Eigenvector Occupation Statistics and Extremes in the Rosenzweig-Porter Model

arXiv:2611.12858 · math-ph, math.MP, math.PR · Submitted 2026-11-09 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematical Physics. The day in mathematical physics: rigorous models of physical systems.

Kai: We're looking at the paper: "Eigenvector Occupation Statistics and Extremes in the Rosenzweig-Porter Model".

Mira: Here's the gist of the paper. Eigenvector occupation statistics and extremes in the Rosenzweig–Porter model study all three aspects at a fixed regular bulk energy: generalized inverse participation sums, multifractal spectrum, and extreme-value statistics of occupation weights The results establish the Poisson convergence of the upper extreme process and thereby prove the Gumbel law for the maximum They also establish the Poisson limit at the lower edge, resulting in a law of the minimum, and jointly with these extreme weights determine the distribution of diagonal energies on which these extreme weights occur Model Setup and Key Parameters The study focuses on the Rosenzweig–Porter (RP) ensemble where the normalized bulk eigenvector occupation probabilities are defined as pi = ui squared of a normalized bulk eigenvector in the GUE Rosenzweig–Porter model AN = BN + τN GN The model is characterized by AN = BN + τN GN, with BN = diag(b1,..., bN) and τ 2 N = N(1−γ) and 1 < γ < 2 For a typical bulk eigenvector of an invariant random-matrix ensemble, the occupation weights are typically pi = O(N − 1/N), while in the non-ergodic delocalized phase of the RP ensemble, an eigenvector occupies a diverging but vanishing fraction of the N available coordinates Derivation of Occupation Statistics The occupation weights behave as a Lorentzian profile multiplied by local mean-one exponential variables The exact one-frame Kac–Rice density is derived using the normalized Kac–Rice Palm law of an eigenvector conditioned on the occurrence of an eigenvalue at energy E This density can be represented as dµKR N,E(p) ∝ 1∆N−1(p)e− Nτ 2 N QE(p) DN E(p) dp The determinant factor DN E(p) is represented as a Hubbard–Stratonovich integral over 2 × 2 Hermitian matrices Inverse Participation Sums and Multifractality The generalized inverse participation sums Iq = Xp qj (record moments of this measure record moments of this measure On the event ΩN, under the quenched Kac–Rice law, we have Iq Γ(q + 1) PN i=1 mi(E) q −→ 1 in probability The number of coordinates with pi ≍ N −α has a two-branch exponent f(α), where the increasing branch comes from the Lorentzian spatial profile and the decreasing branch is generated by the small-value tail of the local exponential factor Extreme Value Statistics The upper and lower extremes have different mechanisms; after rescaling, largest occupations form a marked Poisson process with a Gumbel limit for their maximum The smallest occupations form a second marked Poisson process and N(1+γ)µ 2(E)pmin converges to a mean-one exponential law Furthermore, for every fixed q > 0, the largest component contributes an asymptotically vanishing fraction of the inverse participation sum, showing that these fixed-q statistics do not freeze Model Interpretation and Outlook The self-consistent Lorentzian weights identify an effective resonant set of size KN (E) ≍ N(2−γ) while the local exponential factors determine the fluctuations about that profile The absence-of-freezing corollary makes the relation to REM and logREM phenomenology precise at fixed q Natural next problems include joint occupation laws for two nearby eigenvectors of the same matrix, edge and crossover regimes, and the corresponding statements in the orthogonal and symplectic symmetry classes References

3: N.

Kai: Rosenzweig and C.

Mira: E.

Kai: Porter, “Repulsion of energy levels” in complex atomic spectra, Phys.

Title and authors: Kai: So we're looking at this paper titled Eigenvector Occupation Statistics and Extremes in the Rosenzweig-Porter Model. It sounds pretty technical, but basically it’s about how the probability of finding a specific bulk eigenvector is distributed in this particular type of random matrix ensemble.

Mira: Yeah, it deals with occupation probabilities pi which are like the square magnitude of an eigenvector component, and this model is what they call the Rosenzweig-Porter ensemble. It’s interesting because it sits right between totally random and totally ordered systems in terms of how delocalized those eigenvectors are.

Lev: From my side, I always look at these models and I'm thinking about the computational cost here, especially when you try to simulate this on real hardware with finite N. It’s about finding a way to handle that complexity without just throwing up a huge computational wall.

