Signature of glassy dynamics in dynamic modes decompositions

This paper proposes a model-agnostic signature for detecting glassy dynamics by demonstrating that the gap between oscillatory and decaying modes in the Koopman spectrum vanishes in systems exhibiting algebraic relaxation, a finding validated through both minimal and high-dimensional examples using dynamic mode decomposition.

The Gist

The Big Picture: The "Glass" Problem

Imagine you have a jar of marbles. If you shake it, they settle quickly into a neat pile. That's normal behavior.

Now, imagine a jar filled with honey and marbles. If you shake it, the marbles move, but they get stuck in a messy, tangled mess. They don't settle quickly; they drift very slowly, getting "stuck" in a disordered state. In physics, this messy, slow-moving state is called a Glass.

Scientists have known about this "glassy" behavior for a long time. It happens in materials (like window glass) and in complex systems like networks of neurons or coupled oscillators. The problem? It's incredibly hard to predict when a system will turn "glassy" just by looking at the math. The equations are too messy and high-dimensional.

The New Tool: The "Spectral X-Ray"

The authors of this paper propose a new way to look at these systems using a tool called Dynamic Mode Decomposition (DMD).

Think of a complex system (like a crowd of 10,000 people dancing) as a giant, chaotic song.

• Normal systems (like a marching band) have a clear rhythm. You can easily hear the beat. In math terms, the "song" is made of distinct, separate notes that die out quickly.
• Glassy systems are like a jam session where the instruments are slightly out of sync and the sound fades away very slowly, like a reverb that never quite stops.

The authors used DMD to take a "spectral X-ray" of these systems. This X-ray breaks the complex motion down into its individual "notes" (mathematically called modes).

The Secret Signature: The "Gap"

Here is the magic trick the authors discovered.

When they looked at the "notes" of a normal system, they saw a clear gap between two types of sounds:

1. Oscillating notes: The pure beats that keep going (like a drum).
2. Decaying notes: The sounds that fade away (like a guitar string stopping).

In a normal system, there is a safe distance (a gap) between the fading sounds and the beating sounds. Because of this gap, the system settles down quickly and predictably.

But in a "Glassy" system, that gap disappears.

Imagine the fading sounds creeping closer and closer to the beating sounds until they are practically touching. The authors call this the "Vanishing Gap."

• The Analogy: Think of a staircase.
  • Normal System: There is a big, flat landing between the steps. You can stop easily.
  • Glassy System: The landing disappears. The steps merge into a slippery, continuous slide. Because there is no "landing" (gap) to stop the decay, the system slides down very slowly, following a "power law" (algebraic decay) instead of stopping quickly.

How They Proved It

The team tested this idea in two ways:

1. The Simple Test: They built a tiny, one-dimensional math model. When they cranked up the "glassiness," they watched the gap in the X-ray disappear, and the system started sliding down slowly.
2. The Big Test: They simulated a massive network of 10,000 coupled oscillators (like thousands of fireflies trying to flash in sync).
   • When the connections were weak or simple, the gap remained, and they settled fast.
   • When the connections were complex and "frustrated" (like a puzzle where pieces don't fit), the gap vanished. The system entered a glassy state, and the DMD tool spotted it immediately.

Why This Matters

Before this paper, finding a glassy state was like trying to find a needle in a haystack by guessing. You had to know exactly what to look for.

Now, the authors have a universal detector. You don't need to know the specific rules of the system. You just feed the data into the DMD tool, look for the "Vanishing Gap" in the spectrum, and if it's gone, you know you have found glassy dynamics.

The Takeaway

This research gives scientists a new "order parameter" (a measuring stick). Instead of guessing, they can now quantitatively say: "Look, the gap is gone. The system is glassy."

This is a huge step forward for understanding complex systems, from how materials age to how brain networks might get stuck in certain states. It turns a messy, invisible problem into a clear, visible pattern.

Technical Summary

1. Problem Statement

Quantifying glassy behavior in complex, high-dimensional systems remains a significant scientific challenge.

• Context: Glasses are characterized by rugged energy landscapes, slow relaxation to equilibrium, and spatial disorder. While thermodynamic glassy systems (e.g., spin glasses) are well-studied, dynamical glassy behavior in systems far from equilibrium (such as networks of coupled oscillators) is harder to characterize.
• The Challenge: In dynamical systems, glassy behavior manifests as algebraic (power-law) relaxation rather than the typical exponential relaxation. In high-dimensional systems (like the Daido oscillator model), this behavior is often hidden within the data. Traditional analysis relies on finding specific "order parameters" (like the Kuramoto order parameter) to detect scaling, but this requires a priori knowledge of the system's structure. Without a known order parameter, distinguishing algebraic decay from exponential decay in noisy, high-dimensional data is difficult.
• Goal: The authors aim to develop a model-agnostic, data-driven method to robustly detect and analyze glassy dynamics without requiring prior knowledge of the system's order parameters.

