[Paper Review] On the Structure of Equilibria in Basic Network Formation

Sotiris Nikoletseas, Panagiota N. Panagopoulou+2 more|arXiv (Cornell University)|Jun 7, 2013

Game Theory and Applications10 references4 citations

TL;DR

This paper provides a probabilistic structural characterization of equilibrium graphs in basic network formation games, where nodes swap edges to minimize communication costs. It proves that equilibrium graphs have bounded diameter in terms of the size of the largest $k$-vicinity and confirms a conjecture by [ADHL10] for dense and high-$k$-vicinity graphs, while introducing a new degree-based cost model that admits an exact potential and guarantees equilibrium graphs contain an induced star.

ABSTRACT

We study network connection games where the nodes of a network perform edge swaps in order to improve their communication costs. For the model proposed by Alon et al. (2010), in which the selfish cost of a node is the sum of all shortest path distances to the other nodes, we use the probabilistic method to provide a new, structural characterization of equilibrium graphs. We show how to use this characterization in order to prove upper bounds on the diameter of equilibrium graphs in terms of the size of the largest $k$-vicinity (defined as the the set of vertices within distance $k$ from a vertex), for any $k \geq 1$ and in terms of the number of edges, thus settling positively a conjecture of Alon et al. in the cases of graphs of large $k$-vicinity size (including graphs of large maximum degree) and of graphs which are dense enough. Next, we present a new swap-based network creation game, in which selfish costs depend on the immediate neighborhood of each node; in particular, the profit of a node is defined as the sum of the degrees of its neighbors. We prove that, in contrast to the previous model, this network creation game admits an exact potential, and also that any equilibrium graph contains an induced star. The existence of the potential function is exploited in order to show that an equilibrium can be reached in expected polynomial time even in the case where nodes can only acquire limited knowledge concerning non-neighboring nodes.

Motivation & Objective

• To provide a new structural characterization of swap equilibrium graphs in basic network formation games using the probabilistic method.
• To resolve a conjecture by [ADHL10] on the diameter of equilibrium graphs for graphs with large $k$-vicinity or high edge density.
• To introduce and analyze a new swap-based network creation game where node cost depends on the sum of neighbor degrees.
• To prove the existence of an exact potential function in the new model and establish structural properties of equilibrium graphs.
• To demonstrate that equilibrium can be reached in expected polynomial time even with limited node knowledge.

Proposed method

• The probabilistic method is applied to analyze the structure of equilibrium graphs, focusing on vertex distributions relative to pairs of nodes.
• The paper defines $A_{u,v}(c)$ as the set of vertices whose distances to $u$ and $v$ differ by exactly $c$, enabling a precise characterization of graph symmetry and balance.
• It derives the inequality $\sum_{c=0}^{\infty} c|A_{u,v}(c)| \leq \frac{\delta'_{uv}+1}{\delta'_{uv}-1}n$ for any pair $u,v$, where $\delta'_{uv} = \min\{\deg_{-1}(u), \deg_{-1}(v)\}$, to bound structural imbalances.
• A new network creation game is proposed where a node's cost is the sum of the degrees of its neighbors, leading to a potential function that ensures convergence.
• The existence of an exact potential function is proven, enabling analysis of convergence dynamics and equilibrium stability.
• Contradiction arguments via strategic swaps are used to prove structural properties, such as the absence of long induced cycles and the presence of induced stars.

Experimental results

Research questions

• RQ1 What structural properties characterize equilibrium graphs in the basic network formation game under edge-swapping dynamics?
• RQ2 Does the diameter of equilibrium graphs remain bounded in terms of the size of the largest $k$-vicinity, as conjectured by [ADHL10]?
• RQ3 Can a new network creation game with degree-sum-based cost function admit an exact potential function?
• RQ4 What structural features must equilibrium graphs in the new model possess, particularly regarding induced subgraphs?
• RQ5 Can equilibrium be reached efficiently in the new model even with limited local knowledge?

Key findings

• The paper proves that for any equilibrium graph, $\sum_{c=0}^{\infty} c|A_{u,v}(c)| \leq \frac{\delta'_{uv}+1}{\delta'_{uv}-1}n$, providing a strong structural constraint on equilibrium topology.
• The diameter of equilibrium graphs is bounded in terms of the size of the largest $k$-vicinity, confirming the [ADHL10] conjecture for graphs with large $k$-vicinity or high edge density.
• In the new degree-sum-based network creation game, every equilibrium graph contains an induced star, a structural property not present in the original model.
• The new game admits an exact potential function, implying that any sequence of profitable swaps converges to a pure Nash equilibrium.
• Equilibrium can be reached in expected polynomial time even when nodes have limited knowledge of non-neighboring nodes, due to the existence of the potential function.