Abstract: Glasses are traditionally characterized by their rugged landscape of disordered low-energy states and their slow relaxation towards thermodynamic equilibrium. Far from equilibrium, dynamical forms of glassy behavior with anomalous algebraic relaxation have also been noted, for example, in networks of coupled oscillators. Due to their disordered and high-dimensional nature, such systems have been difficult to study theoretically, but data-driven methods are emerging as a promising alternative that may aid in their analysis. Here, we characterize glassy dynamics using the dynamic mode decomposition, a data-driven spectral computation that approximates the Koopman spectrum. We show that the gap between oscillatory and decaying modes in the Koopman spectrum vanishes in systems exhibiting algebraic relaxation, and thus, we propose a model-agnostic signature for robustly detecting and analyzing glassy dynamics. We demonstrate the utility of our approach through both a minimal example of a one-dimensional ODE and a high-dimensional example of coupled oscillators.
Introduction
Characterizing glassy behavior remains a significant challenge in physics. While "static" glasses like window glass are well-studied, "dynamical" glasses—complex systems like frustrated oscillator networks that fail to settle into a steady state—are more elusive. The defining feature of these systems is their relaxation behavior: while most dynamical systems return to equilibrium exponentially fast, glassy systems exhibit "anomalous algebraic relaxation" (e.g., $1/t$). Detecting this slow crawl in high-dimensional datasets often requires prior knowledge of the correct "order parameter," which is not always available. Nicolaou et al. propose a model-agnostic, data-driven solution using Dynamic Mode Decomposition (DMD) to identify a universal spectral signature for glassy dynamics.
Methods: Data-Driven Koopman Analysis
The authors utilize Dynamic Mode Decomposition (DMD) and its theoretical foundation, Koopman Operator Theory. The Koopman operator represents nonlinear dynamics as a linear, albeit infinite-dimensional, operator. DMD provides a data-driven approximation of this operator, decomposing complex signals into "modes" characterized by a frequency (imaginary part) and a decay rate (real part).
To enhance robustness, the study employs Extended DMD (eDMD) with a dictionary of nonlinear functions and Residual DMD (resDMD). The latter is critical for filtering numerical noise and identifying "spurious" eigenvalues that do not represent physical dynamics. By analyzing the pseudospectra—which indicate the sensitivity of eigenvalues to perturbations—the researchers can visualize characteristic "bulges" toward the imaginary axis in glassy systems.
The Spectral Signature: The "No Gap" Rule
The central finding of this work is the identification of a universal spectral signature: the disappearance of the spectral gap. In standard systems with exponential decay, a clear distance exists between the imaginary axis (purely oscillatory modes) and the decaying modes. In contrast, for systems exhibiting algebraic relaxation, these decaying modes "pile up" against the imaginary axis, effectively closing the gap.
This phenomenon arises because DMD attempts to fit power-law data using sums of exponentials ($e^{\mu t}$). While a single exponential cannot model a power law, a distribution of many exponentials with vanishing decay rates can accurately mimic algebraic relaxation. Thus, the closure of the spectral gap serves as a robust indicator of glassy behavior.
Results and Validation
The approach was validated across multiple systems:
- Minimal 1D ODE: Comparing $\dot{\theta} = -\sin(\theta)^\eta$ for $\eta=1$ (exponential decay) and $\eta=3$ (algebraic decay). The DMD spectrum showed a clear gap for the former and complete closure for the latter.
- The Daido Model: A high-dimensional system of $N=10,000$ coupled phase oscillators with random, frustrated interactions. The method successfully distinguished between "low-rank" regimes (exponential relaxation) and "high-rank" frustrated regimes (glassy dynamics).
- Order Parameter Reconstruction: The authors reconstructed the Kuramoto order parameter with high accuracy using only the DMD modes validated by the residual test, demonstrating the method's ability to extract physically meaningful information from raw data.
Conclusion
By identifying the vanishing spectral gap as a signature of algebraic relaxation, this work provides a powerful, model-agnostic tool for analyzing glassy dynamics in complex systems. Because DMD relies solely on time-series data, it bypasses the need for explicit equations of motion or pre-defined order parameters. Future work may integrate DMD with techniques like Physics-Informed Neural Networks (PINNs) to further improve robustness in noisy, real-world environments.
Summary updated