Dr. Caroline Wormell | Spectral error of EDMD (and other operators)

Item: Dr. Caroline Wormell | Spectral error of EDMD (and other operators)
Rating: 4
Author: PostSingularity Institute

by Dr. Caroline Wormell

Formal & Physical Sciences Mathematics PBKCalculus and mathematical analysis PBKSNumerical analysis

🎙 Dr. Caroline Wormell 👥 2K 📅 August 11, 2026 ⏱ 67 min 👁 39 📄 original study 🧭 2026-08-15

Available in: English (current) Français

Keywords

EDMDKoopman operatorspectral errordynamical systemsdata-driven methods

Summary

This seminar by Dr. Caroline Wormell, presented at the Isaac Newton Institute, addresses the spectral error in Extended Dynamic Mode Decomposition (EDMD) and other operator approximation methods. The talk is divided into two main parts: first, a theoretical framework for understanding the sources of error in approximating the Koopman operator, and second, an application to a specific class of chaotic systems. The speaker introduces the concept of Ruelle-Pollicott resonances as the spectral objects of interest and explains how EDMD aims to approximate them. The core challenge is that EDMD involves two approximations: a projection onto a finite-dimensional function space (dictionary error) and a finite-sample estimation (sampling error). The talk focuses on the sampling error, which is more general and can be analyzed through matrix perturbation theory. The speaker presents a unified error bound for the sampling error, showing that it scales as O(1/sqrt(m)) with the number of data points, but with a constant that may depend on the dictionary size. For the dictionary error, the talk focuses on uniformly expanding maps of the circle, where it is shown that the error decays exponentially with the dictionary size. The speaker also discusses the use of a Hilbert space of analytic functions to achieve this exponential convergence. The talk concludes with a discussion of how to rule out spurious eigenvalues using a norm bound on a certain matrix expression, leading to a spectral exclusion region. The presentation is highly technical and assumes a strong background in dynamical systems and functional analysis.

249 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides a rigorous and novel contribution to the understanding of spectral errors in EDMD. The speaker clearly identifies the two sources of error (dictionary and sampling) and provides a unified framework for analyzing the sampling error. The argumentation is solid, building on established results and extending them to a more general setting. The use of a specific class of maps (uniformly expanding) allows for concrete results with exponential convergence, which is a strong point. The presentation is well-structured, with clear definitions and derivations. However, the talk is highly technical and may be inaccessible to non-specialists, but this is appropriate for the audience. The value lies in the theoretical guarantees provided, which are often lacking in data-driven methods.

Scientific Rigor, Source Quality, Title Accuracy

The talk is scientifically rigorous, with a clear mathematical framework and proofs. The speaker references prior work by herself and colleagues (e.g., Julia and Oscar) and builds on known results in transfer operator theory. The sources cited are primarily academic and are not explicitly listed in the description, but the talk is part of a formal seminar series at the Isaac Newton Institute, which lends credibility. The title accurately reflects the content, focusing on the spectral error of EDMD and other operators. The talk does not include any commercial or promotional content. The description provides links to the institute and the specific seminar page, which are relevant for further information.

244 words

Title / Content Match

The title accurately reflects the content: the talk focuses on the spectral error of Extended Dynamic Mode Decomposition (EDMD) and related operators, with a rigorous mathematical treatment.

Quality & Reliability

8/10

The talk presents original research with rigorous mathematical proofs, set in a formal academic context (INI seminar). The speaker is a recognized researcher. The content is highly technical and assumes advanced knowledge, but the methodology is sound and the presentation is clear for the intended audience.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Isaac Newton Institute for Mathematical Sciences — The talk was given at the Isaac Newton Institute, and the website provides information about the institute and its programs.
Seminar page for the talk — The specific seminar page for this talk, providing details about the event and possibly related materials.

Concurring Sources

Isaac Newton Institute for Mathematical Sciences — The talk is hosted by the institute, which is a reputable research center.

Contribution & Novelties

The talk provides a novel unified analysis of the spectral error in EDMD, separating the contributions of dictionary and sampling errors. The main contribution is a general bound for the sampling error that applies to a wide class of operators, and a specific exponential convergence result for the dictionary error in the case of uniformly expanding maps. This extends previous work by the speaker and colleagues, which was limited to Lebesgue measure, to more general invariant measures.

Pour aller plus loin :

Koopman operator — Foundational concept in dynamical systems theory, central to the talk.
Dynamic mode decomposition — Related data-driven method, EDMD is an extension.
Ruelle-Pollicott resonances — The spectral objects of interest in the talk.
Transfer operator — Related operator theory, used in the analysis.
Central limit theorem for dynamical systems — Assumption used in the talk for statistical properties.

142 words

Radar Profile

The radar profile shows high scores in quantitative information, qualitative information, and technical level, reflecting the dense and rigorous mathematical content. The global reliability score is also high, indicating a trustworthy source. The overall note of 4 stars is consistent with the high technical quality and originality of the research.

Reliability 8/10