CAREER: Many-body Ab initio Potentials and Quantum Dynamics Methods for "First Principles" Simulations in Solution: Hydration, Vibrational Spectroscopy, & Proton Transfer/Trans

加州大学圣地亚哥分校的Francesco Pesani得到了化学部和高级网络基础设施部化学理论、模型和计算方法项目的职业奖的支持，该项目旨在为分子级计算机模拟开发新的理论和计算方法。这样的模拟已经成为化学中的一个强大工具，经常提供对复杂现象的基本见解，否则很难获得这些现象。然而，实现逼真和预测性模拟所需的精度仍然具有挑战性。帕萨尼和他的研究小组正在通过结合各种方法来生成包括溶液中离子在内的许多系统的非常准确的模型，以应对这一挑战。这种新的方法能够以前所未有的精度对水体系进行计算机模拟，提供从大宗离子水合和界面上的离子水合到溶液中质子转移和传输的基本分子过程的信息。新的方法将通过在用于分子模拟的开放和免费的OpenMM软件工具包中的实施向社区提供。这个“参考实现”旨在为社区提供一个完全开放的计算工具，供其他有兴趣在自己的模拟代码中实现它的研究人员使用。此外，这一初步实施是未来开发专门为高性能计算设计的独特软件元素的起点，这将使多体模拟在多核CPU和GPU架构上都具有前所未有的准确性。与拟议的研究活动同时，Pesani建立了一个创新的教育和推广计划，专注于开发一门入门课程，向一年级和二年级的本科生介绍化学中计算方法的使用，以及专门设计的指导活动，旨在通过加州大学圣地亚哥分校的暑期交流计划，促进贫困和传统代表性不足群体的学生在STEM学科中的学习。计算机模拟的真实性和预测力在很大程度上取决于描述分子相互作用和整个系统动力学的准确性。虽然从头算方法原则上可以在不采取特别简化的情况下表征物理化学过程，但相关的计算成本有效地阻止了使用这些方法来模拟现实的凝聚相系统。此外，对实际分子动力学的严格描述通常需要对核运动进行量子力学处理，这进一步增加了与从头计算计算机模拟相关的计算成本。Pesani和他的同事开发的方法试图通过将完全由高度关联的电子结构数据得出的机器学习多体势能面与基于路径积分分子动力学和质心分子动力学的新量子动力学方法相结合来克服这些限制。这些组分的有效集成突破了当前分子动力学技术的界限，为直接与相应光谱测量相结合的凝聚相系统的真实模拟提供了新的机会。虽然该方法的范围要广泛得多，但最初的应用是对溶液中的物理化学过程进行建模，特别侧重于离子水化、线性和非线性振动光谱以及质子转移/传输。新的方法将通过在开放和免费的OpenMM分子模拟软件工具包的C++“参考平台”中的实施向社区提供。具体地说，实现将由一个独立的插件组成，为社区提供这些多体潜力的完全开放的实现。这个插件将包括一套完整的单元测试，涵盖所有能量和力分量以及我们的多体势能的内部函数。还将提供一些测试用例，用于将OpenMM获得的输出能量和力与使用PI的内部实施计算的参考值进行比较。为了方便使用新的多体分子势，该插件还将提供一个Python包装器，该包装器将简化多体分子模拟的设置和运行，从而使所有模拟参数都完全在一个XML文件中定义。该插件将为其他研究人员全面实现我们的多体势在水溶液模拟中的应用提供参考，可供其他软件实现参考。此外，此参考实现将作为未来开发OpenMM工具包的独特软件元素的起点，该工具包专门为多核CPU和GPU架构上的高性能计算而设计。新模拟方法的开发和应用将涉及本科生和研究生以及博士后研究员的培训和教育，他们将在理论、物理和计算化学方面获得坚实的基础。拟议项目的跨学科性质将为学生和博士后提供机会，在分子水平上的物理化学基本定律和凝聚相体系的性质之间建立桥梁和相互联系。在不同学科之间工作的可能性将使学生和博士后为广泛的科学职业生涯做好准备。

Francesco Paesani of the University of California San Diego is supported by a CAREER award from the Chemical Theory, Models, and Computational Methods program in the Chemistry Division and the Division of Advanced Cyberinfrastructure to develop new theoretical and computational approaches for molecular-level computer simulations. Such simulations have become a powerful tool in chemistry, often providing fundamental insights into complex phenomena which are otherwise difficult to obtain. However, achieving the necessary accuracy for realistic and predictive simulations remains challenging. Paesani and his research group are meeting this challenge by combining a variety of approaches to generate very accurate models of many systems including ions in solution. The new methodology enables computer simulations of aqueous systems with unprecedented accuracy, providing information on fundamental molecular processes from ion hydration in bulk and at interfaces to proton transfer and transport in solution. The new methodology will be available to the community through its implementation in the open and free OpenMM software toolkit for molecular simulations. This "reference implementation" aims to provide the community with a completely open computational tool which may be used by other researchers interested in implementing it in their own simulation codes. In addition, this initial implementation is a starting point for future developments of unique software elements specifically designed for high-performance computing which will enable many-body simulations with unprecedented accuracy on both multicore CPU and GPU architectures.In parallel with the proposed research activities, Paesani has established an innovative education and outreach plan focusing on the development of an entry level course that introduces undergraduate students in their freshman and sophomore years to the use of computational methods in chemistry, as well as on mentoring activities specifically designed to promote study in the STEM disciplines among students from underprivileged and