pith. sign in

arxiv: 2605.25736 · v1 · pith:6XPXMYPPnew · submitted 2026-05-25 · ❄️ cond-mat.dis-nn · cond-mat.stat-mech· math.PR

Spectral properties of non-Hermitian real random matrices with long-range correlations

Ulysse Marquis This is my paper

Pith reviewed 2026-06-29 19:30 UTC · model grok-4.3

classification ❄️ cond-mat.dis-nn cond-mat.stat-mechmath.PR
keywords non-Hermitian random matriceslong-range correlationscircular laweigenvalue distributionspectral radiusreal eigenvaluesSaturn effectextended central limit theorem
0
0 comments X

The pith

Long-range correlations cause the eigenvalue spectrum of non-Hermitian real random matrices to spread without bound when the decay exponent falls below 1.

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

The paper studies non-Hermitian real random matrices whose entries are correlated according to a power-law decay |r-r'|^{-α}. It finds that the circular law for eigenvalue placement holds only when α exceeds 1, recovering an effective bounded radius. Below α=1 the radial density broadens with matrix size N and the spectral radius grows as a power of N whose exponents track those appearing in the extended central limit theorem for sums. At α=1 the density becomes self-similar while the radius grows slowly. The same correlations also increase clustering among real eigenvalues and delay the resorption of the Saturn effect.

Core claim

For entry correlations decaying as |r-r'|^{-α}, the circular law is recovered with an effective radius only when α>1; for α<1 the eigenvalue distribution broadens with matrix size and the spectral radius grows according to a power law whose exponents are numerically close to the fluctuation exponents of the extended central limit theorem, while α=1 yields a self-similar density with slowly growing radius; long-range correlations further enhance real-eigenvalue clustering and slow the resorption of the Saturn effect.

What carries the argument

The correlation decay exponent α that controls the transition between bounded and power-law-growing spectral radius.

If this is right

  • For α>1 the spectrum remains confined inside an effective circle whose radius is set by the correlation strength.
  • For α<1 the outer edge of the spectrum grows as a power of matrix size N.
  • The power-law exponents for the spectral radius numerically match the fluctuation exponents of the extended central limit theorem.
  • Real eigenvalues form denser clusters and the Saturn effect disappears more slowly than in the uncorrelated case.
  • The α=1 boundary produces a self-similar eigenvalue density whose radius grows only logarithmically or slower.

Where Pith is reading between the lines

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

  • Physical systems whose interactions or couplings decay with exponent α<1 may exhibit matrix-size-dependent spectral widths rather than fixed radii.
  • Analytic derivations linking the observed powers directly to the extended central limit theorem could replace the current numerical matching.
  • The reported transition suggests a distinct universality class for non-Hermitian matrices with power-law correlations that could be tested in other ensembles.

Load-bearing premise

Finite-size numerical simulations capture the true asymptotic scaling of the spectral radius and eigenvalue density without significant finite-size or boundary artifacts that would alter the reported power-law exponents.

What would settle it

A direct computation on successively larger matrices showing that the spectral radius saturates to a finite value independent of N for α<1, or that the fitted exponents deviate from those of the extended central limit theorem.

Figures

Figures reproduced from arXiv: 2605.25736 by Ulysse Marquis.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Two-dimensional histogram of eigenvalues of 512 [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Distribution of the spectral radius [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Three behaviors emerge: at α > 1, a bounded law with rescaled radius2 re(α) ≈ 1.4 at α = 1.9 is re￾covered (see [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Eigenvalue spacing ratio distributions for [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. a) Expected number of real eigenvalues as a function [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 6
Figure 6. Figure 6: While it decreases notably for α > 1, the de￾cay rate is lower for α = 1 and even more for α < 1, with more than half of the leading eigenvalues being real up to L = 512. This observation raises questions about the appearance of a novel scale for real eigenvalues when there are strong long-range correlations: despite their small proportion O(1/ √ L), they dominate the high-end of the spectrum. IV. DISCUSSI… view at source ↗
read the original abstract

