Selfish routing games with priority lanes

We study selfish routing games where users can choose between regular and priority service for each network edge on their chosen path. Priority users pay an additional fee, but in turn they may travel the edge prior to non-priority users, hence experiencing potentially less congestion. For this model, we establish existence of equilibria for linear latency functions and prove uniqueness of edge latencies, despite potentially different strategic choices in equilibrium. Our main contribution demonstrates that marginal cost pricing achieves system optimality: When priority fees equal marginal externality costs, the equilibrium flow coincides with the socially optimal flow, hence the price of anarchy equals $1$. This voluntary priority mechanism therefore provides an incentive-compatible alternative to mandatory congestion pricing, whilst achieving the same result. We also discuss the limitations of a uniform pricing scheme for the priority option.

💡 Research Summary

The paper introduces a novel nonatomic routing model that captures the growing practice of offering users a voluntary “priority” service on each network link—think express toll lanes on highways, premium security lines at airports, or paid high‑bandwidth channels in communication networks. In the model each original edge e is duplicated into a regular lane (e^R) and a priority lane (e^V). Users may either travel as regular users, incurring the average latency caused by all traffic on the edge, or as priority users, who pay an additional fee ω_e and are allowed to traverse the edge before regular users. The underlying physical latency function is linear, ˆℓ_e(x)=a_e x + b_e, with a_e, b_e ≥ 0.

For a given flow the perceived cost for priority users is
 ˜ℓ_e^V(f_e^V)=½ a_e f_e^V + b_e + ω_e,
while regular users experience
 ˜ℓ_e^R(f_e^R, f_e^V)=a_e f_e^V + ½ a_e f_e^R + b_e.
Thus priority users are unaffected by regular traffic, but regular users suffer an extra delay proportional to the total priority flow.

The authors first address equilibrium existence. The classic potential‑function approach (which works for standard routing games) does not directly apply because the split between regular and priority flow on a given edge may be inconsistent across different paths. Instead, they construct a set‑valued correspondence B that maps any feasible extended flow ˜f to the set of feasible flows that allocate all positive mass only to paths with minimal perceived cost under ˜f. B is non‑empty, convex‑valued, and has a closed graph thanks to continuity of the perceived cost functions. Since the feasible flow set ˜F is compact and convex, Kakutani’s fixed‑point theorem guarantees a flow ˜f with ˜f ∈ B(˜f), i.e., a Wardrop equilibrium of the extended game.

Next, the paper proves a uniqueness result for edge latencies. Although the total‑flow‑to‑cost mapping is discontinuous at the critical flow level 2ω_e/a_e (where users are indifferent between the two lanes), the authors define a single‑valued, non‑decreasing function g_e(x) that coincides with the equilibrium cost on either side of the discontinuity and takes the equilibrium cost ξ_e at the kink. Using this construction they show that every equilibrium satisfies the classic variational inequality ⟨g(f^t), f−f^t⟩ ≥ 0 for all feasible total flows f, where f^t is the total flow vector of the equilibrium. This variational inequality uniquely determines the vector of edge latencies g(f^t), even though the split between regular and priority flow may not be unique. Consequently, all equilibria share the same latency profile.

The core contribution concerns pricing. Two pricing schemes are examined: (i) edge‑specific marginal‑cost (Pigouvian) pricing, where the priority fee ω_e is set exactly equal to the externality imposed by an additional infinitesimal priority user on the total latency of regular users; and (ii) a uniform fee ω applied to every edge. For linear latency functions the marginal externality on edge e equals a_e·f_e (the total flow on e). Setting ω_e = a_e·f_e^t makes each priority user internalize the exact social cost they generate. The authors prove that under this edge‑specific marginal‑cost pricing, every Wardrop equilibrium of the priority game coincides with the socially optimal flow of the original (single‑class) routing problem. Hence the price of anarchy (PoA) is exactly 1.

In contrast, a uniform fee cannot in general achieve optimality. The authors construct a family of Pigou‑type networks (two parallel links, one with constant latency, the other linear) and show that, regardless of the uniform ω chosen, the worst‑case PoA remains 4/3. This demonstrates that price discrimination across edges is essential for eliminating inefficiency in the presence of priority lanes.

The paper concludes with practical implications. Voluntary priority services are already widespread in transportation and communication systems. The theoretical results suggest that, if operators set priority fees equal to the marginal congestion cost on each edge, the self‑selection of users will lead to a socially optimal allocation without any compulsory tolls. However, simplistic uniform pricing policies may fall short, potentially leaving substantial inefficiencies. The work thus bridges the gap between the economics of congestion pricing and the operational reality of tiered service offerings, providing a rigorous foundation for designing incentive‑compatible, efficiency‑maximizing priority mechanisms.