Eminence Grise Coalitions: On the Shaping of Public Opinion

We consider a network of evolving opinions. It includes multiple individuals with first-order opinion dynamics defined in continuous time and evolving based on a general exogenously defined time-varying underlying graph. In such a network, for an arbitrary fixed initial time, a subset of individuals forms an eminence grise coalition, abbreviated as EGC, if the individuals in that subset are capable of leading the entire network to agreeing on any desired opinion, through a cooperative choice of their own initial opinions. In this endeavor, the coalition members are assumed to have access to full profile of the underlying graph of the network as well as the initial opinions of all other individuals. While the complete coalition of individuals always qualifies as an EGC, we establish the existence of a minimum size EGC for an arbitrary time-varying network; also, we develop a non-trivial set of upper and lower bounds on that size. As a result, we show that, even when the underlying graph does not guarantee convergence to a global or multiple consensus, a generally restricted coalition of agents can steer public opinion towards a desired global consensus without affecting any of the predefined graph interactions, provided they can cooperatively adjust their own initial opinions. Geometric insights into the structure of EGC’s are given. The results are also extended to the discrete time case where the relation with Decomposition-Separation Theorem is also made explicit.

💡 Research Summary

The paper investigates opinion dynamics on networks whose interaction topology varies over time, focusing on both continuous‑time and discrete‑time averaging models. Individuals’ opinions are represented by a state vector x(t)∈ℝⁿ that evolves according to the linear differential equation ˙x(t)=A(t)x(t), where each A(t) is a time‑varying intensity matrix: rows sum to zero, off‑diagonal entries are non‑negative, and the matrix is measurable. The associated state‑transition matrix Φ(t,τ) is stochastic (Φ(t,τ)·1ₙ=1ₙ) and invertible for all t≥τ≥0, guaranteeing preservation of the average opinion.

The central concept introduced is the “Eminence Grise Coalition” (EGC). A subset S⊂V of agents is an EGC if, for any desired consensus value x*∈ℝ and any initial opinions of the complement V\S, there exists a choice of initial opinions for agents in S that drives the whole network to the consensus x*·1ₙ as t→∞. Lemma 1 shows that this definition is equivalent to requiring convergence to zero (x*=0) after a suitable translation, exploiting the translation invariance of the linear dynamics.

To quantify the size of the smallest possible EGC, the authors define the null space of the chain at time τ as null_τ(A)={v∈ℝⁿ | lim_{t→∞}Φ(t,τ)v=0}. Lemma 2 proves that the dimension of this null space is independent of τ; this invariant dimension is called the nullity of the chain, nullity(A). The rank of the chain is then defined as rank(A)=N−nullity(A). Theorem 1 establishes that the cardinality of the minimal EGC exactly equals rank(A). The proof proceeds in two directions: (i) constructing an EGC of size rank(A) by selecting a basis of the null space, forming a matrix whose rows are linearly independent, and taking the complement of the corresponding row indices; (ii) showing that any EGC of size h implies nullity(A)≥N−h, which yields rank(A)≤h, thus h=rank(A).

Sections III–VII develop geometric and graph‑theoretic tools to bound rank(A). A geometric interpretation treats the null space as a linear subspace whose orthogonal complement determines the agents that must be controlled. Upper bounds are derived using the infinite‑flow graph (edges that accumulate infinite interaction over time) and the unbounded‑interaction graph (agents receiving unbounded cumulative influence). The number of strongly connected components in the infinite‑flow graph provides a lower bound on rank(A), while extensive unbounded interactions push the rank toward N.

Time‑invariant chains are examined in Section V, where rank(A) coincides with the number of ergodic classes (strongly connected components) of the underlying stochastic matrix. Section VI introduces the class P* of time‑varying chains that, despite variations, retain a structural decomposition that allows exact computation of rank(A). For both classes, the previously derived bounds are shown to be tight.

Section VII characterizes full‑rank chains (rank(A)=N). In such cases the null space contains only the zero vector, meaning that no non‑trivial initial opinion can be driven to zero without controlling every agent. The authors prove that full‑rank occurs precisely when the interaction graph is complete and all edge weights remain strictly positive over time, highlighting the necessity of universal participation for consensus control in highly connected networks.

Section VIII extends the analysis to discrete‑time dynamics x(k+1)=W(k)x(k), where each W(k) is a stochastic matrix. By invoking Sonin’s Decomposition‑Separation Theorem and the concept of “jets,” the authors demonstrate that the number of jets in the jet decomposition equals the rank of the discrete‑time chain, and thus also equals the size of the minimal EGC. This establishes a direct parallel between continuous‑ and discrete‑time settings.

The paper concludes by emphasizing the practical relevance of EGCs: a small, strategically chosen group of agents can steer public opinion to any desired value without altering the underlying interaction structure, merely by setting their initial opinions appropriately. Potential applications include political campaigning, market influence, and coordinated decision‑making. Future work is suggested on nonlinear opinion models, partial information scenarios, and adaptive networks where the topology itself evolves in response to the opinions.