We investigate the spectral properties of non-Hermitian real random matrices whose entries exhibit long-range correlations decaying as~$|r-r'|^{-\alpha}$. We find a progressive breakdown of the circular law, controlled by the decrease of~$\alpha$. In all cases, the radial eigenvalue density decreases away from the origin. At~$\alpha>1$, an effective radius, reminiscent of the circular law, is retrieved, while instead, for~$\alpha<1$, the eigenvalue distribution broadens with matrix size and its spectral radius grows like a power law, with exponents numerically close to the exponents controlling the magnitude of fluctuations in the extended central limit theorem. The case~$\alpha=1$ appears as a case with self-similar eigenvalue density, and slowly growing spectral radius. Long-range correlations also enhance clustering of real eigenvalues and slow the resorption of the Saturn effect. These results reveal a correlation-driven transition and suggest the emergence of a new universality class for correlated non-Hermitian random matrices.

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

3 major / 2 minor

Summary. The manuscript studies the eigenvalue spectra of non-Hermitian real random matrices whose entries have long-range correlations decaying as |r-r'|^(-α). It reports a progressive breakdown of the circular law with decreasing α: for α>1 an effective radius is recovered, while for α<1 the eigenvalue distribution broadens with matrix size N and the spectral radius grows as a power law whose exponents are numerically close to those governing fluctuations in the extended central limit theorem; the α=1 case exhibits self-similar density and slowly growing radius. Long-range correlations are also found to enhance clustering of real eigenvalues and to slow resorption of the Saturn effect, pointing to a correlation-driven transition and a possible new universality class.

Significance. If the reported scalings prove robust, the work identifies a correlation-induced regime in non-Hermitian random matrices that links spectral properties directly to the fluctuation exponents of the extended CLT. The numerical agreement between the spectral-radius exponents and the CLT exponents is a concrete observation that could motivate analytic derivations; the additional remarks on real-eigenvalue clustering and the Saturn effect supply further testable signatures of the long-range regime.

major comments (3)
  1. [Results for α<1 (abstract and corresponding numerical sections)] The central claim that, for α<1, the spectral radius grows as a power law in N with exponents numerically close to the extended-CLT fluctuation exponents rests entirely on finite-N diagonalizations. No range of matrix sizes, number of realizations, fitting procedure, goodness-of-fit metrics, or convergence checks with increasing N are supplied, so it is impossible to determine whether the reported exponents survive the N→∞ limit or are influenced by the N-dependent effective range of the long-range correlations.
  2. [Discussion of the α<1 regime] No analytic bound, scaling argument, or continuum-limit construction is given to control the extrapolation of the observed power-law regime. Because the correlation length grows with N, any apparent power-law window could be a transient crossover whose effective exponent drifts at larger N; without such a control the numerical match to the CLT exponents remains provisional.
  3. [Abstract and § on α>1 case] The abstract states that the radial eigenvalue density decreases away from the origin and that an effective radius is retrieved for α>1, yet the manuscript provides no quantitative definition of this effective radius, no comparison with the standard circular-law radius, and no error estimates on the density profiles, making it difficult to assess how sharply the circular law is recovered or broken.
minor comments (2)
  1. [Introduction] The notation used for the correlation function |r-r'|^(-α) should be introduced with an explicit definition of the indices r,r' (e.g., row/column indices) already in the introduction.
  2. [Figure captions] Figure captions for the eigenvalue-density plots should state the matrix sizes, ensemble sizes, and binning parameters used to generate each panel.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the careful and constructive review. We address each major comment below and indicate the revisions that will be incorporated.

read point-by-point responses

  1. Referee: [Results for α<1 (abstract and corresponding numerical sections)] The central claim that, for α<1, the spectral radius grows as a power law in N with exponents numerically close to the extended-CLT fluctuation exponents rests entirely on finite-N diagonalizations. No range of matrix sizes, number of realizations, fitting procedure, goodness-of-fit metrics, or convergence checks with increasing N are supplied, so it is impossible to determine whether the reported exponents survive the N→∞ limit or are influenced by the N-dependent effective range of the long-range correlations.

