arxiv: 2604.27992 · v1 · submitted 2026-04-30 · ❄️ cond-mat.dis-nn · cond-mat.stat-mech

Recognition: unknown

Discontinuous BBP transitions

Dario Bocchi , Giulio Biroli , Chiara Cammarota , Federico Ricci-Tersenghi

Authors on Pith no claims yet

Pith reviewed 2026-05-07 05:25 UTC · model grok-4.3

classification ❄️ cond-mat.dis-nn cond-mat.stat-mech

keywords BBP transitiondiscontinuous transitionspectral edgeeigenvector overlapfinite-size effectssignal detectionrandom matrix ensemblesWishart ensemble

0 comments

The pith

The overlap between a signal and the leading eigenvector jumps discontinuously at the BBP threshold when noise eigenvalues vanish faster than linearly at the spectral edge.

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

The Baik-Ben Arous-Peche transition marks the point at which low-rank signals embedded in high-dimensional noise become detectable through spectral methods. This paper demonstrates that the transition is not always continuous; when the density of noise eigenvalues drops faster than linearly near the edge of the spectrum, the alignment of the top eigenvector with the true signal appears suddenly rather than growing gradually. Finite-size effects create a window of informative eigenvectors that exists before the infinite-size critical point, so recovery becomes possible at lower signal-to-noise ratios but comes with large fluctuations from one sample to the next. The authors derive this behavior for deformed Gaussian and reweighted Wishart ensembles and contrast it with the smoother case that holds when the edge density vanishes only linearly.

Core claim

In deformed Gaussian and reweighted Wishart ensembles, when the eigenvalue density of the noise vanishes faster than linearly at the spectral edge, the overlap between the leading eigenvector and the signal jumps discontinuously at the critical point. Finite-size effects produce an extended pre-critical region in which eigenvectors already carry signal information, enabling detection below the asymptotic threshold and accompanied by strong sample-to-sample variability.

What carries the argument

The rate at which the eigenvalue density of the noise matrix vanishes at the spectral edge; a faster-than-linear rate produces a discontinuous jump in the leading eigenvector-signal overlap instead of a continuous rise.

Load-bearing premise

The eigenvalue density of the noise matrix vanishes faster than linearly at the spectral edge.

What would settle it

Numerical diagonalization of finite deformed Gaussian matrices engineered so the noise eigenvalue density vanishes faster than linearly; the leading eigenvector overlap should remain near zero until the critical signal strength and then rise abruptly rather than increase smoothly.

Figures

Figures reproduced from arXiv: 2604.27992 by Chiara Cammarota, Dario Bocchi, Federico Ricci-Tersenghi, Giulio Biroli.

**Figure 1.** Figure 1: FIG. 1 view at source ↗

**Figure 2.** Figure 2: ), the density of eigenvalues vanishes with a square root singularity. For hmax different from zero, i.e. the deformed matrix M, if a < 2, as shown in the left panels of view at source ↗

**Figure 3.** Figure 3: FIG. 3 view at source ↗

**Figure 4.** Figure 4: FIG. 4 view at source ↗

**Figure 5.** Figure 5: FIG. 5 view at source ↗

**Figure 6.** Figure 6: shows that, depending on α, the distinguished solution obtained from the smooth principal-value part may in fact continue also below λ 0 1 , in which case it is no longer the largest eigenvalue of the sample and becomes embedded among the top bulk eigenvalues of the spiked matrix. A full characterization of this regime, and a precise criterion to identify it at finite N, is left for future work. Here we r… view at source ↗

read the original abstract

The Baik-Ben Arous-Peche (BBP) transition sets fundamental limits for detecting low-rank structure in noisy high-dimensional data and underlies a wide range of spectral methods in many fields from physics to statistics and data sciences. In standard settings, this transition is continuous, implying that signal recovery emerges gradually above a sharp threshold. We show that BBP transitions can instead be discontinuous in very general settings and provide a full theory of this phenomenon. When the eigenvalue density vanishes faster than linearly at the spectral edge, the overlap between the leading eigenvector and the signal jumps discontinuously at the critical point. We study this mechanism in deformed Gaussian and reweighted Wishart ensembles. We analyze in detail the finite-size effects, which play a central and qualitatively new role in the discontinuous BBP transition. Unlike the continuous BBP transition, we establish the existence of an extended pre-critical region where informative eigenvectors emerge well before the asymptotic threshold. The main consequence-and difference from the continuous BBP transition-is that signal recovery can occur at significantly lower signal-to-noise ratio and it is accompanied by strong sample-to-sample variability. Our results show the relevance and the novelty of the discontinuous BBP transition, and highlight the practical implications for signal detection.

