Research map / Probabilistic and scientific modeling / Molecules and proteins
Sampling-free Inference for Ab-Initio Potential Energy Surface Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use a nuclear-geometry GNN surrogate trained online on intermediate noisy VMC energies, avoiding inference-time Monte Carlo integration | PESNet++ · C2H5OH | inference time per geometry: 2.7 hours -> 118 microseconds | Helped |
| Add restricted neural wave functions with doubly occupied orbitals (ablation) | PESNet++ · N2 | MAE to experimental results: 46%; 1.21 milli-hartree | Helped |
| Replace block-diagonal orbital matrices with dense orbital matrices (ablation) | PESNet++ · N2 | energy error | Helped |
| Add a permutation-invariant Jastrow factor (ablation) | PESNet++ · N2 | energy error | Helped |
| Use careful initialization, rescaling after activations, and SiLU to reduce dead neurons (ablation) | PESNet++ · neural wave function training | dead neuron proportion: >40% -> <10% | Helped |
| Add the nucleus-relative electron coordinate transform during geometry perturbations (ablation) | PESNet++ · H2-HF full 6D surface | final energy standard deviation: 17% lower; 3 out of 5 runs diverge without the transform | Helped |
| Increase surrogate training steps from one to five (ablation) | PlaNet · N2 | absolute MAE (milli-hartree): 0.57(21) -> 0.33(7) | Helped |
| Increase surrogate training steps from five to ten (ablation) | PlaNet · N2 | absolute MAE (milli-hartree): 0.33(7) -> 0.32(6) | No clear effect |
| Use a fixed high EMA decay of 0.9999 instead of adaptive decay (ablation) | PlaNet · H4 | absolute MAE (milli-hartree): 0.0072(7) -> 0.167(30) | Hurt |
| Use adaptive EMA decay instead of fixed decay 0.99 (ablation) | PlaNet · N2 | absolute MAE (milli-hartree): 0.57(15) -> 0.33(7) | Helped |
| Train on a three-dimensional rather than two-dimensional H2-HF surface (ablation) | PlaNet · H2-HF two-dimensional slice | MAE (milli-hartree): 0.28(9) -> 1.02(34) | Hurt |
| Train on the full six-dimensional rather than two-dimensional H2-HF surface (ablation) | PlaNet · H2-HF two-dimensional slice | MAE (milli-hartree): 0.28(9) -> 2.39(25) | Hurt |
Closest papers by meaning
- The Importance of Being Scalable: Improving the Speed and Accuracy of Neural Network Interatomic Potentials Across Chemical Domains (NeurIPS 2024 · similarity 0.87)
- ELoRA: Low-Rank Adaptation for Equivariant GNNs (ICML 2025 · similarity 0.86)
- FlashTP: Fused, Sparsity-Aware Tensor Product for Machine Learning Interatomic Potentials (ICML 2025 · similarity 0.86)
- Physics-Informed Weakly Supervised Learning For Interatomic Potentials (ICML 2025 · similarity 0.86)
- Efficient and Scalable Density Functional Theory Hamiltonian Prediction through Adaptive Sparsity (ICML 2025 · similarity 0.85)
- Higher-Rank Irreducible Cartesian Tensors for Equivariant Message Passing (NeurIPS 2024 · similarity 0.85)
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.