Data-Driven Discovery and Rigorous Certification of Periodic Solutions in Infinite-Dimensional Dynamical Systems
In this work, we consider a standard regression technique for time series analysis which fits a low-dimensional proxy system to a higher (even infinite) dimensional target model. We then propose a novel diagnostic that exploits a-posteriori computer-assisted proofs, and try to certify the existence of a true periodic solution of the target, nearby. When successful, this certification procedure provides mathematically rigorous information about how well an invariant structure in the proxy approximates a corresponding invariant structure in the target model. To illustrate the utility of the procedure, we develop surrogates using time series data from the Mackey--Glass delay differential equation, and for the Kuramoto--Sivashinsky (KS) partial differential equation, using periodic boundary conditions for forecasting and Neumann boundary conditions for certification. In both test cases we extract invariant dynamics from low-dimensional proxies and use this to validate the existence of corresponding structures in the targets.
Authors
- Aaron Loh (ORCID: https://orcid.org/0009-0002-3375-7563)
- J. D. Mireles James
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-12
- DOI
- https://doi.org/10.48550/arxiv.2611.16706
- Primary Topic
- Model Reduction and Neural Networks
- Type
- preprint