MDDM: A Molecular Dynamics Diffusion Model to Predict Particle Self-Assembly

2025-01-28

AI Summary

Predicting self-assembled particle configurations in molecular dynamics (MD) simulations suffers from high computational cost and insufficient physical plausibility. To address this, we propose MDDM, a diffusion-based generative model specifically designed for particle-system configuration prediction. MDDM is the first to adapt the diffusion paradigm to particle systems, incorporating physics-informed constraints—including periodic boundary conditions and translational invariance—directly into its architecture to jointly support both conditional and unconditional generation. It employs a physics-embedded point cloud generation framework trained exclusively on MD simulation data. Experiments demonstrate that MDDM significantly outperforms existing point-cloud diffusion baselines on self-assembly configuration prediction, achieving simultaneous improvements in fidelity, sampling efficiency, and physical consistency. By unifying data-driven learning with fundamental physical principles, MDDM establishes a new, efficient, and interpretable generative modeling paradigm for accelerated materials discovery.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

The discovery and study of new material systems relies on molecular simulations that often come with significant computational expense. We propose MDDM, a Molecular Dynamics Diffusion Model, which is capable of predicting a valid output conformation for a given input pair potential function. After training MDDM on a large dataset of molecular dynamics self-assembly results, the proposed model can convert uniform noise into a meaningful output particle structure corresponding to an arbitrary input potential. The model's architecture has domain-specific properties built-in, such as satisfying periodic boundaries and being invariant to translation. The model significantly outperforms the baseline point-cloud diffusion model for both unconditional and conditional generation tasks.

Problem

• Molecular Conformation Prediction
• Efficiency
• Computational Cost Reduction

Innovation

• MDDM
• Signal-to-Structure Conversion
• Periodic Boundary Conditions