    Authors: We agree that the numerical details must be expanded for the claims to be assessable. The revised manuscript will contain a dedicated numerical-methods subsection specifying the matrix sizes (N=128 to 4096), number of realizations (500–2000 per N), least-squares fitting protocol on log-log plots, R² and residual diagnostics, and explicit convergence plots of the fitted exponent versus maximum N. revision: yes

  2. Referee: [Discussion of the α<1 regime] No analytic bound, scaling argument, or continuum-limit construction is given to control the extrapolation of the observed power-law regime. Because the correlation length grows with N, any apparent power-law window could be a transient crossover whose effective exponent drifts at larger N; without such a control the numerical match to the CLT exponents remains provisional.

    Authors: We acknowledge that no analytic control is supplied and that the power-law regime is observed only at finite N. The revision will add an explicit paragraph stating the provisional nature of the extrapolation, sketching a heuristic scaling argument that relates the correlation decay to the extended-CLT exponents, and noting that a rigorous continuum-limit analysis lies outside the present numerical scope. revision: partial

  3. Referee: [Abstract and § on α>1 case] The abstract states that the radial eigenvalue density decreases away from the origin and that an effective radius is retrieved for α>1, yet the manuscript provides no quantitative definition of this effective radius, no comparison with the standard circular-law radius, and no error estimates on the density profiles, making it difficult to assess how sharply the circular law is recovered or broken.

    Authors: We accept the need for a quantitative definition. The revised text will define the effective radius (e.g., radius enclosing 99 % of eigenvalues), compare it to the circular-law value √N, and display all radial densities with standard-error bars obtained from the ensemble. revision: yes

standing simulated objections not resolved
  • A rigorous analytic bound or continuum-limit construction establishing the power-law growth of the spectral radius for α<1 in the N→∞ limit.

Circularity Check

0 steps flagged

No circularity: claims rest on direct numerical diagonalization without fitted-parameter loops or self-citation reduction

full rationale

The paper reports eigenvalue statistics obtained from explicit matrix diagonalization of finite-N ensembles with prescribed long-range correlations. The central observations—for α<1 the spectral radius grows as a power law whose exponent is numerically close to an extended-CLT fluctuation exponent—are presented as simulation outputs, not as predictions derived from any fitted model or ansatz that would reduce to the input data by construction. No equations are given that define a quantity in terms of itself, no parameter is fitted on a subset and then relabeled a prediction, and the abstract contains no load-bearing self-citation. The finite-size extrapolation concern raised by the skeptic is a question of numerical convergence, not a circularity in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; all such elements remain unknown.

pith-pipeline@v0.9.1-grok · 5699 in / 1122 out tokens · 47348 ms · 2026-06-29T19:30:11.367221+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

42 extracted references · 31 canonical work pages · 3 internal anchors

  1. [1]

    Flow phenomena in rocks: from continuum models to fractals, percolation, cellular automata, and simulated annealing

    Sahimi, M. Flow phenomena in rocks: from continuum models to fractals, percolation, cellular automata, and simulated annealing. Rev. Mod. Phys..65, 1393-1534 (1993,10), https://link.aps.org/doi/10.1103/RevModPhys.65.1393

    work page doi:10.1103/revmodphys.65.1393 1993

  2. [2]

    & Garcı´ a-Ojalvo, J

    Sagu´ es, F., Sancho, J. & Garcı´ a-Ojalvo, J. Spatiotemporal order out of noise. Rev. Mod. Phys..79, 829-882 (2007,7), https://link.aps.org/doi/10.1103/RevModPhys.79.829

    work page doi:10.1103/revmodphys.79.829 2007

  3. [3]