Kai: Exactly, and what they are doing is focusing on a fixed bulk energy E and looking at three things: generalized inverse participation sums, the multifractal spectrum, and the extreme-value statistics of those occupation weights. It’s trying to get a complete picture at that specific energy level.

Mira: That's right, it aims to establish some Poisson convergence for the upper extreme process and also a Poisson limit at the lower edge. That means they are proving laws for both the largest and smallest occupations under different limits.

Lev: That’s crucial because those extremes dictate how much computational effort you need to budget if you were trying to build an error correcting code based on these states.

Kai: So, they’re essentially mapping out the statistical landscape of these eigenvector occupations right at a fixed energy, which is a big step for understanding the delocalized phase of random matrices.

Mira: It sets up a framework where you can analyze how the typical behavior changes when you look specifically at the most extreme components, whether they are huge or tiny.

Lev: And that connection between extremes and the underlying structure is what makes it relevant for practical applications in quantum information theory.

The paper's summary: Kai: So, diving into the actual findings of Eigenvector Occupation Statistics and Extremes in the Rosenzweig-Porter Model, they derived some very specific mathematical tools to describe these occupation weights. They start by using a normalized one-eigenpair Kac-Rice law conditioned on an eigenvalue at a regular bulk energy E.

Mira: That’s the starting point, and what they get is an exact finite N projective density for those occupation probabilities pi. Then, they take that density and analyze it by reducing the problem to a Hubbard-Stratonovich integral over two by two Hermitian matrices.

Lev: Reducing it to a saddle-point analysis via that integral is the part I find most interesting from an engineering viewpoint, because once you have that saddle point solution, you get the structure of the occupation field itself.

Kai: And what they found is that this occupation field settles into a self-consistent Lorentzian profile with an effective resonant size KN E which scales as N to the power of two minus gamma, and then there are these local exponential fluctuations around that main profile.

Mira: That Lorentzian profile tells us something about the typical extent of delocalization in this non-ergodic phase, while those local exponentials capture the smaller fluctuations. They also calculated a two-branch thickened amplitude counting spectrum f alpha which comes from two different physical origins.

Lev: The two branches are key here because one comes from diagonal energies at a distance of bi minus E, and the other comes from a local GUE Porter-Thomas factor, which is like noise you get locally.

Kai: And they also confirmed that the generalized inverse participation sums Iq will converge to one in probability as N goes to infinity for any fixed q greater than zero.

Mira: That convergence result means that for any fixed q, the largest component contributes a vanishing fraction of those sums as N gets large, so these fixed-q statistics don't freeze onto just the single biggest component.

Lev: So they are showing that even when you look at specific moments, the system still has enough complexity to be dynamic at this scale.

The paper's improvements: Kai: Now, let’s talk about what they suggest as improvements for their own model or approach in Eigenvector Occupation Statistics and Extremes in the Rosenzweig-Porter Model. They are focusing on how to make their analysis more precise.

Mira: They are suggesting that they can rigorously derive exact finite N projective densities for these eigenvector occupation probabilities by using a Kac-Rice law conditioned on an eigenvalue at a regular bulk energy E, which is what they’ve done but emphasizing the exactness for finite N.

Lev: From a hardware perspective, making it work exactly at finite N is always harder than just getting an asymptotic result; it means their method has to be robust enough to handle the actual constraints of a real system size.

Kai: They also propose using the saddle-point method to perform the asymptotic analysis of those generalized inverse participation sums Iq by reducing it to that Hubbard-Stratonovich integral over two by two Hermitian matrices.

Mira: That reduction is a way to get an exact limit for different values of q, and they are trying to evaluate it precisely rather than just guessing the limiting behavior.

Lev: Evaluating that through the saddle point method means they have a concrete path to calculate those limits, which is something I’d need if I was designing the error correction scheme based on these statistical properties.

Kai: And another improvement they highlight is computing that two-branch thickened amplitude counting spectrum f alpha by explicitly identifying the physical origins of both its branches.

Mira: They identify one branch coming from diagonal energies at a distance of bi minus E, and the other from a local GUE Porter-Thomas factor, which helps them distinguish between the structural spectrum and that small amplitude exponential tail.

Lev: That distinction is important for me because it tells us whether we’re looking at global structure or just local noise in the eigenvector distribution.

