An arXiv preprint gives a conditional answer to a central problem in spectral graph theory: when a weighted graph is split into connected pieces, can one partition actually attain the lowest possible spectral energy? Under stated compactness, Poincaré-type and path conditions, the analysis establishes connected minimizers for the Dirichlet, Neumann and boundaryless versions of the problem.
The setting is a simple, connected weighted graph that may be finite or countably infinite; its vertex and edge weights are positive and locally finite. A partition uses all the vertices, divides them into k nonempty, pairwise disjoint sets, and requires each induced subgraph to be connected.
The question behind the score
The paper compares three eigenvalue-based energies—Dirichlet, Neumann and boundaryless—and treats convergence of subgraphs point by point. That matters because a lowest value can be an infimum, approached by candidate partitions, without being attained by any one of them.
Existence comes with conditions
The authors lay out a one-way hierarchy of sufficient assumptions. With finite total vertex measure, one assumption implies another, followed by canonical compactifiability and then compact embedding of the relevant function space.
When the compact-embedding condition holds, a connected k-partition attains the Dirichlet spectral infimum for every p in the stated range, 1≤p≤∞, and every natural-number k. The corresponding Neumann and boundaryless results require their respective Poincaré-type and path assumptions.
The paper also gives conditional comparisons and lower bounds. With finite total vertex measure and finite total length, the boundaryless minimum is no greater than the Neumann minimum and has an explicit positive lower bound; corresponding lower bounds are given for the Dirichlet and Neumann minima.
A ladder shows why the boundary matters
Those abstract statements are illustrated by finite two-ladder graphs with every vertex and edge weight equal to 1, divided into k=2 cells. For the Dirichlet energy, the unique minimizing configuration depends on parity: the paper identifies one shape when the ladder length is n=2m and another when it is n=2m+1.
The boundaryless problem has path-shaped minimizers, including horizontal and L-shaped configurations. The Neumann minimum has the same value as the boundaryless minimum in this example, but only the horizontal path configuration minimizes Neumann energy.
Infinite size raises a separate problem
On the infinite two-ladder, finite total vertex measure and finite total length imply that all three functionals admit a minimizing k-partition when k≥2 and 1≤p≤∞. This is the well-posed case for the infinite example, and it depends on the stated summability conditions.
With unit vertex and edge weights instead, the boundaryless and Neumann quantities are not well-defined on an infinite cluster. Under the paper’s generalized Dirichlet-infimum convention, the partition infimum is 0 for every k≥2 and 1≤p≤∞, and is stated to be attained only for k=2.
A conditional mathematical result
The result has a narrow reach. The general existence theorems are conditional; the graph model is restricted to simple connected graphs with positive, locally finite weights, and the ladder classifications apply only to the stated two-ladder geometry, weights and partition setting. The paper does not establish that the displayed lower bounds are sharp in general.
The document is an arXiv version 1 preprint dated 20 Aug 2026. It is a mathematical analysis based on proofs, inequalities and worked graph examples, not empirical validation or statistical uncertainty analysis.
Paper data and sources
Original title: Spectral minimal partitions of combinatorial graphs
Authors: Matthias Hofmann, James B. Kennedy, Delio Mugnolo, Marvin Plümer
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text