DEQuify your force field: More efficient simulations using deep equilibrium models

2025-09-10

AI Summary

Existing machine learning force fields achieve higher accuracy than classical force fields but fail to explicitly encode the temporal continuity of state evolution and the strong prior of structural similarity between adjacent configurations in molecular dynamics. This work introduces, for the first time, a continuous-time evolution prior into equivariant force field design. We propose an implicit equivariant architecture based on Deep Equilibrium Models (DEQs), enabling intermediate feature reuse and implicit gradient computation while preserving SE(3) equivariance and ensuring memory-efficient training. Our method achieves a joint 10–20% improvement in both simulation speed and accuracy on the MD17, MD22, and OC20-200k benchmarks. Moreover, it significantly reduces GPU memory consumption, facilitating training of larger systems and more complex models.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Machine learning force fields show great promise in enabling more accurate molecular dynamics simulations compared to manually derived ones. Much of the progress in recent years was driven by exploiting prior knowledge about physical systems, in particular symmetries under rotation, translation, and reflections. In this paper, we argue that there is another important piece of prior information that, thus fa,r hasn't been explored: Simulating a molecular system is necessarily continuous, and successive states are therefore extremely similar. Our contribution is to show that we can exploit this information by recasting a state-of-the-art equivariant base model as a deep equilibrium model. This allows us to recycle intermediate neural network features from previous time steps, enabling us to improve both accuracy and speed by $10%-20%$ on the MD17, MD22, and OC20 200k datasets, compared to the non-DEQ base model. The training is also much more memory efficient, allowing us to train more expressive models on larger systems.

Problem

• Improving molecular dynamics simulation accuracy and efficiency
• Exploiting continuity in molecular system simulations
• Recycling neural network features for speed and memory

Innovation

• Deep equilibrium models for molecular dynamics
• Recycling features from previous time steps
• Improving accuracy and speed simultaneously