CRII: ACI: Algorithms and Tools to Facilitate the Development of High Fidelity Reactive Molecular Dynamics Models

原子模拟能够在时空尺度上理解、分析和设计复杂的系统，而这些系统不容易通过实验观察得到。传统的分子动力学技术已经在从材料设计到生物物理系统的应用领域取得了巨大的成功。反应分子动力学的最新发展已将这些技术的应用范围大大扩展到涉及化学键活性的系统。这些反应性原子模拟的核心是原子相互作用的模型，该模型描述了该领域的化学结构及其时间演化。该项目旨在建立一个网络基础设施，以促进原子相互作用的高分辨率模型的开发，以及它们相关的参数化。它将通过新颖的算法和软件工具、参考数据集以及在不同应用领域的全面验证来实现这些目标。由此产生的基础设施可以使计算科学家以时间和成本效益的方式开发高保真度的反作用力场。因此，反应原子模拟的准确性和范围有可能得到显著提高。本项目的智力贡献还得到教育和外联工作的补充，这些工作的重点是通过开发和传播教学模块、出版物和演讲，进行跨学科教育。外联工作的重点是在研究生阶段招募代表性不足的少数民族，并在早期阶段将本科生纳入研究工作。因此，这项研究与美国国家科学基金会促进科学进步和促进国家健康、繁荣和福利的使命是一致的。本项目重点研究用于开发和优化反应分子动力学（MD）技术力场的算法和软件。这些模拟的保真度主要取决于原子间势的功能形式和参数化。传统上，原子间势的开发涉及大量的领域专业知识，并且是高度耗时和劳动密集型的。该项目旨在设计一个强大的新型网络基础设施，该基础设施集成了参考数据集的提取、模型选择和优化算法以及全面验证，以实现复杂反应原子模型的快速原型化和部署。特别是，参考数据集的选择是为了覆盖范围以及它们对原子间势的影响，使用新的程序来最大限度地减少对量子力学筛选的要求。优化力场的过程利用了这个参考集，并引入了一种基于直方图的参数采样和选择优化技术。最后，提出了一种新的API，可以将合适的力场从高级数学描述转换为高效的并行软件。通过自动化每个劳动密集型步骤，所提出的框架旨在为高保真反应MD模型的开发形成一个首创的环境。考虑到反应MD模型从材料建模到生物物理模拟的适用性，所提出的网络基础设施可以帮助推进先进材料设计和药物发现的最新技术。

Atomistic simulations enable understanding, analysis, and design of complex systems at spatio-temporal scales not easily accessible to experimental observation. Conventional molecular dynamics techniques have been applied with great success in application domains ranging from materials design to biophysical systems. Recent developments in reactive molecular dynamics have significantly extended the application scope of these techniques to systems that involve chemical bond activity. The core of these reactive atomistic simulations is a model for atomic interactions that describes the chemical structure of the domain, as well as its time evolution. This project aims to build a cyberinfrastructure to facilitate the development of highly resolved models of atomic interactions, along with their associated parameterizations. It will accomplish these goals through novel algorithms and software tools, reference datasets, and comprehensive validation in diverse application domains. The resulting infrastructure can enable computational scientists to develop high fidelity reactive force fields in a time and cost effective way. As a result, the accuracy and scope of reactive atomistic simulations have the potential to be increased significantly. The intellectual contributions of the project are complemented by education and outreach efforts that focus on interdisciplinary education through development and dissemination of instructional modules, publications, and presentations. Outreach efforts focus on recruitment of underrepresented minorities at the graduate level, and integration of undergraduate students into research efforts at an early stage. Therefore, this research aligns with the NSF mission to promote the progress of science and to advance the national health, prosperity and welfare.This project focuses on algorithms and software for the development and optimization of force fields for reactive molecular dynamics (MD) techniques. The fidelity of these simulations is critically dependent on the functional forms and parameterizations of the inter-atomic potentials. Traditionally, the development of inter-atomic potentials has involved significant domain expertise, and is highly time and labor-intensive. This project aims to design a powerful new cyberinfrastructure that integrates extraction of reference datasets, model selection and optimization algorithms, and comprehensive validation, to enable rapid prototyping and deployment of complex reactive atomistic models. In particular, reference datasets are selected for coverage as well as their impact on the inter-atomic potential, using novel procedures to minimize the requirements on quantum mechanical screenings. The process of optimizing the force field leverages this reference set and introduces a histogram based optimization technique for parameter sampling and selection. Finally, a novel API is proposed to enable the translation of suitable force fields from high-level mathematical descriptions into efficient parallel software. By automating each labor-intensive step along the way, the proposed framework aims to form a first-of-its-kind environment for development of high fidelity reactive MD models. Given the applicability of reactive MD models from materials modeling to biophysical simulations, the proposed cyberinfrastructure can help advance the state-of-the-art in advance materials design and drug discovery.