Mixture of Experts Framework in Machine Learning Interatomic Potentials for Atomistic Simulations

2026-04-28

AI Summary

This work addresses the challenge of balancing accuracy and computational efficiency in large-scale molecular simulations using machine learning interatomic potentials (MLIPs). The authors propose a multi-fidelity mixture-of-experts framework based on the E(3)-equivariant Allegro architecture, which partitions the simulation domain according to chemical complexity and assigns models of varying capacity to each region. A co-training strategy enforces consistency constraints on both energy and forces across model interfaces, ensuring physical continuity. This approach represents the first application of mixture-of-experts methodology to atomistic simulations and effectively resolves mechanical mismatch between models under static partitioning. Validation on a Pt+CO catalytic system demonstrates strict energy conservation, accurate reproduction of bulk mechanical properties—including the equation of state and bulk modulus—and prediction accuracy comparable to full high-fidelity simulations, with over a two-fold acceleration in computational speed.

Abstract

First-principles atomistic simulations are essential for understanding complex material phenomena but are fundamentally limited by their computational cost. While Machine Learning Interatomic Potentials (MLIPs) have drastically improved cost for a given accuracy, their inference cost remains a bottleneck for massive systems or long timescales. To address this, we introduce a multifidelity "Mixture-of-Experts" framework based on the E(3)-equivariant Allegro architecture. Our method spatially partitions the simulation domain into a chemically complex region (e.g., reactive interfaces) and a simple region (e.g., bulk lattice), assigning models of varying capacity to each. Among the challenges in such static domain decomposition, the mechanical mismatch between models at the interface is particularly critical, as it can generate artificial stress fields and instability. We address this challenge with a co-training strategy in which the loss function includes agreement constraints -- penalties on per-atom energy and force discrepancies between models evaluated on shared bulk environments -- forcing the independent models to learn a consistent physical description of the bulk material. We validate this approach on a realistic Pt+CO catalytic system, demonstrating that the co-trained models maintain exact energy conservation, align their bulk mechanical response (e.g., equation of state and bulk modulus), and achieve predictive accuracy comparable to a full high-fidelity simulation at more than twice the computational speed.

Problem

• Machine Learning Interatomic Potentials
• Mixture of Experts
• Atomistic Simulations
• Mechanical Mismatch
• Multifidelity Modeling

Innovation

• Mixture-of-Experts
• Machine Learning Interatomic Potentials
• E(3)-equivariance
• multifidelity modeling
• co-training