Pith. sign in

REVIEW 2 major objections 3 minor 47 references

Spectrally Robust Covariance Shrinkage for Hotelling's $T^2$ in High Dimensions

T0 review · 2 major / 3 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A practical shrinker nearly maximizes Hotelling T2 power across spectra.

desk verdict A genuinely useful theory paper whose central approximation theorem is internally inconsistent for its own isotropic prior. read the letter →

arxiv 2502.02006 v4 pith:LFHC7CJB submitted 2025-02-04 math.ST math.PRstat.MEstat.TH

Benjamin D. Robinson , Van Latimer This is my paper · ORCID
classification math.STmath.PRstat.MEstat.TH MSC 62H1560B20
keywords mean-shiftdetectionnonlinearcovarianceshrinkageHotellingT2testhigh-dimensionalstatisticsrandommatrixtheoryHilberttransformsub-Gaussianconcentration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper addresses a classic failure: Hotelling's $T^2$ test loses power and becomes inconsistent when the dimension $p$ grows with sample size $n$. The authors claim to fix this by replacing the sample precision matrix with a nonlinear eigenvalue shrinkage chosen to maximize a detection criterion. They characterize the optimal limiting shrinker as the solution of a variational problem and construct a finite-sample estimator that attains the same asymptotic detection performance. If correct, this yields a shrinkage rule for mean-shift testing that is near-optimal across essentially arbitrary population spectra, not just spiked or well-conditioned ones.

What carries the argument

The load-bearing object is the deterministic variance functional $\sigma_\infty^2(f)=\int_F [\Gamma f(x)]^2 x\delta(x)\,d\mu_\infty(x)$, with $\Gamma f = f - \phi\pi(H[f\delta w] - H[\delta w]f)$. The paper shows the unobservable detection criterion $U_n(f_n)=\mu_n' f_n(S_n)\mu_n/(\tilde\sigma_n\sqrt p)$ concentrates around $U_\infty(f)$, and that the maximizer of $U_\infty$ is obtained by inverting the operator $\Gamma' M_a \Gamma$ via Hilbert-transform identities, yielding $f^*$ in (5.9). The finite-sample version replaces the spectral density, its Hilbert transform, and the prior density by kernel-smoothed empirical estimates, with Theorem 5 giving the $O_\prec(n^{-2/3})$ $L^1$ rate that makes the plug-in asymptotically achieve the optimum.

What would settle it

Generate data satisfying the training model but with $\Omega_n = p^{1/2} vv'$ for a fixed eigenvector $v$ of $\Sigma_n$, so the prior is rank-one and aligned with the population eigenbasis. If the empirical power of the proposed SRHT falls below that of the naive diagonal test, or if $p^{-3/2}\operatorname{tr}(\Omega_n f_n(S_n))$ does not concentrate to $\int f_n\,d\omega_\infty$, then the concentration assumption (5.8) is violated and the optimality conclusion of Theorem 8 does not hold.

Watch

Extended reading notes

Core claim

The central discovery is an explicit, spectrally robust formula for the shrinker inside the shrinkage-regularized Hotelling $T^2$ statistic (SRHT). Under a maximum-entropy Gaussian prior $\mu_n \sim N(0,\Omega_n)$ on the unknown mean shift, with $p/n\to\phi\in(0,1)$ and $\|\mu_n\|^2=o_P(\sqrt p)$, the paper defines the deterministic detection criterion $U_\infty(f)=\int f\,d\omega_\infty / \sigma_\infty(f)$ and proves it is maximized by the function $f^*$ of (5.9), built from Hilbert transforms of the prior density $h$ and the limiting spectral density. It then constructs a data-dependent regular function $f_n$ and proves $U_n(f_n)=U_\infty(f^*)+o_P(1)$. For Gaussian data, power at any fixed significance level is asymptotically maximized; for sub-Gaussian data, the Hanson-Wright lower bound on power is asymptotically saturated. The proof relies on a new $O_\prec(n^{-2/3})$ error rate in $L^1(\mu_n)$ for the nonlinear shrinkage eigenvalues, extending local random matrix laws to the quantities entering the test statistic.

