Diagonal of Pseudoinverse of Graph Laplacian: Fast Estimation and Exact Results

TL;DR

This paper introduces a nearly linear time algorithm to efficiently estimate the diagonal entries of the pseudoinverse of a graph Laplacian, which are crucial for various network analysis metrics, and provides exact formulas for specific graph types.

Contribution

The paper presents a novel, theoretically guaranteed approximation algorithm for the diagonal of the Laplacian pseudoinverse, enabling scalable analysis of large graphs, along with exact solutions for certain network classes.

Findings

Algorithm achieves near-linear time complexity.

High accuracy demonstrated on real-world networks.

Exact formulas verified for Koch networks and recursive trees.

Abstract

The diagonal entries of pseudoinverse of the Laplacian matrix of a graph appear in many important practical applications, since they contain much information of the graph and many relevant quantities can be expressed in terms of them, such as Kirchhoff index and current flow centrality. However, a na\"{\i}ve approach for computing the diagonal of a matrix inverse has cubic computational complexity in terms of the matrix dimension, which is not acceptable for large graphs with millions of nodes. Thus, rigorous solutions to the diagonal of the Laplacian matrices for general graphs, even for particluar graphs are much less. In this paper, we propose a theoretically guaranteed estimation algorithm, which approximates all diagonal entries of the pseudoinverse of a graph Laplacian in nearly linear time with respect to the number of edges in the graph. We execute extensive experiments on real-life networks, which indicate that our algorithm is both efficient and accurate. Also, we determine exact expressions for the diagonal elements of pseudoinverse of the Laplacian matrices for Koch networks and uniform recursive trees, and compare them with those obtained by our approximation algorithm. Finally, we use our algorithm to evaluate the Kirchhoff index of three deterministic model networks, for which the Kirchhoff index can be rigorously determined. These results further show the effectiveness and efficiency of our algorithm.