Physics-informed fine-tuning of a Stokes-flow surrogate.
Status: Ongoing research. Results below are from the specified reduced-budget experiment, not the upstream full-budget configuration.
Experimental setup
Reduced local study with paired evaluation on two test meshes; 4,000 baseline iterations and 10,000 physics-stage iterations.
Results achieved
| Metric | Result |
|---|---|
| Velocity relative L² error | 0.188 → 0.109 (−41.8%) |
| Momentum residual RMS | 0.898 → 0.292 (−67.5%) |
| Divergence residual RMS | 0.610 → 0.063 (−89.6%) |
| Boundary residual RMS | 0.0033 → 0.0155 (increased) |
Experimental figures
Figures are drawn directly from the project's public GitHub repository.
Next steps
Verify provenance and evaluate the remaining eight meshes; examine boundary trade-offs.
For reproducibility, consult the repository's configuration files, scripts, saved results and scientific audit.
Repository, methods and results ↗ ← All PhysicsNeMo projects