First-principles multiway spectral partitioning of graphs

  title={First-principles multiway spectral partitioning of graphs},
  author={Marianna Riolo and Mark E. J. Newman},
We consider the minimum-cut partitioning of a graph into more than two parts using spectral methods. While there exist well-established spectral algorithms for this problem that give good results, they have traditionally not been well motivated. Rather than being derived from first principles by minimizing graph cuts, they are typically presented without direct derivation and then proved after the fact to work. In this paper, we take a contrasting approach in which we start with a matrix… 

Figures from this paper

Spectral methods for network community detection and graph partitioning

  • M. Newman
  • Computer Science
    Physical review. E, Statistical, nonlinear, and soft matter physics
  • 2013
It is shown that with certain choices of the free parameters appearing in these spectral algorithms the algorithms for all three problems are identical, and hence there is no difference between the modularity- and inference-based community detection methods, or between either and graph partitioning.

Optimization via Low-rank Approximation for Community Detection in Networks

This work proposes a general approach for maximizing a function of a network adjacency matrix over discrete labels by projecting the set of labels onto a subspace approximating the leading eigenvectors of the expected adjACency matrix.

Optimization via Low-rank Approximation, with Applications to Community Detection in Networks

This work proposes a general approach for maximizing a function of a network adjacency matrix over discrete labels by projecting the set of labels onto a subspace spanned by leading eigenvectors of the adjACency matrix and proves a general result on this method and demonstrates its consistency for label estimation in each case.

Signed Laplacian for spectral clustering revisited

This work revisits comparing the standard and signed Laplacians and argues that the former is more natural than the latter, also showing that the negative spectrum is actually beneficial, for spectral clustering of signed graphs.

On spectral partitioning of signed graphs

It is observed that negative eigenvalues are beneficial for spectral partitioning of signed graphs, making the Fiedler vector easier to compute.


It is shown that extremal eigenvalues of M provide an exact relaxation of the modularity measure Q, in the sense that the maximum eigenvalue of M is equal to the maximum value of Q, however at the price of being more challenging to be computed than those of M .

A link-elimination partitioning approach for application graph mapping in reconfigurable computing systems

A link-elimination partitioning approach for application graphs to reduce computations and reach an optimal solution faster as well as improving quality of solutions and time overhead.

De la détection de la structure communautaire des réseaux complexes

This thesis presents the basis of a unified theory of community detection, which builds upon recent advances of the spectral theory of complex networks and proves that a large class of detection algorithm is equivalent to an optimization problem that can be solved approximately though the spectral decomposition of a very general cost matrix.

A spectral method to detect community structure based on distance modularity matrix

An algorithm to detect community structure of networks is proposed by using spectra of distance modularity matrix that focuses on the distance of vertices within communities, rather than the most weakly connected vertex pairs or number of edges between communities.

Spectral Clustering Methods for Multiplex Networks



Spectral Partitioning: The More Eigenvectors, The Better

This work maps each graph vertex to a vector in d-dimensional space, where d is the number of eigenvectors, such that these vectors constitute an instance of the vector partitioning problem.

A tutorial on spectral clustering

This tutorial describes different graph Laplacians and their basic properties, present the most common spectral clustering algorithms, and derive those algorithms from scratch by several different approaches.

How Good is Recursive Bisection?

The most commonly used p-way partitioning method is recursive bisection (RB). It first divides a graph or a mesh into two equal-sized pieces, by a "good" bisection algorithm, and then recursively

Algorithms for graph partitioning on the planted partition model

The NP-hard graph bisection problem is to partition the nodes of an undirected graph into two equal-sized groups so as to minimize the number of edges that cross the partition. The more general graph


It is shown that lower bounds on separator sizes can be obtained in terms of the eigenvalues of the Laplacian matrix associated with a graph, which can be used to compute good separators in grid graphs.

On Spectral Clustering: Analysis and an algorithm

A simple spectral clustering algorithm that can be implemented using a few lines of Matlab is presented, and tools from matrix perturbation theory are used to analyze the algorithm, and give conditions under which it can be expected to do well.

Multiway Spectral Clustering: A Margin-based Perspective

The margin-based perspective illuminates both the relaxation and rounding aspects of spectral clustering, providing a unified analysis of existing algorithms and guiding the design of new algorithms.

Normalized cuts and image segmentation

  • Jianbo ShiJ. Malik
  • Computer Science
    Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition
  • 1997
This work treats image segmentation as a graph partitioning problem and proposes a novel global criterion, the normalized cut, for segmenting the graph, which measures both the total dissimilarity between the different groups as well as the total similarity within the groups.

On clusterings-good, bad and spectral

Two results regarding the quality of the clustering found by a popular spectral algorithm are presented, one proffers worst case guarantees whilst the other shows that if there exists a "good" clustering then the spectral algorithm will find one close to it.