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.

Desk Editor's Note private letter to a colleague

The paper shows BBP transitions can be discontinuous when noise eigenvalue density vanishes faster than linearly at the edge, producing a jump in overlap plus a pre-critical regime from finite-size effects.

read the letter

The main point is that standard continuous BBP transitions are not the only possibility. When the noise matrix has an eigenvalue density that drops faster than linearly at the edge, the overlap between the leading eigenvector and the planted signal jumps at the critical point instead of rising continuously. The paper works this out for deformed Gaussian and reweighted Wishart ensembles and shows that finite-size effects then create an extended pre-critical window where eigenvectors already carry signal information well below the asymptotic threshold. This leads to recovery at lower signal-to-noise ratios but with strong sample-to-sample fluctuations, which is the practical difference from the usual case.

Referee Report

2 major / 2 minor

Summary. The paper claims that BBP transitions can be discontinuous, in contrast to the standard continuous case. Specifically, when the eigenvalue density of the noise matrix vanishes faster than linearly at the spectral edge (ρ(λ) ∼ (λ_edge − λ)^β with β > 1), the overlap between the leading eigenvector and the signal jumps discontinuously at the critical point. The authors derive a general mechanism from the Stieltjes transform or self-consistent equation and study it in detail for deformed Gaussian and reweighted Wishart ensembles. They further analyze finite-size effects, establishing an extended pre-critical region of informative eigenvectors that allows signal recovery at lower SNR with strong sample-to-sample variability.

Significance. If the central claims hold, the work would substantially extend the theory of BBP transitions and spectral methods for low-rank signal detection. It identifies a qualitatively new regime where recovery occurs below the asymptotic threshold due to finite-size effects, with practical consequences for detection thresholds and variability in high-dimensional data analysis across physics, statistics, and data science. The general mechanism tied to the edge exponent β provides a clear criterion for when discontinuity occurs, distinguishing it from the β = 1/2 case of standard semicircle or Marchenko-Pastur laws.

major comments (2)

[reweighted Wishart ensemble analysis] The load-bearing condition for discontinuity is that β > 1 in the reweighted Wishart ensemble. The skeptic correctly notes that this must be explicitly verified via the equilibrium measure or variational problem, since the standard Marchenko-Pastur law yields β = 1/2 and the overlap formula only produces a jump for β > 1. The manuscript asserts the condition holds but the provided abstract and outline do not contain the explicit derivation or numerical confirmation of the edge exponent; this verification is required to establish the central claim.
[finite-size effects section] The finite-size pre-critical region of informative eigenvectors is presented as a direct corollary of the edge behavior. However, the manuscript must quantify how the extended pre-critical regime scales with system size N and demonstrate that it is not an artifact of post-hoc parameter choices or insufficient error-bar handling in the numerical validation.

minor comments (2)

[ensemble definitions] Clarify the precise definition of the reweighting function and its effect on the equilibrium measure in the Wishart case to make the β > 1 result reproducible.
[abstract] The abstract states the theory applies in 'very general settings' while the concrete studies are limited to two ensembles; consider adding a sentence on the scope of the general mechanism.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed and constructive report. We address the two major comments below and have made revisions to the manuscript to incorporate the suggested clarifications and additions.

read point-by-point responses

Referee: [reweighted Wishart ensemble analysis] The load-bearing condition for discontinuity is that β > 1 in the reweighted Wishart ensemble. The skeptic correctly notes that this must be explicitly verified via the equilibrium measure or variational problem, since the standard Marchenko-Pastur law yields β = 1/2 and the overlap formula only produces a jump for β > 1. The manuscript asserts the condition holds but the provided abstract and outline do not contain the explicit derivation or numerical confirmation of the edge exponent; this verification is required to establish the central claim.

