Research map / Learning theory and algorithms / Combinatorial and search optimization
A Unified Hard-Constraint Framework for Solving Geometrically Complex PDEs
Songming Liu, Zhongkai Hao, Chengyang Ying, Hang Su, Jun Zhu, Ze Cheng
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 the unified hard-constraint framework instead of FBPINN | PDE solution neural networks · 2D battery pack, average over t∈[0,1] | MAE of T: 0.0343 -> 0.0221 | Improved |
| Use the unified hard-constraint framework instead of FBPINN | PDE solution neural networks · 2D battery pack, average over t∈[0,1] | MAPE of T: 14.74% -> 5.10% | Improved |
| Use the unified hard-constraint framework instead of the strongest listed baseline | Airfoil Navier-Stokes solution · Airfoil | MAE of u1: 0.4018 -> 0.2689 | Improved |
| Use the unified hard-constraint framework instead of FBPINN | Airfoil Navier-Stokes solution · Airfoil | MAE of u2: 0.0563 -> 0.0435 | Improved |
| Use the unified hard-constraint framework instead of PINN-LA | Airfoil Navier-Stokes solution · Airfoil | MAE of p: 0.2652 -> 0.2032 | Improved |
| Use the unified hard-constraint framework instead of PINN-LA | PDE solution neural networks · High-dimensional heat equation, average over t∈[0,1] | MAE of u: 0.0235 -> 0.0026 | Improved |
| Add extra fields to the PDE reformulation (ablation) | Vanilla PINN · Poisson's equation and nonlinear Schrödinger equation | Moving variance of PDE-loss gradient magnitude | Improved |
| Increase the spatial and temporal hardness parameters (ablation) | HC on the 2D battery-pack heat equation · 2D battery pack, average over t∈[0,1] | MAE of T: 0.3492 / 48.21% -> 0.0221 / 5.10% (β_s=1, β_t=1 to β_s=5, β_t=10); 0.0221 -> 0.0329 (β_s=5, β_t=10 to β_s=10, β_t=10) | Mixed |
Closest papers by meaning
- Structure-Preserving Learning Improves Geometry Generalization in Neural PDEs (ICML 2026 · similarity 0.88)
- UGrid: An Efficient-And-Rigorous Neural Multigrid Solver for Linear PDEs (ICML 2024 · similarity 0.86)
- Solving Differential Equations with Constrained Learning (ICLR 2025 · similarity 0.86)
- Graph Neural PDE Solvers with Conservation and Similarity-Equivariance (ICML 2024 · similarity 0.85)
- Towards General Neural Surrogate Solvers with Specialized Neural Accelerators (ICML 2024 · similarity 0.85)
- A Neural-Preconditioned Poisson Solver for Mixed Dirichlet and Neumann Boundary Conditions (ICML 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.
