Fisher Ng | Searcharxiv

arXiv2026

Stable Takens' Embedding Theorem for Non-Uniformly-Sampled Linear Systems

Takens' time-delay embedding theorem provides conditions under which delay-coordinate maps, formed using uniformly-sampled time series of trajectories evolving on attractors of dynamical systems, can faithfully represent the dynamics of the original system. Nonlinear systems can be highly sensitive, and Takens' theorem does not provide guarantees about the stability of time-delay embeddings. In the linear setting, statements about the stability of time-delay embeddings are more tractable and have been proven for delay-coordinate maps with evenly-spaced delays. In many experimental applications, however, time series data may be non-uniformly-sampled, especially in systems with multiple timescales or when using event-based rather than time-based sampling techniques. In this paper, we extend the theorems for the stable linear Takens' embeddings to the setting where the delay-coordinate maps involve unevenly-spaced delays. We pose a conjecture about the rank of generalized Vandermonde matrices that capture the temporal structure of time-delay embeddings. We prove that, provided the conjecture holds, existing theorems about stable linear Takens' embeddings readily extend to unevenly-sampled settings, and the quality of the embedding converges to the same asymptotic bounds when using a large number of delays in the delay-coordinate map, a result which is supported by numerical simulations.

math.DS↗

arXiv2026

Linear Recurrent Neural Networks as Time-Delay Embeddings

Sequence models, and particularly Linear Recurrent Neural Networks (LRNNs) of the form $\mathbf{h}_{k+1} = \mathbf{W} \mathbf{h}_{k} + \mathbf{y}_k + \mathbf{b}$, are widely applicable in time-series analysis for dynamical systems, yet, as black-box algorithms, much is unknown about why they perform well. In this work, we leverage Takens' embedding theorem, which provides conditions under which partially observed time series organized into delay-coordinate vectors can faithfully represent the original system's dynamics, as a theoretical framework for explaining how and why sequence models preserve and reconstruct dynamical systems. For LRNNs, concatenating output states into delay-coordinate vectors gives rise to a ``delay" matrix $\mathbb{M}_{n,m}\in \mathbb{C}^{(nm) \times (n+1)m}$: a block matrix consisting of identity matrices $\mathbf{I} \in \mathbb{R}^{m \times m}$ repeated $n$ times along the main diagonal and weight matrices $\mathbf{W} \in \mathbb{C}^{m \times m}$ featured $n$ times along the super-diagonal. $\mathbb{M}_{n,m}$ relates the delay-coordinates of the input time series to those of the LRNN output states, and, for $\mathbb{M}_{n,m}$ to be an embedding, it must be full row-rank. We provide explicit conditions for $\mathbb{M}_{n,m}$ to be full row-rank and prove the condition number of $\mathbb{M}_{n,m}$ and determinant of $\mathbb{M}_{n,m} \mathbb{M}_{n,m}^*$--measures of embedding stability--are bounded independent of $n$, at least for certain ranges of $\mathbf{W}$'s singular values: namely, when $\sigma_{\max}(\mathbf{W}) \le 1$. This result explains why the spectrum of $\mathbf{W}$ for trained LRNNs tends to converge to within the unit circle.

math.DS↗