Research map / Probabilistic and scientific modeling / Differential equations and PDEs
MUSA-PINN: Multi-scale Weak-form Physics-Informed Neural Networks for Fluid Flow in Complex Geometries
Weizheng Zhang, Xunjie Xie, Hao Pan, Xiaowei Duan, Bingteng Sun, Qiang Du, Lin Lu
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 weak-form flux-balance constraints over multi-scale control volumes | PINN-PE style strong-form PINN · TPMS Diamond channel | relative l2 error on velocity: 90.18% -> 12.46% | Improved |
| use all three scales instead of large-only (ablation) | MUSA-PINN · TPMS Diamond | relative l2 error on velocity: 14.59% -> 12.46% | Improved |
| use all three scales instead of medium-only or small-only (ablation) | MUSA-PINN · TPMS Diamond | relative l2 error on velocity: 87.72% / 89.99% -> 12.46% | Improved |
| add weak momentum stage after continuity warm-up (ablation) | MUSA-PINN · TPMS Diamond | relative l2 error on pressure: 44.19% -> 26.55% | Improved |
| add weak momentum on top of continuity, velocity (ablation) | MUSA-PINN · TPMS Diamond | relative l2 error on velocity: 13.28% -> 12.46% | Improved |
| use two-stage schedule instead of joint training from scratch (ablation) | MUSA-PINN · TPMS Diamond | relative l2 error on velocity: 95.31% -> 12.46% | Improved |
| use weak momentum only without continuity (ablation) | MUSA-PINN · TPMS Diamond | relative l2 error on velocity: 12.46% -> 95.87% | Hurt |
| keep strong-form loss alongside weak-form terms (ablation) | MUSA-PINN · TPMS Diamond | relative l2 error on velocity: 43.92% -> 12.46% | Improved |
| use MUSA-PINN instead of best baseline on Gyroid at Re = 400 | MUSA-PINN · Gyroid Re=400 | relative l2 error on velocity: 69.33% -> 21.08% | Improved |
| use MUSA-PINN on easy Primitive channel versus baselines | MUSA-PINN · TPMS Primitive (P), velocity | relative l2 error: 2.60% -> 1.42% | Improved |
| use weak-form hp-VPINN baseline instead of strong-form PINN | hp-VPINN · TPMS Gyroid, velocity | relative l2 error: 83.19% -> 94.10% (vs PINN-PE) | Hurt |
| increase MLP capacity from 256x5 to 256x8 (ablation) | MUSA-PINN · Liquid cooling plate | relative l2 error on velocity: 14.91% -> 14.18% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CoPINN — CoPINN: Cognitive Physics-Informed Neural Networks (ICML 2025) · better on 1 of 1 shared result
- RoPINN — RoPINN: Region Optimized Physics-Informed Neural Networks (NeurIPS 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Coupled Integral PINN for Discontinuity (NeurIPS 2026 · similarity 0.87)
- RoPINN: Region Optimized Physics-Informed Neural Networks (NeurIPS 2024 · similarity 0.87)
- PINNsFormer: A Transformer-Based Framework For Physics-Informed Neural Networks (ICLR 2024 · similarity 0.87)
- PINNacle: A Comprehensive Benchmark of Physics-Informed Neural Networks for Solving PDEs (NeurIPS 2024 · similarity 0.87)
- Parameterized Physics-informed Neural Networks for Parameterized PDEs (ICML 2024 · similarity 0.86)
- Dual Cone Gradient Descent for Training Physics-Informed Neural Networks (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.
