Spectral Approximation for Chaotic and Mixing Systems

This direction is split into other directions: continuous spectrum analysis. Its last version stays readable and citable here.

Version 1 of 133 papers45 conceptsfrozen 2026-10-04

In the study of nonlinear dynamics, the Koopman operator offers a tantalizing promise: we can trade the difficulty of nonlinear equations for the simplicity of linear ones. By "lifting" a system’s state into a higher-dimensional space of observables, we can represent complex evolution as a linear transformation.

However, for decades, this approach hit a mathematical wall when applied to the most interesting systems in nature—those that are chaotic, turbulent, or stochastic. The problem lies in the spectrum. While simple systems have a "point spectrum" (discrete eigenvalues), chaotic systems possess a continuous spectrum. Traditional data-driven methods, designed to find discrete notes, fail to capture the "roar" of the continuum.

The Discrete Wall

Most popular algorithms for system identification, such as Dynamic Mode Decomposition (DMD), assume that a system's behavior can be decomposed into a finite set of frequencies and growth rates. As Shaowu Pan, Steven L. Brunton, and colleagues note in their work on PyKoopman, these frameworks struggle with chaotic systems because they rely on the assumption of a discrete, pointwise spectrum.

Imagine trying to describe a symphony using only five specific notes. If the music stays within those notes, your model is perfect. But if the music includes a violin slide or the crashing white noise of a cymbal, your five-note model will produce "spectral pollution"—spurious eigenvalues that don't correspond to physical reality. This is exactly what happens when we apply standard Koopman methods to turbulent fluid flows or weather patterns. As Yoshihiko Susuki and Atsushi Okazaki demonstrate in Extracting transient Koopman modes from short-term weather simulations, sharp transitions and sudden bursts in data are often signatures of the continuous spectrum that finite sets of discrete modes simply cannot capture.

From Eigenvalues to Spectral Measures

To solve this, researchers are shifting from finding individual eigenvalues to approximating spectral measures. Instead of asking "What are the exact frequencies?", they ask "What is the density of frequencies across the spectrum?"

Alex Townsend and Matthew J. Colbrook pioneered this shift in Rigorous data-driven computation of spectral properties of Koopman operators. Rather than identifying discrete points, their algorithms compute smoothed approximations of spectral measures. This allows them to represent the "density" of the operator's spectrum, providing a mathematically rigorous way to model systems where energy is spread across a continuum of frequencies rather than concentrated in a few modes.

This perspective is essential for stochastic systems. In Mean Field Control by Stochastic Koopman Operator, Yuhan Zhao and co-authors argue that including the "absolutely continuous component" of the spectral measure is the only way to accurately capture the inherent randomness of real-world multi-agent systems. Without it, the model ignores the very stochasticity it aims to control.

New Mathematical Foundations

Capturing the continuum requires moving beyond standard Hilbert spaces ($L^2$), where the continuous spectrum lacks proper eigenfunctions. Researchers are reaching for more sophisticated mathematical toolkits:

  1. Rigged Hilbert Spaces (Gelfand Triplets): Hayato Chiba and colleagues use these to uncover "hidden" generalized eigenvalues in symbolic dynamical systems, allowing them to calculate the exact rate at which a system "mixes" or reaches equilibrium.
  2. Reproducing Kernel Hilbert Spaces (RKHS): Nicolas Boullé and Matthew J. Colbrook have developed convergent methods that leverage the structure of RKHS to bypass the need for massive datasets, providing optimal computational complexity for spectral approximation.
  3. Hilbert $C^*$-modules: In Koopman spectral analysis of skew-product dynamics, Dimitrios Giannakis and collaborators replace scalar eigenvalues with "eigenoperators." This allows them to decompose systems with complex internal structures, like measure-preserving fluid flows, more effectively than any scalar-based method.

Learning the Continuum

The latest frontier involves neural networks designed specifically to "hunt" for these continuous structures. The ResKoopNet framework, for instance, doesn't just filter noise; it minimizes spectral residuals to actively search for the complete spectrum. This makes it uniquely capable of resolving the complex spectral densities found in turbulence and neural signals.

Similarly, Daniel Dylewsky and J. Nathan Kutz introduced Principal Component Trajectories (PCT), which treats nonlinear dynamics as a superposition of time-delayed orbits. By doing so, they can model systems with continuous spectra as forced linear systems, even when the forcing signal is unknown.

Why It Matters: From Oceans to Weather

This isn't just abstract math; it changes how we understand the physical world. In oceanography, Antoine Venaille and Yohei Onuki showed that disentangling the continuous spectrum is the key to understanding how ocean tides drive deep-water mixing. Without the continuum, models miss the intense velocity gradients that actually move heat and nutrients through the abyss.

In meteorology, researchers like Sebastian Springer are finding that the atmosphere lacks a "spectral gap" between slow and fast processes. Their work on agnostic detection of weather patterns suggests that weather regimes like "blockings" exist as a continuum rather than discrete states.

By embracing the continuous spectrum, we are finally moving past the "five-note" approximation of the world, developing a language that can hear the full, complex roar of nonlinear reality.

Papers in this version

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_spectral_approximation_for_chaotic_and_m_49b9479b,
  title = {Spectral Approximation for Chaotic and Mixing Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spectral-approximation-for-chaotic-and-mixing-systems/rdr_49b9479b712f4e7c83267d482284d88c},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_spectral_approximation_for_chaotic_and_m_49b9479b_v1,
  title = {Spectral Approximation for Chaotic and Mixing Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/spectral-approximation-for-chaotic-and-mixing-systems/rdr_49b9479b712f4e7c83267d482284d88c/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}