Lower threshold ground state energy and testability of minimal balanced cut density

TL;DR

This paper introduces a new lower threshold version of the ground state energy for weighted graphs and graphons, demonstrating its convergence properties and implications for testing balanced multiway cut densities in clustering.

Contribution

It defines the lower threshold ground state energy and proves its convergence properties, enabling testability of minimal balanced cut densities in graph clustering.

Findings

Convergence of higher threshold energies implies convergence of lower threshold energies.

The new energy concept allows for testability of minimal balanced multiway cut densities.

Provides a framework connecting graph limits with clustering problems.

Abstract

Lov\'asz and his coauthors defined the notion of microcanonical ground state energy $\hat{\mathcal{E}}_\mathbb{a} (G,J)$ -- borrowed from the statistical physics -- for weighted graphs $G$, where $\mathbb{a}$ is a probability distribution on $\{1,...,q\}$ and $J$ is a symmetric $q \times q$ matrix with real entries. We define a new version of the ground state energy, $\hat{\mathcal{E}}^c (G,J)=\inf_{\mathbb{a}\in A_c}\hat{\mathcal{E}}_\mathbb{a} (G,J)$, called lower threshold ground state energy, where $A_c = \{\mathbb{a} :\, a_i\ge c,\,i=1,\dots, q \}$. Both types of energies can be extended for graphons $W$, the limit objects of convergent sequences of simple graphs. In the main result of the paper it is stated that if $0\leq c_1<c_2 \leq 1$, then the convergence of the sequences $(\hat{\mathcal{E}}^{c_2/q} (G_n,J))$ for each $J$ implies convergence of the sequences $(\hat{\mathcal{E}}^{c_1/q} (G_n,J))$ for each $J$. As a byproduct one can derive in a natural way the testability of minimum balanced multiway cut densities, that is one of the fundamental problems of cluster analysis.