Physics-Informed Inference Time Scaling for Solving High-Dimensional PDE via Defect Correction
Zexi Fan 1 , Yan Sun 2 , Shihao Yang 3 , Yiping Lu* 4 1 Peking University 2 Visa Inc. 3 Georgia Institute of Technology 4 Northwestern University [email protected],[email protected], [email protected],[email protected] · 2025
Read Physics-Informed Inference Time Scaling for Solving High-Dimensional PDE via Defect Correction by Zexi Fan 1 , Yan Sun 2 , Shihao Yang 3 , Yiping Lu* 4 1 Peking University 2 Visa Inc. 3 Georgia Institute of Technology 4 Northwestern University [email protected],[email protected], [email protected],[email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-16172.txt.
Introduction
Zexi Fan and co-authors wrote a paper from 2025 on fixing the answers of trained solvers for hard partial differential equations, without retraining them. Data-driven solvers are fast but unreliable in high dimensions and come with weak error guarantees. The team's framework, SCaSML, treats the trained solver's error itself as a new equation. That defect equation keeps the shape of the original problem, so classic random simulators can solve it cheaply, and its solution patches the first answer.
The authors prove the patched error shrinks at a faster rate, ending bounded by the product of the two methods' errors. On test equations up to one hundred sixty dimensions, the patch cut the error of several solver types by twenty to eighty percent. Neural solvers and Gaussian process solvers both gained. Code is public. The pitch is reliability bought at run time rather than in training.
Catalog record
| CREATOR | Zexi Fan 1 , Yan Sun 2 , Shihao Yang 3 , Yiping Lu* 4 1 Peking University 2 Visa Inc. 3 Georgia Institute of Technology 4 Northwestern University [email protected],[email protected], [email protected],[email protected] |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 171 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-16172