Research map / Probabilistic and scientific modeling / Differential equations and PDEs
A composable machine-learning approach for steady-state simulations on high-resolution grids
Rishikesh Ranade, Derek Christopher Hill, Lalit Ghule, Jay Pathak
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 |
|---|---|---|---|
| subdomain resolution 32^2 to 64^2 (ablation) | CoMLSim · 2-D linear Poisson | mean absolute error: 0.015 -> 0.011 | Improved |
| subdomain resolution 64^2 to 128^2 (ablation) | CoMLSim · 2-D linear Poisson | mean absolute error: 0.011 -> 0.029 | Hurt |
| subdomain resolution 64^2 to 128^2 (ablation) | CoMLSim · 2-D linear Poisson | iterations to convergence: 162 -> 89 | Improved |
| use Gauss Seidel instead of Point Jacobi (ablation) | CoMLSim · 2-D linear Poisson | time to converge (seconds): 4.3 -> 72 | Hurt |
| use Gauss Seidel instead of Point Jacobi (ablation) | CoMLSim · 2-D linear Poisson | mean absolute error: 0.013 -> 0.017 | No clear effect |
| replace multiple subdomains by a single subdomain (no local learning) (ablation) | CoMLSim · 2-D linear Poisson | mean absolute error: 0.011 -> 0.072 | Hurt |
| CoMLSim instead of FNO (best baseline) | 2-D linear Poisson · 2-D linear Poisson | L1 error: 0.031 -> 0.011 | Improved |
| CoMLSim instead of UNet on chip cooling | 3-D chip cooling · 3-D chip cooling | L-infinity error: 95.21 -> 15.2 | Improved |
| initialize with coarse-grid solution instead of zero or noise (ablation) | CoMLSim · 2-D linear Poisson | convergence speed | Improved |
| increase out-of-distribution source Gaussians up to 60 | CoMLSim and baselines · linear and non-linear Poisson | error | Hurt |
| scale grid from 1024^2 up to 8192^2 without retraining | CoMLSim · 2-D linear Poisson | error | No clear effect |
Closest papers by meaning
- P$^2$C$^2$Net: PDE-Preserved Coarse Correction Network for efficient prediction of spatiotemporal dynamics (NeurIPS 2024 · similarity 0.88)
- Compute-optimal data scaling for neural surrogates via multi-fidelity training (NeurIPS 2026 · similarity 0.87)
- Mesh-Free Convolution: Learned Spectral Attenuation and Transport (NeurIPS 2026 · similarity 0.86)
- MultiPDENet: PDE-embedded Learning with Multi-time-stepping for Accelerated Flow Simulation (ICML 2025 · similarity 0.86)
- CALM-PDE: Continuous and Adaptive Convolutions for Latent Space Modeling of Time-dependent PDEs (NeurIPS 2025 · similarity 0.86)
- Physics-informed Reduced Order Modeling of Time-dependent PDEs via Differentiable Solvers (NeurIPS 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.
