Kemeny's constant for several families of graphs and real-world networks

Journal Article (2020)
Author(s)

Robert Kooij (TU Delft - Network Architectures and Services)

J.L.A. Dubbeldam (TU Delft - Mathematical Physics)

Research Group
Network Architectures and Services
Copyright
© 2020 Robert Kooij, J.L.A. Dubbeldam
DOI related publication
https://doi.org/10.1016/j.dam.2020.05.033
More Info
expand_more
Publication Year
2020
Language
English
Copyright
© 2020 Robert Kooij, J.L.A. Dubbeldam
Research Group
Network Architectures and Services
Volume number
285
Pages (from-to)
96-107
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

The linear relation between Kemeny's constant, a graph metric directly linked with random walks, and the effective graph resistance in a regular graph has been an incentive to calculate Kemeny's constant for various networks. In this paper we consider complete bipartite graphs, (generalized) windmill graphs and tree networks with large diameter and give exact expressions of Kemeny's constant. For non-regular graphs we propose two approximations for Kemeny's constant by adding to the effective graph resistance term a linear term related to the degree heterogeneity in the graph. These approximations are exact for complete bipartite graphs, but show some discrepancies for generalized windmill and tree graphs. However, we show that a recently obtained upper-bound for Kemeny's constant in Wang et al. (2017) based on the pseudo inverse Laplacian gives the exact value of Kemeny's constant for generalized windmill graphs. Finally, we have evaluated Kemeny's constant, its two approximations and its upper bound, for 243 real-world networks. This evaluation reveals that the upper bound is tight, with average relative error of only 0.73%. In most cases the upper bound clearly outperforms the other two approximations.