From Consensus to Robust Clustering: Multi-Agent Systems with Nonlinear Interactions

TL;DR

The paper develops a unified theory for the shift from global consensus to robust clustering in multi-agent networks with nonlinear, cooperative interactions. By linking the signal Lipschitz constant $K$ to the second-largest eigenvalue $ obreak\lambda_{N-1}$ of the row-stochastic adjacency, it derives a sharp threshold $K \lambda_{N-1}<1$ that guarantees all equilibria lie on the synchronization manifold; when the threshold is violated, non-synchronized equilibria (NFSE) can emerge. It then introduces an Input-to-State Stability framework to quantify the robustness of clusters against network perturbations, showing that cluster coherence depends on perturbation heterogeneity rather than its magnitude, and provides a practical bound via a residual perturbation term. This theory is validated on Zachary's Karate Club network, where the model reproduces a robust, two-cluster partition under NFSE conditions and demonstrates how ISS bounds govern intra-cluster cohesion. The results offer a principled design and analysis method for modularity in complex networks, with implications for robust consensus, community detection, and distributed control under nonlinear interactions.

Abstract

This paper establishes a theoretical framework to describe the transition from consensus to stable clustering in multi-agent systems with nonlinear, cooperative interactions. We first establish a sharp threshold for consensus. For a broad class of non-decreasing, Lipschitz-continuous interactions, an explicit inequality linking the interaction's Lipschitz constant to the second-largest eigenvalue of the normalized adjacency matrix of the interaction graph confines all system equilibria to the synchronization manifold. This condition is shown to be a sharp threshold, as its violation permits the emergence of non-synchronized equilibria. We also demonstrate that such clustered states can only arise if the interaction law itself possesses specific structural properties, such as unstable fixed points. For the clustered states that emerge, we introduce a formal framework using Input-to-State Stability (ISS) theory to quantify their robustness. This approach allows us to prove that the internal cohesion of a cluster is robust to perturbations from the rest of the network. The analysis reveals a fundamental principle: cluster coherence is limited not by the magnitude of external influence, but by its heterogeneity across internal nodes. This unified framework, explaining both the sharp breakdown of consensus and the quantifiable robustness of the resulting modular structures, is validated on Zachary's Karate Club network, used as a classic benchmark for community structure.

Figures (5)

• Figure 1: Illustration of Proposition \ref{['prop:invariance_of_hypercubes_defined_by_fixed_points']}.
• Figure 2: Illustration of the attraction basins for dynamics restricted to FSM (i.e., $\dot{x}\mathbf{1} = (s(x) - x) \mathbf{1}$) for two different signal functions $s(x)$. The first plots show the function $s(x)$ and the identity line $y=x$ and exhibits the fixed points of $s(x)$. The stable fixed points are represented by a black circle and the unstable ones by a white circle. The second plots show the basins of attraction of the different equilibria over $\mathcal{S}$.
• Figure 3: Bifurcation diagram for the $N=5$ line graph with $s(x)=\tanh(Kx)$ for $K \in [0.5,3.5]$, illustrating the distinction between stability on the invariant manifold $\mathcal{M}$ and stability in the full state space $\mathcal{X}$. The plot shows the equilibrium state $x_1^*$ as a function of the gain parameter $K$. Stability on $\mathcal{M}$ (lines): The FSE at $x_1^*=0$ loses stability at $K=\sqrt{2}$ (dashed line), creating two branches of NFSE that are stable on the manifold (solid lines). Stability in $\mathcal{X}$ (red/blue colors): The FSE becomes unstable in the full space at $K=\sqrt{2}$ (red dashed line). Crucially, the new NFSE branches are also born unstable in $\mathcal{X}$ (solid red lines), despite their stability on $\mathcal{M}$. They only become stable in the full space (solid blue lines) after the transversal stabilization bifurcation at $K_{\mathrm{stab}} \approx 2.463$. The inset phase portraits visualize the dynamics \ref{['eq:dynamics_on_line_N5']} on $\mathcal{M}$ at three representative values of $K$.
• Figure 4: Visual validation of the sharp synchronization threshold on various network topologies. The figure displays the final equilibrium state, with node colors representing the agent state $x_i^* \in [-1,1]$. For diverse topologies (columns 1-3), the top panels confirm that meeting the condition $K\lambda_{N-1} < 1$ guarantees convergence to an FSE, as predicted by Corollary \ref{['coro:attractivity_of_S']}. In contrast, the bottom panels demonstrate that violating this threshold leads to an NFSE, characterized by robust clustering. Column 4 highlights the special case of highly connected graphs (e.g., Star and Complete), which are intrinsically robust to disagreement and always synchronize. All simulations use the signal function $s(x) = \max(-1, \min(1, Kx))$, with identical initial conditions for top and bottom panels to ensure a fair comparison.
• Figure 5: Numerical validation of the ISS framework on Zachary's Karate Club network, a canonical example of a social network with a strong, empirically verified community structure. The simulation uses a signal function $s(x)= \max(-1, \min(1, Kx))$ (with $K=1.2$) chosen to violate the threshold for full synchronization ($K\lambda_{N-1} > 1$), forcing the emergence of a stable, clustered equilibrium (NFSE).

Theorems & Definitions (28)

• Lemma 1: Perron-Frobenius bulloLecturesNetworkSystems2018
• Definition 1
• Lemma 2
• Definition 2
• Proposition 1
• Lemma 3
• Proposition 2
• Corollary 1
• Lemma 4
• Theorem 1: Stability of FSE