A team led by Los Alamos National Laboratory has developed machine-learning interatomic potentials that enable time- and cost-saving simulations, a powerful tool for everything from drug design to materials discovery. Photo courtesy of LANL
LANL News Release
The ability to simulate the behavior of systems at the atomic level represents a powerful tool for everything from drug design to materials discovery. A team led by Los Alamos National Laboratory researchers has developed machine-learning interatomic potentials that predict the molecular energies and forces acting on atoms, making simulations time- and cost-saving compared to current computational methods. Enables.
“Machine learning capabilities increasingly offer an effective alternative to computationally expensive simulations that attempt to represent complex physical systems at the atomic scale,” said Benjamin Nibgen, a Los Alamos chemical physicist and coauthor of a recent Nature Chemistry paper. The author, describing the work, said: “A general reactive machine-learning interatomic capability, applicable to a wide range of reactive chemistry without the need for refitting, would greatly benefit chemistry and materials science.”
Bridging the gap in effective simulation
Efficient simulations for molecular dynamics in chemistry have traditionally been constructed with physics-based computational models, including classical force fields or quantum mechanics. While quantum mechanical models are accurate and generally applicable, they are computationally very expensive. In contrast, classical force fields are computationally efficient, but of relatively low accuracy and applicable only to limited range of systems. ANI-1xnr, the team’s transformational machine learning model, fills the speed, accuracy and generality gap that has existed in chemistry for decades. (Machine learning is an application of artificial intelligence where computer programs “learn” through training.)
ANI-1xnr represents the first reactive machine learning interatomic potential general—applicable to many different chemical systems—to perform large-scale reactive atomistic simulations with physics-based computational models. Can compete. ANI-1xnr was developed using an automated workflow that performed reactive molecular dynamics simulations on a wide range of chemical systems containing the elements carbon, hydrogen, nitrogen, and oxygen.
ANI-1xnr is capable of studying systems ranging from carbon phase transitions to combustion to prebiotic chemistry. The team validated them by comparing them with results from experiments and conventional computational techniques.
In this workflow, nanoreactor simulations automatically sample the reaction chemical space without relying on human intuition. A nanoreactor is a special class of nuclear simulation in which chemical reactions occur by colliding molecules at high speeds. Active learning uses the machine learning potential, ANI-1xnr, to simulate nanoreactor dynamics and subselect structures with high uncertainty. Case studies such as the carbon phase transition of carbon and methane combustion test the generality of the resulting model ANI-1xnr. Thanks to LANL
Interatomic potential of a change
“ANI-1xnr does not require domain specialization or refitting for each new use case, enabling scientists from different domains to study unknown chemistry,” said the Los Alamos computational scientist and co-author of the paper. Richard Messerly said. “The general application of ANI-1xnr is transformative, representing an important step toward replacing long-standing modeling techniques to study reaction chemistry at scale.”
The dataset used by the team and the ANI-1xnr code have been made publicly available to the research community.
Paper: “Exploring the Frontiers of Condensed Phase Chemistry with General Reactive Machine Learning Potential.” Nature Chemistry. DOI: 10.1038/s41557-023-01427-3
Funding: This work was supported by the DOE Office of Science, Basic Energy Sciences Chemical Sciences, Geosciences, and Biosciences Division and the Laboratory Directed Research and Development Program at Los Alamos. The work at Los Alamos was performed in part at the Center for Nonlinear Studies and the Center for Integrated Nanotechnologies, a DOE Office of Science user facility. This research utilized resources provided by the Los Alamos Institutional Computing Program.