{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T15:52:33Z","timestamp":1791474753268,"version":"4.3.4"},"reference-count":59,"publisher":"Institute of Mathematical Statistics","issue":"1","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Ann. Statist."],"published-print":{"date-parts":[[2021,2,1]]},"abstract":"<jats:p>This paper is concerned with the interplay between statistical asymmetry and spectral methods. Suppose we are interested in estimating a rank-1 and symmetric matrix $\\boldsymbol{M}^{\\star }\\in \\mathbb{R}^{n\\times n}$, yet only a randomly perturbed version $\\boldsymbol{M}$ is observed. The noise matrix $\\boldsymbol{M}-\\boldsymbol{M}^{\\star }$ is composed of independent (but not necessarily homoscedastic) entries and is, therefore, not symmetric in general. This might arise if, for example, when we have two independent samples for each entry of $\\boldsymbol{M}^{\\star }$ and arrange them in an asymmetric fashion. The aim is to estimate the leading eigenvalue and the leading eigenvector of $\\boldsymbol{M}^{\\star }$. We demonstrate that the leading eigenvalue of the data matrix $\\boldsymbol{M}$ can be $O(\\sqrt{n})$ times more accurate (up to some log factor) than its (unadjusted) leading singular value of $\\boldsymbol{M}$ in eigenvalue estimation. Moreover, the eigen-decomposition approach is fully adaptive to heteroscedasticity of noise, without the need of any prior knowledge about the noise distributions. In a nutshell, this curious phenomenon arises since the statistical asymmetry automatically mitigates the bias of the eigenvalue approach, thus eliminating the need of careful bias correction. Additionally, we develop appealing nonasymptotic eigenvector perturbation bounds; in particular, we are able to bound the perturbation of any linear function of the leading eigenvector of $\\boldsymbol{M}$ (e.g., entrywise eigenvector perturbation). We also provide partial theory for the more general rank-$r$ case. The takeaway message is this: arranging the data samples in an asymmetric manner and performing eigendecomposition could sometimes be quite beneficial.<\/jats:p>","DOI":"10.1214\/20-aos1963","type":"journal-article","created":{"date-parts":[[2021,1,28]],"date-time":"2021-01-28T22:00:38Z","timestamp":1611871238000},"page":"435-458","source":"Crossref","is-referenced-by-count":13,"title":["Asymmetry helps: Eigenvalue and eigenvector analyses of asymmetrically perturbed low-rank matrices"],"prefix":"10.1214","volume":"49","author":[{"given":"Yuxin","family":"Chen","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Chen","family":"Cheng","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jianqing","family":"Fan","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"108","reference":[{"key":"31","unstructured":"Jain, P. and Netrapalli, P. (2015). Fast exact matrix completion with finite samples. In <i>Conference on Learning Theory<\/i> 1007\u20131034."},{"key":"56","unstructured":"Xia, D. (2016). Statistical inference for large matrices. Ph.D. thesis, Georgia Institute of Technology."},{"key":"33","doi-asserted-by":"publisher","unstructured":"Johnstone, I. M. and Lu, A. Y. (2009). On consistency and sparsity for principal components analysis in high dimensions. <i>J. Amer. Statist. Assoc.<\/i> <b>104<\/b> 682\u2013693.","DOI":"10.1198\/jasa.2009.0121"},{"key":"3","doi-asserted-by":"publisher","unstructured":"Baik, J. and Silverstein, J. W. (2006). Eigenvalues of large sample covariance matrices of spiked population models. <i>J. Multivariate Anal.<\/i> <b>97<\/b> 1382\u20131408.","DOI":"10.1016\/j.jmva.2005.08.003"},{"key":"6","doi-asserted-by":"publisher","unstructured":"Benaych-Georges, F. and Nadakuditi, R. R. (2011). The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices. <i>Adv. Math.<\/i> <b>227<\/b> 494\u2013521.","DOI":"10.1016\/j.aim.2011.02.007"},{"key":"32","doi-asserted-by":"publisher","unstructured":"Johnstone, I. M. (2001). On the distribution of the largest eigenvalue in principal components analysis. <i>Ann. Statist.<\/i> <b>29<\/b> 295\u2013327.","DOI":"10.1214\/aos\/1009210544"},{"key":"15","doi-asserted-by":"publisher","unstructured":"Cand\u00e8s, E. J. and Recht, B. (2009). Exact matrix completion via convex optimization. <i>Found. Comput. Math.<\/i> <b>9<\/b> 717\u2013772.","DOI":"10.1007\/s10208-009-9045-5"},{"key":"24","doi-asserted-by":"publisher","unstructured":"Davis, C. and Kahan, W. M. (1970). The rotation of eigenvectors by a perturbation. III. <i>SIAM J. Numer. Anal.