Koopman Mode Decomposition of Thermodynamic Dissipation in Nonlinear Langevin Dynamics

This paper employs Koopman mode decomposition to establish a general framework that linearizes nonlinear Langevin dynamics, thereby decomposing thermodynamic dissipation into interpretable, frequency-dependent contributions from individual oscillatory modes and demonstrating how these modes govern energy loss in phenomena like coherent resonance within the noisy FitzHugh-Nagumo model.

The Gist

Imagine you are watching a complex dance performance. The dancers (representing particles in a system like a neuron or a chemical reaction) are moving in rhythmic, swirling patterns. But this isn't a ballet in a quiet studio; it's happening in a hurricane. The wind (noise) is constantly pushing them around, trying to throw them off balance.

To keep dancing in a circle instead of just tumbling chaotically, the dancers must constantly burn energy. In physics, this energy loss is called dissipation.

The big question this paper answers is: How much energy does it take to keep this dance going, and which specific moves are costing the most?

Here is the breakdown of their discovery, using simple analogies.

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

For a long time, scientists could measure the total energy being burned by the system (the total dissipation). But because the system is nonlinear (meaning the rules change depending on how fast or hard the dancers move), it was like looking at a black box. You knew energy was going in and heat was coming out, but you couldn't tell which specific rhythm or pattern was responsible for the cost.

It's like knowing your car is using a lot of gas, but not knowing if it's the engine revving high, the air conditioning running, or the tires dragging on the road.

2. The Solution: The "Koopman Magic Mirror"

The authors used a mathematical trick called Koopman Mode Decomposition.

Think of the dance floor as a chaotic, swirling mess. The Koopman method is like a magic mirror that reflects the dance. In the real world, the dancers move in complicated, curved paths. But in the mirror's reflection, the chaos is transformed into a set of simple, straight-line movements and pure, perfect sine waves.

Instead of seeing one messy dancer, the mirror reveals that the dance is actually a symphony of many different instruments playing at once. Some instruments are playing a slow, deep bass (low frequency), while others are playing a fast, high-pitched violin (high frequency).

3. The Big Discovery: The "Frequency Tax"

Once they broke the dance down into these individual "instruments" (modes), they found a beautiful, simple rule for how much energy each one costs:

Energy Cost = (How Fast You Spin)² × (How Big Your Spin Is)

• Frequency (How Fast): This is the most important part. Because the cost is proportional to the square of the speed, fast rhythms are incredibly expensive. If you double the speed of a rhythm, the energy cost quadruples.
• Intensity (How Big): This is how wide the dance move is. A big, sweeping move costs more than a tiny wiggle.

The Analogy: Imagine you are spinning a hula hoop.

• If you spin it slowly, it's easy.
• If you spin it fast, you have to work much harder (the square law).
• If you make the hoop huge, you also have to work harder.
• The paper proves that you can calculate the exact "fatigue" (entropy production) of the system just by adding up the fatigue of each individual hula hoop spin.

4. Real-World Examples: The Brain and the Heart

The authors tested this on a model of a neuron (the FitzHugh–Nagumo model), which acts like a tiny biological oscillator (like a heartbeat or a nerve firing).

Scenario A: The "Hopf Bifurcation" (The Dance Floor Shrinks)
Imagine the dance floor suddenly gets smaller. The dancers are forced to stop doing big, wild loops and start doing tiny, tight circles.

• Old View: Scientists just saw the total energy drop.
• New View: The authors showed why it dropped. The "fast, big instruments" stopped playing. The system switched to a single, quiet, low-energy instrument. They could see exactly which "notes" in the symphony disappeared as the system changed state.

Scenario B: "Coherent Resonance" (The Goldilocks Zone)
Sometimes, adding a little bit of noise (wind) actually helps the dancers stay in rhythm. This is called coherent resonance.