Conclusion: Kai: So, wrapping up on Eigenvector Occupation Statistics and Extremes in the Rosenzweig-Porter Model, what this paper really contributes is showing a very clear statistical picture of these occupation weights at a fixed bulk energy E. They confirm that we have established Poisson convergence for the upper extreme process and the Gumbel law for their maximum, while also finding a Poisson limit at the lower edge leading to a law of the minimum.

Mira: And they tie it all together by showing how those extreme weights determine the distribution of diagonal energies on which these extremes occur, which gives a complete picture of where that statistical behavior manifests. It’s about connecting the extreme fluctuations back to where those states actually live in the energy landscape.

Lev: For us, it means we can predict the scaling behavior for error correction overhead based on these extreme statistics, even with the finite N constraints they mentioned.

Kai: To summarize, this paper provides rigorous methods—the exact density calculation and the saddle-point reduction—to analyze inverse participation sums and multifractality in this non-ergodic phase of random matrices. It’s a solid piece of work for understanding localization transitions.

Mira: It confirms that the system exhibits these specific scaling behaviors, like the effective resonant size KN E scaling with N to the power of two minus gamma, driven by those self-consistent Lorentzian weights and local exponential factors.

Lev: And we need to be careful about what they didn't cover here. They focus on fixed bulk energy E; so applying these results to dynamic scenarios where the energy itself is changing might require a different approach or some new machinery.

Kai: Right, so that’s the main point—we have a solid understanding of how eigenvectors behave in this specific regime, and now we can move on to looking at other aspects of random matrices.

Yan V. Fyodorov, Mariya Shcherbina, Tatyana Shcherbina

King’s College London · University of Bristol · B. Verkin Institute for Low Temperature Physics and Engineering · University of Wisconsin–Madison

math-ph, math.MP, math.PR

Submitted: 2026-11-09

Updated: 2026-11-09

Comments: 37 pp

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 89/100

The gist: Eigenvector occupation statistics and extremes in the Rosenzweig–Porter model study all three aspects at a fixed regular bulk energy: generalized inverse participation sums…

Key concepts

Rosenzweig–Porter Model
A specific random matrix ensemble characterized by eigenvector occupation probabilities ($ ho_i = |u_i|^2$) that behave in a non-ergodic delocalized phase. The model is defined by the relation $A_N = B_N + au^N G_N$, where $ au^2_N = N^{1-\gamma}$ and $1 < \gamma < 2$, leading to specific scaling of occupation weights.
Inverse Participation Sums ($I_q$)
These are record moments of the measure related to eigenvector occupation. The study shows that under the quenched Kac–Rice law, these sums converge in probability: $I_q o 1$ as $N o \infty$. This helps characterize the multifractal spectrum of how coordinates are occupied by eigenvectors.
Extreme-Value Statistics
The paper analyzes the behavior of the largest and smallest eigenvector occupation weights. It finds that largest occupations follow a Gumbel distribution (Poisson process limit), while smallest occupations follow an exponential law. These extremes are crucial for understanding the fluctuations in the system's structure.
Multifractal Spectrum $f(\alpha)$
This spectrum describes how the number of coordinates with occupation $ ho_i$ near a certain value scales. The study identifies two branches: one related to diagonal energies and another from local GUE Porter-Thomas factors, allowing for a detailed distinction between structural and local fluctuations.

Terminology

Summary

Eigenvector occupation statistics and extremes in the Rosenzweig–Porter model study all three aspects at a fixed regular bulk energy: generalized inverse participation sums, multifractal spectrum, and extreme-value statistics of occupation weights The results establish the Poisson convergence of the upper extreme process and thereby prove the Gumbel law for the maximum They also establish the Poisson limit at the lower edge, resulting in a law of the minimum, and jointly with these extreme weights determine the distribution of diagonal energies on which these extreme weights occur

Model Setup and Key Parameters

The study focuses on the Rosenzweig–Porter (RP) ensemble where the normalized bulk eigenvector occupation probabilities are defined as pi = ui squared of a normalized bulk eigenvector in the GUE Rosenzweig–Porter model AN = BN + τN GN The model is characterized by AN = BN + τN GN, with BN = diag(b1,..., bN) and τ 2 N = N(1−γ) and 1 < γ < 2 For a typical bulk eigenvector of an invariant random-matrix ensemble, the occupation weights are typically pi = O(N − 1/N), while in the non-ergodic delocalized phase of the RP ensemble, an eigenvector occupies a diverging but vanishing fraction of the N available coordinates

