pith. sign in

arxiv: 2606.11263 · v1 · pith:YC6WPNTEnew · submitted 2026-06-09 · 🧮 math.ST · cs.LG· cs.NA· math.NA· math.PR· stat.TH

Geometric bias in eigenspace perturbation under random heterogeneous noise

Fengkai Liu , Ke Wang , Wanjie Wang This is my paper

Pith reviewed 2026-06-27 11:38 UTC · model grok-4.3

classification 🧮 math.ST cs.LGcs.NAmath.NAmath.PRstat.TH
keywords spectral perturbationheterogeneous noiseeigenvector biasquadratic vector equationlocal lawsrandom matricesDavis-Kahan
0
0 comments X

The pith

Heterogeneous noise variances cause a deterministic geometric bias in empirical eigenvectors that standard perturbation theory overlooks.

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

The paper shows that when noise has varying variances across entries, the eigenvectors estimated from data have a fixed directional shift determined by how the signal aligns with the variance pattern. Classical theorems like Davis-Kahan only give worst-case bounds that ignore this structure and treat noise as uniform. By solving a quadratic vector equation for the bias and proving local laws, the authors obtain sharper bounds that split the error into signal strength, random noise, and this geometric term. This matters because in real data like genomics or networks, noise is rarely uniform, so ignoring the bias can lead to misinterpreting the recovered directions.

Core claim

Under sparse random noise with arbitrary inhomogeneous variance profile, the empirical leading eigenvectors exhibit a systematic geometric bias term in their perturbation expansion, which can be characterized using the solution to the quadratic vector equation and is invisible to operator-norm based bounds such as Davis-Kahan.

What carries the argument

The quadratic vector equation (QVE) that governs the bias and the fine-grained isotropic local laws that control the fluctuations.

If this is right

  • If the claim holds, then perturbation bounds for eigenspaces can be refined to include an explicit bias correction term based on the variance profile.
  • The operator and 2-to-infinity norm bounds become near-optimal by separating the three contributions.
  • Classical bounds remain valid but loose when the noise variance profile correlates with the signal geometry.
  • Methods relying on eigenvector stability, such as spectral clustering, must account for this bias in heterogeneous settings.

Where Pith is reading between the lines

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

  • If the bias is deterministic and predictable, one could design a correction step to debias the eigenvectors using an estimate of the variance profile.
  • This geometric bias might appear in other random matrix models beyond sparse noise, such as in covariance estimation with heteroscedastic errors.
  • The approach may extend to non-sparse noise if the local laws can be adapted.

Load-bearing premise

That the quadratic vector equation and fine-grained isotropic local laws hold for signal-plus-noise matrices with arbitrary inhomogeneous sparse random noise variance profiles.

What would settle it

Generate a low-rank signal matrix plus sparse noise with known heterogeneous variances, compute the empirical eigenvectors, and check whether the observed deviation from the true eigenvectors matches the predicted geometric bias term up to the stochastic fluctuation size.

Figures

Figures reproduced from arXiv: 2606.11263 by Fengkai Liu, Ke Wang, Wanjie Wang.

Figure 1
Figure 1. Figure 1: Geometric bias under varying spectral gaps. Setup: dimension [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
read the original abstract

Spectral methods rely fundamentally on the stability of principal eigenspaces under random perturbations. Classically, this stability is quantified by the Davis-Kahan and Wedin theorems, which bound the eigenspace error using the operator norm of the noise and the relevant spectral gaps. While these worst-case bounds are sharp for arbitrary deterministic perturbations, they can be wasteful in the low-rank signal-plus-random-noise setting, as they fail to capture the fine-grained interaction between the signal geometry and the noise distribution. In this paper, we study the spectral perturbation of signal-plus-noise matrices corrupted by sparse, random noise with an arbitrary, inhomogeneous variance profile. We demonstrate that under heterogeneous noise variances, the empirical eigenvectors suffer a systematic, deterministic geometric bias that is entirely invisible to classical perturbation bounds. By leveraging the Quadratic Vector Equation (QVE) and establishing fine-grained isotropic local laws, we derive near-optimal, non-asymptotic perturbation bounds for the leading eigenspaces in the operator and $2\to\infty$ norms. The bounds separate the usual signal-to-noise contribution, stochastic fluctuations, and structured geometric bias terms determined by the alignment between the signal eigenspaces and the row-wise variance profile.