• Too little wind: The dancers are stuck in one spot.
• Too much wind: The dancers are thrown everywhere.
• Just right: The dancers find a perfect groove.
• The Discovery: The authors found that at this "perfect" noise level, the system doesn't just use one rhythm. It uses a broad spectrum of many different frequencies working together. It's like a full orchestra playing in perfect harmony. When the noise is too high or too low, the orchestra shrinks to just one or two instruments. This explains why the system is most efficient at that specific "sweet spot."

Why Does This Matter?

This paper gives us a new way to look at life. Living things (like our brains and hearts) are full of these noisy, rhythmic oscillations. They have to burn energy to keep going.

This framework allows scientists to say: "Oh, this specific brain rhythm is costing us 40% of our energy budget because it's spinning too fast."

It turns the messy, invisible thermodynamics of life into a clear, readable score sheet, showing us exactly which "moves" are the most expensive and how nature optimizes them. It connects the rhythm of life directly to the cost of living.

Technical Summary

1. Problem Statement

Nonlinear oscillations are ubiquitous in complex systems far from equilibrium (e.g., biological rhythms, neuronal firing). Maintaining these oscillations against thermal noise requires continuous thermodynamic dissipation (entropy production). While the relationship between oscillation and dissipation is understood in linear systems, it remains a significant challenge in nonlinear systems.

The core difficulty lies in the fact that entropy production is a scalar quantity. When parameters change (e.g., noise intensity or bifurcation parameters), the total entropy production rate changes, but it is impossible to discern how specific oscillatory characteristics—such as frequency, amplitude, or coherence across elements—contribute to this change. Existing methods often fail to decompose the dissipation of general nonlinear phenomena like limit cycles, bifurcations, and coherent resonance into interpretable, mode-by-mode components.

2. Methodology

The authors propose a framework combining Stochastic Thermodynamics with Koopman Mode Decomposition (KMD) to linearize and decompose nonlinear dynamics.

A. Theoretical Framework

1. Geometric Decomposition of Entropy Production:
   The total entropy production rate ($\sigma_t$) is decomposed into an excess part (due to relaxation to steady state) and a housekeeping part ($\sigma^{hk}_t$). The housekeeping part represents the dissipation required to maintain the steady state against non-conservative forces. In a steady state, $\sigma_t = \sigma^{hk}_t$.
   $$\sigma^{hk}_t = \langle (\nu^{hk}_t)^\top D_t^{-1} \nu^{hk}_t \rangle_t$$
   where $\nu^{hk}_t$ is the "housekeeping" component of the local mean velocity, and $D_t$ is the noise diffusion matrix.

2. Virtual Deterministic Dynamics:
   To apply KMD, the authors introduce a virtual deterministic dynamical system driven solely by the housekeeping local mean velocity field:
   $$dx_s = \nu^{hk}_t(x_s) ds$$
   Crucially, $\nu^{hk}_t$ includes both the deterministic drift and the diffusion-induced force ($-D_t \nabla \ln p_t(x)$). This ensures the virtual dynamics capture the effective stochastic flows that sustain oscillations, even in the presence of noise.

3. Koopman Mode Decomposition (KMD):
   The nonlinear virtual dynamics are recast as linear evolution in an infinite-dimensional function space via the Koopman generator ($K$). The identity function (state $x$) is expanded using the eigenfunctions ($\phi_k$) and eigenvalues ($\lambda_k$) of $K$:
   $$x_s = \sum_k e^{\lambda_k s} \phi_k(x_0) v_k$$
   where $v_k$ are the Koopman modes. For the steady-state housekeeping dynamics, the eigenvalues are purely imaginary ($\lambda_k = 2\pi i \chi_k$), representing oscillatory modes with frequency $\chi_k$.

B. The Main Result: Mode-wise Dissipation

The authors derive a fundamental relationship decomposing the housekeeping entropy production rate into a sum of positive contributions from each oscillatory mode:
$$\sigma^{hk}_t = \sum_k \sigma^{hk,(k)}_t$$
$$\sigma^{hk,(k)}_t = (2\pi)^2 \chi_k^2 J_k$$
Where:

• $\chi_k$: The frequency of the $k$-th mode.
• $J_k$: The intensity of the $k$-th mode, defined as the $L_2$ norm of the mode under the metric $D_t^{-1} p_t(x)$.