traditionally underrepresented groups through the development of a summer exchange program at UC-San Diego. Both the realism and the predicting power of a computer simulation strongly depend on the accuracy with which the molecular interactions and the overall system dynamics are described. Although ab initio methods can, in principle, enable the characterization of physicochemical processes without resorting to ad hoc simplifications, the associated computational cost effectively prevents the use of these methods to model realistic condensed-phase systems. Furthermore, a rigorous description of the actual molecular dynamics often requires a quantum-mechanical treatment of the nuclear motion, which further increases the computational cost associated with ab initio computer simulations. The methods developed by Paesani and coworkers seeks to overcome these limitations by combining machine-learning many-body potential energy surfaces derived entirely from highly-correlated electronic structure data with novel quantum-dynamical approaches based on path-integral molecular dynamics and centroid molecular dynamics. The efficient integration of these components pushes the boundaries of current molecular dynamics techniques and provides new opportunities for realistic simulations of condensed-phase systems in direct connection with corresponding spectroscopic measurements. Although much broader in scope, the initial application of the methodology is to modeling physicochemical processes in solution, with a specific focus on ion hydration, linear and nonlinear vibrational spectroscopy, and proton transfer/transport. The new methodology will be made available to the community through its implementation in the C++ "reference platform" of the open and free OpenMM software toolkit for molecular simulations. Specifically, implementation will consist of an independent plug-in to provide the community with a completely open implementation of these many-body potentials. This plug-in will include a complete suite of unit tests that cover all energy and force components as well as the inner functions of our many-body potentials. A number of test cases will also be made available for comparing output energies and forces obtained with OpenMM with the reference values calculated with the PI's in house implementation. To facilitate the use of the new many-body potentials, the plug-in will also offer a Python wrapper that will simplify both setting up and running many-body molecular simulations to the point where all simulation parameters will be entirely defined in an XML file. This plug-in will thus provide other researchers with a comprehensive implementation of our many-body potentials for aqueous simulations, which can be used as a reference for the implementation in other software. In addition, this reference implementation will serve as a starting point for future developments of unique software elements for the OpenMM toolkit, specifically designed for high-performance computing on both multicore CPU and GPU architectures. The development and application of the new simulation methodology will involve the training and education of undergraduate and graduate students as well as postdoctoral fellows, who will acquire a solid foundation in theoretical, physical, and computational chemistry. The interdisciplinary nature of the proposed project will provide an opportunity for students and postdocs to establish bridges and inter-connections between the fundamental laws of physical chemistry at a molecular level and the properties of condensed-phase systems. The possibility to work at the interface of different disciplines will prepare both students and postdocs for a wide range of scientific careers.