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

2 major / 2 minor

Summary. The paper studies spectral perturbation for low-rank signal-plus-noise matrices corrupted by sparse random noise with arbitrary inhomogeneous variance profiles. It claims that empirical eigenvectors exhibit a systematic deterministic geometric bias arising from alignment between signal eigenspaces and the row-wise variance profile; this bias is invisible to classical Davis-Kahan/Wedin bounds. Leveraging the quadratic vector equation (QVE) and fine-grained isotropic local laws, the authors derive near-optimal non-asymptotic bounds on leading eigenspace error in the operator and 2→∞ norms, separating the usual signal-to-noise term, stochastic fluctuations, and the structured geometric bias.

Significance. If the local laws hold under the stated assumptions, the work offers a finer decomposition of eigenspace error sources than classical worst-case bounds, which could improve analysis of spectral methods in settings with heterogeneous noise (e.g., certain covariance estimation or PCA tasks). The explicit separation of a deterministic geometric component and the focus on 2→∞ norms are strengths; the attempt to obtain non-asymptotic, near-optimal rates is also noteworthy.

major comments (2)
  1. [Abstract and local-laws section] Abstract and the section establishing the isotropic local laws: the central claim that a deterministic geometric bias term can be extracted and is invisible to Davis-Kahan/Wedin rests on the QVE and fine-grained isotropic local laws applying directly to signal-plus-noise matrices with arbitrary inhomogeneous sparse variance profiles. Standard QVE theory requires regularity conditions (uniform boundedness of entries, row/column sum control, or Lipschitz continuity of the profile) for uniqueness of the fixed-point solution and error control in the local law. The manuscript must explicitly state and verify these conditions for the claimed 'arbitrary' profiles; without them the bias term is not rigorously justified.
  2. [Main theorem on perturbation bounds] Main theorem on the perturbation bounds (likely the result separating the three error sources): the near-optimality assertion and the claim that the geometric bias is 'entirely invisible' to classical bounds require the explicit form of the bias term (how it depends on the alignment between signal subspaces and the variance profile) to be displayed and compared against the operator-norm bound. If this term is only implicit in the QVE solution, the separation into signal-to-noise, stochastic, and geometric components cannot be verified as load-bearing.
minor comments (2)
  1. [Introduction] Clarify the precise notion of 'near-optimal' (with respect to which minimax rate or information-theoretic lower bound) already in the introduction, rather than only in the abstract.
  2. [Assumptions section] Notation for the variance profile matrix should be introduced with an explicit display equation early in the assumptions section to avoid ambiguity when referring to row-wise inhomogeneity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and valuable comments. We address each major comment below and outline the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract and local-laws section] Abstract and the section establishing the isotropic local laws: the central claim that a deterministic geometric bias term can be extracted and is invisible to Davis-Kahan/Wedin rests on the QVE and fine-grained isotropic local laws applying directly to signal-plus-noise matrices with arbitrary inhomogeneous sparse variance profiles. Standard QVE theory requires regularity conditions (uniform boundedness of entries, row/column sum control, or Lipschitz continuity of the profile) for uniqueness of the fixed-point solution and error control in the local law. The manuscript must explicitly state and verify these conditions for the claimed 'arbitrary' profiles; without them the bias term is not rigorously justified.

    Authors: We agree that the regularity conditions underlying the QVE must be stated explicitly to justify the applicability to the variance profiles considered. The manuscript assumes the variance profile satisfies the standard conditions for QVE uniqueness and local law error bounds, including bounded maximum entry size, controlled row and column sums, and appropriate regularity for the sparse inhomogeneous case. To address this, we will add an explicit statement of these assumptions in the local laws section, along with a verification that they hold for the class of profiles under consideration. This will clarify that 'arbitrary' refers to arbitrary profiles within this regularity class, under which the geometric bias is rigorously derived. revision: yes

  2. Referee: [Main theorem on perturbation bounds] Main theorem on the perturbation bounds (likely the result separating the three error sources): the near-optimality assertion and the claim that the geometric bias is 'entirely invisible' to classical bounds require the explicit form of the bias term (how it depends on the alignment between signal subspaces and the variance profile) to be displayed and compared against the operator-norm bound. If this term is only implicit in the QVE solution, the separation into signal-to-noise, stochastic, and geometric components cannot be verified as load-bearing.

