Machine Learning Breakthrough Cuts Material Simulation Time in Half

Researchers have developed a new method to accelerate materials simulations using machine learning. Emil Annevelink and Varun Shankar from Physics Inverted Materials, Inc. introduced triplet envelope functions to enhance efficiency in machine learning interatomic potentials (MLIPs). Their approach doubles training speed, triples memory efficiency, and maintains accuracy for common simulations.

Machine learning interatomic potentials (MLIPs) have revolutionized materials science over the past decade. They provide near-density functional theory (DFT) accuracy while reducing computational costs. However, their high resource demand has limited broader use in large-scale simulations.

The team concentrated on refining the radial envelope function, drawing inspiration from empirical methods like the modified embedded atom method (MEAM). By implementing triplet envelope functions, they streamlined local atomic neighborhoods more effectively. Tests on solid and liquid materials with 5Å and 8Å cutoffs confirmed the gains: faster training, lower memory usage, and stable results.

A 72-qubit superconducting processor aided in developing higher-order envelope functions for MLIPs. This advancement not only bolstered performance but also enabled modelling open structures with larger atomic distances. The method upholds accuracy while minimizing computational overhead.

Despite these advancements, industries like nuclear energy, aerospace, and automotive manufacturing still rely on classical empirical potentials. MLIPs face obstacles in scaling to systems with billions of atoms, real-time simulations, and integration with older high-performance computing clusters.

The new triplet envelope functions pave the way for scalable and efficient materials modelling. They reduce training time, decrease memory requirements, and retain precision for common simulation setups. This progress could potentially expand MLIP adoption in industrial applications where computational limits have been a barrier.