Research map / Probabilistic and scientific modeling / Molecules and proteins
Derivative Informed Learning of Exchange-Correlation Functionals
Eike Eberhard, Luca Thiede, Abdulrahman Aldossary, Andreas Burger, Nicholas Gao, Vignesh C Bhethanabotla, Alan Aspuru-Guzik, Stephan Günnemann
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 derivative-informed loss (gradient and Hessian supervision) to energy+density loss | four ML-XC architectures · B3LYP distillation, QM40 OOD | total-energy MAE: decreases by 66% on average | Improved |
| add DI-Loss | EG-XC · B3LYP distillation, QM40 OOD | total energy error, L2 density: total energy error cut by 80%; L2[rho] lowered by 8% | Improved |
| add DI-Loss | semi-local ML-XC functionals · B3LYP distillation, QM40 OOD | E_rho, dipole and L2 density errors: E_rho 1.2 -> 0.8 mEh; dipole and L2 not uniformly improved | Mixed |
| add gradient term only (without Hessian) to E+rho (ablation) | ML-XC functionals · TDDFT excitation MAE | mean excitation MAE: raises MAE by 9--14% for XCdiff | Hurt |
| add Hessian term on top of gradient term (ablation) | XCdiff · TDDFT excitation MAE | mean excitation MAE: reduced by 28%, 24%, 24%, 19% for 7, 9, 12, 14 heavy atoms; 35% for 5 | Improved |
| add DI-Loss | XCdiff · TDDFT fifth excitation, 7 heavy atoms | MAE (meV): 356 -> 233 meV | Improved |
| add DI-Loss | XCdiff · TDDFT first excitation, 14 heavy atoms | MAE (meV): 462 -> 386 meV | Improved |
| initialize B3LYP with distilled ML-XC density instead of SCAN density | B3LYP SCF · QM9 9 heavy atoms | relative iteration count: about 65% -> about 50% of MINAO iterations | Improved |
| add derivative supervision to distillation for SCF initialization (ablation) | distilled ML-XC · QM9 B3LYP RIC | relative iteration count | No clear effect |
| add Hessian term (ablation) | distilled functionals · SCF iterations vs B3LYP reference | SCF iterations: distilled need ~15% more iterations; Hessian recovers part | Improved |
| initialize B3LYP with six NNmGGA cycles | B3LYP on 35-heavy-atom QM40 molecule · QM40 | walltime speedup: 1.35x | Improved |
Closest papers by meaning
- Learning Generalized Residual Exchange-Correlation-Uncertain Functional for Density Functional Theory (AAAI 2025 · similarity 0.90)
- Learning Equivariant Non-Local Electron Density Functionals (ICLR 2025 · similarity 0.89)
- Colour me shocked: Exact Molecular Hessians from MLIPs in O(N) time using sparse differentiation! (NeurIPS 2026 · similarity 0.86)
- Physics-Informed Weakly Supervised Learning For Interatomic Potentials (ICML 2025 · similarity 0.86)
- The Importance of Being Scalable: Improving the Speed and Accuracy of Neural Network Interatomic Potentials Across Chemical Domains (NeurIPS 2024 · similarity 0.86)
- Enhancing the Scalability and Applicability of Kohn-Sham Hamiltonians for Molecular Systems (ICLR 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.
