pith. sign in

arxiv: 2604.11682 · v1 · submitted 2026-04-13 · 🧮 math.PR · math-ph· math.MP

Semilocalization for inhomogeneous random graphs

Thomas Buc-d'Alch\'e , Antti Knowles This is my paper

Pith reviewed 2026-05-10 15:11 UTC · model grok-4.3

classification 🧮 math.PR math-phmath.MP
keywords inhomogeneous random graphseigenvector semilocalizationspectral edgesdegree sequencespruning procedurerandom treesadjacency matrix
0
0 comments X

The pith

In random graphs with bounded degree moments, eigenvectors near spectral edges concentrate mass on small sets of resonant vertices.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that for adjacency matrices of inhomogeneous random graphs built from degree sequences with bounded mean and variance, eigenvectors associated with eigenvalues near the spectrum edges are semilocalized. Their l2-mass gathers around a limited collection of vertices whose local structure resonates with the eigenvalue value. For the largest and smallest eigenvalues, this tightens further to localization on exactly one vertex. The result matters because it clarifies how degree inhomogeneity shapes the extreme spectral behavior of large networks, where uniform random graph models fail to capture varying connectivity.

Core claim

We show that near the edges of the spectrum, the eigenvectors are semilocalized in the sense that their mass concentrates around a small set of resonant vertices. For the extremal eigenvalues, we establish localization around a single vertex. This follows from a new economical pruning procedure that extracts a forest from the original graph and couples its adjacency matrix to that of random trees with independent edges, yielding effective estimates even under strong degree inhomogeneity.

What carries the argument

The economical pruning procedure, which extracts a forest from the graph whose adjacency matrix couples locally to random trees with independent edges.

If this is right

  • Semilocalization of eigenvectors holds near the spectrum edges.
  • Single-vertex localization occurs for the largest and smallest eigenvalues.
  • The pruning-to-tree coupling produces explicit control on eigenvector entries despite high degree variation.
  • Spectral edge behavior reduces to local tree-like structures around resonant vertices.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Extreme eigenvalues are likely governed by isolated high-degree vertices or small clusters.
  • The same pruning technique may apply to other sparse inhomogeneous matrix ensembles.
  • Numerical checks on simulated graphs can directly measure the support size of computed eigenvectors.

Load-bearing premise

The empirical degree sequence has bounded mean and variance so that the pruning procedure can extract a forest that couples effectively to random trees.

What would settle it

Construct a large inhomogeneous random graph from a bounded-moment degree sequence, compute an eigenvector for an eigenvalue near the spectral edge, and check whether its mass remains spread over many vertices without concentrating on a small resonant set.

Figures

Figures reproduced from arXiv: 2604.11682 by Antti Knowles, Thomas Buc-d'Alch\'e.

Figure 2.1
Figure 2.1. Figure 2.1: A down-up path. 9 [PITH_FULL_IMAGE:figures/full_fig_p009_2_1.png] view at source ↗
Figure 2.2
Figure 2.2. Figure 2.2: A representation of the event EDk,r. The simple paths can share a vertex, but cannot be edge-intersecting. Definition 2.8. Two paths γ and γ ′ are edge-intersecting if there exists i and j such that γi = γ ′ j and either γi−1 = γ ′ j−1 or γi−1 = γ ′ j+1. Two paths γ and γ ′ are edge-disjoint if they are not edge-intersecting. We shall consider the following event concerning edge-disjoint paths EDk,r(x) =… view at source ↗
Figure 2.3
Figure 2.3. Figure 2.3: Construction of the paths γˆ and γˆ ′ . The paths γ and γ ′ are edge-intersecting, respectively from y to z and y ′ to z ′ . where V˜ c is the vertex set of the connected component G˜ c of G˜. Since each connected component has at least two vertices, we have # n G˜ c connected component with V˜ c odd o ⩽ k. This means that doing so, we find at least k pairs: 2k distinct vertices {y˜i , z˜i} and k simple … view at source ↗
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
read the original abstract