Load-bearing premise

The load-bearing premise is the concentration assumption (5.8): for every reasonably smooth shrinker, the normalized trace $p^{-3/2}\operatorname{tr}(\Omega_n f_n(S_n))$ must converge to its deterministic limit. This requires the signal prior to be correctly specified and asymptotically delocalized across the sample eigenbasis; a misspecified or low-rank-aligned prior breaks the optimality guarantee.

Editorial extensions

If this is right

  • Under Gaussian data, for any fixed significance level, the proposed SRHT asymptotically maximizes power among regular shrinkers.
  • Under sub-Gaussian data, the test asymptotically attains the best power allowed by the Hanson-Wright concentration bound, uniformly over the population spectra covered by the regularity assumptions.
  • The $O_\prec(n^{-2/3})$ $L^1$ rate for the nonlinear shrinkage eigenvalues makes the null standardization and power analysis of the SRHT valid, and the same rates can standardize other quadratic-form tests.
  • Empirically, at significance $10^{-4}$, the rule reports up to 50% power gains over leading linear and nonlinear competitors in simulations and matches the best competitor on the measured sensor data.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The variational formulation is not tied to mean-shift detection: replacing $h$ by the appropriate objective density, the same $f^*$ formula should transfer to portfolio optimization or radar detection problems that share the quadratic-form criterion.
  • The optimality claim depends on knowing $\Omega_n$; estimating the prior from the test sample, or hedging with a mixture of ignorant and covariance-matched priors, is a natural extension that the paper does not develop.
  • The paper leaves the singular regime $p/n \ge 1$ open; extending the variational solution to the zero-eigenvalue subspace of $S_n$ is the natural next step toward full-spectrum optimality.
  • Since the class of near-optimal shrinkers appears flexible, averaging within that class could reduce finite-sample variance without hurting asymptotic power.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 3 minor

Summary. The paper studies covariance shrinkage for Hotelling's T^2 test in the regime p/n -> phi in (0,1), under sub-Gaussian data with a known maximum-entropy signal prior. It defines the shrinkage-regularized Hotelling T^2 (SRHT) statistic, proves asymptotic size control (Theorem 6), derives a variational characterization of the asymptotically optimal limiting shrinker in terms of Hilbert transforms (Theorem 7), and proposes a finite-sample approximation claimed to achieve the same asymptotic detection performance (Theorem 8). Empirical comparisons on synthetic data and the CRAWDAD UMich/RSS data set are presented, reporting power gains over several linear and nonlinear competitors.

Significance. If the main results were fully correct, the paper would be a substantial contribution: it gives an explicit, data-driven nonlinear shrinker targeting detection power for T^2 under very general spectral distributions, builds on recent local random matrix laws, and includes a nontrivial variational solution that is elegant and potentially influential. The L^1 convergence rates for the Ledoit-Wolf eigenvalue estimates and the deterministic variance formula are useful in their own right, and the empirical claims are concrete and falsifiable. However, the central practical claim (Theorem 8) currently contains an internal inconsistency for the isotropic prior, so the paper is not ready for publication in its present form.

