Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Deep Equilibrium Based Neural Operators for Steady-State PDEs
Tanya Marwah, Ashwini Pokle, J Zico Kolter, Zachary Chase Lipton, Jianfeng Lu, Andrej Risteski
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 |
|---|---|---|---|
| weight-tie the FNO block (FNO-WT, 1 block) instead of unique FNO++ layers | FNO++ · Darcy Flow clean | relative L2 error: 0.0064 -> 0.0055 | Improved |
| solve for the fixed point with a root solver (FNO-DEQ) instead of unrolling weight-tied layers | FNO-WT · Navier-Stokes nu=0.001 clean | relative L2 error: 0.151 -> 0.128 | Improved |
| replace FNO-WT with FNO-DEQ | FNO-WT · Darcy Flow | relative L2 error: 0.0055 -> 0.0055 | No clear effect |
| weight-tied FNO-DEQ (1 block) vs FNO++ with 4 blocks (4x parameters) | FNO++ · Navier-Stokes nu=0.01 clean | relative L2 error: 0.130 -> 0.088 | Improved |
| increase FNO depth from 2 to 4 blocks | FNO · Darcy Flow clean | relative L2 error: 0.0105 -> 0.2550 | Hurt |
| increase FNO++ depth from 1 to 4 blocks | FNO++ · Darcy Flow clean | relative L2 error: 0.0075 -> 0.0064 | Improved |
| increase FNO++ depth from 1 to 4 blocks | FNO++ · Navier-Stokes nu=0.001 clean | relative L2 error: 0.199 -> 0.151 | Improved |
| use shallow FNO-DEQ (single FNO layer per block) instead of deep (ablation) | FNO-DEQ · Darcy Flow noisy observations | relative L2 error: 0.0112 -> 0.0136 | Hurt |
| use shallow FNO-WT instead of deep (ablation) | FNO-WT · Darcy Flow clean | relative L2 error: 0.0055 -> 0.0057 | No clear effect |
| use shallow FNO++ instead of deep FNO++ 1 block (ablation) | FNO++ · Darcy Flow noisy observations | relative L2 error: 0.0145 -> 0.0402 | Hurt |
| add noise to inputs (var 0.001) in training | FNO-DEQ · Navier-Stokes nu=0.001 | relative L2 error: 0.128 -> 0.144 | Hurt |
| increase fixed-point solver steps from 2 to 4 to 8 at Darcy (ablation) | FNO-DEQ · Darcy Flow clean | test error: 0.0777 -> 0.00567 | Improved |
Closest papers by meaning
- Derivative-enhanced Deep Operator Network (NeurIPS 2024 · similarity 0.88)
- Global convergence of adjoint-optimized neural PDEs (NeurIPS 2026 · similarity 0.87)
- RIGNO: A Graph-based Framework For Robust And Accurate Operator Learning For PDEs On Arbitrary Domains (NeurIPS 2025 · similarity 0.87)
- BENO: Boundary-embedded Neural Operators for Elliptic PDEs (ICLR 2024 · similarity 0.87)
- Active Learning for Neural PDE Solvers (ICLR 2025 · similarity 0.87)
- Pretraining Codomain Attention Neural Operators for Solving Multiphysics PDEs (NeurIPS 2024 · similarity 0.87)
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.
