Cost Sharing over Combinatorial Domains: Complement-Free Cost Functions and Beyond

TL;DR

This paper advances the understanding of combinatorial cost sharing mechanisms by introducing trace-monotonicity, designing ascending mechanisms for symmetric submodular valuations, and analyzing their budget balance and social cost approximation.

Contribution

It introduces trace-monotonicity applicable to iterative mechanisms and provides new parameterizations of cost functions for improved mechanism design in combinatorial sharing.

Findings

Mechanism is weakly group-strategyproof under certain conditions.

Achieves $O(1)$ budget balance and $O(H_n)$ social cost approximation.

Improves analysis of Moulin's Sequential Mechanism for general valuations.

Abstract

We study mechanism design for combinatorial cost sharing. Imagine that multiple items or services are available to be shared among a set of interested agents. The outcome of a mechanism in this setting consists of an assignment, determining for each item the set of players who are granted service, together with respective payments. Although there are several works studying specialized versions of such problems, there has been almost no progress for general combinatorial cost sharing domains until recently \cite{DobzinskiO17}. The main goal of our work is to further understand this interplay in terms of budget balance and social cost approximation. Towards this, we provide a refinement of cross-monotonicity (trace-monotonicity) that is applicable to iterative mechanisms. The trace here refers to the order in which players become finalized. On top of this, we also provide two parameterizations of cost functions which capture the behavior of their average cost-shares. Based on our trace-monotonicity property, we design a scheme of ascending cost sharing mechanisms which is applicable to the combinatorial cost sharing setting with symmetric submodular valuations. Using our first cost function parameterization, we identify conditions under which our mechanism is weakly group-strategyproof, $O(1)$-budget-balanced and $O(H_n)$-approximate with respect to the social cost. Finally, we consider general valuation functions and exploit our second parameterization to derive a more fine-grained analysis of the Sequential Mechanism introduced by Moulin. This mechanism is budget balanced by construction, but in general only guarantees a poor social cost approximation of $n$. We identify conditions under which the mechanism achieves improved social cost approximation guarantees.