Research map / Probabilistic and scientific modeling / Molecules and proteins
Gold-standard solutions to the Schrödinger equation using deep learning: How much physics do we need?
Leon Gerard, Michael Scherbela, Philipp Marquetand, 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 |
|---|---|---|---|
| dense determinants instead of block-diagonal (ablation) | FermiNet-based DL-VMC · ethene, N2, K atom | energy error | Improved |
| tuned hyperparameters (higher norm constraint, lower lr, faster decay) (ablation) | FermiNet-based DL-VMC · ethene | energy error: 3.7 mHa | Improved |
| SchNet-like embedding and local invariant input features (ablation) | FermiNet-based DL-VMC · ethene | energy error: 1.4 mHa | Improved |
| physics-inspired envelope initialization omega=Z/n instead of 1 (ablation) | FermiNet-based DL-VMC · K atom | energy error | Improved |
| all changes combined vs FermiNet (ablation) | FermiNet-based DL-VMC · ethene, N2, K atom | energy error: ~3-20x | Improved |
| use ~50% fewer Monte Carlo walkers with tuned hyperparameters (ablation) | FermiNet-based DL-VMC · optimization throughput | time per optimization: ~2x | Improved |
| dense to block-diagonal determinants toward PauliNet-like model (ablation) | our architecture · NH3 | energy error | Hurt |
| PauliNet-like embedding replacing our embedding (ablation) | our architecture · NH3 | energy error | Hurt |
| CASSCF envelopes instead of exponential envelopes (ablation) | our architecture · NH3 | energy error | Hurt |
| remove electron-nucleus difference vector, keep only distance (ablation) | our architecture · NH3 | energy error | Hurt |
| smooth radial basis functions plus explicit electron-electron cusp term (ablation) | our architecture · NH3 | energy error | Improved |
| increase pre-training steps beyond 20k (ablation) | our architecture · NH3 | energy error | Hurt |
Closest papers by meaning
- Quadratic Quantum Variational Monte Carlo (NeurIPS 2024 · similarity 0.85)
- Neural Pfaffians: Solving Many Many-Electron Schrödinger Equations (NeurIPS 2024 · similarity 0.85)
- Enhancing the Scalability and Applicability of Kohn-Sham Hamiltonians for Molecular Systems (ICLR 2025 · similarity 0.85)
- Light Schrödinger Bridge (ICLR 2024 · similarity 0.85)
- High-order Equivariant Flow Matching for Density Functional Theory Hamiltonian Prediction (NeurIPS 2025 · similarity 0.85)
- A Theoretical Framework for an Efficient Normalizing Flow-Based Solution to the Electronic Schrödinger Equation (AAAI 2025 · similarity 0.84)
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.
