[KongchangAI]
Benchmark

UniHam

UniHam is the first SOC-complete large-scale Hamiltonian dataset and benchmark. It contains over 100,000 complex-valued Hermitian Hamiltonians obtained from DFT calculations, covering 72 elements, and is used to evaluate the Hamiltonian prediction capability of foundation models for quantum materials.

Timeline (last 90 days)

Nov 10

UniHam constructed a dataset containing 100,000+ complex-valued Hermitian Hamiltonians obtained from DFT calculations.

Unverified50%
Nov 10

The UniHam dataset covers 72 elements.

Unverified50%
Nov 10

The UniHam dataset treats spin-orbit coupling (SOC) as a default configuration rather than an option.

Unverified50%
Nov 10

UniHam designed three complementary evaluation dimensions: Hamiltonian reconstruction accuracy, out-of-distribution (OOD) generalization, and downstream property prediction capability.

Unverified50%
Nov 10

On the UniHam benchmark, two representative frontier models show significant differences in compositional OOD robustness.

Unverified50%
Nov 10

The experiments exposed failure modes of systematic inaccuracy in SOC and specific-element-related scenarios.

Unverified50%
Nov 10

UniHam open-sourced the dataset and benchmark.

Unverified50%
Nov 10

The UniHam dataset covers multiple lattice types and space group families.

Unverified50%
Nov 10

UniHam is positioned as a reproducible, SOC-complete benchmark aimed at sharpening comparisons between models and accelerating the development of next-generation foundation models for quantum materials.

Unverified50%

All Facts (9)

Source Articles