Coalitions in nonatomic network congestion games

This work shows that the formation of a finite number of coalitions in a nonatomic network congestion game benefits everyone. At the equilibrium of the composite game played by coalitions and individuals, the average cost to each coalition and the individuals’ common cost are all lower than in the corresponding nonatomic game (without coalitions). The individuals’ cost is lower than the average cost to any coalition. Similarly, the average cost to a coalition is lower than that to any larger coalition. Whenever some members of a coalition become individuals, the individuals’ payoff is increased. In the case of a unique coalition, both the average cost to the coalition and the individuals’ cost are decreasing with respect to the size of the coalition. In a sequence of composite games, if a finite number of coalitions are fixed, while the size of the remaining coalitions goes to zero, the equilibria of these games converge to the equilibrium of a composite game played by the same fixed coalitions and the remaining individuals.

💡 Research Summary

The paper investigates the impact of introducing a finite number of coalitions into a nonatomic network congestion game. In the classical setting, an infinite population of infinitesimal users selects routes on a directed graph, each edge having a non‑decreasing, continuous cost function that depends on the total flow. The equilibrium concept is the Wardrop equilibrium, where every user experiences the minimal possible travel cost given the current flow distribution.

The authors extend this framework by allowing some of the infinitesimal users to group together into coalitions (also called “players with mass”). A coalition controls a positive amount of flow and can internally reallocate its members across routes so as to minimize the coalition’s average cost. Individual users remain infinitesimal and continue to follow the Wardrop condition. The resulting game, which contains both coalitions and individuals, is termed a “composite game,” and its equilibrium is called a composite equilibrium.

The main contributions are fourfold. First, the authors prove that at any composite equilibrium the average cost incurred by each coalition and the common cost incurred by the individuals are strictly lower than the cost in the corresponding pure nonatomic game (i.e., the game without coalitions). The presence of coalitions thus improves overall efficiency because the coalition can internally smooth out congestion. Second, the individuals’ cost is lower than the average cost of any coalition; this follows from the fact that an infinitesimal player can react to marginal changes in flow more precisely than a mass player. Third, the average cost of a coalition is monotone in its size: larger coalitions experience higher average costs than smaller ones. Intuitively, as a coalition grows it loses flexibility in reallocating its flow, which reduces the benefit of internal coordination. Fourth, when a subset of a coalition’s members become independent individuals, the individuals’ cost rises, confirming that the coalition’s internal coordination is beneficial to the rest of the population.

The analysis relies on convexity of the total cost function, variational inequalities, and the Karush‑Kuhn‑Tucker (KKT) conditions. Each coalition solves a convex optimization problem that minimizes its average cost subject to flow conservation; individuals satisfy the Wardrop variational inequality. By coupling the KKT conditions of all coalitions with the Wardrop condition, the authors establish existence and uniqueness of the composite equilibrium under standard assumptions (continuous, non‑decreasing, differentiable edge cost functions). The monotonicity results are derived from the sign of the second derivative of the edge cost functions and from the structure of the optimality conditions.

A further contribution concerns the limiting behavior when many coalitions become vanishingly small. The authors consider a sequence of composite games in which a fixed finite set of coalitions retains a positive mass while the remaining coalitions’ masses tend to zero. They prove that the equilibria of these games converge to the equilibrium of a composite game that contains only the fixed coalitions and a continuum of individuals. This result shows that the model is robust to the granularity of coalition formation: as coalitions fragment, the system smoothly approaches the pure nonatomic regime.

The paper concludes with a discussion of practical implications. In transportation networks, encouraging certain vehicle groups (e.g., buses, high‑occupancy vehicles) to coordinate their routes can lower overall congestion and simultaneously reduce travel times for private drivers. In communication networks, service providers that manage traffic for large user groups can achieve lower latency for both the managed groups and the remaining users. The theoretical findings thus provide a solid foundation for policies that promote coalition‑based coordination in congested networks.

Overall, the work establishes that coalition formation in nonatomic congestion games is universally beneficial, quantifies how coalition size influences cost, and demonstrates convergence properties when coalitions shrink. These insights open avenues for future research on dynamic coalition formation, incomplete information, and multi‑objective extensions.