Research map / Architectures and efficiency / RNNs and normalization
PGD-NO: A Neural Operator with Precomputed Geometry Decomposition for 3D Million-Scale Physics Simulations
Weiheng Zhong, Jing Bi, Victor Oancea, Hadi Meidani
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 precomputed surface geometry tokens instead of grid-pooling tokens (ablation) | PGD-NO-v1 · Heat Sink | L2 relative error (%): 9.43 -> 2.47 | Improved |
| add precomputed surface geometry tokens instead of grid-pooling tokens (ablation) | PGD-NO-v1 · JEB | L2 relative error (%): 62.5 -> 37.2 | Improved |
| add precomputed surface geometry tokens instead of grid-pooling tokens (ablation) | PGD-NO-v1 · Aircraft | L2 relative error (%): 35.4 -> 18.2 | Improved |
| add precomputed surface geometry tokens instead of grid-pooling tokens (ablation) | PGD-NO-v1 · CFD-VOL | L2 relative error (%): 36.7 -> 12.2 | Improved |
| increase geometry token layers from 2 to 12 (ablation) | PGD-NO-v1 · CFD-VOL | L2 relative error (%): 17.9 -> 11.1 | Improved |
| increase geometry token layers from 2 to 12 (ablation) | Transolver++ · CFD-VOL | L2 relative error (%): 17.8 -> 15.1 | Improved |
| increase number of geometry tokens from 32 to 64 (ablation) | PGD-NO · JEB | L2 relative error (%): 65.4 -> 41.5 | Improved |
| increase number of geometry tokens from 128 to 256 (ablation) | PGD-NO · JEB | L2 relative error (%): 37.2 -> 35.4 | Improved |
| add attention layer on geometry tokens (v2 vs v1) | PGD-NO · JEB | L2 relative error (%): 39.4 -> 37.2 | Improved |
| add attention layer on geometry tokens (v2 vs v1) | PGD-NO · Aircraft | L2 relative error (%): 4.77 -> 5.37 | Hurt |
| use transposed segmentation matrix hierarchical mapping (v3 vs v1) | PGD-NO · Aircraft | L2 relative error (%): 4.77 -> 4.66 | Improved |
| compare PGD-NO-v1 with Transolver++ at 12M points, full sampling | PGD-NO-v1 · CFD-VOL (12M points) | L2 relative error (%): 15.8 -> 12.2 | Improved |
Closest papers by meaning
- From Cheap Geometry to Expensive Physics: Elevating Neural Operators via Latent Shape Pretraining (ICLR 2026 · similarity 0.90)
- Transolver++: An Accurate Neural Solver for PDEs on Million-Scale Geometries (ICML 2025 · similarity 0.88)
- Geometry Aware Operator Transformer as an efficient and accurate neural surrogate for PDEs on arbitrary domains (NeurIPS 2025 · similarity 0.88)
- 3DID: Direct 3D Inverse Design for Aerodynamics with Physics-Aware Optimization (NeurIPS 2025 · similarity 0.86)
- Learning a Neural Solver for Parametric PDEs to Enhance Physics-Informed Methods (ICLR 2025 · similarity 0.85)
- Neural Physical Simulation with Multi-Resolution Hash Grid Encoding (AAAI 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.
