Collaborative Research:CDS&E:D3SC:Topology, Rare-event Simulation, and Machine Learning as Routes to Predicting Molecular Crystal Structures and Understanding Their Phase Behav

点击翻译按钮获取中文摘要

Mark Tuckerman of New York University and Jerome Delhommelle of the University of North Dakota are supported by an award from the Chemical Theory, Models and Computational Methods program in the Division of Chemistry to develop computational methods and software to study molecular crystals. Ordered arrays of molecules forming structures known as molecular crystals play an essential role in the pharmaceutical, agrochemical, electronics, and defense industries. In many instances, a given chemical compound may have more than one crystal structure, a phenomenon known as polymorphism. A crystal may also contain impurities, the most important among these being water. Such structures are referred to as crystal hydrates. The ability of these materials to function in a desired manner may depend on which structure, pure or impure, they form. If a well-engineered molecular crystal converts to another form or if it absorbs impurities over time., its performance may be seriously degraded. Such transformations can, for example, cause drugs to fail or insecticides to lose their potency. On the other hand, polymorphism and hydrate formation in molecular crystals are features that can be exploited to enhance the performance of these material. Utilizing advances in high-performance computing and artificial intelligence, the theoretical molecular sciences are currently poised to drive new directions in molecular crystal engineering. Computational approaches have the potential to highlight potential pitfalls associated with structural and compositional variability before expensive experiments are performed or large investments in manufacturing a particular material are made. With the aim of realizing this potential, Professors Tuckerman and Delhommelle propose to create new computational approaches and software components for rapidly predicting polymorphic structures in molecular crystals and understanding the transitions between structures. Broad dissemination of these tools and their incorporation into the materials design and engineering processes will affect a reduction in time between concept and realization of crystal systems with desired optimal properties and will catalyze the creation of new course materials for enhancing STEM education. The basic properties of organic molecular materials in the solid state are often strongly influenced by the details of their crystal structures and the existence of polymorphs and/or impurities such as water. Experimental determination of these structures is costly and time-consuming, which places increased importance on the role of theory and computation and the leveraging of advances in high-performance computing machine learning methods. The aim of this project is to develop a suite of new methods and software tools for the prediction of organic molecular crystal structures, including multiple polymorphs, elucidation of the mechanisms and thermodynamics of polymorphic and solid-liquid phase transitions, and the mapping of favored locations for water molecules in stoichiometric and non-stoichiometric crystal hydrates. The proposed developments bring together techniques of topological analysis, machine learning, enhanced molecular dynamics, thermodynamics, and solvation theories. The main goals of the project are (1) to create a topological theory for crystal structure generation based on solely on molecular order parameters, thus bypassing the need to parameterize an intermolecular interaction model, (2) to develop new entropy- and path-based collective variables, aided by machine learning , for studying polymorphic transitions via state-of-the-art enhanced sampling techniques, and (3) to devise new theoretical and computational techniques for mapping the locations of water molecules in non-stoichiometric crystal hydrates. Broad dissemination of these tools and methods and their incorporation into crystal engineering pipelines could indicate fruitful directions in materials design, thus effecting a reduction in time between concept and realization of systems with desired properties and lead to the creation of new learning modules for graduate level courses in topics such as statistical mechanics, science of materials, and machine learning in the molecular sciences.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.