Researchhamiltonian modelsso2 equivariantmolecular dynamicsmd17
QHFlow2 Delivers Superior Hamiltonian Energy And Force Accuracy
9.1
Researchers present QHFlow2, an SO(2)-equivariant machine-learning Hamiltonian model with a two-stage edge update, submitted to arXiv on Feb 18, 2026. QHFlow2 achieves 40% lower Hamiltonian error than prior models and is first to reach NequIP-level force accuracy on MD17/rMD17, while yielding up to 20× lower energy MAE. On QH9 it reduces energy error versus MACE by up to 20× and shows consistent scaling with capacity and data.
Key Points
- 1Achieves 40% lower Hamiltonian error than previous best model with fewer parameters
- 2Reaches NequIP-level force accuracy on MD17/rMD17 and up to 20× lower energy MAE
- 3Enables direct energy and force computation from predicted Hamiltonians, improving MD simulation fidelity
Scoring Rationale
High novelty and strong benchmark gains drive score; limitation is evaluation on standard datasets and preprint status.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
