GOALI: Elements: Atomistic Tool for Open, Machine-learned Identification and Characterization (ATOMIC)

Atomistic simulations, whereby the motion of collections of atoms through space and time is simulated, are widely used in science to understand phenomena such as chemical reactions, deformation of materials, and folding of proteins. The use of atomistic simulations has seen exponential growth in the last 30-years, with nearly 20,000 published scientific works in the year 2023 alone.

A major challenge in employing atomistic simulations is efficiently analyzing the fine-scale processes occurring in the simulation. Often scientists are interested in identifying when and where specific atomic structures/configurations form. Unfortunately, to date, there is a lack of highly flexible tools for efficient identification and analysis of atomic structures in simulation data.

This greatly hinders the impact of atomistic simulations in science since discoveries are locked inside difficult-to-interpret datasets. The goal of this research project is to create open-source software, along with associated documentation and tutorials, for analyzing atomic structures which leverage modern machine-learning and algorithmic techniques.

The current paradigms for atomic structure analysis are either narrow in scope or based on simple heuristics that are difficult to generalize. The goal of this research project is to employ a data-driven, machine-learning-enabled scheme for the identification of atomic structures that are readily extensible. To accomplish this goal, four software packages are being developed to provide the core of the ATOMIC tool: (1) an analyzer tool that applies the ATOMIC classifier to user datasets designed for implementation in commonly used analysis software, (2) a machine-learning tool which trains and tests classifiers for chosen atomic structures, (3) a campaign manager which manages disparate computing resources for training, and (4) a web portal where users can request training of classifiers for their structures of interest.

This three-year project focuses on developing classifiers for crystalline point defects as a set of exemplar structures. The developed software framework can then be extended to other classes of structures in future efforts. Collaboration with the industrial partner OVITO GmbH enables native integration into the widely used OVITO analysis tool.

This award by the NSF Office of Advanced Cyberinfrastructure is jointly supported by the Division of Materials Research and the Division of Civil, Mechanical and Manufacturing Innovation.

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.