究金属表面的反应动力学。方法为实现研究目标，将使用以下方法：量子化学计算将用于绘制现实化学系统的能量景观和摩擦张量[2，5]机器学习将应用于电子结构理论的表示，以最大限度地减少量子化学产生的错误基于量子化学计算开发依赖于内存的摩擦内核创建位置和频率-相关Langevin积分算法开发的方法将被集成到最近开发的动力学Julia代码[3]并使用真实的化学系统进行测试。重要性这项研究有可能大大提高我们对金属表面化学反应动力学的理解，并为电子动力学研究人员提供更好的仿真工具完全依赖位置和频率的Langevin动力学的发展将提供这些反应动力学的更准确和完整的图像，这可以在以下领域具有重要意义：1更好地了解反应机理：它提供了原子和电子水平上的反应机理和途径的详细信息，这对于理解和优化催化过程是重要的。新催化剂的设计：它可以通过预测特定反应的活性和选择性，以及提供有关优选结合位点和与底物相互作用的信息，来帮助设计新的金属催化剂。减少实验：通过模拟反应条件，模拟可以帮助减少耗时和昂贵的实验，并可以为实验室无法获得的条件提供预测。

Chemical reactions at metal surfaces are the backbone of various industries, from catalysis to electrochemistry and surface science. Yet, these reactions are not as simple as they seem. The dynamics of these reactions are often affected by electronic excitations, making them a puzzle to unravel. In recent years, Langevin dynamics has been used as a key to unlock the secrets of chemical reaction dynamics at metal surfaces. However, the commonly used Markov approximation [1] in Langevin dynamics has been found to be lacking, failing to capture the important physics of these reactions. Recent studies have shown that the Markov approximation does not take into account the effect of memory, which is crucial in describing the dynamics of reactions at metal surfaces.Research objectivesTo uncover the hidden dynamics of chemical reactions at metal surfaces, our goal is to develop fully position- and frequency-dependent Langevin dynamics [4]. This will involve developing memory-dependent friction kernels Langevin integration algorithms, and improving machine learning surrogate models. This approach will enable us to study the dynamics of reactions at metal surfaces in a more realistic and accurate way.MethodologyTo achieve the research objectives, the following methods will be used:Quantum chemistry calculations will be used to map energy landscapes and friction tensors for realistic chemical systems [2, 5]Machine learning will be applied in representation of electronic structure theory to minimize errors that arise from quantum chemistryDevelop memory-dependent friction kernels based on quantum chemistry calculationsCreate position- and frequency-dependent Langevin integration algorithmsThe developed methods will be integrated into recently developed dynamics Julia code [3] and tested using realistic chemical systemsSignificanceThe proposed research has the potential to significantly advance our understanding of chemical reaction dynamics at metal surfaces and provide the electronic dynamics researchers a better simulation tool. The development of fully position- and frequency-dependent Langevin dynamics will provide a more accurate and complete picture of the dynamics of these reactions, which can have important implications in fields as follows:1A better insight into reaction mechanisms: It provides detailed information on the reaction mechanisms and pathways at the atomic and electronic level, which is important forunderstanding and optimizing catalytic processes. Design of new catalysts: It can help in the design of new metal catalysts by predicting the activity and selectivity for specific reactions, and by providing information on thepreferred binding sites and the interactions with the substrate.Reduced experimentation: By simulating the reaction conditions, simulations can help to reduce the need for time-consuming and expensive experimentation, and can providepredictions for conditions that are not accessible in the laboratory.