UniHam: A Large-Scale SOC-Complete Dataset and Benchmark for Hamiltonian Learning in Materials
Organizations: Sun Yat-Sen University, Guangzhou, China
Abstract
Accurate prediction of electronic Hamiltonians would enable broad property inference while avoiding the high computational cost of Density Functional Theory (DFT). However, progress toward general-purpose materials foundation models is limited by a data bottleneck: existing Hamiltonian datasets are typically small, lack structural diversity, and often omit essential relativistic physics such as spin--orbit coupling (SOC). We therefore construct UniHam, a large-scale Hamiltonian dataset and benchmark suite comprising 100,000+ DFT-computed complex-valued Hermitian Hamiltonians with full SOC, covering 72 elements and a wide range of crystal geometries and symmetries (spanning diverse lattice types and space-group families). Building on UniHam, we benchmark two representative state-of-the-art models under a standardized protocol and introduce complementary evaluation metrics that jointly assess three dimensions: (i) Hamiltonian reconstruction accuracy, (ii) out-of-distribution (OOD) generalization across composition/symmetry shifts, and (iii) the ability to support downstream property prediction from the predicted Hamiltonians. Experiments on UniHam demonstrate that the proposed benchmark and metrics effectively differentiate model capabilities, revealing intrinsic SOC- and element-dependent failure modes, large variations in compositional OOD robustness, and the necessity of spectral-level evaluation to assess whether Hamiltonian predictions reliably support downstream electronic-structure properties. Overall, UniHam provides a reproducible, SOC-complete benchmark that can sharpen model comparisons and accelerate the development of next-generation foundation models for quantum materials.
Figures & tables
| Feature | UniHam (Ours) | Mat-HAM-SOC | DeepH | HamGNN |
| ( Yin et al., 2025 ) | ( Wang et al., 2024 ) | ( Zhong et al., 2024 ) | ||
| Structures | 103,121 | 17,000 | 12,062 | 44,000 |
| Elements | 72 | 68 | 24 (per. 1–4) | NR |
| Lanthanide ( ) | 9 (La–Lu) | Excluded | Excluded | NR |
| Full SOC | All | All | No | No |
| Heavy elem. ( ) | 66.6% | 50% | 0% | NR |
| Task | Training Protocol | Primary Metrics |
| Recon-ID | Random split (8:1:1) | |
| OOD-Size | Evaluate on larger structures using Recon-ID model | |
| OOD-Comp | Retrain after removing specified elements from training set | |
| Prop-ID | Use Recon-ID model; diagonalize predicted Hamiltonians | Band gap, , |
| Model | Test Set MAE (meV) | ||
| Real | Imag. | Total | |
| DeepH-pack | 4.49 | 0.385 | 2.44 |
| HamGNN | 2.96 | 0.0412 | 1.50 |
| SOC Regime | N | HamGNN | DeepH-pack |
| Weak ( ) | 200 | 1.10 | 1.90 |
| Light-3 ( ) | 401 | 1.42 | 2.17 |
| Medium-4 ( ) | 614 | 1.38 | 2.38 |
| Strong-5 /4 ( ) | 785 | 1.74 | 2.76 |
| Model | Setting | In | RE | NG |
| DeepH-pack | ID | 2.89 | 3.14 | 5.09 |
| OOD | 22.49 | 22.73 | 31.48 | |
| HamGNN | ID | 1.40 | 2.04 | 3.06 |
| OOD | 5.71 | 5.39 | 5.93 |
| Elem. | DeepH-pack | HamGNN | |
| Pr ( ) | 107 | 19.59 | 6.80 |
| Nd ( ) | 228 | 21.96 | 5.96 |
| Sm ( ) | 262 | 22.36 | 4.42 |
| Dy ( ) | 121 | 24.21 | 5.25 |
| Ho ( ) | 127 | 26.15 | 5.35 |
| Band gap (eV) | 0–1 | 1–2 | 2–3 | 3–4 | 4–5 | 5–6 | 6+ |
| Count | 267 | 216 | 124 | 65 | 31 | 12 | 5 |
| Fraction (%) | 37.1 | 30.0 | 17.2 | 9.0 | 4.3 | 1.7 | 0.7 |
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
| Property | Value |
| Total structures | 103,121 |
| Unique compositions | 50,077 |
| Elements covered | 72 / Heavy ( ): 41 |
| Structures with heavy elements | 66.6% |
| Atoms per cell (mean/median/range) | 30 / 20 / 1–444 |
| Space groups covered | 225 / 230 |
| Hyperparameter | DeepH-pack | HamGNN |
| Architecture | DeepH-E3 (Eagle) | HamGNN |
| Message-passing layers | 3 | 3 |
| Radial cutoff | 10.0 Å | 26.0 Bohr |
| Radial basis | Gaussian | 64 Bessel |
| Attention heads | 2 | — |
| Delta learning ( ) | No | Yes |