    Authors: The explicit form of the geometric bias is obtained by solving the QVE for the heterogeneous variance profile and comparing to the homogeneous case; it takes the form of a deterministic shift in the eigenvector directions proportional to the projection of the signal vectors onto the variance profile. We will revise the main theorem section to display this explicit expression and include a direct comparison showing that the classical Davis-Kahan bound, which depends only on the operator norm of the noise, does not capture this alignment-dependent term. This makes the separation into the three components explicit and verifies that the geometric bias is indeed invisible to classical bounds. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected; derivation relies on external QVE and local laws

full rationale

The paper's central derivation leverages the Quadratic Vector Equation and establishes fine-grained isotropic local laws to separate signal-to-noise, stochastic, and geometric bias terms in the perturbation bounds. No steps reduce by construction to fitted parameters, self-definitions, or load-bearing self-citations whose content is unverified within the paper. The assumptions on the variance profile are stated as part of the setup for applying these tools, without the target bounds being tautological to the inputs. This is the standard case of a paper building on established external theory.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Central claim rests on applicability of QVE and isotropic local laws to the heterogeneous sparse noise model; these are imported from prior random matrix literature rather than derived here.

axioms (2)
  • domain assumption Noise entries are independent with row-wise variance profile that can be arbitrary
    Stated directly in abstract as the setting for the signal-plus-noise matrix.
  • domain assumption Quadratic vector equation and fine-grained isotropic local laws hold for this inhomogeneous noise model
    Invoked to derive the perturbation bounds (abstract paragraph on QVE and local laws).

pith-pipeline@v0.9.1-grok · 5749 in / 1290 out tokens · 26624 ms · 2026-06-27T11:38:50.160839+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

96 extracted references · 3 linked inside Pith

  1. [1]

    E. Abbe, J. Fan, and K. Wang. Anℓ p theory of PCA and spectral clustering.The Annals of Statistics, 50(4):2359–2385, 2022. 97

    2022

  2. [2]

    E. Abbe, J. Fan, K. Wang, and Y . Zhong. Entrywise eigenvector analysis of random matrices with low expected rank.The Annals of Statistics, 48(3):1452–1474, 2020

    2020

  3. [3]

    Agterberg

    J. Agterberg. Distributional theory and statistical inference for linear functions of eigenvec- tors with small eigengaps.arXiv preprint arXiv:2308.02480, 2023

    Pith/arXiv arXiv 2023

  4. [4]

    Agterberg, Z

    J. Agterberg, Z. Lubberts, and C. E. Priebe. Entrywise estimation of singular vectors of low-rank matrices with heteroskedasticity and dependence.IEEE Trans. Inform. Theory, 68(7):4618–4650, 2022

    2022

  5. [5]

    L. V . Ahlfors and L. V . Ahlfors.Complex analysis, volume 3. McGraw-Hill New York, 1979

    1979

  6. [6]

    Ajanki, L

    O. Ajanki, L. Erd ˝os, and T. Kr¨uger.Quadratic vector equations on complex upper half-plane, volume 261. American Mathematical Society, 2019

    2019

  7. [7]

    Allez and J.-P

    R. Allez and J.-P. Bouchaud. Eigenvector dynamics under free addition.Random Matrices Theory Appl., 3(3):1450010, 17, 2014

    2014

  8. [8]

    N. Alon, M. Krivelevich, and B. Sudakov. Finding a large hidden clique in a random graph. Random Structures & Algorithms, 13(3-4):457–466, 1998

    1998

  9. [9]

    Arias-Castro and N

    E. Arias-Castro and N. Verzelen. Community detection in dense random networks.The Annals of Statistics, 42(3):940–969, 2014

    2014

  10. [10]