We analyse the eigenvectors of the adjacency matrix of a random inhomogeneous graph constructed from a specified degree sequence. We assume that the empirical degree sequence has bounded mean and variance. We show that near the edges of the spectrum, the eigenvectors are semilocalized in the sense that their mass concentrates around a small set of resonant vertices. For the extremal eigenvalues, we establish localization around a single vertex. In order to obtain effective estimates in the presence of highly inhomogeneous degrees, we introduce a new economical pruning procedure that carefully extracts a forest from the original graph, whose adjacency matrix is compared to that of the original graph using a suitably constructed local coupling to random trees with independent edges.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper claims that for inhomogeneous random graphs with empirical degree sequence having bounded mean and variance, the eigenvectors of the adjacency matrix exhibit semilocalization near the spectrum edges, concentrating their mass around small sets of resonant vertices. For the extremal eigenvalues, the eigenvectors localize around a single vertex. The proof introduces an economical pruning procedure to extract a forest from the graph and uses local coupling to random trees with independent edges to analyze the structure.

Significance. If these results are rigorously established, the paper would make a significant contribution to the spectral analysis of random graphs with inhomogeneous degrees. This extends localization results to settings without bounded maximum degree, which is relevant for modeling real-world networks. The new pruning procedure could be useful in other contexts involving perturbations of graph spectra. The local coupling to trees provides a solid foundation for the analysis when the pruning is controlled.

major comments (1)
  1. [Economical pruning procedure] The central claim depends on the pruning producing a forest whose adjacency matrix is close enough to the original to preserve eigenvector mass concentration around resonant vertices. However, with only bounded mean and variance (not maximum degree), high-degree vertices may lead to dependencies that the pruning does not fully eliminate. A quantitative error bound on the perturbation (e.g., in operator norm or on the relevant quadratic forms) that scales appropriately with the distance to the bulk spectrum is necessary to control the effect near the edges. Without this, the semilocalization may not hold uniformly.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for identifying a point that merits further clarification. We address the major comment below and will incorporate additional quantitative details in the revision.

read point-by-point responses

  1. Referee: [Economical pruning procedure] The central claim depends on the pruning producing a forest whose adjacency matrix is close enough to the original to preserve eigenvector mass concentration around resonant vertices. However, with only bounded mean and variance (not maximum degree), high-degree vertices may lead to dependencies that the pruning does not fully eliminate. A quantitative error bound on the perturbation (e.g., in operator norm or on the relevant quadratic forms) that scales appropriately with the distance to the bulk spectrum is necessary to control the effect near the edges. Without this, the semilocalization may not hold uniformly.

    Authors: The economical pruning procedure is constructed precisely to eliminate cycles and long-range dependencies while retaining the local neighborhoods of resonant vertices; high-degree vertices are handled by removing excess edges in a controlled manner that does not affect the local tree-like structure. The subsequent comparison to independent-edge random trees is performed via a local coupling whose total-variation distance is bounded using only the first two moments of the degree sequence. This coupling directly yields error estimates on the quadratic forms of the adjacency matrices that decay exponentially with graph distance and are uniform in the spectral parameter as long as one remains a fixed distance away from the bulk edge. Nevertheless, to make the dependence on the distance to the bulk explicit and to address uniformity concerns, we will add a dedicated lemma stating the operator-norm perturbation bound (restricted to the relevant eigenspace) together with its scaling in the spectral gap. This addition will not alter the main arguments but will render the control fully quantitative. revision: yes

Circularity Check

0 steps flagged

No significant circularity; new pruning and external coupling arguments

full rationale

The paper's central claims on semilocalization of eigenvectors near spectrum edges rely on a newly introduced economical pruning procedure that extracts a forest, followed by a local coupling of its adjacency matrix to random trees with independent edges. These steps are compared directly to the original graph using external random tree models under the assumption of bounded mean and variance in the degree sequence. No derivation step reduces by construction to fitted inputs, self-definitions, or self-citation chains; the argument remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The result rests on the bounded-moment assumption for degrees and the validity of the new pruning procedure plus local coupling to trees; no free parameters or invented physical entities are introduced.

axioms (1)
  • domain assumption Empirical degree sequence has bounded mean and variance
    Explicitly stated in the abstract as the setting in which the semilocalization holds.
invented entities (1)
  • Economical pruning procedure no independent evidence
    purpose: Extracts a forest from the inhomogeneous graph for comparison via local coupling to random trees
    New technical tool introduced to obtain effective estimates despite high degree inhomogeneity.

