🔬 condensed matter

Bridging the Gap between Koopmanism and Response Theory: Using Natural Variability to Predict Forced Response

This paper demonstrates how combining the fluctuation-dissipation theorem with Koopman analysis and the Extended Dynamical Mode Decomposition (EDMD) algorithm enables the accurate prediction of a system's forced response to perturbations by deconstructing response operators into interpretable modes derived from natural variability.

Original authors: Niccolò Zagli, Matthew Colbrook, Valerio Lucarini, Igor Mezić, John Moroney

Published 2026-06-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Niccolò Zagli, Matthew Colbrook, Valerio Lucarini, Igor Mezić, John Moroney

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict how a complex machine, like a weather system or a chaotic pendulum, will react if you give it a little push. Usually, to know this, you have to actually push it and watch what happens. But what if you could predict the reaction just by watching the machine wiggle around on its own, without ever touching it?

This paper is about a new way to do exactly that. It bridges two powerful mathematical ideas to create a "crystal ball" for predicting how complex systems respond to changes, using only data from their natural, unforced movements.

Here is the breakdown of their approach using simple analogies:

1. The Problem: The "Black Box" of Complexity

Complex systems (like the climate or turbulent fluids) are messy. They have many moving parts interacting in unpredictable ways.

The Old Way: To see how the system reacts to a change (like a temperature rise), scientists used a formula called the Fluctuation-Dissipation Theorem (FDT). Think of this as a recipe that says, "If you know how the system jiggles naturally, you can calculate how it will move when pushed."
The Flaw: While the old recipe works, it's like looking at a smoothie. You know the ingredients are in there, but you can't tell which specific fruit (strawberry vs. banana) is responsible for the sweetness. It doesn't tell you which part of the system is reacting. It lacks "interpretability."

2. The New Tool: "Koopmanism" (The Magic Lens)

The authors introduce a concept called Koopmanism.

The Analogy: Imagine a chaotic dance floor where everyone is moving randomly. It looks impossible to predict. But Koopmanism is like a special pair of glasses that changes your perspective. Instead of watching the dancers (the system's state), you watch the patterns of the music (observables).
The Magic: Even though the dancers are moving non-linearly (chaotically), the patterns of their movement follow simple, straight-line rules (linear rules). This allows mathematicians to break the complex chaos down into a set of distinct "modes" or "vibrations," much like how a guitar string vibrates in specific harmonics.

3. The Breakthrough: Deconstructing the Reaction

The paper's main achievement is combining the "recipe" (FDT) with the "magic lens" (Koopmanism).

The Result: Instead of getting a blurry smoothie of a prediction, they can now say: "If you push the system, 40% of the reaction comes from the 'slow, deep rumble' mode, and 60% comes from the 'fast, high-pitched squeak' mode."
Why it matters: This makes the prediction interpretable. You can see exactly which internal "vibrations" of the system are driving the response. This helps scientists understand why a system might be sensitive to a change or when it might be approaching a "tipping point" (a sudden, catastrophic shift).

4. How They Did It: The "EDMD" Algorithm

To make this work with real data (not just theory), they used a computer algorithm called Extended Dynamic Mode Decomposition (EDMD).

The Analogy: Imagine you have a video of a chaotic system moving naturally. You don't know the equations governing it. You feed this video into the EDMD algorithm.
The Process: The algorithm acts like a super-powered microscope. It looks at the video frames, identifies the hidden "modes" of vibration (the eigenvalues and eigenvectors), and then uses those modes to reconstruct the "Green's function."
The Green's Function: Think of this as the system's "fingerprint of reaction." It tells you exactly how the system responds to a push at any given time.

5. The Proof: Three Test Cases

The authors tested this method on three different mathematical models to prove it works:

A Chaotic Map (The Spinning Top): A simple, one-dimensional system that spins wildly. They showed the method could predict how it reacts to a uniform push.
A Stretched Cat Map (The Fractal Cat): A two-dimensional system that stretches and folds space like dough. This one is tricky because its natural movement is concentrated in a "fractal" shape (like a crumpled piece of paper). Even here, the method worked, successfully predicting responses to complex, uneven pushes.
A Double-Well Potential (The Valley Hopper): A system that sits in one of two valleys and occasionally jumps to the other. This is a classic equilibrium system. They showed the method could distinguish between the "fast" movements within a valley and the "slow" movements of jumping between valleys.

The Bottom Line

The paper claims that by using Koopman analysis and the EDMD algorithm, we can take a single long video of a complex system moving naturally, break it down into its fundamental "vibrational modes," and use that information to accurately predict how the system will react to future forces.

This doesn't just give a number; it gives a story. It tells us which specific internal rhythms of the system are responsible for the reaction, offering a much clearer view of how complex systems behave and how close they might be to a critical change.

