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

Michigan State University, East Lansing MI

Investigators

Abstract

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.