major comments (2)
  1. [Theorem 3, Section 4.1] For the allowed isotropic prior Omega_n = p^{1/2} I_p, h is identically 1, so H = 0 and (5.9) reduces to f* = g^2/a with a(x) = x delta(x) w(x). Appendix A shows that delta is positive, finite, and continuous on a neighborhood of F, while w(x) is asymptotic to kappa(x)^{1/2} near the spectral edge; consequently f*(x) is asymptotic to kappa(x)^{-1/2} and is unbounded on F. Definition 3 requires every regular f_n to be uniformly bounded in sup norm, and a uniformly bounded sequence cannot converge in L^2(w) to an unbounded function. Appendix G nevertheless asserts that ||(f_n - f*)w||_2 -> 0, and Theorem 8 asserts that f_n is regular and satisfies U_n(f_n) = U_infty(f*) + o_P(1). Thus Theorem 8 cannot hold as stated for the simplest prior that Section 5.2 explicitly says satisfies assumption [TEST]. This is a load-bearing issue for the paper's central optimality claim, not a cosmetic gap; it requires either relaxing the regularity definition and reworking the CLT and Hanson-Wright arguments, or proving a direct convergence U_n(f_n) -> U_infty(f*) that avoids L^2(w) convergence to an unbounded target.
  2. [Theorem 3, Section 4.1] The local Ledoit-Peche law is central to the paper: it underpins Theorem 5, Lemma 5.1, and ultimately the consistency argument of Theorem 8. Its proof is only sketched, with a reference to [LR23] and [DLY24] and no verification that the hypotheses of those references match [TRAIN] in the precise form needed here, especially for the small-scale laws (4.4)-(4.5). The paper should either include a complete proof or state the result with an explicit citation and a careful check that the cited results apply verbatim to the present setting.
minor comments (3)
  1. [Section 6] There is a typo in Section 6: 'comptetiors' should be 'competitors'.
  2. [Throughout] The data set name is rendered inconsistently: 'C RAWDAD' with a space appears in several places and 'CRAWDAD' elsewhere; please standardize.
  3. [Theorem 6] The constants c and C in Theorem 6 are not made precise: c is described as 'known absolute' and C as depending only on moments of W, but the Hanson-Wright inequality as cited involves the sub-Gaussian norm; specifying the dependence explicitly would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the optimal shrinker is derived from an explicit variational problem under a stated concentration assumption, and its finite-sample approximation is proved consistent rather than fitted to the target.

full rationale

The derivation chain is self-contained in the relevant sense: (i) Lemmas 5.1-5.2 and Theorem 7(a) convert the empirical detection criterion U_n into the deterministic functional U_infinity under Assumption [TEST], equation (5.8). This is an explicit conditional concentration assumption, supplied for Omega=p^{1/2}I and Omega=p^{1/2}Sigma by the small-scale laws of Theorems 1 and 3; it is not a fitted quantity or a definition of the target result. (ii) Theorem 7(b) solves the variational problem max_f U_infinity by Cauchy-Schwarz and operator inversion in Appendix F, giving the closed-form f* in (5.9). (iii) Theorem 8 defines a plug-in estimator matching each term of f* and proves in Appendix G that ||(f_n-f*)w||_2 -> 0 in probability using Lemmas C.1 and E.1 and condition (5.11); no target quantity is used as an input, and no fitted parameter is renamed as a prediction. The self-citations to [LR23], [RMH21], and [RML+22] are auxiliary: [LR23] is cross-referenced alongside external works [DLY24] and [LP24] for the local Ledoit-Peche law, while the other two appear as context and motivation. There is no self-citation chain forcing the optimality conclusion, no uniqueness theorem imported from the authors' prior work, and no ansatz smuggled in solely by citation. The possible unboundedness of f* near spectral edges because w(x) is of order kappa(x)^{1/2} is a correctness and regularity concern about Definition 3 and Theorem 8, not a circularity, and does not affect this verdict.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters are fit to the target data: the kernel and bandwidth are fixed by the theory, and the signal prior h is a user input rather than a fitted constant. The axioms are the standard sub-Gaussian random matrix model, the regularity of the limiting spectrum, the spectral concentration of the signal prior, and external results from random matrix theory and Hilbert transform identities. No new physical or mathematical entities are introduced.

