Eigenvector centrality

Eigenvector centrality is a measure of the influence of a node in a network that assigns relative scores to all nodes based on the principle that connections to high-scoring nodes contribute more to the score of the node in question than equal connections to low-scoring nodes. Formally, it is the principal eigenvector of the adjacency matrix of the graph — the eigenvector associated with the largest eigenvalue. This recursive definition captures a deep structural intuition: a node is important not merely because it has many connections, but because it is connected to other important nodes.

The concept was introduced by Phillip Bonacich in 1972 as a way to model social influence in networks. Unlike degree centrality, which simply counts connections, eigenvector centrality weights each connection by the centrality of the node at the other end. A node with a few connections to highly central nodes can have higher eigenvector centrality than a node with many connections to peripheral nodes. This makes eigenvector centrality a measure of \'\'\'functional centrality\'\'\' — the capacity to influence and be influenced by the network\'s core — rather than merely positional prominence.

Mathematical Foundations

For an undirected graph with adjacency matrix A, the eigenvector centrality scores x satisfy the eigenvalue equation:

Ax = λx

where λ is the largest eigenvalue of A. By the Perron-Frobenius theorem, this eigenvalue is real, positive, and simple for connected graphs, and its corresponding eigenvector has strictly positive entries. This guarantees a unique centrality ranking (up to scaling) for any connected network.

The centrality scores can also be computed iteratively through the power method: starting with an initial vector of ones, repeatedly multiply by the adjacency matrix and normalize. This process converges to the principal eigenvector because the largest eigenvalue dominates the spectrum. The convergence rate depends on the \'\'\'spectral gap\'\'\' — the ratio between the largest and second-largest eigenvalues — which is why eigenvector centrality is deeply connected to the mixing properties of random walks on the graph.

For directed graphs, the Perron-Frobenius theorem applies only if the graph is strongly connected. In practice, directed networks like the web are not strongly connected, which is why PageRank modifies the adjacency matrix by adding a damping factor that ensures irreducibility. PageRank is, in essence, a regularized variant of eigenvector centrality designed for directed graphs with dangling nodes and disconnected components.

Interpretation and Applications

Eigenvector centrality reveals a network\'s \'\'\'core-periphery structure\'\'\'. High-centrality nodes form a densely interconnected core that dominates the network\'s dynamics. In social networks, these are the individuals whose opinions propagate most effectively. In citation networks, they are the foundational papers that subsequent research continually references. In financial networks, they are the institutions whose distress would propagate most rapidly through interlocking obligations.

The measure is particularly useful for identifying \'\'\'influential spreaders\'\'\' in epidemic models. A node with high eigenvector centrality is not merely connected to many others; it is connected to others who are themselves well-connected, creating multi-hop pathways for contagion. This is why vaccination strategies that target high-eigenvector-centrality nodes can be more effective than strategies that target high-degree nodes — the eigenvector captures the network\'s \'\'\'resonant modes\'\'\', the pathways along which influence amplifies rather than dissipates.

In economics, eigenvector centrality measures the \'\'\'systemic importance\'\'\' of firms in production networks. A firm is systemically important not because it has many suppliers or customers, but because it occupies a position in the supply chain where disruptions would cascade through the network\'s core. The input-output tables of national economies, when treated as adjacency matrices, reveal that eigenvector centrality predicts which sectors are most critical for macroeconomic stability better than simpler measures like sector size or direct connectivity.

Limitations and Extensions

Eigenvector centrality has two significant limitations. First, it is defined only for connected graphs; disconnected components have independent centrality scores that are not comparable across components. Second, it is sensitive to the presence of high-degree nodes: a single node with extremely high degree can dominate the principal eigenvector, compressing the scores of all other nodes and making the ranking less informative.

Several extensions address these limitations. \'\'\'Katz centrality\'\'\' adds a damping parameter that attenuates the contribution of distant connections, preventing the measure from being dominated by long cycles. \'\'\'Closeness centrality\'\'\', by contrast, measures how close a node is to all others in terms of path length, providing a complementary perspective that emphasizes reach rather than recursive influence. Betweenness centrality measures brokerage; eigenvector centrality measures embeddedness in the core. No single measure captures everything, and the choice of centrality measure is a theoretical commitment about what kind of structural power matters in the domain under study.

Eigenvector centrality is often treated as a technical refinement of degree centrality — a way to weight connections by the importance of neighbors. But this understates its significance. Eigenvector centrality reveals that a network has a \'\'native hierarchy\'\' that is not imposed by an external ranking but emergent from its topology. The principal eigenvector is not a constructed metric; it is a structural property of the graph, as real as its diameter or its clustering coefficient. The claim that a