![[2509.19877] Advancing Universal Deep Learning for Electronic-Structure Hamiltonian Prediction of Materials](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2509.19877] Advancing Universal Deep Learning for Electronic-Structure Hamiltonian Prediction of Materials
Summary
This article presents advancements in deep learning techniques for predicting electronic-structure Hamiltonians in materials, addressing challenges in generalization and computational efficiency.
Why It Matters
The research highlights significant improvements in the predictive capabilities of deep learning models for electronic structures, which can lead to faster and more accurate material design. This is crucial for industries relying on material properties, such as electronics and energy.
Key Takeaways
- Introduction of NextHAM, a neural network for Hamiltonian prediction.
- Utilization of zeroth-step Hamiltonians as informative descriptors enhances model accuracy.
- Development of a novel training objective to mitigate error amplification in predictions.
- Creation of the Materials-HAM-SOC dataset, improving coverage and quality for training.
- Experimental results show NextHAM outperforms traditional methods in accuracy and efficiency.
Computer Science > Machine Learning arXiv:2509.19877 (cs) [Submitted on 24 Sep 2025 (v1), last revised 19 Feb 2026 (this version, v4)] Title:Advancing Universal Deep Learning for Electronic-Structure Hamiltonian Prediction of Materials Authors:Shi Yin, Zujian Dai, Xinyang Pan, Lixin He View a PDF of the paper titled Advancing Universal Deep Learning for Electronic-Structure Hamiltonian Prediction of Materials, by Shi Yin and 3 other authors View PDF HTML (experimental) Abstract:Deep learning methods for electronic-structure Hamiltonian prediction has offered significant computational efficiency advantages over traditional DFT methods, yet the diversity of atomic types, structural patterns, and the high-dimensional complexity of Hamiltonians pose substantial challenges to the generalization performance. In this work, we contribute on both the methodology and dataset sides to advance universal deep learning paradigm for Hamiltonian prediction. On the method side, we propose NextHAM, a neural E(3)-symmetry and expressive correction method for efficient and generalizable materials electronic-structure Hamiltonian prediction. First, we introduce the zeroth-step Hamiltonians, which can be efficiently constructed by the initial charge density of DFT, as informative descriptors of neural regression model in the input level and initial estimates of the target Hamiltonian in the output level, so that the regression model directly predicts the correction terms to the target ground tr...