assumptions (5)
  • domain assumption Data model [TRAIN 1]: X_n = Σ_n^{1/2} W_n with i.i.d. one-sub-Gaussian entries W.
    Entrywise independence in a fixed basis is required for the anisotropic local law and Hanson-Wright concentration; the paper acknowledges it rules out some dependence structures.
  • domain assumption Regularity of the limiting population spectrum [TRAIN 5] (absolutely continuous with density bounded above and below on compact support).
    Invoked for edge behavior and local laws from KY17; excludes spikes in the main theory, though the authors argue finite-rank perturbations are admissible via DLY24.
  • domain assumption Signal prior [TEST]: μ_n ~ N(0,Ω_n) with p^{-3/2} tr(Ω_n f_n(S_n)) → ∫ f_n dω_∞ for regular f_n, equation (5.8).
    This concentration of the signal prior against the sample eigenbasis is load-bearing for the surrogate criterion U_∞; without it, the optimal shrinker is not defined by the spectral density alone.
  • standard math Knowles-Yin anisotropic local law and Ledoit-Péché law with stated rates (Theorems 1 and 3) hold under [TRAIN].
    External results from KY17 and LP11/DLY24/self-cited LR23 are used for all subsequent small-scale laws.
  • standard math Hilbert transform identity H[bf - B Hf] = Bf + bHf from [CL77].
    Used in Appendix F to invert the operator Γ'.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Spectrally Robust Covariance Shrinkage for Hotelling's $T^2$ in High Dimensions." pith.science (2026). https://pith.science/paper/LFHC7CJB

@misc{pith2026250202006,
  author       = {Pith},
  title        = {Pith review of: Spectrally Robust Covariance Shrinkage for Hotelling's $T^2$ in High Dimensions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LFHC7CJB}},
  note         = {Machine review of arXiv:2502.02006}
}
abstract

We investigate covariance shrinkage for Hotelling's $T^2$ in the regime where the data dimension $p$ and the sample size $n$ grow in a fixed ratio -- without assuming that the population covariance matrix is spiked or well-conditioned. When $p/n\to\phi \in (0,1)$, we propose a practical finite-sample shrinker that, for any maximum-entropy signal prior and any fixed significance level, (a) asymptotically maximizes power under Gaussian data, and (b) asymptotically saturates the Hanson--Wright lower bound on power in the more general sub-Gaussian case. Our approach is to formulate and solve a variational problem characterizing the optimal limiting shrinker, and to show that our finite-sample method consistently approximates this limit by extending recent local random matrix laws. Empirical studies on simulated and real-world data, including the Crawdad UMich/RSS data set, demonstrate up to a $50\%$ gain in power over leading linear and nonlinear competitors at a significance level of $10^{-4}$.

Figures

Figures reproduced from arXiv: 2502.02006 by the authors.

Figure 1
Figure 1. Marcenko–Pastur density ˇ w(x) = dµ∞(x)/dx (red) for p/n = 1/5 and π∞ = δ1 ver￾sus histogram of eigenvalues of Sn for n = 5000, p = 1000, and Σn = Ip, and Gaussian data. Due the error bounds of roughly 1/n in the small-scale law of Theorem 1, a fairly high￾resolution histogram matches the well-known density of w(x) reasonably well. and, further Im[ ˘m(x)] = πw(x). Another fact is that the real part of m˘ (x) is give… view at source ↗
Figure 2
Figure 2. A plot showing Z˜ n of (5.1) to be roughly standard normal when both the test datum y and the reference sample {xi} 400 i=1 are a coloring matrix times a vector of standard Gaussian random variables. Here, the dimension is 200, the population covariance is the sum of a matrix with Unif[0, 1] eigenvalues and rank-40 matrix with eigenvalues 10(40−j)/10 for j = 0, ...39. where σ˜n = ˜σn(fn), as defined in Lemma 5.2. Th… view at source ↗
Figure 3
Figure 3. Scree plot of population covariance matrix chosen to generate artificial data, [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Domination by Proposed in well-conditioned and roughly-spiked cases. [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]
Figure 5
Figure 5. Figure 5: Domination by Proposed even when condition number and model order are large. [PITH_FULL_IMAGE:figures/full_fig_p023_5.png]
Figure 6
Figure 6. Figure 6: p = 182, covariance-matched prior, measured data. Our detector approximately ties with LW22 and Hotelling for best on the CRAWDAD data set for n = 300 and 200. (Hotelling coincides with LW for n = 300; curves overlap.) The lower performance and non-convex ROC curves of…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 43 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION format.url url empty "" url if FUNCTION article output.bibitem format.authors "author" output.check author format.key output output.year.check new.block format.title "title" output.check new.block crossref missing format.jour.vol output format.article.crossref output.nonnull format.pages output if ne...