    Bai and J

    Z. Bai and J. Yao. Central limit theorems for eigenvalues in a spiked population model.Ann. Inst. H. Poincar´e Probab. Statist., 44(3):447–474, 2008

    2008

  11. [11]

    J. Baik, G. B. Arous, S. P ´ech´e, et al. Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices.The Annals of Probability, 33(5):1643–1697, 2005

    2005

  12. [12]

    Baik and J

    J. Baik and J. W. Silverstein. Eigenvalues of large sample covariance matrices of spiked population models.Journal of Multivariate Analysis, 97(6):1382–1408, 2006

    2006

  13. [13]

    Balakrishnan, M

    S. Balakrishnan, M. Kolar, A. Rinaldo, A. Singh, and L. Wasserman. Statistical and com- putational tradeoffs in biclustering. InNIPS 2011 Workshop on Computational Trade-offs in Statistical Learning, volume 4, 2011

    2011

  14. [14]

    A. S. Bandeira and R. van Handel. Sharp nonasymptotic bounds on the norm of random matrices with independent entries.The Annals of Probability, 44(4):2479–2506, 2016

    2016

  15. [15]

    Z. Bao, K. M. Cheong, J. Lee, and Y . Li. Signal detection from spiked noise via asym- metrization.arXiv preprint arXiv:2504.19450, 2025

    arXiv 2025

  16. [16]

    Z. Bao, X. Ding, J. Wang, and K. Wang. Statistical inference for principal components of spiked covariance matrices.The Annals of Statistics, 50(2):1144–1169, 2022

    2022

  17. [17]

    Z. Bao, X. Ding, and K. Wang. Singular vector and singular subspace distribution for the matrix denoising model.The Annals of Statistics, 49(1):370–392, 2021. 98

    2021

  18. [18]

    Bao and D

    Z. Bao and D. Wang. Eigenvector distribution in the critical regime of BBP transition.Prob- ability Theory and Related Fields, pages 1–81, 2021

    2021

  19. [19]

    Benaych-Georges, N

    F. Benaych-Georges, N. Enriquez, and A. Micha ¨ıl. Eigenvectors of a matrix under random perturbation.Random Matrices Theory Appl., 10(2):2150023, 2021

    2021

  20. [20]

    Benaych-Georges, A

    F. Benaych-Georges, A. Guionnet, and M. Maida. Fluctuations of the extreme eigenvalues of finite rank deformations of random matrices.Electron. J. Probab., 16:1621–1662, 2011

    2011

  21. [21]

    Benaych-Georges and R

    F. Benaych-Georges and R. R. Nadakuditi. The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices.Advances in Mathematics, 227(1):494–521, 2011

    2011

  22. [22]

    Benaych-Georges and R

    F. Benaych-Georges and R. R. Nadakuditi. The singular values and vectors of low rank per- turbations of large rectangular random matrices.Journal of Multivariate Analysis, 111:120– 135, 2012

    2012

  23. [23]

    L. Benigni. Eigenvectors distribution and quantum unique ergodicity for deformed wigner matrices.Annales de l’Institut Henri Poincar´e-Probabilit´es et Statistiques, 56(4):2822–2867, 2020

    2020

  24. [24]

    Bhardwaj and V

    A. Bhardwaj and V . Vu. Matrix perturbation: Davis-Kahan in the infinity norm. InPro- ceedings of the 2024 Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 880–934. SIAM, 2024

    2024

  25. [25]

    Bhatia.Matrix analysis, volume 169 ofGraduate Texts in Mathematics

    R. Bhatia.Matrix analysis, volume 169 ofGraduate Texts in Mathematics. Springer-Verlag, New York, 1997

    1997

  26. [26]

    Bhattacharya, A

    B. Bhattacharya, A. Chakrabarty, and R. S. Hazra. Outlier eigenvalues and eigen- vectors of generalized wigner matrices with finite-rank perturbations.arXiv preprint arXiv:2601.10204, 2026

    arXiv 2026

  27. [27]

    Bloemendal, A

    A. Bloemendal, A. Knowles, H.-T. Yau, and J. Yin. On the principal components of sample covariance matrices.Probability Theory and Related Fields, 164(1-2):459–552, 2016

