On the Role of Consistency Between Physics and Data in Physics-Informed Neural Networks
Nicolás Becerra-Zuniga, Lucas Lacasa, Eusebio Valero, Gonzalo Rubio · 2026
Read On the Role of Consistency Between Physics and Data in Physics-Informed Neural Networks by Nicolás Becerra-Zuniga, Lucas Lacasa, Eusebio Valero, Gonzalo Rubio (2026) 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-2602-10611.txt.
Description
Nicolás Becerra-Zuniga and three co-authors ask how much physics-informed neural networks can learn from imperfect data in a 2026 paper. These networks solve partial differential rules by combining scarce labels with real constraints. In practice, however, training data carry measurement noise, discretization error, and modeling assumptions that clash with the governing rules.
The paper systematically analyzes that clash and presents a consistency barrier: an intrinsic lower bound on error arising from data-to-equation mismatch. No amount of training can beat the barrier while the inconsistency stands. The result reframes expectations for physics-informed learning, separating errors the model can fix from errors baked into the data itself.
Catalog record
| CREATOR | Nicolás Becerra-Zuniga, Lucas Lacasa, Eusebio Valero, Gonzalo Rubio |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-10611