  2. [2]

    Asymptotic theory for principal component analysis

    Theodore Wilbur Anderson. Asymptotic theory for principal component analysis. The Annals of Mathematical Statistics , 34(1):122--148, 1963

    work page 1963

  3. [3]

    Andrew C. Berry. The accuracy of the G aussian approximation to the sum of independent variates. Transactions of the American Mathematical Society , 49(1):122--136, 1941

    work page 1941

  4. [4]

    Lectures on the local semicircle law for W igner matrices

    Florent Benaych-Georges and Antti Knowles. Lectures on the local semicircle law for W igner matrices. arXiv preprint arXiv:1601.04055 , 2016

    arXiv 2016

  5. [5]

    Effect of high dimension: B y an example of a two sample problem

    Zhidong Bai and Hewa Saranadasa. Effect of high dimension: B y an example of a two sample problem. Statistica Sinica , pages 311--329, 1996

    work page 1996

  6. [6]

    Silverstein, et al

    Zhi-Dong Bai, Jack W. Silverstein, et al. No eigenvalues outside the support of the limiting spectral distribution of large-dimensional sample covariance matrices. The Annals of Probability , 26(1):316--345, 1998

    work page 1998

  7. [7]

    Bergin and P.M

    J.S. Bergin and P.M. Techau. High-fidelity site-specific radar simulation: KASSPER '02 workshop datacube. Information Systems Laboratories, Inc., Vienna, VA, Technical Report ISL-SCRD-TR-02-105 , 2002

    work page 2002

  8. [8]

    On sample eigenvalues in a generalized spiked population model

    Zhidong Bai and Jianfeng Yao. On sample eigenvalues in a generalized spiked population model. Journal of Multivariate Analysis , 106:167--177, 2012

    work page 2012

Show all 47 references
  1. [9]

    Carton-Lebrun

    C. Carton-Lebrun. Product properties of H ilbert transforms. Journal of Approximation Theory , 21(4):356--360, 1977

    1977

  2. [10]

    Robust spiked random matrices and a robust G-MUSIC estimator

    Romain Couillet. Robust spiked random matrices and a robust G-MUSIC estimator. Journal of Multivariate Analysis , 140:139--161, 2015

    2015

  3. [11]

    A two-sample test for high-dimensional data with applications to gene-set testing

    Song Xi Chen and Ying-Li Qin. A two-sample test for high-dimensional data with applications to gene-set testing. The Annals of Statistics , 38(2):808--835, 2010

    2010

  4. [12]

    Eldar, and Alfred O

    Yilun Chen, Ami Wiesel, Yonina C. Eldar, and Alfred O. Hero. Shrinkage algorithms for MMSE covariance estimation. IEEE Transactions on Signal Processing , 58(10):5016--5029, 2010

    2010

  5. [13]

    Yilun Chen, Ami Wiesel, and Alfred O. Hero. Robust shrinkage estimation of high-dimensional covariance matrices. IEEE Transactions on Signal Processing , 59(9):4097--4107, 2011

    2011

  6. [14]

    Donoho, Matan Gavish, and Iain M

    David L. Donoho, Matan Gavish, and Iain M. Johnstone. Optimal shrinkage of eigenvalues in the spiked covariance model. Annals of Statistics , 46(4):1742, 2018

    2018

  7. [15]

    Eigenvector distributions and optimal shrinkage estimators for large covariance and precision matrices

    Xiucai Ding, Yun Li, and Fan Yang. Eigenvector distributions and optimal shrinkage estimators for large covariance and precision matrices. arXiv preprint arXiv:2404.14751 , 2024

    2024 arXiv

  8. [16]

    Dey and C

    Dipak K. Dey and C. Srinivasan. Estimation of a covariance matrix under S tein's loss. The Annals of Statistics , pages 1581--1591, 1985

    1985

  9. [17]