    & Hohenberg, P

    Cross, M. & Hohenberg, P. Pattern formation outside of equilibrium.Rev. Mod. Phys..65, 851-1112 (1993,7), https://link.aps.org/doi/10.1103/RevModPhys.65.851

    work page doi:10.1103/revmodphys.65.851 1993

  4. [4]

    H. E. Stanley, Introduction to Phase Transitions and Critical Phenomena, Oxford University Press, New York (1987)

    1987

  5. [5]

    & Stanley, H

    Podobnik, B. & Stanley, H. Detrended Cross-Correlation Analysis: A New Method for Analyzing Two Nonstation- ary Time Series.Physical Review Letters.100(2008,2), http://dx.doi.org/10.1103/PhysRevLett.100.084102

    work page doi:10.1103/physrevlett.100.084102 2008

  6. [6]

    Drozdz, S., Jarosz, P., Kwapien, J., Skupien, M., & Watorek, M. (2025). Detrended Cross-Correlations and Their Random Matrix Limit: An Example from the Cryptocurrency Market. Entropy, 27(12), 1236. https://doi.org/10.3390/e27121236

    work page doi:10.3390/e27121236 2025

  7. [7]

    Spectral Graph Theory

    Chung, F. Spectral Graph Theory. (American Mathemat- ical Society,1997)

    1997

  8. [8]

    & Vespignani, A

    Barrat, A., Barthelemy, M. & Vespignani, A. Dynamical Processes on Complex Networks. (Cambridge University Press,2008)

    2008

  9. [9]

    Potters M, Bouchaud J-P

    1. Potters M, Bouchaud J-P. A First Course in Random Matrix Theory: For Physicists, Engineers and Data Sci- entists. Cambridge University Press; 2020

    2020

  10. [10]

    Log-Gases and Random Matrices

    Forrester, P. Log-Gases and Random Matrices. (Prince- ton University Press,2010)

    2010

  11. [11]

    & Castillo, I

    Rogers, T. & Castillo, I. Cavity approach to the spectral density of non-Hermitian sparse matrices.Phys. Rev. E. 79, 012101 (2009,1)

    2009

  12. [12]

    & Bouchaud, J

    Patil, N., Aguirre-L´ opez, F. & Bouchaud, J. The spec- tral boundary of block structured random matrices. Journal Of Physics: Complexity.5, 035001 (2024,7), 7 https://doi.org/10.1088/2632-072X/ad5cba

    work page doi:10.1088/2632-072x/ad5cba 2024

  13. [13]

    & Schomerus, H

    Aceituno, P., Rogers, T. & Schomerus, H. Uni- versal hypotrochoidic law for random matrices with cyclic correlations.Physical Review E.100(2019,7), http://dx.doi.org/10.1103/PhysRevE.100.010302

    work page doi:10.1103/physreve.100.010302 2019

  14. [14]

    & Galla, T

    Baron, J., Jewell, T., Ryder, C. & Galla, T. Eigenvalues of Random Matrices with Gen- eralized Correlations: A Path Integral Ap- proach.Physical Review Letters.128(2022,3), http://dx.doi.org/10.1103/PhysRevLett.128.120601

    work page doi:10.1103/physrevlett.128.120601 2022

  15. [15]

    & Samukhin, A

    Dorogovtsev, S., Goltsev, A., Mendes, J. & Samukhin, A. Spectra of complex net- works.Phys. Rev. E.68, 046109 (2003,10), https://link.aps.org/doi/10.1103/PhysRevE.68.046109

    work page doi:10.1103/physreve.68.046109 2003

  16. [16]

    Statistical Ensembles of Com- plex, Quaternion, and Real Matrices.Journal Of Mathematical Physics.6, 440-449 (1965,3), https://doi.org/10.1063/1.1704292

    Ginibre, J. Statistical Ensembles of Com- plex, Quaternion, and Real Matrices.Journal Of Mathematical Physics.6, 440-449 (1965,3), https://doi.org/10.1063/1.1704292

    work page doi:10.1063/1.1704292 1965

  17. [17]

