Towards A Universally Transferable Acceleration Method for Density Functional Theory

2025-09-29

AI Summary

Existing deep learning approaches for accelerating density functional theory (DFT) rely on predicting Hamiltonian matrices to initialize self-consistent field (SCF) calculations; however, Hamiltonians are numerically unstable and exhibit poor transferability across molecular systems, basis sets, and exchange-correlation (XC) functionals. Method: We propose the first E(3)-equivariant neural network framework that directly predicts the electron density in a compact auxiliary basis—bypassing Hamiltonian or density matrix prediction—and serves as a transferable SCF initial guess. Contribution/Results: Trained solely on small molecules, our model generalizes robustly to systems with up to 60 atoms. It achieves strong cross-system, cross-basis-set, and cross-XC-functional transferability—surpassing prior Hamiltonian- or density-matrix-based paradigms. Experiments show an average reduction of 33.3% in SCF iterations, with consistent acceleration across system sizes. This significantly enhances both the efficiency and universality of DFT computations.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Humans and AI: Other Foundations of Human Computation & AI
• Data Mining & Knowledge Management: Scalability, Parallel & Distributed Systems

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Recently, sophisticated deep learning-based approaches have been developed for generating efficient initial guesses to accelerate the convergence of density functional theory (DFT) calculations. While the actual initial guesses are often density matrices (DM), quantities that can convert into density matrices also qualify as alternative forms of initial guesses. Hence, existing works mostly rely on the prediction of the Hamiltonian matrix for obtaining high-quality initial guesses. However, the Hamiltonian matrix is both numerically difficult to predict and intrinsically non-transferable, hindering the application of such models in real scenarios. In light of this, we propose a method that constructs DFT initial guesses by predicting the electron density in a compact auxiliary basis representation using E(3)-equivariant neural networks. Trained on small molecules with up to 20 atoms, our model is able to achieve an average 33.3% self-consistent field (SCF) step reduction on systems up to 60 atoms, substantially outperforming Hamiltonian-centric and DM-centric models. Critically, this acceleration remains nearly constant with increasing system sizes and exhibits strong transferring behaviors across orbital basis sets and exchange-correlation (XC) functionals. To the best of our knowledge, this work represents the first and robust candidate for a universally transferable DFT acceleration method. We are also releasing the SCFbench dataset and its accompanying code to facilitate future research in this promising direction.

Problem

• Developing transferable initial guesses for DFT convergence acceleration
• Overcoming Hamiltonian prediction limitations in deep learning DFT methods
• Creating universally applicable acceleration across system sizes and functionals

Innovation

• Predicts electron density using E(3)-equivariant neural networks
• Employs compact auxiliary basis representation for transferability
• Achieves significant SCF step reduction across system sizes