    On the L iapunoff limit of error in the theory of probability

    Carl-Gustav Esseen. On the L iapunoff limit of error in the theory of probability. Ark. Mat. Astr. Fys. , 28(1--19), 1942

    1942

  10. [18]

    Hero III, Neal Patwari, and Kumar Sricharan

    Alfred O. Hero III, Neal Patwari, and Kumar Sricharan. Crawdad umich/rss, 2022

    2022

  11. [19]

    Johnstone and Arthur Yu Lu

    Iain M. Johnstone and Arthur Yu Lu. On consistency and sparsity for principal components analysis in high dimensions. Journal of the American Statistical Association , 104(486):682--693, 2009

    2009

  12. [20]

    Johnstone

    Iain M. Johnstone. On the distribution of the largest eigenvalue in principal components analysis. Annals of Statistics , pages 295--327, 2001

    2001

  13. [21]

    High-dimensional covariance matrix estimation with application to H otelling’s tests

    Dong Kai. High-dimensional covariance matrix estimation with application to H otelling’s tests. 2015

    2015

  14. [22]

    Anisotropic local laws for random matrices

    Antti Knowles and Jun Yin. Anisotropic local laws for random matrices. Probability Theory and Related Fields , 169(1):257--352, 2017

    2017

  15. [23]

    An adaptable generalization of H otelling's T^2 test in high dimension

    Haoran Li, Alexander Aue, Debashis Paul, Jie Peng, and Pei Wang. An adaptable generalization of H otelling's T^2 test in high dimension. The Annals of Statistics , 48(3):1815--1847, 2020

    2020

  16. [24]

    Eigenvectors of some large sample covariance matrix ensembles

    Olivier Ledoit and Sandrine P \'e ch \'e . Eigenvectors of some large sample covariance matrix ensembles. Probability Theory and Related Fields , 151(1-2):233--264, 2011

    2011

  17. [25]

    Eigenvector overlaps in large sample covariance matrices and nonlinear shrinkage estimators

    Zeqin Lin and Guangming Pan. Eigenvector overlaps in large sample covariance matrices and nonlinear shrinkage estimators. arXiv preprint arXiv:2404.18173 , 2024

    2024 arXiv

  18. [26]

    Robinson

    Van Latimer and Benjamin D. Robinson. The local L edoit- P \' e ch\' e law. arXiv preprint arXiv:2302.13708 , 2023

    2023 arXiv

  19. [27]

    A well-conditioned estimator for large-dimensional covariance matrices

    Olivier Ledoit and Michael Wolf. A well-conditioned estimator for large-dimensional covariance matrices. Journal of Multivariate Analysis , 88(2):365--411, 2004

    2004

  20. [28]

    Direct nonlinear shrinkage estimation of large-dimensional covariance matrices

    Olivier Ledoit and Michael Wolf. Direct nonlinear shrinkage estimation of large-dimensional covariance matrices. Technical report, Working Paper, 2017

    2017

  21. [29]

    Nonlinear shrinkage of the covariance matrix for portfolio selection: M arkowitz meets G oldilocks

    Olivier Ledoit and Michael Wolf. Nonlinear shrinkage of the covariance matrix for portfolio selection: M arkowitz meets G oldilocks. The Review of Financial Studies , 30(12):4349--4388, 2017

    2017

  22. [30]

    Optimal estimation of a large-dimensional covariance matrix under S tein's loss

    Olivier Ledoit and Michael Wolf. Optimal estimation of a large-dimensional covariance matrix under S tein's loss. Bernoulli , 24(4B):3791--3832, 2018

    2018

  23. [31]

    Analytical nonlinear shrinkage of large-dimensional covariance matrices

    Olivier Ledoit and Michael Wolf. Analytical nonlinear shrinkage of large-dimensional covariance matrices. The Annals of Statistics , 48(5):3043--3065, 2020

    2020

  24. [32]

    Quadratic shrinkage for large covariance matrices

    Olivier Ledoit and Michael Wolf. Quadratic shrinkage for large covariance matrices. Bernoulli , 28(3):1519--1547, 2022