Derivation of Occupation Statistics

The occupation weights behave as a Lorentzian profile multiplied by local mean-one exponential variables The exact one-frame Kac–Rice density is derived using the normalized Kac–Rice Palm law of an eigenvector conditioned on the occurrence of an eigenvalue at energy E This density can be represented as dµKR N,E(p) ∝ 1∆N−1(p)e− Nτ 2 N QE(p) DN E(p) dp The determinant factor DN E(p) is represented as a Hubbard–Stratonovich integral over 2 × 2 Hermitian matrices

Inverse Participation Sums and Multifractality

The generalized inverse participation sums Iq = Xp qj (record moments of this measure record moments of this measure On the event ΩN, under the quenched Kac–Rice law, we have Iq Γ(q + 1) PN i=1 mi(E) q −→ 1 in probability The number of coordinates with pi ≍ N −α has a two-branch exponent f(α), where the increasing branch comes from the Lorentzian spatial profile and the decreasing branch is generated by the small-value tail of the local exponential factor

Extreme Value Statistics

The upper and lower extremes have different mechanisms; after rescaling, largest occupations form a marked Poisson process with a Gumbel limit for their maximum The smallest occupations form a second marked Poisson process and N(1+γ)µ 2(E)pmin converges to a mean-one exponential law Furthermore, for every fixed q > 0, the largest component contributes an asymptotically vanishing fraction of the inverse participation sum, showing that these fixed-q statistics do not freeze

Model Interpretation and Outlook

The self-consistent Lorentzian weights identify an effective resonant set of size KN (E) ≍ N(2−γ) while the local exponential factors determine the fluctuations about that profile The absence-of-freezing corollary makes the relation to REM and logREM phenomenology precise at fixed q Natural next problems include joint occupation laws for two nearby eigenvectors of the same matrix, edge and crossover regimes, and the corresponding statements in the orthogonal and symplectic symmetry classes

References

[3] N. Rosenzweig and C. p. 0 E. Porter, “Repulsion of energy levels” in complex atomic spectra, Phys. p. 2 Rev. 120 (1960), 1698–1714

[5] A D Mirlin and F Evers, Multifractality and critical fluctuations at the Anderson transition, Phys. Rev. B 62 (2000), 7920–7933

[6] H Kunz and B Shapiro, Transition from Poisson to Gaussian unitary statistics: the two-point correlation function, Phys. Rev. E 58 (1998), 400–407

[7] V E Kravtsov, I M Khaymovich, E Cuevas, and M Amini, A random matrix model with localization and ergodic transitions, New J. Phys. 17 (2015), 122002

[8] D Facoetti, P Vivo, and G Biroli, From non-ergodic eigenvectors to local resolvent statistics and back: a random matrix perspective, Europhys. Lett. 115 (2016), 47003

[9] K Truong and A Ossipov, Eigenvectors under a generic perturbation: non-perturbative results from the random matrix approach, Europhys. Lett. 116 (2016), 37002

[10] C Monthus, Multifractality of eigenstates in the delocalized non-ergodic phase of some random matrix models: Wigner–Weisskopf approach, J. Phys. A: Math. Theor. 50 (2017), 295101

[11] E Bogomolny and M Sieber, Eigenfunction distribution for the Rosenzweig–Porter model, Phys. Rev. E 98 (2018), 032139

[12] G De Tomasi, M Amini, S Bera, I M Khaymovich, and V E Kravtsov, Survival probability in generalized Rosenzweig–Porter random matrix ensemble, SciPost Phys. 6 (2019), 014

[13] I M Khaymovich and V E Kravtsov, Dynamical phases in a “multifractal” Rosenzweig–Porter model, SciPost Phys. 11 (2021), 045

[14] D Venturelli, L F Cugliandolo, G Schehr, and M Tarzia, Replica approach to the generalized Rosenzweig–Porter model, SciPost Phys. 14 (2023), 110

[15] B L Altshuler and V E Kravtsov, Random Cantor sets and mini-bands in local spectrum of quantum systems, Ann. Phys. 456 (2023), 169300

[16] P von Soosten and S Warzel, Non-ergodic delocalization in the Rosenzweig–Porter model, Lett. Math. Phys. 109 (2019), 905–922

[17] L Benigni, Eigenvectors distribution and quantum unique ergodicity for deformed Wigner matrices, Ann Inst Henri Poincaré Probab Stat. 56 (2020), 2822–2867