    2016

  28. [28]

    Brailovskaya and R

    T. Brailovskaya and R. Van Handel. Extremal random matrices with independent entries and matrix superconcentration inequalities.The Annals of Probability, 54(2):669–704, 2026

    2026

  29. [29]

    Brennan, G

    M. Brennan, G. Bresler, and W. Huleihel. Reducibility and computational lower bounds for problems with planted sparse structure. InConference On Learning Theory, pages 48–166. PMLR, 2018

    2018

  30. [30]

    Butucea and Y

    C. Butucea and Y . I. Ingster. Detection of a sparse submatrix of a high-dimensional noisy matrix.Bernoulli, 19(5B):2652–2688, 2013

    2013

  31. [31]

    Butucea, Y

    C. Butucea, Y . I. Ingster, and I. A. Suslina. Sharp variable selection of a sparse submatrix in a high-dimensional noisy matrix.ESAIM: Probability and Statistics, 19:115–134, 2015. 99

    2015

  32. [32]

    C. Cai, G. Li, Y . Chi, H. V . Poor, and Y . Chen. Subspace estimation from unbalanced and incomplete data matrices:ℓ 2,8 statistical guarantees.The Annals of Statistics, 49(2):944– 967, 2021

    2021

  33. [33]

    T. T. Cai, T. Liang, and A. Rakhlin. Computational and statistical boundaries for submatrix localization in a large noisy matrix.The Annals of Statistics, 45(4):1403–1430, 2017

    2017

  34. [34]

    T. T. Cai and A. Zhang. Rate-optimal perturbation bounds for singular subspaces with appli- cations to high-dimensional statistics.The Annals of Statistics, 46(1):60–89, 2018

    2018

  35. [35]

    E. J. Cand `es and Y . Plan. Matrix completion with noise.Proceedings of the IEEE, 98:925– 936, 2010

    2010

  36. [36]

    E. J. Cand `es and B. Recht. Exact matrix completion via convex optimization.Found. Com- put. Math., 9(6):717–772, 2009

    2009

  37. [37]

    J. Cape, M. Tang, and C. E. Priebe. The two-to-infinity norm and singular subspace geometry with applications to high-dimensional statistics.The Annals of Statistics, 47(5):2405–2439, 2019

    2019

  38. [38]

    Capitaine

    M. Capitaine. Limiting eigenvectors of outliers for Spiked Information-Plus-Noise type ma- trices.S ´eminaire de Probabilit´es XLIX, pages 119–164, 2018

    2018

  39. [39]

    Capitaine and C

    M. Capitaine and C. Donati-Martin. Non universality of fluctuations of outlier eigenvectors for block diagonal deformations of wigner matrices.ALEA: Latin American Journal of Probability and Mathematical Statistics, 18(1):129–165, 2021

    2021

  40. [40]

    Chang and J

    J. Chang and J. Cape. Extreme value theory for singular subspace estimation in the matrix denoising model.arXiv preprint arXiv:2507.19978, 2025

    arXiv 2025

  41. [41]

    Y . Chen, Y . Chi, J. Fan, and C. Ma. Spectral methods for data science: A statistical perspec- tive.Foundations and Trends® in Machine Learning, 14(5):566–806, 2021

    2021

  42. [42]

    Y . Chen, J. Fan, C. Ma, and K. Wang. Spectral method and regularized MLE are both optimal for top-K ranking.The Annals of Statistics, 47(4):2204–2235, 2019

    2019

  43. [43]

    Chen and J

    Y . Chen and J. Xu. Statistical-computational tradeoffs in planted problems and submatrix localization with a growing number of clusters and submatrices.The Journal of Machine Learning Research, 17(1):882–938, 2016

    2016

  44. [44]

    Cheng, Y

    C. Cheng, Y . Wei, and Y . Chen. Tackling small eigen-gaps: Fine-grained eigenvector estima- tion and inference under heteroscedastic noise.IEEE Transactions on Information Theory, 67(11):7380–7419, 2021

    2021

  45. [45]

