Research map / Probabilistic and scientific modeling / Molecules and proteins
Variational Monte Carlo on a Budget — Fine-tuning pre-trained Neural Wavefunctions
Michael Scherbela, Leon Gerard, Philipp Grohs
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 |
|---|---|---|---|
| use improved architecture on prior dataset (ablation) | pre-trained wavefunction · 3 heavy atoms zero-shot | energy error: 76% | Improved |
| use improved architecture on prior dataset (ablation) | pre-trained wavefunction · 3 heavy atoms after 4k fine-tuning | energy error: 35% | Improved |
| add normal-mode distortion data augmentation (ablation) | pre-trained wavefunction · 3 heavy atoms fine-tuned | energy error: 12% | Improved |
| enlarge training set to 98 molecules with torsional conformers (ablation) | pre-trained wavefunction · 3 heavy atoms zero-shot | energy error: 98% | Improved |
| enlarge training set to 98 molecules (ablation) | pre-trained wavefunction · 3 heavy atoms fine-tuned | energy error | Improved |
| increase determinants from 4 to 8 (ablation) | pre-trained wavefunction · 3 heavy atoms fine-tuned | energy error: 8% | Improved |
| increase pre-training steps from 128k to 256k (ablation) | pre-trained wavefunction · 3 heavy atoms fine-tuned | energy error: 21% | Improved |
| fine-tune 700 steps per geometry | pre-trained wavefunction · bicyclobutane dis-TS | relative energy error: 183.8 -> 94.5 | Improved |
| fine-tune 4000 steps across geometries | pre-trained wavefunction · rotated propadiene | energy variation: below 1.6 mHa | Improved |
| train longer with rotation augmentation (170k to 256k epochs) | pre-trained wavefunction · rotated propadiene | energy variation: ±3 -> ±1 | Improved |
| fine-tune 4k steps over 10 geometries | pre-trained wavefunction · twisted propadiene | max deviation along path: 40 -> 2 | Improved |
| add pre-training | neural wavefunction · 3 heavy atoms energy vs steps | energy error: 1-2 orders lower up to 16k steps | Mixed |
Closest papers by meaning
- Quadratic Quantum Variational Monte Carlo (NeurIPS 2024 · similarity 0.90)
- Enhancing the Scalability and Applicability of Kohn-Sham Hamiltonians for Molecular Systems (ICLR 2025 · similarity 0.88)
- TITAN: A Trajectory-Informed Technique for Adaptive Parameter Freezing in Large-Scale VQE (NeurIPS 2025 · similarity 0.86)
- Neural Pfaffians: Solving Many Many-Electron Schrödinger Equations (NeurIPS 2024 · similarity 0.86)
- PALQO: Physics-informed model for Accelerating Large-scale Quantum Optimization (NeurIPS 2025 · similarity 0.86)
- High-order Equivariant Flow Matching for Density Functional Theory Hamiltonian Prediction (NeurIPS 2025 · similarity 0.86)
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.
