Fair Tree Connection Games with Topology-Dependent Edge Cost
How do rational agents self-organize when trying to connect to a common target? We study this question with a simple tree formation game which is related to the well-known fair single-source connection game by Anshelevich et al. (FOCS'04) and selfish spanning tree games by Gourv\`es and Monnot...
Gespeichert in:
Hauptverfasser: | , , , , |
---|---|
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | How do rational agents self-organize when trying to connect to a common
target? We study this question with a simple tree formation game which is
related to the well-known fair single-source connection game by Anshelevich et
al. (FOCS'04) and selfish spanning tree games by Gourv\`es and Monnot
(WINE'08). In our game agents correspond to nodes in a network that activate a
single outgoing edge to connect to the common target node (possibly via other
nodes). Agents pay for their path to the common target, and edge costs are
shared fairly among all agents using an edge. The main novelty of our model is
dynamic edge costs that depend on the in-degree of the respective endpoint.
This reflects that connecting to popular nodes that have increased internal
coordination costs is more expensive since they can charge higher prices for
their routing service.
In contrast to related models, we show that equilibria are not guaranteed to
exist, but we prove the existence for infinitely many numbers of agents.
Moreover, we analyze the structure of equilibrium trees and employ these
insights to prove a constant upper bound on the Price of Anarchy as well as
non-trivial lower bounds on both the Price of Anarchy and the Price of
Stability. We also show that in comparison with the social optimum tree the
overall cost of an equilibrium tree is more fairly shared among the agents.
Thus, we prove that self-organization of rational agents yields on average only
slightly higher cost per agent compared to the centralized optimum, and at the
same time, it induces a more fair cost distribution. Moreover, equilibrium
trees achieve a beneficial trade-off between a low height and low maximum
degree, and hence these trees might be of independent interest from a
combinatorics point-of-view. We conclude with a discussion of promising
extensions of our model. |
---|---|
DOI: | 10.48550/arxiv.2009.10988 |