Epidemic threshold in directed networks

Abstract

Epidemics have so far been mostly studied in undirected networks. However, many real-world networks, such as the social network Twitter and the WWW networks, upon which information, emotion or malware spreads, are shown to be directed networks, composed of both unidirectional links and bidirectional links. We define the directionality as the percentage of unidirectional links. The epidemic threshold for the susceptible-infected-susceptible (SIS) epidemic has been proved to be 1/lambda_{1} in directed networks by N-intertwined Mean-field Approximation, where lambda_{1}, also called as spectral radius, is the largest eigenvalue of the adjacency matrix. Here, we propose two algorithms to generate directed networks with a given degree distribution, where the directionality can be controlled. The effect of directionality on the spectral radius lambda_{1}, principal eigenvector x_{1}, spectral gap lambda_{1}-|lambda_{2}|) and algebraic connectivity |mu_{N-1}| is studied. Important findings are that the spectral radius lambda_{1} decreases with the directionality, and the spectral gap and the algebraic connectivity increase with the directionality. The extent of the decrease of the spectral radius depends on both the degree distribution and the degree-degree correlation rho_{D}. Hence, the epidemic threshold of directed networks is larger than that of undirected networks, and a random walk converges to its steady-state faster in directed networks than in undirected networks with degree distribution.