Resilience of the slow component in timescale separated synchronized oscillators

Technical Summary: Resilience of the slow component in timescale separated synchronized oscillators

Problem Statement
Physiological and physical networks often consist of numerous coupled oscillators evolving on multiple distinct timescales. While the synchronization dynamics of such systems are well-studied, the specific resilience of a "slow" component when the system is divided into fast and slow subsystems remains under-explored. In many scenarios, such as physiological networks or power grids, a subset of oscillators may evolve significantly slower than others due to intrinsic characteristics or external perturbations. This timescale separation alters how noise and input signals are transmitted between components compared to a system where all oscillators evolve on a single timescale. The central problem addressed is quantifying the resilience of the slow component in the presence of time-correlated noise, specifically determining how the separation of timescales affects the variance of phase deviations from a synchronized state.

Methodology
The author employs a theoretical framework based on Kuramoto phase oscillators, though the approach is noted to apply generally to coupled dynamical systems near a stable fixed point. The methodology proceeds as follows:

1. Model Formulation: The system is modeled as a set of $N$ coupled oscillators divided into a fast set ($F$) and a slow set ($S$). The timescale separation is parameterized by a small parameter $\epsilon$, where the dynamics of the fast component are scaled by $\epsilon^{-1}$ relative to the slow component.
2. Dimensional Reduction: To analyze the slow component, the author applies the Mori-Zwanzig formalism. This allows for the elimination of the fast variables, resulting in a reduced dynamical equation for the slow component. This reduction is shown to be mathematically equivalent to a Kron reduction of the system's Jacobian matrix.
3. Linearization and Noise Analysis: The dynamics are linearized around a stable fixed point (phase-locked state). The system is subjected to time-correlated noise with distinct correlation times ($\tau_S$ and $\tau_F$) for the slow and fast components, respectively.
4. Variance Calculation: By solving the resulting linear stochastic differential equations, the author derives a closed-form expression for the variance of phase deviations in the slow component. This expression depends on the eigenmodes and eigenvalues of the reduced Jacobian ($J_{red}$) and a coupling term ($\Gamma_{\alpha\beta}$) that describes how noise from the fast component is transmitted to the slow one.
5. Analytical Limits and Numerical Validation: The theory is analyzed in specific network topologies (e.g., disconnected fast oscillators, all-to-all coupling, layered networks) and in limiting cases of noise correlation times (short vs. long). These analytical predictions are validated against numerical simulations of the full non-linear system on various network structures (Watts-Strogatz, Erdős-Rényi).

Key Contributions and Results

• Reduced Dynamics and Noise Transmission: The paper establishes that the resilience of the slow component is governed by a reduced Jacobian ($J_{red} = J_{SS} - J_{SF}J_{FF}^{-1}J_{FS}$). The noise transmission from the fast to the slow component is not merely additive; it involves a complex mixing of eigenmodes of the reduced system, weighted by the coupling structure between the two components.
• Reversal of Robustness: A primary finding is that the separation of timescales can fundamentally alter which oscillators are most vulnerable. Oscillators that exhibit small variance (high robustness) in a single-timescale system can become the most vulnerable (exhibiting large variance) when timescale separation is introduced, provided they are well-connected to the fast component. This contrasts with standard intuition where high connectivity usually confers robustness.
• Impact of Coupling Topology:
  • Disconnected Fast Oscillators: The variance depends on the location of the slow oscillators on the slowest eigenmodes of the reduced Jacobian.
  • All-to-All from Fast to Slow: If fast oscillators with inter-component connections are connected homogeneously to all slow oscillators, the noise transmission term vanishes, resulting in no additional variance from the fast component (only a global phase shift).
  • All-to-All from Slow to Fast: Conversely, if a slow oscillator connects homogeneously to all fast oscillators, the variance in the slow component is amplified. The magnitude of this amplification depends on the size of the fast component but is independent of the internal network structure of the fast component.
• Layered Networks: In layered network structures where fast and slow components form distinct layers with one-to-one inter-layer connections, the variance is found to be largely insensitive to timescale separation when the noise correlation time is homogeneous and long. In this regime, the variance with and without timescale separation differs only by a small constant.
• Noise Correlation Time Regimes: The amplitude of phase excursions is shown to depend critically on the ratio between the noise correlation time and the system's eigenvalues. In the limit of short noise correlation times, the total variance scales inversely with the coupling strength ($K^{-1}$), whereas for long correlation times, it scales as $K^{-2}$.

Significance
The paper claims that understanding timescale separation is crucial for assessing the resilience of coupled phase oscillators, particularly in physiological networks where different timescales are ubiquitous. The work demonstrates that the "most robust" oscillators in a homogeneous timescale system are not necessarily the most robust when timescales are separated; in fact, their connectivity to a fast subsystem can render them the most fragile. This insight suggests that resilience assessments must account for the specific timescale architecture of the network. The findings are presented as applicable to any coupled dynamical system evolving near a stable fixed point, offering a theoretical tool to predict how noise propagates and amplifies in complex, multi-timescale biological and physical systems. The author notes that future work should extend these findings to multiple timescale separations and regimes beyond small fluctuations.