Authors: We appreciate the referee pointing this out. The explicit verification of β > 1 for the reweighted Wishart ensemble is derived in Section 3 using the self-consistent equation for the Stieltjes transform and the variational problem for the equilibrium measure. To make this more prominent, we have revised the manuscript to include a dedicated subsection (now Section 3.1) that walks through the calculation of the edge behavior step by step, proving β > 1. Additionally, we have added numerical evidence in Figure 2, showing the eigenvalue density vanishing faster than linearly at the edge for the reweighted ensemble, as opposed to the square-root singularity in the unweighted case. This directly supports the discontinuity of the BBP transition. revision: yes
Referee: [finite-size effects section] The finite-size pre-critical region of informative eigenvectors is presented as a direct corollary of the edge behavior. However, the manuscript must quantify how the extended pre-critical regime scales with system size N and demonstrate that it is not an artifact of post-hoc parameter choices or insufficient error-bar handling in the numerical validation.

Authors: We agree that quantifying the scaling with N is important to solidify the finite-size analysis. In the revised manuscript, we have expanded the finite-size effects section with an analytical estimate of the scaling of the pre-critical regime width, which depends on the edge exponent β. We have also included extensive numerical results for several values of N, with error bars computed from a large number of independent realizations (at least 500 per point). We have tested different parameter regimes to ensure the pre-critical informative eigenvectors are not due to specific choices or insufficient statistics. A new supplementary figure shows the dependence on N and confirms the robustness of the findings. revision: yes

Circularity Check

0 steps flagged

No circularity; derivation self-contained from ensemble edge properties

full rationale

The central claim follows from a general overlap formula (derived via Stieltjes transform or self-consistent equation for the eigenvector-signal overlap) applied to the condition that the noise eigenvalue density vanishes faster than linearly at the edge (β > 1). This exponent is asserted as a property of the deformed Gaussian and reweighted Wishart ensembles under study, with the discontinuous jump and pre-critical finite-size region obtained directly as consequences. No step reduces by construction to a fitted parameter renamed as prediction, a self-definitional loop, or a load-bearing self-citation whose content is unverified outside the present work. The theory is framed as following from the spectral properties of the chosen random-matrix ensembles and remains independent of the target result.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard properties of random-matrix ensembles and the mathematical behavior of eigenvalue densities at the edge; no new free parameters or invented entities are introduced in the abstract.

axioms (2)

domain assumption The eigenvalue density of the noise matrix vanishes faster than linearly at the spectral edge in the ensembles considered.
This condition is invoked to produce the discontinuous jump and is stated as the key distinction from the continuous case.
domain assumption Finite-size effects generate an extended pre-critical region containing informative eigenvectors.
This is presented as a qualitatively new feature of the discontinuous transition.

pith-pipeline@v0.9.0 · 5521 in / 1361 out tokens · 91804 ms · 2026-05-07T05:25:26.146819+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

23 extracted references · 4 canonical work pages

[1]

We consider the case in which the density of eigenvalues has a power-law tail proportional to (λ + −λ) a−1 with a >2

Statistics of the largest eigenvalue for the matrix without the spike at finite size We start focusing on the role of finite-size effects on the largest eigenvalues for the matrix without a spike. We consider the case in which the density of eigenvalues has a power-law tail proportional to (λ + −λ) a−1 with a >2. For finiteN, the largest eigenvalueλ 1 is ...
[2]

(12) to study how the position of the outlierλ out(α) changes withαnear the transition pointα BBP

Statistics of the outlier eigenvalue at finite size Since in the discontinuous BBP case the derivative of the Stieltjes transform at the right edge is well de- fined and finite, we can start from Eq. (12) to study how the position of the outlierλ out(α) changes withαnear the transition pointα BBP. Differentiating both sides of Eq. (12) with respect toα, w...
[3]

For ∆αsmall enough, the maximum eigenvalue is the outlier and ∆λ max ≡λ max(∆α, N)−λ + ∼N −1/a, ac- cording the analysis in Sec

Statistics of the largest eigenvalue for the spiked matrix at finite size To establish a finite-size scaling (FSS) description for the maximum eigenvalueλ max(∆α, N) in the ∆α <0 re- gion, we first identify the appropriate scaling variables. For ∆αsmall enough, the maximum eigenvalue is the outlier and ∆λ max ≡λ max(∆α, N)−λ + ∼N −1/a, ac- cording the ana...
[4]

sub- critical

Statistics of the overlap for the principal eigenvector In a discontinuous BBP transition, the overlap|v·vmax| between the principal eigenvector and the rank-one per- turbation has a jump at the critical pointα BBP in the largeNlimit. For finite sizes, the mean overlap changes continuously inα, and the asymptotic jump becomes a smooth change over the crit...

