ADAPT: Lightweight, Long-Range Machine Learning Force Fields Without Graphs

2025-09-28

AI Summary

Current graph neural network (GNN)-based machine learning force fields (MLFFs) suffer from two key limitations in point-defect simulations: over-smoothing and inadequate representation of long-range interactions. To address these, we propose a lightweight, graph-free MLFF that treats atoms as tokens and directly ingests their 3D Cartesian coordinates—eliminating explicit graph construction and associated information loss. Leveraging a Transformer encoder, our method explicitly models pairwise atomic interactions across the entire system, enabling accurate capture of long-range physics without distance-based cutoffs. The resulting model achieves both high accuracy and low computational overhead: on a silicon point-defect dataset, it reduces energy and force prediction errors by approximately 33% compared to state-of-the-art GNN-based force fields, while significantly accelerating inference. This work establishes a new paradigm for efficient, high-fidelity atomistic simulation of defective materials.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: (Large) Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Point defects play a central role in driving the properties of materials. First-principles methods are widely used to compute defect energetics and structures, including at scale for high-throughput defect databases. However, these methods are computationally expensive, making machine-learning force fields (MLFFs) an attractive alternative for accelerating structural relaxations. Most existing MLFFs are based on graph neural networks (GNNs), which can suffer from oversmoothing and poor representation of long-range interactions. Both of these issues are especially of concern when modeling point defects. To address these challenges, we introduce the Accelerated Deep Atomic Potential Transformer (ADAPT), an MLFF that replaces graph representations with a direct coordinates-in-space formulation and explicitly considers all pairwise atomic interactions. Atoms are treated as tokens, with a Transformer encoder modeling their interactions. Applied to a dataset of silicon point defects, ADAPT achieves a roughly 33 percent reduction in both force and energy prediction errors relative to a state-of-the-art GNN-based model, while requiring only a fraction of the computational cost.

Problem

• Modeling long-range atomic interactions in point defects
• Addressing oversmoothing in graph neural network force fields
• Reducing computational cost of first-principles defect simulations

Innovation

• Replaces graph representations with coordinate-based formulation
• Models all pairwise atomic interactions explicitly
• Uses Transformer encoder for lightweight long-range modeling