Research map / Probabilistic and scientific modeling / Molecules and proteins
Learning from the Electronic Structure of Molecules across the Periodic Table
Manasa Kaniselvan, Benjamin Kurt Miller, Meng Gao, Juno Nam, Daniel S. Levine
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 |
|---|---|---|---|
| add node-parity expansion in the Hamiltonian head (ablation) | HELM · nabla2DFT 2k test | energy error from H (meV): 329,923.53 -> 1,530.83 | Improved |
| add l=0 component scaling on top of node-parity expansion (ablation) | HELM · nabla2DFT 2k test | energy error from H (meV): 1,530.83 -> 706.39 | Improved |
| add l=0 scaling on top of no parity expansion (ablation) | HELM · nabla2DFT 2k test | Hamiltonian error (1e-6 Eh): 324.35 -> 120.18 | Improved |
| Hamiltonian pretraining with frozen backbone before energy training | HELM · nabla2DFT 2k test | energy MAE (meV): 791.02 -> 324.64 | Improved |
| Hamiltonian pretraining then finetuning on energies | HELM · nabla2DFT 2k test | energy MAE (meV): 791.02 -> 266.23 | Improved |
| Hamiltonian pretraining with frozen backbone | HELM · OMol_all_5k | energy MAE (meV): 5976.40 -> 3255.69 | Improved |
| finetune the pretrained backbone instead of keeping it frozen (ablation) | HELM · OMol_CSH_58k tests | energy MAE (meV): 3255.69 -> 3926.12 | Hurt |
| finetune the pretrained backbone instead of keeping it frozen (ablation) | HELM · nabla2DFT 2k test | energy MAE (meV): 324.64 -> 266.23 | Improved |
| Hamiltonian pretraining on OMol_CSH_58k, in-distribution energy training split | HELM · OMol_CSH_58k train | energy MAE (meV): 353.15 -> 361.98 | No clear effect |
| use HELM architecture for Hamiltonian prediction instead of previous models | HELM · nabla2DFT | H error: ~3-5x lower than the previous best | Improved |
Closest papers by meaning
- Learning the Electronic Hamiltonian of Large Atomic Structures (ICML 2025 · similarity 0.92)
- EDBench: Large-Scale Electron Density Data for Molecular Modeling (NeurIPS 2025 · similarity 0.91)
- Benchmarking Compositional Generalisation for Machine Learning Interatomic Potentials (NeurIPS 2026 · similarity 0.90)
- Enhancing the Scalability and Applicability of Kohn-Sham Hamiltonians for Molecular Systems (ICLR 2025 · similarity 0.89)
- Physics-Informed Weakly Supervised Learning For Interatomic Potentials (ICML 2025 · similarity 0.89)
- Higher-Rank Irreducible Cartesian Tensors for Equivariant Message Passing (NeurIPS 2024 · similarity 0.89)
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.