Technical Summary: Bridging the Gap between Koopmanism and Response Theory

Problem Statement
The central challenge addressed is the prediction of a complex system's forced response to perturbations using only data from its natural, unperturbed variability. While the Fluctuation-Dissipation Theorem (FDT) provides a theoretical foundation for linking free fluctuations to forced responses, particularly near equilibrium, it often lacks interpretability. Specifically, standard response formulas do not decompose the system's response into contributions from specific modes of natural variability. This limitation hinders the identification of feedback mechanisms across scales and the understanding of critical behaviors, such as tipping points, where response operators diverge due to slow correlation decay. Furthermore, in dissipative chaotic systems far from equilibrium, the classical FDT is not strictly valid, and constructing rigorous linear response theories is mathematically difficult.

Methodology
The authors propose a framework that synthesizes Linear Response Theory with Koopman Operator Theory (Koopmanism). The core methodology involves:

Operator Formalism: The system is modeled as a stochastic process governed by Stochastic Differential Equations (SDEs). The statistical properties are analyzed via the Markov semigroup $P_t$ and its generator, the Kolmogorov operator $K_0$ . The response of an observable to a perturbation is expressed via Green's functions, which are shown to be related to time-lagged correlation functions of the unperturbed system.
Spectral Decomposition: The authors utilize the spectral properties of $K_0$ (specifically the Ruelle-Pollicott resonances) to decompose correlation and Green's functions into sums of exponentially decaying terms. Each term corresponds to a specific eigenmode of the system, thereby linking the response directly to intrinsic relaxation rates and oscillatory modes.
Data-Driven Estimation (EDMD): To compute these spectral properties from data without knowing the governing equations, the paper employs the Extended Dynamic Mode Decomposition (EDMD) algorithm. EDMD approximates the action of the Kolmogorov operator on a finite-dimensional subspace spanned by a dictionary of observable functions.
Reconstruction Strategies: The paper details three methods to reconstruct response operators from EDMD outputs:
- Method 1 & 2: Direct estimation of the invariant density $\rho_0$ and the perturbation operator action to compute the response observable $\Gamma$ (or $\gamma$ ).
- Method 3: A more robust approach that estimates the projection of the response observable directly from time-series data using ergodic averages, avoiding the need for explicit density estimation and regularization of singular distributions.

Key Contributions

Theoretical Synthesis: The paper establishes a rigorous link between the spectral decomposition of the Kolmogorov operator and the Green's function of linear response theory. It demonstrates that the response operator can be deconstructed into a sum of terms, each associated with a specific mode of natural variability (eigenfunction of $K_0$ ).
Algorithmic Implementation: It provides explicit formulas for computing correlation functions and Green's functions using EDMD, translating abstract operator theory into practical numerical recipes.
Validation on Diverse Systems: The framework is tested on three distinct mathematical models:
1. A one-dimensional uniformly expanding chaotic map (deterministic).
2. A stochastically forced, two-dimensional generalized Arnold's cat map (chaotic with a singular invariant measure).
3. A two-dimensional stochastic gradient flow with a double-well potential (equilibrium system).

Results

Accuracy: In all three test cases, the EDMD-reconstructed correlation functions and Green's functions show excellent agreement with direct numerical simulations.
Interpretability: The spectral decomposition successfully isolates the contributions of different eigenmodes. For the double-well potential, the method clearly separates the slow inter-well dynamics (associated with the subdominant eigenvalue) from fast intra-well dynamics.
Robustness: The study demonstrates that the spectral reconstruction is robust to variations in trajectory length and dictionary size. Notably, Method 3 allows for accurate response estimation with relatively small datasets (e.g., $M=500$ points) compared to the massive ensembles ( $10^6$ ) required for direct numerical response experiments.
Handling Singular Measures: The paper shows that while estimating response for systems with fractal attractors (singular measures) is challenging due to density estimation issues, alternative formulations (Method 2 and 3) can successfully capture the response properties.

Significance and Claims
The authors claim this work provides the first evidence that Koopman analysis can be effectively used to interpret and predict the response of complex systems to perturbations. By decomposing the response into specific modes of variability, the method offers a significant improvement in interpretability over standard FDT approaches. This capability is crucial for:

Assessing the resilience of complex systems.
Identifying the physical processes responsible for critical transitions and tipping points.
Understanding how specific forcings resonate with underlying modes of variability.

The paper maintains a modest tone regarding future implications, acknowledging that the method currently assumes full state observation. It suggests that extending this framework to coarse-grained or partially observed systems (potentially using time-delay embeddings or Markov state models) is a necessary direction for future work, rather than claiming immediate applicability to all partially observed real-world scenarios. The authors emphasize that while EDMD provides strong convergence properties, rigorous error control (e.g., via ResDMD) is generally recommended to ensure the physical meaningfulness of the spectral terms.