Oscillatory State-Space Models as Inductive Biases for Physics-Informed Neural PDE Solvers
Abhishek ChandraKTH Royal Institute of TechnologyStockholm, Sweden, Taniya Kapoor Wageningen University & ResearchWageningen, The Netherlands [email protected] · 2026
Read Oscillatory State-Space Models as Inductive Biases for Physics-Informed Neural PDE Solvers by Abhishek ChandraKTH Royal Institute of TechnologyStockholm, Sweden, Taniya Kapoor Wageningen University & ResearchWageningen, The Netherlands [email protected] (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-2606-02623.txt.
Description
Chandra and Kapoor add structured time dynamics to physics-informed neural PDE solvers. Such solvers learn solutions from the governing equations. Capturing time evolution accurately stays hard. Sequence models can parameterize time, but they are general-purpose. They do not encode the modal structure of PDE solutions. Their memory also scales poorly with sequence length and resolution.
The new approach builds in oscillatory state-space dynamics. A linear oscillator drives time evolution. A PDE-aware spectral basis handles space. Spatial differentiation then has closed forms. Boundary conditions are enforced consistently. Tests cover forward, inverse, and high-dimensional problems. Cases reach 100 spatial dimensions. Accuracy improves over recent sequence-model solvers. Memory use falls as well. The authors argue for physics-aligned designs in neural PDE solvers.
Catalog record
| CREATOR | Abhishek ChandraKTH Royal Institute of TechnologyStockholm, Sweden, Taniya Kapoor Wageningen University & ResearchWageningen, The Netherlands [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02623