Universal deep learning method speeds up electronic structure calculations
Physical structure as an organizing principle for universal Hamiltonian learning
Artificial Intelligence
Summary
Calculating the electronic properties of materials helps scientists and engineers design new materials, but traditional methods are slow and limited to small systems. The authors created a new machine learning framework called UniTB that combines physics knowledge with deep learning to predict these properties efficiently and accurately across many elements. UniTB can handle large and complex materials faster than usual while keeping results physically meaningful. This approach makes it easier to screen thousands of materials for practical applications like electronics and optics.
What this means in practice
- •For materials scientists: Accelerate screening of large sets of crystals for electronic and optical properties without costly simulations.
- •For computational chemists: Use a universal, physics-informed neural model to predict Hamiltonians for disordered and new-composition materials faster than traditional methods.
Authors
Mingkang Liu, Zhanghao Zhouyin, Guanghui Wang, Daye Zheng, Lei Shen, Qiangqiang Gu
Abstract
Density functional theory (DFT) gives the electronic structure of materials, but its cost limits large-supercell calculations and materials screening. Machine-learned Hamiltonians bypass the self-consistent cycle but are usually system-specific; universal ones could cover new compositions and disordered solids. Such models must generalize across chemistry and keep downstream properties physical, while staying inexpensive for large cells. Here we introduce UniTB, a universal E(3) equivariant Hamiltonian representation framework that integrates physical priors rooted in density-functional tight-binding theory with deep learning. UniTB's network, which corrects a prior Hamiltonian, generalizes across chemistry through a mixture of experts routed by that prior and a charge-equilibration head as in self-consistent-charge tight binding. A projection bounds band shifts from the prior, suppressing spurious states and keeping downstream properties physical. We provide the fastest SO(2) tensor-product GPU kernels for the network. The variant UniTB-Flash assembles its prior from two-center tables at $O(N)$ cost for $N$ atoms, against $O(N\log N)$ on a grid. Trained on 29,242 crystals, UniTB reaches a 0.48 meV matrix-element error on nonmagnetic test crystals spanning 68 elements. Without amorphous training data, both models agree with DFT optics and conductivity of amorphous Ge-Sb-Te. UniTB-Flash ranks candidates with no significant difference from UniTB, whose property values are more accurate; together they screen 16,195 crystals for five application domains of industrial relevance.