pith-pipeline@v0.9.0 · 5404 in / 1183 out tokens · 57625 ms · 2026-05-10T15:11:17.394288+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages

  1. [1]

    Abrahams, editor

    E. Abrahams, editor. 50 years of A nderson Localization , volume 24. World Scientific, 2010

    work page 2010

  2. [2]

    J. Alt, R. Ducatez, and A. Knowles. Delocalization transition for critical E rd o s- R \' e nyi graphs. Comm.\ Math.\ Phys. , 388(1):507--579, 2021

    work page 2021

  3. [3]

    J. Alt, R. Ducatez, and A. Knowles. Extremal eigenvalues of critical E rd o s- R \' e nyi graphs. Ann. Prob. , 49(3):1347--1401, 2021

    work page 2021

  4. [4]

    J. Alt, R. Ducatez, and A. Knowles. The completely delocalized region of the E rd o s- R \' e nyi graph. Electron. Commun. Probab. , 27:Paper No. 10, 9, 2022

    work page 2022

  5. [5]

    J. Alt, R. Ducatez, and A. Knowles. Poisson statistics and localization at the spectral edge of sparse E rd o s-- R \' e nyi graphs. Ann.\ Probab. , 51(1):277--358, 2023

    work page 2023

  6. [6]

    J. Alt, R. Ducatez, and A. Knowles. Localized phase for the E rd o s-- R \'e nyi graph. Comm. Math. Phys. , 405(1):9, 2024

    work page 2024

  7. [7]

    Anderson

    P.W. Anderson. Absence of diffusion in certain random lattices. Phys. Rev. , 109(5):1492, 1958

    work page 1958

  8. [8]

    Spectral radii of sparse random matrices

    Florent Benaych-Georges, Charles Bordenave, and Antti Knowles. Spectral radii of sparse random matrices. Ann. Inst. Henri Poincar \'e , Probab. Stat. , 56(3):2141--2161, 2020

    work page 2020

  9. [9]

    Bollob \'a s, S

    B. Bollob \'a s, S. Janson, and O. Riordan. The phase transition in inhomogeneous random graphs. Random Structures & Algorithms , 31(1):3--122, 2007

    work page 2007

  10. [10]

    Boucheron, G

    S. Boucheron, G. Lugosi, and P. Massart. Concentration inequalities: A nonasymptotic theory of independence . Oxford university press, 2013

    work page 2013

  11. [11]

    Chung and L

    F. Chung and L. Lu. Connected components in random graphs with given expected degree sequences. Ann. Comb. , 6(2):125--145, 2002

    work page 2002

  12. [12]

    Erd o s, A

    L. Erd o s, A. Knowles, H.-T. Yau, and J. Yin. Spectral statistics of E rd o s- R \'enyi graphs I : Local semicircle law. Ann. Prob. , 41:2279--2375, 2013

    work page 2013

  13. [13]

    Evers and A.D

    F. Evers and A.D. Mirlin. Anderson transitions. Rev. Mod. Phys. , 80(4):1355, 2008

    work page 2008

  14. [14]

    Y. He, A. Knowles, and M. Marcozzi. Local law and complete eigenvector delocalization for supercritical E rd o s- R \' e nyi graphs. Ann. Prob. , 47(5):3278--3302, 2019

    work page 2019

  15. [15]

    Hiesmayr and T

    E. Hiesmayr and T. McKenzie. The spectral edge of constant degree E rd o s-- R \'e nyi graphs. Random Structures & Algorithms , 66(3):e70011, 2025

    work page 2025

  16. [16]

    Random Graphs and Complex Networks

    Remco van der Hofstad. Random Graphs and Complex Networks . Cambridge University Press, 2016

    work page 2016

  17. [17]

    Lee and T.V

    P.A. Lee and T.V. Ramakrishnan. Disordered electronic systems. Rev. Mod. Phys. , 57(2):287, 1985

    work page 1985

  18. [18]

    Lagendijk, B

    A. Lagendijk, B. Van Tiggelen, and D.S. Wiersma. Fifty years of A nderson localization. Phys. Today , 62(8):24--29, 2009

    work page 2009