Two-Population Kuramoto–Sakaguchi Model

• The two-population KS model is a framework that describes interacting oscillator networks with distinct intra- and inter-population couplings and phase frustration parameters.
• It employs analytical reductions and effective potentials to predict dynamical regimes such as phase-locking, drifting, chimera, and fragmentation under both deterministic and stochastic forcing.
• The model’s applications span neural circuits, coupled oscillatory systems, and decision-making networks, illustrating how topology and heavy-tailed noise shape synchronization outcomes.

The two-population Kuramoto–Sakaguchi (KS) model generalizes the classical Kuramoto synchronization framework to describe two interacting oscillator populations subject to both intra- and inter-population couplings, each with distinct phase frustration (Sakaguchi) parameters. This formalism enables rigorous analysis of competitive or cooperative synchronization, phase lag-induced frustration, network structure effects, and stochastic influences—including heavy-tailed noise—across diverse contexts such as neural networks, coupled oscillatory circuits, and collective decision-making systems. Recent advances have revealed a range of dynamical phenomena, including phase-locked, drifting, chimeric, fragmented, and noise-restored regimes, which are accessible through both analytical reductions (e.g., to effective ratchet potentials) and high-performance numerical simulation.

1. Model Formulation and Governing Equations

The archetypal two-population KS model considers two networks—often denoted Blue ($\mathcal{B}$, size $N$) and Red ($\mathcal{R}$, size $M$)—with intra-network adjacency matrices $\mathcal{B}_{ij}$, $\mathcal{R}_{ij}$, and inter-network adjacency $\mathcal{A}^{(BR)}$ (with adjoint $\mathcal{A}^{(RB)}$). The phases $(\beta_i,\, i\in \mathcal{B})$ and $(\varphi_j,\, j\in \mathcal{R})$ evolve via

$$\begin{aligned} d\beta_i &= \left[\omega_i - \zeta_B \sum_{j\in \mathcal{B}} \mathcal{B}_{ij} \sin(\beta_i-\beta_j) - \zeta_{BR} \sum_{k\in \mathcal{R}} \mathcal{A}^{(BR)}_{ik} \sin(\beta_i-\varphi_k-\phi)\right]\,dt + dL_i(t),\ d\varphi_j &= \left[\nu_j - \zeta_R \sum_{k\in \mathcal{R}} \mathcal{R}_{jk} \sin(\varphi_j-\varphi_k) - \zeta_{RB} \sum_{i\in \mathcal{B}} \mathcal{A}^{(RB)}_{ji} \sin(\varphi_j-\beta_i-\psi)\right]\,dt + d\tilde{L}_j(t), \end{aligned}$$

where

• $\zeta_B,\zeta_R$ are intra-network couplings, $\zeta_{BR},\zeta_{RB}$ inter-network couplings
• $\phi$, $\psi$ are Sakaguchi phase-lags (frustrations)
• $L_i(t),\tilde{L}_j(t)$ are independent noise processes (typically tempered-stable Lévy, but reducible to Gaussian for $\alpha=2$)

Key order parameters are $$O_B(t) = \frac{1}{N} |\sum_{i\in \mathcal{B}} e^{i\beta_i}|$$ and $O_R(t)$ similarly, with centroid phases $B(t) = \frac{1}{N}\sum_{i} \beta_i$, $P(t) = \frac{1}{M}\sum_{j} \varphi_j$, and centroid difference $\Delta(t)=B-P$.

2. Analytical Reductions and Effective Potentials

In the cluster-locked regime (internal synchronization within each population), the system admits a linearization onto Laplacian modes, yielding a two-cluster ansatz: $$\beta_i = B + b_i,\quad \varphi_j = P + p_j,\quad \sum_i b_i = \sum_j p_j = 0,$$ with small $b,p$. Projecting onto the centroid mode reduces the deterministic dynamics (noise off) to a low-dimensional nonlinear oscillator: $$\dot{\Delta} = -V'(\Delta),\quad V(\Delta) = -\mu \Delta - \sqrt{C^2 + S^2} \cos(\Delta - \rho),$$ with

$$\begin{aligned} \mu &= \bar{\omega} - \bar{\nu},\ C &= d_T^{(BR)} \left( \frac{\zeta_{BR}\cos\phi}{N} + \frac{\zeta_{RB}\cos\psi}{M} \right),\ S &= d_T^{(BR)} \left( \frac{\zeta_{BR}\sin\phi}{N} - \frac{\zeta_{RB}\sin\psi}{M} \right),\ \rho &= \mathrm{atan2}(S, C). \end{aligned}$$

The sign of ${\cal K} = C^2 + S^2 - \mu^2$ determines synchronized (locked, ${\cal K} > 0$) vs. drifting (${\cal K} < 0$) phases for $\Delta$. These fixed-point formulas and threshold surfaces (for example, $K_c(\alpha_{12}, \alpha_{21})$) provide explicit analytical predictions for phase-locking onset and stability in general networks (Kalloniatis et al., 2015).

3. Stochastic Effects and Noise-Restoration Mechanisms

When exogenous stochasticity is present, especially in the form of tempered-stable Lévy noise (parameterized by stable index $\alpha$, tempering $\lambda$, skewness $\theta$, scale $\sigma^2$), the centroid dynamics satisfy a fractional Fokker–Planck equation on the circle: $$\left[\sigma^2 \,\partial_x^{\alpha,\theta,\lambda} + \partial_x V'(x)\right] P_{st}(x) = 0,$$ where $\partial_x^{\alpha,\theta,\lambda}$ is the (tempered, fractional) Riesz derivative. The stationary current $J_{st}$ yields the mean drift $\langle \dot{\Delta} \rangle$.