Key Insight: The thermodynamic cost of a specific oscillatory mode is proportional to the square of its frequency and its intensity. High-frequency oscillations are significantly more "expensive" in terms of entropy production.

C. Numerical Implementation

The authors applied this framework to the noisy FitzHugh–Nagumo (FHN) model (a canonical model for neuronal excitability).

• Data-Driven Estimation: They used Dynamic Mode Decomposition (DMD), specifically Hankel DMD combined with Physics-Informed DMD (PiDMD), to extract Koopman eigenvalues, eigenfunctions, and modes from time-series data of the virtual dynamics.
• Validation: The sum of the decomposed mode contributions was verified to match the true housekeeping entropy production rate calculated via direct numerical integration.

3. Key Results and Applications

A. Frequency-Resolved Dissipation

The decomposition successfully revealed that the total entropy production is not a monolithic quantity but a superposition of costs from different frequencies.

• Frequency vs. Cost: As the time constant $\tau$ in the FHN model was varied, the system shifted to lower frequencies. The total dissipation decreased, and the decomposition showed this was driven by a shift in dominant modes to lower frequencies (since cost $\propto \chi^2$), rather than a change in amplitude.

B. Analysis of Bifurcations

The framework was applied to study the system near a Hopf bifurcation (controlled by input current $I$).

• Pre-bifurcation: At low $I$, the system exhibits large limit cycles. The entropy production is supported by a broad spectrum of frequencies.
• Post-bifurcation: As $I$ increases past the bifurcation point, the system transitions to a stable fixed point with small noise-induced loops. The decomposition revealed intermittent dropouts of specific frequency bands.
• Dominance: Near the stable fixed point ($I \approx 2.4$), the dissipation became dominated by a single frequency mode, corresponding to the linear stability analysis of the local mean velocity field. This provided a quantitative picture of how the "structure" of dissipation collapses as the system loses its oscillatory character.

C. Coherent Resonance

The authors investigated coherent resonance, where an intermediate noise intensity maximizes the temporal coherence of oscillations.

• Optimal Noise: At the noise intensity where coherence is maximal, the entropy production rate peaks.
• Spectral Broadening: The decomposition showed that this peak is supported by a broad spectrum of frequencies, all contributing significantly to the total dissipation.
• Non-Optimal Noise: At low or high noise levels, the dissipation is dominated by specific, narrow frequency modes. This suggests that the "efficiency" of coherent resonance relies on the collective contribution of a wide range of oscillatory modes.

4. Significance and Implications

• Interpretability: The work provides the first mode-by-mode picture of thermodynamic dissipation in nonlinear systems. It moves beyond scalar entropy rates to explain which oscillatory features drive the cost.
• Thermodynamic Constraints: The derived inequality $J_k \leq \sigma_t / (2\pi \chi_k)^2$ establishes a fundamental trade-off: sustaining high-frequency oscillations with high intensity requires a disproportionately large thermodynamic cost (entropy production). This offers a potential explanation for why biological oscillators (like neural rhythms) operate at specific frequencies.
• Generalizability: While demonstrated on the FHN model, the framework is general for overdamped nonlinear Langevin systems. It bridges the gap between nonlinear dynamics (Koopman theory) and nonequilibrium thermodynamics.
• Future Directions: The authors note that while the method assumes a finite number of modes (diagonalizability), it can be extended to systems with continuous spectra (e.g., chaos) using advanced DMD techniques. They also highlight the potential for applying this to experimental data to infer thermodynamic constraints in real biological systems.

In summary, this paper establishes a rigorous mathematical link between the spectral properties of nonlinear oscillations and their thermodynamic cost, offering a powerful new tool for analyzing energy efficiency in complex, noisy, non-equilibrium systems.