Machine Learning Hamiltonians are Accurate Energy-Force Predictors
Seongsu Kim, Chanhui Lee, Yoonho Kim, Seongjun Yun, Honghui Kim, Nayoung Kim, Changyoung Park, Sehui Han, Sungbin Lim, Sungsoo Ahn · 2026
Read Machine Learning Hamiltonians are Accurate Energy-Force Predictors by Seongsu Kim, Chanhui Lee, Yoonho Kim, Seongjun Yun, Honghui Kim, Nayoung Kim, Changyoung Park, Sehui Han, Sungbin Lim, Sungsoo Ahn (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-16897.txt.
Description
Seongsu Kim and co-authors benchmark machine learning Hamiltonians as direct energy-force predictors. Such models have gained traction as fast approximations of electronic structures such as orbitals and electron densities, also enabling direct evaluation of energies and forces from predictions. Yet existing models were judged mainly by reconstruction scores, leaving their energy-force skill unclear.
The new benchmark computes energies and forces directly from predicted Hamiltonians. Within it, the authors propose QHFlow2, with an SO(2)-equivariant backbone and a two-stage edge update. It cuts Hamiltonian error by forty percent against the prior best model with fewer parameters. On MD17 and rMD17, it is the first Hamiltonian model to reach NequIP-level force accuracy, and on QH9 it cuts energy error by up to a fifth against MACE. Hamiltonian accuracy gains translate into better energy and force scores.
Catalog record
| CREATOR | Seongsu Kim, Chanhui Lee, Yoonho Kim, Seongjun Yun, Honghui Kim, Nayoung Kim, Changyoung Park, Sehui Han, Sungbin Lim, Sungsoo Ahn |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 103 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-16897