betaFeedback

WorksComputer Science

A Combinatorial Characterization of Immunity to the Strong Braess Paradox

Abstract

In a nonatomic congestion game, a strong Braess paradox occurs when reducing resource costs, population demands, or both increases at least one population's equilibrium cost. We characterize immunity when each population's available strategies form a fixed finite family of nonempty subsets of a common resource set. Immunity means that no such paradox can arise for any nonnegative continuous nondecreasing resource cost functions common to all populations and any positive demand vector. Our characterization resolves the open question posed by Fujishige et al.(2017) without varying the resource embedding. A menu is positively exposed if it consists of all available strategies of minimum total weight for some strictly positive resource weights. For a single population, we prove that immunity holds if and only if the column matroid of the augmented incidence matrix is series-parallel for every positively exposed menu. For multiple populations, we consider one menu for each population, all exposed by the same positive weights. The associated dual matroid includes one labeled element for each population to track its demand. Immunity holds if and only if, for every such collection of menus, each marked resource block is series-parallel and all demand labels retained in that block are parallel. We establish these characterizations through an affine response criterion and a blockwise ray lemma. Our results recover the immunity of matroid base systems under arbitrary resource sharing and the series-parallel routing criterion for complete acyclic path systems. Whenever a strategy system is not immune, we construct a strong Braess paradox with all demands equal to one and affine resource costs having strictly positive slopes and intercepts, by lowering just one intercept while keeping it positive. When strategies are listed explicitly, nonimmunity has polynomial-size certificates.

Paper details

vlnc:W_nfb5g7dx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
arXiv nonexclusive license
Usage

views · 4

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Generative AI tools

View statement
Looking for related papers…