    Comtet.Advanced Combinatorics: The art of finite and infinite expansions

    L. Comtet.Advanced Combinatorics: The art of finite and infinite expansions. Springer Science & Business Media, 2012

    2012

  46. [46]

    Dadon, W

    M. Dadon, W. Huleihel, and T. Bendory. Detection and recovery of hidden submatrices. IEEE Transactions on Signal and Information Processing over Networks, 2024. 100

    2024

  47. [47]

    Davis and W

    C. Davis and W. M. Kahan. The rotation of eigenvectors by a perturbation. III.SIAM J. Numer. Anal., 7:1–46, 1970

    1970

  48. [48]

    Dekel, O

    Y . Dekel, O. Gurel-Gurevich, and Y . Peres. Finding hidden cliques in linear time with high probability.Combinatorics, Probability and Computing, 23(1):29–49, 2014

    2014

  49. [49]

    Eldridge, M

    J. Eldridge, M. Belkin, and Y . Wang. Unperturbed: spectral analysis beyond Davis-Kahan. InAlgorithmic Learning Theory, pages 321–358. PMLR, 2018

    2018

  50. [50]

    J. Fan, Y . Fan, X. Han, and J. Lv. Asymptotic theory of eigenvectors for large random matrices.Journal of the American Statistical Association, 117(538):996–1009, 2022

    2022

  51. [51]

    J. Fan, W. Wang, and Y . Zhong. Anℓ8 eigenvector perturbation bound and its application to robust covariance estimation.Journal of Machine Learning Research, 18:207–207, 2018

    2018

  52. [52]

    Feige and D

    U. Feige and D. Ron. Finding hidden cliques in linear time. InDiscrete Mathematics and Theoretical Computer Science, pages 189–204. Discrete Mathematics and Theoretical Com- puter Science, 2010

    2010

  53. [53]

    L. S. Ferreira and F. L. Metz. BBP transition and the leading eigenvector of the spiked wigner model with inhomogeneous noise.arXiv preprint arXiv:2604.18523, 2026

    Pith/arXiv arXiv 2026

  54. [54]

    I. C. Gohberg and E. I. Sigal. An operator generalization of the logarithmic residue theorem and the theorem of Rouch´e.Mathematics of the USSR-Sbornik, 13(4):603–625, 1971

    1971

  55. [55]

    Guionnet, J

    A. Guionnet, J. Ko, F. Krzakala, and L. Zdeborov ´a. Low-rank matrix estimation with inho- mogeneous noise.Information and Inference: A Journal of the IMA, 14(2):iaaf010, 2025

    2025

  56. [56]

    Hajek, Y

    B. Hajek, Y . Wu, and J. Xu. Achieving exact cluster recovery threshold via semidefinite programming.IEEE Transactions on Information Theory, 62(5):2788–2797, 2016

    2016

  57. [57]

    Y . He, A. Knowles, and R. Rosenthal. Isotropic self-consistent equations for mean-field random matrices.Probab. Theory Related Fields, 171(1-2):203–249, 2018

    2018

  58. [58]

    D. Hong, L. Balzano, and J. A. Fessler. Asymptotic performance of PCA for high- dimensional heteroscedastic data.Journal of Multivariate Analysis, 167:435–452, 2018

    2018

  59. [59]

    D. Hong, K. Gilman, L. Balzano, and J. A. Fessler. HePPCAT: Probabilistic PCA for data with heteroscedastic noise.IEEE Transactions on Signal Processing, 69:4819–4834, 2021

    2021

  60. [60]

    D. Hong, F. Yang, J. A. Fessler, and L. Balzano. Optimally weighted PCA for high- dimensional heteroscedastic data.SIAM Journal on Mathematics of Data Science, 5(1):222– 250, 2023

    2023

  61. [61]

    Keshavan, A

    R. Keshavan, A. Montanari, and S. Oh. Matrix completion from noisy entries.Advances in Neural Information Processing Systems, 22, 2009

    2009

  62. [62]

    Kolar, S

    M. Kolar, S. Balakrishnan, A. Rinaldo, and A. Singh. Minimax localization of structural information in large noisy matrices.Advances in Neural Information Processing Systems, 24, 2011. 101

    2011

  63. [63]