    Mukhanov, H.A

    Bouchaud, J. & Georges, A. Anomalous diffusion in disordered media: Statistical mechanisms, mod- els and physical applications.Physics Reports.195, 127-293 (1990),https://www.sciencedirect.com/ science/article/pii/037015739090099N

    work page arXiv 1990

  18. [18]

    & Stanley, H

    Makse, H., Havlin, S., Schwartz, M. & Stanley, H. Method for generating long-range correlations for large systems.Physical Review E.53, 5445-5449 (1996,5), http://dx.doi.org/10.1103/PhysRevE.53.5445

    work page doi:10.1103/physreve.53.5445 1996

  19. [19]

    Pan, Zhao Q

    Victor Y. Pan, Zhao Q. Chen: The Complexity of the Matrix Eigenproblem. STOC 1999: 507-516

    1999

  20. [20]

    Random matrices: the circu- lar law

    Tao, Terence; Vu, Van. Random matrices: the circu- lar law. Commun. Contemp. Math. 10 (2008), no. 2, 261–307

    2008

  21. [21]

    & Sommers, H

    Lehmann, N. & Sommers, H. Eigenvalue statistics of ran- dom real matrices.Phys. Rev. Lett..67, 941-944 (1991,8), https://link.aps.org/doi/10.1103/PhysRevLett.67.941

    work page doi:10.1103/physrevlett.67.941 1991

  22. [22]

    Edelman, A. The Probability that a Random Real Gaussian Matrix has k Real Eigenvalues, Re- lated Distributions, and the Circular Law.Jour- nal Of Multivariate Analysis.60, 203-232 (1997), https://www.sciencedirect.com/science/ article/pii/S0047259X9691653X

    1997

  23. [23]

    & Shub, M

    Edelman, A., Kostlan, E. & Shub, M. How Many Eigenvalues of a Random Matrix are Real?.Journal Of The American Mathematical Society.7, 247-267 (1994), http://www.jstor.org/stable/2152729

    work page arXiv 1994

  24. [24]

    & Akemann, G

    Kanzieper, E. & Akemann, G. Statistics of Real Eigenvalues in Ginibre’s Ensemble of Random Real Matrices.Phys. Rev. Lett..95, 230201 (2005,11), https://link.aps.org/doi/10.1103/PhysRevLett.95.230201

    work page doi:10.1103/physrevlett.95.230201 2005

  25. [25]

    Correlation Functions of Asymmetric Real Matrices

    Borodin, A. & Sinclair, C. Correlation Func- tions of Asymmetric Real Matrices. (2008), https://arxiv.org/abs/0706.2670

    work page internal anchor Pith review Pith/arXiv arXiv 2008

  26. [26]

    & Nagao, T

    Forrester, P. & Nagao, T. Eigenvalue Statistics of the Real Ginibre Ensem- ble.Phys. Rev. Lett..99, 050603 (2007,8), https://link.aps.org/doi/10.1103/PhysRevLett.99.050603

    work page doi:10.1103/physrevlett.99.050603 2007

  27. [27]

    Will a Large Complex System be Stable?

    MAY, R. Will a Large Complex System be Stable?. Nature 238, 413–414 (1972). https://doi.org/10.1038/238413a0

    work page doi:10.1038/238413a0 1972

  28. [28]

    & Sinclair, C

    Rider, B. & Sinclair, C. Extremal laws for the real Gini- bre ensemble.The Annals Of Applied Probability.24 (2014,8), http://dx.doi.org/10.1214/13-AAP958

    work page doi:10.1214/13-aap958 2014

  29. [29]

    Jinho Baik, Thomas Bothner ”The largest real eigen- value in the real Ginibre ensemble and its relation to the Zakharov–Shabat system,” The Annals of Applied Probability, Ann. Appl. Probab. 30(1), 460-501, (Febru- ary 2020)

    2020

  30. [30]

