Geometric Dictionary Learning of Dynamical Systems with Optimal Transport
Abstract
Learning dynamical systems through operator-theoretic representations provides a powerful framework for analyzing complex dynamics, as spectral quantities such as eigenvalues and invariant structures encode characteristic time scales and long-term behavior. However, dynamical operators are typically estimated independently for each system, preventing the discovery of shared structure across related dynamics. To address this limitation, we posit that related dynamical systems lie near a low-dimensional manifold in spectral operator space. Based on this hypothesis, we introduce DOODL (Dynamical OperatOr Dictionary Learning), a framework that learns a dictionary of characteristic spectral dynamics whose combinations approximate this manifold and yield compact, interpretable embeddings of individual systems. Beyond representation learning, DOODL enables fast and interpretable operator estimation from short and partially observed trajectories by constraining the estimation to the learned operator manifold. Experiments on metastable Langevin dynamics and turbulent plasma simulations demonstrate that DOODL scales to highly complex multiscale regimes while capturing characteristic spectral structure governing the dynamics rather than merely fitting trajectories, achieving errors one to two orders of magnitude lower than independent operator estimation methods in challenging low-data regimes.
Related
- Sparse Dictionary Learning for Measure-Valued Data via Linearized Optimal Transport: Efficient Algorithms and Sample Complexity
- Geometric Hybrid Dynamical Systems: Part I - Modeling and Stability
- Geometric Methods for Stochastic Dynamical Systems: A Concise Introduction
- Geometric Stability of Dynamical Systems with Self-Generated State-Space Memory
- From Gradient Descent to Potential Theory: A Geometric Dictionary for Binary Classification
BibTeX
@misc{germain2026geometric,
title = {{Geometric Dictionary Learning of Dynamical Systems with Optimal Transport}},
author = {Thibaut Germain and Sami Chemlal and Rémi Flamary and Vladimir R. Kostic and Karim Lounici},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2605.18276},
eprint = {2605.18276},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.18276},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Geometric Dictionary Learning of Dynamical Systems with Optimal Transport
AU - Thibaut Germain
AU - Sami Chemlal
AU - Rémi Flamary
AU - Vladimir R. Kostic
AU - Karim Lounici
PY - 2026
DA - 2026/09/08
PB - arXiv
DO - 10.48550/arxiv.2605.18276
UR - https://arxiv.org/abs/2605.18276
AB - Learning dynamical systems through operator-theoretic representations provides a powerful framework for analyzing complex dynamics, as spectral quantities such as eigenvalues and invariant structures encode characteristic time scales and long-term behavior. However, dynamical operators are typically estimated independently for each system, preventing the discovery of shared structure across related dynamics. To address this limitation, we posit that related dynamical systems lie near a low-dimensional manifold in spectral operator space. Based on this hypothesis, we introduce DOODL (Dynamical OperatOr Dictionary Learning), a framework that learns a dictionary of characteristic spectral dynamics whose combinations approximate this manifold and yield compact, interpretable embeddings of individual systems. Beyond representation learning, DOODL enables fast and interpretable operator estimation from short and partially observed trajectories by constraining the estimation to the learned operator manifold. Experiments on metastable Langevin dynamics and turbulent plasma simulations demonstrate that DOODL scales to highly complex multiscale regimes while capturing characteristic spectral structure governing the dynamics rather than merely fitting trajectories, achieving errors one to two orders of magnitude lower than independent operator estimation methods in challenging low-data regimes.
ER -