    Koltchinskii and D

    V . Koltchinskii and D. Xia. Perturbation of linear forms of singular vectors under Gaussian noise. InHigh dimensional probability VII, volume 71 ofProgr. Probab., pages 397–423. Springer, [Cham], 2016

    2016

  64. [64]

    L. Lei. Unifiedℓ 2Ñ8 eigenspace perturbation theory for symmetric random matrices.arXiv preprint arXiv:1909.04798, 2019

    arXiv 1909

  65. [65]

    G. Li, C. Cai, H. V . Poor, and Y . Chen. Minimax estimation of linear functions of eigenvec- tors in the face of small eigen-gaps.IEEE Transactions on Information Theory, 71(2):1200– 1247, 2024

    2024

  66. [66]

    L ¨offler, A

    M. L ¨offler, A. Y . Zhang, and H. H. Zhou. Optimality of spectral clustering in the gaussian mixture model.The Annals of Statistics, 49(5):2506–2530, 2021

    2021

  67. [67]

    Y . Luo, R. Han, and A. R. Zhang. A Schatten-qlow-rank matrix perturbation analysis via perturbation projection error bound.Linear Algebra and its Applications, 630:225–240, 2021

    2021

  68. [68]

    Ma and Y

    Z. Ma and Y . Wu. Computational barriers in minimax submatrix detection.The Annals of Statistics, pages 1089–1116, 2015

    2015

  69. [69]

    McSherry

    F. McSherry. Spectral partitioning of random graphs. InProceedings 42nd IEEE Symposium on Foundations of Computer Science, pages 529–537. IEEE, 2001

    2001

  70. [70]

    Montanari, D

    A. Montanari, D. Reichman, and O. Zeitouni. On the limitation of spectral methods: From the gaussian hidden clique problem to rank-one perturbations of gaussian tensors.Advances in Neural Information Processing Systems, 28, 2015

    2015

  71. [71]

    O’Rourke, V

    S. O’Rourke, V . Vu, and K. Wang. Eigenvectors of random matrices: A survey.Journal of Combinatorial Theory, Series A, 144:361 – 442, 2016. Fifty Years of the Journal of Combinatorial Theory

    2016

  72. [72]

    O’Rourke, V

    S. O’Rourke, V . Vu, and K. Wang. Random perturbation of low rank matrices: improving classical bounds.Linear Algebra Appl., 540:26–59, 2018

    2018

  73. [73]

    O’Rourke, V

    S. O’Rourke, V . Vu, and K. Wang. Matrices with Gaussian noise: optimal estimates for singular subspace perturbation.IEEE Transactions on Information Theory, 70(3):1978– 2002, 2024

    1978

  74. [74]

    O’Rourke and N

    S. O’Rourke and N. Williams. Pairing between zeros and critical points of random polyno- mials with independent roots.Trans. Amer. Math. Soc., 371(4):2343–2381, 2019

    2019

  75. [75]

    D. Paul. Asymptotics of sample eigenstructure for a large dimensional spiked covariance model.Statistica Sinica, 17(4):1617–1642, 2007

    2007

  76. [76]

    Tran and V

    P. Tran and V . Vu. New matrix perturbation bounds via combinatorial expansion I: Perturba- tion of eigenspaces.arXiv e-prints, pages arXiv–2409, 2024

    2024

  77. [77]

    Tran and V

    P. Tran and V . Vu. New matrix perturbation bounds with relative norm: Perturbation of eigenspaces.arXiv preprint arXiv:2409.20207, 2024. 102

    arXiv 2024

  78. [78]

    Vershynin.High-dimensional probability: An introduction with applications in data sci- ence, volume 47

    R. Vershynin.High-dimensional probability: An introduction with applications in data sci- ence, volume 47. Cambridge university press, 2018

    2018

  79. [79]

    V on Luxburg, M

    U. V on Luxburg, M. Belkin, and O. Bousquet. Consistency of spectral clustering.The Annals of Statistics, 36(2):555–586, 2008

    2008

  80. [80]

    V . Vu. Singular vectors under random perturbation.Random Structures & Algorithms, 39(4):526–538, 2011

    2011

Showing first 80 references.