    2022

  25. [33]

    Finite sample size effect on minimum variance beamformers: O ptimum diagonal loading factor for large arrays

    Xavier Mestre and Miguel \'A ngel Lagunas. Finite sample size effect on minimum variance beamformers: O ptimum diagonal loading factor for large arrays. IEEE Transactions on Signal Processing , 54(1):69--82, 2005

    2005

  26. [34]

    Mar c enko and Leonid Andreevich Pastur

    Vladimir A. Mar c enko and Leonid Andreevich Pastur. Distribution of eigenvalues for some sets of random matrices. Mathematics of the USSR-Sbornik , 1(4):457, 1967

    1967

  27. [35]

    Muirhead

    Robb J. Muirhead. Aspects of Multivariate Statistical Theory . John Wiley & Sons, 2009

    2009

  28. [36]

    Optshrink: A n algorithm for improved low-rank signal matrix denoising by optimal, data-driven singular value shrinkage

    Raj Rao Nadakuditi. Optshrink: A n algorithm for improved low-rank signal matrix denoising by optimal, data-driven singular value shrinkage. IEEE Transactions on Information Theory , 60(5):3002--3018, 2014

    2014

  29. [37]

    High-dimensional linear models: A random matrix perspective

    Jamshid Namdari, Debashis Paul, and Lili Wang. High-dimensional linear models: A random matrix perspective. Sankhya A , 83(2):645--695, 2021

    2021

  30. [38]

    On the local regularity of the H ilbert transform

    Yifei Pan, Jianfei Wang, and Yu Yan. On the local regularity of the H ilbert transform. arXiv preprint arXiv:2308.11947 , 2023

    2023 arXiv

  31. [39]

    G. M. Pan and Wang Zhou. Central limit theorem for H otelling's T^2 statistic under large dimension. The Annals of Applied Probability , pages 1860--1910, 2011

    1910

  32. [40]

    Robinson, Robert Malinas, and Alfred O

    Benjamin D. Robinson, Robert Malinas, and Alfred O. Hero. Space-time adaptive detection at low sample support. IEEE Transactions on Signal Processing , 69:2939--2954, 2021

    2021

  33. [41]

    Robinson, Robert Malinas, Van Latimer, Beth Morrison, and Alfred O

    Benjamin D. Robinson, Robert Malinas, Van Latimer, Beth Morrison, and Alfred O. Hero. An improvement on the H otelling T^2 test using the L edoit- W olf nonlinear shrinkage estimator. In 2022 30th European Signal Processing Conference (EUSIPCO) , pages 2106--2110, 2022

    2022

  34. [42]

    Hanson-- W right inequality and sub-gaussian concentration

    Mark Rudelson and Roman Vershynin. Hanson-- W right inequality and sub-gaussian concentration. Electron. Commun. Probab. , pages 1--9, 2013

    2013

  35. [43]

    Silverstein and Z

    Jack W. Silverstein and Z. D. Bai. On the empirical distribution of eigenvalues of a class of large dimensional random matrices. Journal of Multivariate Analysis , 54(2):175--192, 1995

    1995

  36. [44]

    Silverstein, Sang-Il Choi, et al

    Jack W. Silverstein, Sang-Il Choi, et al. Analysis of the limiting spectral distribution of large dimensional random matrices. Journal of Multivariate Analysis , 54(2):295--309, 1995

    1995

  37. [45]

    Silverstein

    Jack W. Silverstein. Strong convergence of the empirical distribution of eigenvalues of large dimensional random matrices. Journal of Multivariate Analysis , 55(2):331--339, 1995

    1995

  38. [46]

    Estimation of a covariance matrix, R ietz lecture

    Charles Stein. Estimation of a covariance matrix, R ietz lecture. In 39th Annual Meeting IMS, Atlanta, GA, 1975 , 1975

    1975

  39. [47]

    A distribution-free M -estimator of multivariate scatter

    David E Tyler. A distribution-free M -estimator of multivariate scatter. The Annals of Statistics , pages 234--251, 1987

    1987

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.