Empirically, varying $\alpha$ away from the Gaussian limit ($\alpha=2$) toward heavier tails destroys phase locking monotonically if $\lambda=0$, but introducing moderate tempering ($\lambda>0$) restores order for $\alpha \ll 1$ by suppressing extreme jumps. This generates non-monotonic transitions between noisy, periodic, and locked regimes that are accurately captured both analytically—through metastability restoration via the ratchet mechanism—and numerically, through order-parameter and fitness landscape diagnostics (Kalloniatis et al., 2021).

4. Network Topology, Synchronization, and Fragmentation

The ability of each population to achieve internal phase-locking is governed by the algebraic connectivity (smallest nonzero Laplacian eigenvalue $\lambda_2$) and the spread of natural frequencies $\sigma_\omega$, requiring $K_{11}\lambda_2^{(1)} \gtrsim \sigma_\omega^{(1)}$, and analogously for the second population. Cross-network synchronization thresholds and stability further depend on inter-network degree distributions (e.g., $d_T^{(BR)}$), asymmetry in couplings, and frustration parameters.

Beyond global locking, fragmentation regimes arise when one population splits into internally synchronized subgroups with distinct centroids. The reduced linearized dynamics for these situations admit a generalized set of centroid separation equations, with existence and stability again controlled by discriminants of the relevant coupling and phase lag parameters (Kalloniatis et al., 2015). In certain stochastic regimes, adding noise to the more tightly coupled network can induce re-synchronization in a fragmented competitor (Holder et al., 2016).

5. Dynamical Regimes: Chimeras, Bifurcations, and Multistability

The two-population KS model exhibits a diversity of collective states, including:

• Phase-locked states: Both populations are synchronized with a fixed centroid separation $\Delta_*$
• Drifting states: Centroids exhibit steady drift due to insufficient coupling or unfavorable phase-lag configuration (${\cal K}<0$)
• Chimera states: One population remains synchronized ($r_\sigma \approx 1$), the other incoherent ($r_\sigma \ll 1$); which is which depends on initial conditions and network size
• Traveling clusters: With asymmetric couplings and lags, stable traveling (rotating) two-cluster solutions emerge, with explicit analytical formulas for speed and separation (Choi et al., 2017)
• Bistability and fragmentation: Coexistence of multiple collective states (e.g., zero-lag, $\pi$-state, traveling waves) within the same parameter regime, depending on initial conditions and finite size effects (Sonnenschein et al., 2015)
• Metastability and rare transitions: In finite-size deterministic models, apparent random transitions between symmetric chimera states occur, governed by Poissonian statistics with an Arrhenius dependence of mean switching time on system size, analyzable via small-noise Kramers theory (Irvine et al., 14 Oct 2025)

The table below summarizes some principal regimes and their control parameters as observed in numerical and analytical studies:

Regime	Control Parameter(s)	Stability Criterion / Transition
Phase-locked (synchrony)	${\cal K}>0$	Synchrony threshold: $C^2+S^2 - \mu^2 > 0$
Drifting	${\cal K}<0$	Loss of locking: $C^2+S^2 - \mu^2 < 0$
Chimera	$K>\kappa>0$, $N$ finite	One $r_\sigma \simeq 1$, other $r_\sigma \sim N^{-1/2}$
Traveling clusters	$K_\pm$, $\alpha_\pm,\ p$	Both $\Delta_\pm>0$, $|\delta|<\pi$, see Eq. (3.5) (Choi et al., 2017)
Fragmented	coupling/lag/degree tuning	Discriminant for multi-cluster fixed points

6. Extensions: Noise, Generalized Networks, and Biological Applications

Generalizations of the basic two-population KS model include:

• Tempered-stable Lévy noise: Captures heavy-tailed exogenous shocks; tempering parameter $\lambda$ enables interpolation between purely stable and Gaussian, controlling ergodicity and variance (Kalloniatis et al., 2021).
• Arbitrary network topologies: General intra- and inter-population connectivity (e.g., tree vs. Erdős–Rényi), with Laplacian spectral gap directly entering synchronization thresholds (Kalloniatis et al., 2015, Holder et al., 2016).
• Excitation-Inhibition (E-I) feedback: Modeling coupled E-I neuronal circuits yields analytically tractable KS-type equations, with parameter regimes admitting oscillatory, phase-leading, or bifurcating rhythms (Montbrió et al., 2018).
• Higher-dimensional order parameters: Complex- or real-valued 2D extensions enable description of generalized oscillator ensembles with richer chimera and bifurcation structures (Lee et al., 2023).

7. Computational and Numerical Methodology

Numerical studies use time-integration schemes such as Euler–Maruyama for stochastic differential equations, with custom routines to sample stable and tempered-stable distributed increments, and parallelization strategies (e.g., MPI) for large networks. Key observables include instantaneous and time-averaged order parameters, fitness landscapes over phase-lag space (optimized using Bayesian methods), and the 0–1 test for noise vs. periodic/steady behavior. Analytical predictions for switching times, order-parameter distributions, and multimodal phase statistics have been experimentally validated against simulations across model classes (Kalloniatis et al., 2021, Irvine et al., 14 Oct 2025).

In conclusion, the two-population Kuramoto–Sakaguchi model constitutes a flexible, technically rich framework exhibiting highly nontrivial synchronization, competition, and noise-induced ordering transitions in coupled oscillator networks, with explicit analytic and computational tools available for dissecting these effects across deterministic and stochastic regimes (Kalloniatis et al., 2021).