2. Methodology

The paper proposes using Dynamic Mode Decomposition (DMD), a data-driven spectral method that approximates the Koopman operator, to identify signatures of glassy dynamics.

• Koopman Spectrum Analysis:

  • The Koopman operator is an infinite-dimensional linear operator governing the evolution of observables in a nonlinear system. Its spectrum consists of eigenvalues $\mu$.
  • Exponential Decay: In standard systems, there is a spectral gap between purely imaginary eigenvalues (oscillatory modes) and eigenvalues with negative real parts (decaying modes). The relaxation is dominated by the leading decaying mode.
  • Algebraic Decay (Glassy Signature): The authors hypothesize that in glassy systems, this gap vanishes. Instead, there is an accumulation of decaying modes approaching the imaginary axis.
  • Mechanism: Mathematically, a power-law decay ($t^{-\alpha}$) can be represented as an infinite sum of exponentially decaying modes ($e^{-\lambda t}$) where the amplitudes and decay rates scale specifically. If the DMD spectrum shows an accumulation of modes near the imaginary axis with specific amplitude scaling, it indicates algebraic relaxation.

• Algorithmic Implementation:

  • Extended DMD (EDMD): Uses a dictionary of observable functions (e.g., Fourier modes) to approximate the Koopman operator.
  • Residual DMD (resDMD): To mitigate spectral noise and spurious eigenvalues common in high-dimensional data, the authors employ resDMD. This filters modes based on a residual error threshold ($\epsilon$), ensuring only true Koopman modes are analyzed.
  • Pseudospectra: They analyze $\epsilon$-pseudospectra to identify neighborhoods close to eigenvalues, helping to distinguish true spectral features from numerical noise.

3. Key Contributions

1. Identification of a Universal Signature: The paper establishes that the vanishing of the gap between oscillatory and decaying modes in the Koopman spectrum is a robust signature of glassy dynamics.
2. Data-Driven Order Parameter: The authors introduce a new order parameter, $\eta$, defined as the average negative real part of the DMD eigenvalues that pass the resDMD criteria:
   $$\eta \equiv -\langle \text{Re}(\mu_i) \rangle |_{\epsilon_i < 5 \times 10^{-8}}$$
   This parameter quantitatively signals the onset of glassy dynamics (where $\eta$ approaches zero from the positive side as the gap vanishes).
3. Model-Agnostic Approach: Unlike previous methods requiring specific knowledge of the system's physics or order parameters, this DMD-based approach works directly on trajectory data, making it applicable to a wide range of disordered systems.

4. Results

The authors validated their approach using two examples:

• Minimal Example (1D ODE):

  • System: $\dot{\theta} = -\sin(\theta)\zeta$.
  • Case $\zeta=1$ (Exponential): The DMD spectrum showed a clear gap between the imaginary axis and the decaying modes.
  • Case $\zeta=3$ (Algebraic): The gap vanished. The spectrum showed an accumulation of decaying modes approaching the imaginary axis. The mode amplitudes scaled with the decay rate in a manner consistent with the theoretical decomposition of a power law into exponentials.

• High-Dimensional Example (Coupled Oscillator Glass):

  • System: Generalized Daido model with $N=10,000$ oscillators, random couplings, and frustration.
  • Low-Rank Regime ($K=2$): Exhibited exponential relaxation. The DMD spectrum was confined to the imaginary axis (oscillatory) with no significant accumulation of decaying modes near the axis.
  • High-Rank Regime ($K=N$): Exhibited algebraic relaxation (glassy behavior). The DMD spectrum revealed a dense accumulation of decaying modes near the imaginary axis that did not converge to the axis as the dictionary size increased.
  • Order Parameter $\eta$: Plotting $\eta$ against the rank of the adjacency matrix ($K$) revealed a clear transition. For low $K$, $\eta$ was non-zero (exponential). For high $K$, $\eta$ dropped significantly, signaling the onset of glassy dynamics. The transition point shifted depending on the coupling constant $J$.

5. Significance

• Theoretical Insight: The work bridges the gap between the abstract theory of Koopman operators and the practical observation of glassy dynamics. It confirms that algebraic relaxation is fundamentally a spectral phenomenon involving the continuous accumulation of decay rates.
• Practical Utility: The proposed method provides a tool for analyzing complex systems (e.g., neural networks, other disordered oscillator systems) where the underlying order parameters are unknown. It allows researchers to detect phase transitions (like the "volcano transition" in oscillator glasses) purely from data.
• Future Directions: The authors suggest that this framework can be extended to other data-driven techniques (e.g., physics-informed constraints, sparse identification) and applied to broader classes of non-equilibrium systems, potentially revolutionizing how glassy dynamics are detected in experimental and simulation data.

In summary, the paper successfully demonstrates that Dynamic Mode Decomposition can serve as a powerful, model-free diagnostic tool for identifying glassy dynamics by detecting the spectral signature of a vanishing gap in the Koopman spectrum.