A Large-Scale Atomic Interaction Model Based on Matter Wave Theory

Zihan Zhao, Kuanping Gong, Yongquan Jiang, Fuhua Li

Keywords: atomic interaction prediction, large-scale atomic interaction model, deep potential atomic model, wave–particle duality, matter waves

Abstract: Accurate and rapid prediction of atomic interactions constitutes a fundamental challenge in materials science. Traditional numerical methods face persistent limitations in balancing computational accuracy with efficiency. In contrast, AI-based large-scale atomic interaction models efficiently learn characteristic patterns of atomic configurations, enabling high-speed simulations while preserving accuracy. This offers a novel paradigm for molecular dynamics simulations and accelerated discovery of new materials and pharmaceuticals. To advance beyond current performance limits, this work proposes a matter wave theory-based large-scale atomic interaction model. First, we explicitly encode quantum mechanical matter wave theory into the neural network architecture, designing a quantum-inspired matter wave network as the core module. This innovation fundamentally enhances physical representation by effectively capturing atomic wave-particle duality. Subsequently, comprehensive error evaluation across multiple datasets (including Perovskite Oxides) demonstrates that our proposed Matter Wave Deep Potential Atomic model achieves root mean square errors of 0.5 meV/atom for energy and 28.7 meV/Å for force. These represent reductions of 16% and 8%, respectively, compared to state-of-the-art models including Deep Potential Atomic. Finally, as a standalone, general-purpose module, the matter wave network readily integrates with other advanced atomic interaction models. This adaptability will propel molecular dynamics simulation capabilities and expedite materials design and pharmaceutical discovery, thereby generating significant societal value.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 12111