Research map / Architectures and efficiency / RNNs and normalization
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
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace QHFlow with QHFlow2 (SO(2) eSEN backbone plus two-stage pair update) | QHFlow · QH9-stable-id | Hamiltonian MAE (uHa): 23.0 -> 9.2 (QHFlow2-m, 43M) | Improved |
| replace QHFlow with QHFlow2 on downstream energy | QHFlow · QH9-stable-id | total energy MAE (meV): 2.37 -> 0.69 (QHFlow2-m) | Improved |
| scale QHFlow2 from 12M to 990M parameters (ablation) | QHFlow2 · QH9-stable-id | Hamiltonian MAE (uHa): 16.3 -> 4.6 | Improved |
| scale QHFlow2 from 12M to 990M parameters (ablation) | QHFlow2 · QH9-stable-id | force MAE (meV/A): 3.1 -> 1.2 (non-monotonic: 2.1, 1.4 in between) | Mixed |
| increase training set size (ablation) | QHFlow2-m · MD benchmark (salicylic acid, aspirin) | Hamiltonian, orbital, energy and force MAE | Improved |
| use QHFlow2 Hamiltonian predictor instead of MLIP baselines | NequIP · MD17/rMD17 (six molecules) | energy MAE: up to 20x lower energy MAE | Improved |
| use QHFlow2 Hamiltonian predictor instead of NequIP | NequIP · MD17/rMD17 (six molecules) | force MAE | No clear effect |
| add SO(3) tensor-product refinement as second stage of pair update (ablation) | QHFlow2 · off-diagonal Hamiltonian blocks, cutoff/radial-basis sensitivity | Hamiltonian off-diagonal accuracy | Improved |
| initialize SCF with predicted Hamiltonian instead of MinAO | DFT SCF · QH9-stable-iid test (300 molecules) | SCF cycles ratio: QHFlow2-l 2.18x end-to-end speedup | Improved |
| scale QHNet beyond about 40M parameters (ablation) | QHNet · QH9 | training stability | Hurt |
Closest papers by meaning
- The dark side of the forces: assessing non-conservative force models for atomistic machine learning (ICML 2025 · similarity 0.88)
- Enhancing the Scalability and Applicability of Kohn-Sham Hamiltonians for Molecular Systems (ICLR 2025 · similarity 0.88)
- High-order Equivariant Flow Matching for Density Functional Theory Hamiltonian Prediction (NeurIPS 2025 · similarity 0.88)
- Physics-Informed Weakly Supervised Learning For Interatomic Potentials (ICML 2025 · similarity 0.88)
- EDBench: Large-Scale Electron Density Data for Molecular Modeling (NeurIPS 2025 · similarity 0.87)
- Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
