Network Design problems typically ask for a minimum cost subnetwork from a given host network. This classical point-of-view assumes a central authority enforcing the optimum solution. But how should networks be designed to cope with selfish agents that own parts of the network? In this setting, minimum cost networks may be very unstable in that agents will deviate from a proposed solution if this decreases their individual cost. Hence, designed networks should be both efficient in terms of total cost and stable in terms of the agents' willingness to accept the network.We study this novel type of Network Design problem by investigating the creation of (𝛽, 𝛾)-networks, that are in 𝛽-approximate Nash equilibrium and have a total cost of at most 𝛾 times the optimal cost, for the recently proposed Euclidean Generalized Network Creation Game by Bilò et al. [11]. There, 𝑛 agents corresponding to points in Euclidean space create costly edges among themselves to optimize their centrality in the created network. Our main result is a simple O (𝑛 2 )-time algorithm that computes a (𝛽, 𝛽)-network with low 𝛽 for any given set of points. Moreover, on integer grid point sets or random point sets our algorithm achieves a low constant 𝛽. Besides these results for the Euclidean model, we discuss a generalization of our algorithm to instances with arbitrary, even non-metric, edge lengths. Moreover, in contrast to these algorithmic results, we show that no such positive results are possible when focusing on either optimal networks, i.e., (𝛽, 1)-networks, or perfectly stable networks, i.e., (1, 𝛾)-networks, as in both cases NP-hard problems arise, there exist instances with very unstable optimal networks, and there are instances for perfectly stable networks with high total cost. Along the way, we significantly improve several results from Bilò et al. and we asymptotically resolve their conjecture about the Price of Anarchy by providing a tight bound.
CCS CONCEPTS• Theory of computation → Network formation; Quality of equilibria; Algorithmic game theory.