    On the distribution of the largest real eigenvalue for the real Ginibre ensemble

    Poplavskyi, M., Tribe, R. & Zaboronski, O. On the distri- bution of the largest real eigenvalue for the real Ginibre ensemble. (2016), https://arxiv.org/abs/1603.05849

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  31. [31]

    Cipolloni, G., Erd˝ os, L., Schr¨ oder, D. & Xu, Y. Directional extremal statistics for Ginibre eigenval- ues.Journal Of Mathematical Physics.63(2022,10), http://dx.doi.org/10.1063/5.0104290

    work page doi:10.1063/5.0104290 2022

  32. [32]

    Cipolloni, G., Erd˝ os, L., Schr¨ oder, D. & Xu, Y. On the rightmost eigenvalue of non-Hermitian random matrices. (2023), https://arxiv.org/abs/2206.04448

    work page arXiv 2023

  33. [33]

    & Forrester, P

    Byun, S. & Forrester, P. Progress on the study of the Ginibre ensembles II: GinOE and GinSE. (2023), https://arxiv.org/abs/2301.05022

    work page arXiv 2023

  34. [34]

    Fyodorov, & B.A

    Y.V. Fyodorov, & B.A. Khoruzhenko, Nonlinear analogue of the May-Wigner instability transition, Proc. Natl. Acad. Sci. U.S.A. 113 (25) 6827-6832, https://doi.org/10.1073/pnas.1601136113 (2016)

    work page doi:10.1073/pnas.1601136113 2016

  35. [35]

    Allesina, S., Grilli, J., Barab´ as, G. et al. Predicting the stability of large structured food webs. Nat Commun 6, 7842 (2015). https://doi.org/10.1038/ncomms8842

    work page doi:10.1038/ncomms8842 2015

  36. [36]

    & Bouchaud, J

    Moran, J. & Bouchaud, J. May’s instability in large economies.Physical Review E.100(2019,9), http://dx.doi.org/10.1103/PhysRevE.100.032307

    work page doi:10.1103/physreve.100.032307 2019

  37. [37]

    & Prosen, T

    S´ a, L., Ribeiro, P. & Prosen, T. Complex Spac- ing Ratios: A Signature of Dissipative Quan- tum Chaos.Physical Review X.10(2020,4), http://dx.doi.org/10.1103/PhysRevX.10.021019

    work page doi:10.1103/physrevx.10.021019 2020

  38. [38]

    How many zeros of a random polynomial are real?

    Edelman, A. & Kostlan, E. How many ze- ros of a random polynomial are real?. (1995), https://arxiv.org/abs/math/9501224

    work page internal anchor Pith review Pith/arXiv arXiv 1995

  39. [39]

    & Schehr, G

    Majumdar, S., Pal, A. & Schehr, G. Extreme value statistics of correlated random variables: A pedagog- ical review.Physics Reports.840pp. 1-32 (2020,1), http://dx.doi.org/10.1016/j.physrep.2019.10.005

    work page doi:10.1016/j.physrep.2019.10.005 2020

  40. [40]

    & Som- mers, H

    Sompolinsky, H., Crisanti, A. & Som- mers, H. Chaos in Random Neural Net- works.Phys. Rev. Lett..61, 259-262 (1988,7), https://link.aps.org/doi/10.1103/PhysRevLett.61.259

    work page doi:10.1103/physrevlett.61.259 1988

  41. [41]

    & Bengio, Y

    Pascanu, R., Mikolov, T. & Bengio, Y. On the diffi- culty of training recurrent neural networks.Proceedings Of The 30th International Conference On International Conference On Machine Learning - Volume 28. pp. III- 1310–III–1318 (2013)

    2013

  42. [42]

    Universal sum and product rules for random matrices.Journal Of Mathematical Physics.51(2010,9), http://dx.doi.org/10.1063/1.3481569

    Rogers, T. Universal sum and product rules for random matrices.Journal Of Mathematical Physics.51(2010,9), http://dx.doi.org/10.1063/1.3481569

    work page doi:10.1063/1.3481569 2010