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

原子模拟使人们能够在时空尺度上理解、分析和设计复杂系统，而这对于实验观测来说是不容易实现的。传统的分子动力学技术在从材料设计到生物物理系统的应用领域都取得了巨大的成功。反应分子动力学的最新发展使这些技术的应用范围大大扩展到涉及化学键活性的体系。这些反应性原子模拟的核心是一个原子相互作用模型，该模型描述了结构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，以促进开发高度分辨的原子相互作用模型及其相关的参数。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一种时间和成本高效的方式开发高保真的反应力场。因此，反应原子模拟的准确性和范围有可能显著提高。该项目的智力贡献得到了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、出版物和演示文稿来开展跨学科教育。外联工作的重点是在研究生一级招聘人数不足的少数群体，并在早期阶段将本科生纳入研究工作。因此，这项研究与NSF促进科学进步、促进国民健康、繁荣和福祉的使命是一致的。本项目专注于反应分子动力学(MD)技术力场的开发和优化的算法和软件。这些模拟的保真度在很大程度上取决于原子间势的泛函形式和参数。传统上，原子间相互作用势的开发涉及大量的领域专业知识，并且是高度时间和劳动密集型的。该项目旨在设计一个强大的新的网络基础设施，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综合验证，以实现复杂反应原子模型的快速原型和部署。特别是，为了覆盖范围及其对原子间势的影响，选择了参考数据集，使用新的程序将对量子力学筛选的要求降至最低。优化力场的过程利用了该参考集，并引入了基于直方图的优化技术来进行参数采样和选择。最后，提出了一种新的API，它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。通过自动化过程中的每个劳动密集型步骤，建议的框架旨在形成首个此类环境，用于开发高保真反应性MD模型。鉴于反应性MD模型从材料建模到生物物理模拟的适用性，拟议的网络基础设施可以帮助推动先进材料设计和药物发现方面的最先进水平。

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.