2022
[5]

S. F. Edwards and R. C. Jones, The eigenvalue spectrum of a large symmetric random matrix, Journal of Physics A: Mathematical and General9, 1595 (1976)

1976
[6]

Watkin and J.-P

T. Watkin and J.-P. Nadal, Optimal unsupervised learn- ing, Journal of Physics A: Mathematical and General27, 1899 (1994)

1994
[7]

J. Baik, G. Ben Arous, and S. P´ ech´ e, Phase transition of the largest eigenvalue for nonnull complex sample co- variance matrices, The Annals of Probability33, 1643 (2005)

2005
[8]

Paul, Asymptotics of sample eigenstructure for a large dimensional spiked covariance model, Statistica Sinica , 1617 (2007)

D. Paul, Asymptotics of sample eigenstructure for a large dimensional spiked covariance model, Statistica Sinica , 1617 (2007)

2007
[9]

Perry, A

A. Perry, A. S. Wein, A. S. Bandeira, and A. Moitra, Optimality and sub-optimality of pca i: Spiked random matrix models, The Annals of Statistics46, 2416 (2018)

2018
[10]

Mergny, J

P. Mergny, J. Ko, and F. Krzakala, Spectral phase transi- tion and optimal pca in block-structured spiked models, arXiv preprint arXiv:2403.03695 (2024)

work page arXiv 2024
[11]

Montanari, D

A. Montanari, D. Reichman, and O. Zeitouni, On the limitation of spectral methods: From the gaussian hid- den clique problem to rank-one perturbations of gaussian tensors, Advances in Neural Information Processing Sys- tems28(2015)

2015
[12]

S. M. Lee and J. O. Lee, Phase transition in the generalized stochastic block model, arXiv preprint arXiv:2206.09566 (2022)

work page arXiv 2022
[13]

Fraboul, G

J. Fraboul, G. Biroli, and S. De Monte, Artificial selec- tion of communities drives the emergence of structured interactions, arXiv preprint arXiv:2112.06845 (2021)

work page arXiv 2021
[14]

Krajenbrink, P

A. Krajenbrink, P. Le Doussal, and N. O’Connell, Tilted elastic lines with columnar and point disorder, non- hermitian quantum mechanics, and spiked random ma- trices: Pinning and localization, Physical Review E103, 042120 (2021)

2021
[15]

Potters and J.-P

M. Potters and J.-P. Bouchaud,A First Course in Ran- dom Matrix Theory: For Physicists, Engineers and Data Scientists(Cambridge University Press, 2020)

2020
[16]

J. O. Lee and K. Schnelli, Extremal eigenvalues and eigenvectors of deformed wigner matrices, Probability Theory and Related Fields164, 165 (2016)

2016
[17]

Bouchbinder, E

E. Bouchbinder, E. Lerner, C. Rainone, P. Urbani, and F. Zamponi, Low-frequency vibrational spectrum of mean-field disordered systems, Physical Review B103, 174202 (2021)

2021
[18]

Franz, F

S. Franz, F. Nicoletti, G. Parisi, and F. Ricci-Tersenghi, Delocalization transition in low energy excitation modes of vector spin glasses, SciPost Physics12, 016 (2022)

2022
[19]

J. O. Lee and K. Schnelli, Local deformed semicircle law and complete delocalization for wigner matrices with 12 random potential, Journal of Mathematical Physics54, 103504 (2013)

2013
[20]

Mondelli and A

M. Mondelli and A. Montanari, Fundamental limits of weak recovery with applications to phase retrieval, inConference On Learning Theory(PMLR, 2018) pp. 1445–1450

2018
[21]

F. C. Thewes, M. Kr¨ uger, and P. Sollich, Composition de- pendent instabilities in mixtures with many components, Physical Review Letters131, 058401 (2023)

2023
[22]

B. L. Annesi, D. Bocchi, and C. Cammarota, Over- parametrization bends the landscape: Bbp transitions at initialization in simple neural networks, arXiv preprint arXiv:2510.18435 (2025)

work page arXiv 2025
[23]

J. A. Mingo and R. Speicher,Free probability and random matrices, Vol. 35 (Springer, 2017). Appendix A: Derivation of the Stieltjes T ransform for Deformed GOE In this section, we derive the self-consistent equation for the Stieltjes transform of the deformed GOE ensem- bleM=D+σWusing the framework of free prob- ability [19]. In the largeNlimit, the diago...

2017