[18] F Benaych-Georges, N Enriquez, and A Michaïl, Eigenvectors of a matrix under random perturbation, Random Matrices Theory Appl. 10 (2021), 2150023

[19] A Mészáros and B Virág, Eigenvectors of the square grid plus GUE, Commun Math Phys. 405 (2024), 2

[20] G Cipolloni, L Erdős, and J Henheik, On a Rosenzweig–Porter-type model, arXiv:2607.02446 (2026)

[30] T Shcherbina, On the correlation function of the characteristic polynomials of the Hermitian Wigner ensemble, Commun. Math. Phys. 308 (2011), 1–21

[31] T Shcherbina, On the second mixed moment of the characteristic polynomials of the 1D band matrices, Commun. Math. Phys. 328 (2014), 45–82

[32] M Shcherbina and T Shcherbina, Characteristic polynomials for 1D random band matrices from the localization side, Commun. Math. Phys. 351 (2017), 1009–1044

[33] The authors acknowledge assistance from ChatGPT (OpenAI) for performing consistency checks of the proofs, for critical reading and improving the clarity of the presentation, and designing numerical checks

[34] The authors acknowledge assistance from ChatGPT (OpenAI) for performing consistency checks of the proofs, for critical reading and improving the clarity of the presentation, and designing numerical checks

[35] The authors acknowledge assistance from ChatGPT (OpenAI) for performing consistency checks of the proofs, for critical reading and improving the clarity of the presentation, and designing numerical checks

[36] The authors acknowledge assistance from ChatGPT (OpenAI) for performing consistency checks of the proofs, for critical reading and improving the clarity of the presentation, and designing numerical checks

[37] The authors acknowledge assistance from ChatGPT (OpenAI) for performing consistency checks of the proofs, for critical reading and improving the clarity of the presentation, and designing numerical checks

[30] T Shcherbina, On the correlation function of the characteristic polynomials of the Hermitian Wigner ensemble, Commun. Math. Phys. 308 (2011), 1–21

[31] T Shcherbina, On the second mixed moment of the characteristic polynomials of the 1D band matrices, Commun. Math. Phys. 328 (2014), 45–82

[32] M Shcherbina and T Shcherbina, Characteristic polynomials for 1D random band matrices from the localization side, Commun. Math. Phys. 351 (2017), 1009–1044

[37] The authors acknowledge assistance from ChatGPT (OpenAI) for performing consistency checks of the proofs, for critical reading and improving the clarity of the presentation, and designing numerical checks

[30] T Shcherbina, On the correlation function of the characteristic polynomials of the Hermitian Wigner ensemble, Commun. Math. Phys. 308 (2011), 1–21

[31] T Shcherbina, On the second mixed moment of the characteristic polynomials of the 1D band matrices, Commun. Math. Phys. 328 (2014), 45–82

[32] M Shcherbina and T Shcherbina, Characteristic polynomials for 1D random band matrices from the localization side, Commun. Math. Phys.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved system can rigorously derive exact finite-N projective densities for eigenvector occupation probabilities by leveraging a Kac-Rice law conditioned on an eigenvalue at a regular bulk energy E, as shown in Section 3.

  2. Improved system can perform asymptotic analysis of generalized inverse participation sums Iq by reducing the problem to a Hubbard-Stratonovich integral over 2x2 Hermitian matrices and evaluating it via the saddle point method to determine exact limits for different values of q.

  3. Improved system can compute the two-branch thickened amplitude counting spectrum f(alpha) by identifying the physical origins of its branches: one from diagonal energies at distance bi - E and another from a local GUE Porter-Thomas factor, as detailed in Theorem 2.2.

  4. Improved system can analyze extreme-value statistics for occupation weights by proving the Poisson point-process limits for the largest components (Gumbel fluctuations) and the smallest components (exponential limit on scale N minus 1 minus gamma), as established in Corollary 2.1 and Corollary 2.3.

  5. Improved system can confirm that fixed q statistics do not freeze onto the largest component by demonstrating that p q max over Iq approaches zero in probability for every fixed q > 0, as shown in Corollary 2.2, showing the largest component carries an asymptotically vanishing fraction of every fixed positive inverse participation sum Iq.

  6. Improved system can distinguish between the two branches of multifractality by calculating the amplitude exponent alphaN from KN (E) and identifying its relationship to both the structural spectrum and the small-amplitude exponential tail.

Sources

Related papers