Research map / Probabilistic and scientific modeling / Dynamical systems
Simplifying Hamiltonian and Lagrangian Neural Networks via Explicit Constraints
Marc Finzi, Ke Alexander Wang, Andrew G Wilson
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 Cartesian coordinates with explicit constraints instead of generalized coordinates with implicit constraints | CHNN and CLNN · Physical-system benchmarks | trajectory prediction accuracy: 10 to 100 times more accurate | Improved |
| Use Cartesian coordinates with explicit constraints | CHNN and CLNN · 2-pendulum | geometric average relative error: 20 trajectories; HNN: 200 trajectories; NeuralODE: 1600 trajectories | Improved |
| Use Cartesian coordinates with explicit constraints | CHNN and CLNN · Gyroscope | trajectory prediction accuracy: 100 times more data efficient or 260 times more accurate | Improved |
| Use Cartesian coordinates with explicit constraints | CHNN and CLNN · N-pendulum systems | geometric mean relative error | Improved |
| Use Cartesian coordinates with explicit constraints | CHNN and CLNN · 3D systems | geometric mean relative error | Improved |
| Use Cartesian coordinates with explicit constraints | CHNN and CLNN · Physical-system benchmarks | data efficiency: 100 times less data | Improved |
| Remove explicit constraints from the Cartesian-coordinate model (ablation) | CHNN and CLNN · 3-Pendulum | geometric mean relative error | Hurt |
| Remove distance constraints successively, starting from the bottom (ablation) | CHNN and CLNN · 3-Pendulum | geometric mean relative error: >100 times better than HNN; removing even a single constraint makes performance slightly worse than HNN | Hurt |
| Use Cartesian coordinates rather than generalized coordinates | CHNN · N-Pendulum | effective dimensionality | Improved |
Closest papers by meaning
- Unsupervised Learning of Lagrangian Dynamics from Images for Prediction and Control (NeurIPS 2020 · similarity 0.89)
- Learning Articulated Rigid Body Dynamics with Lagrangian Graph Neural Network (NeurIPS 2022 · similarity 0.88)
- A Riemannian Framework for Learning Reduced-order Lagrangian Dynamics (ICLR 2025 · similarity 0.87)
- Learning Hamiltonian Dynamics at Scale: A Differential-Geometric Approach (ICML 2026 · similarity 0.87)
- Learned Lagrangian Models of PDEs via Euler–Lagrange Residual Minimization (NeurIPS 2026 · similarity 0.87)
- Physics-Informed Regularization for Domain-Agnostic Dynamical System Modeling (NeurIPS 2024 · 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.