<\/i> <b>7<\/b> 1\u201346.","DOI":"10.1137\/0707001"},{"key":"58","doi-asserted-by":"publisher","unstructured":"Yin, Y. Q., Bai, Z. D. and Krishnaiah, P. R. (1988). On the limit of the largest eigenvalue of the large-dimensional sample covariance matrix. <i>Probab. Theory Related Fields<\/i> <b>78<\/b> 509\u2013521.","DOI":"10.1007\/BF00353874"},{"key":"30","doi-asserted-by":"publisher","unstructured":"F\u00fcredi, Z. and Koml\u00f3s, J. (1981). The eigenvalues of random symmetric matrices. <i>Combinatorica<\/i> <b>1<\/b> 233\u2013241.","DOI":"10.1007\/BF02579329"},{"key":"2","doi-asserted-by":"publisher","unstructured":"Bai, Z. and Yao, J. (2008). Central limit theorems for eigenvalues in a spiked population model. <i>Ann. Inst. Henri Poincar\u00e9 Probab. Stat.<\/i> <b>44<\/b> 447\u2013474.","DOI":"10.1214\/07-AIHP118"},{"key":"17","doi-asserted-by":"publisher","unstructured":"Capitaine, M., Donati-Martin, C. and F\u00e9ral, D. (2009). The largest eigenvalues of finite rank deformation of large Wigner matrices: Convergence and nonuniversality of the fluctuations. <i>Ann. Probab.<\/i> <b>37<\/b> 1\u201347.","DOI":"10.1214\/08-AOP394"},{"key":"29","doi-asserted-by":"crossref","unstructured":"F\u00e9ral, D. and P\u00e9ch\u00e9, S. (2007). The largest eigenvalue of rank one deformation of large Wigner matrices. <i>Comm. Math. Phys.<\/i> <b>272<\/b> 185\u2013228.","DOI":"10.1007\/s00220-007-0209-3"},{"key":"44","doi-asserted-by":"publisher","unstructured":"P\u00e9ch\u00e9, S. (2006). The largest eigenvalue of small rank perturbations of Hermitian random matrices. <i>Probab. Theory Related Fields<\/i> <b>134<\/b> 127\u2013173.","DOI":"10.1007\/s00440-005-0466-z"},{"key":"46","doi-asserted-by":"publisher","unstructured":"Renfrew, D. and Soshnikov, A. (2013). On finite rank deformations of Wigner matrices II: Delocalized perturbations. <i>Random Matrices Theory Appl.<\/i> <b>2<\/b> 1250015, 36.","DOI":"10.1142\/S2010326312500153"},{"key":"7","doi-asserted-by":"publisher","unstructured":"Benaych-Georges, F. and Nadakuditi, R. R. (2012). The singular values and vectors of low rank perturbations of large rectangular random matrices. <i>J. Multivariate Anal.<\/i> <b>111<\/b> 120\u2013135.","DOI":"10.1016\/j.jmva.2012.04.019"},{"key":"52","doi-asserted-by":"publisher","unstructured":"Vu, V. (2011). Singular vectors under random perturbation. <i>Random Structures Algorithms<\/i> <b>39<\/b> 526\u2013538.","DOI":"10.1002\/rsa.20367"},{"key":"54","doi-asserted-by":"publisher","unstructured":"Wang, R. (2015). Singular vector perturbation under Gaussian noise. <i>SIAM J. Matrix Anal. Appl.<\/i> <b>36<\/b> 158\u2013177.","DOI":"10.1137\/130938177"},{"key":"55","doi-asserted-by":"publisher","unstructured":"Wedin, P. (1972). Perturbation bounds in connection with singular value decomposition. <i>BIT<\/i> <b>12<\/b> 99\u2013111.","DOI":"10.1007\/BF01932678"},{"key":"51","doi-asserted-by":"publisher","unstructured":"Tropp, J. A. (2015). An introduction to matrix concentration inequalities. <i>Found. Trends Mach. Learn.<\/i> <b>8<\/b> 1\u2013230.","DOI":"10.1561\/2200000048"},{"key":"37","doi-asserted-by":"crossref","unstructured":"Koltchinskii, V. and Xia, D. (2016). Perturbation of linear forms of singular vectors under Gaussian noise. In <i>High Dimensional Probability VII<\/i>. <i>Progress in Probability<\/i> <b>71<\/b> 397\u2013423. Springer, Cham.","DOI":"10.1007\/978-3-319-40519-3_18"},{"key":"43","doi-asserted-by":"publisher","unstructured":"O\u2019Rourke, S., Vu, V. and Wang, K. (2018). Random perturbation of low rank matrices: Improving classical bounds. <i>Linear Algebra Appl.<\/i> <b>540<\/b> 26\u201359.","DOI":"10.1016\/j.laa.2017.11.014"},{"key":"1","doi-asserted-by":"publisher","unstructured":"Abbe, E., Fan, J., Wang, K. and Zhong, Y. (2020). Entrywise eigenvector analysis of random matrices with low expected rank. <i>Ann. Statist.<\/i> <b>48<\/b> 1452\u20131474.","DOI":"10.1214\/19-AOS1854"},{"key":"4","unstructured":"Bao, Z., Ding, X. and Wang, K. (2018). Singular vector and singular subspace distribution for the matrix denoising model. Preprint. Available at <a href=\"arXiv:1809.10476\">arXiv:1809.10476<\/a>."},{"key":"5","doi-asserted-by":"publisher","unstructured":"Bauer, F. L. and Fike, C. T. (1960). Norms and exclusion theorems. <i>Numer. Math.<\/i> <b>2<\/b> 137\u2013141.","DOI":"10.1007\/BF01386217"},{"key":"8","doi-asserted-by":"publisher","unstructured":"Benaych-Georges, F. and Rochet, J. (2016). Outliers in the single ring theorem. <i>Probab. Theory Related Fields<\/i> <b>165<\/b> 313\u2013363.","DOI":"10.1007\/s00440-015-0632-x"},{"key":"9","doi-asserted-by":"publisher","unstructured":"Bordenave, C. and Capitaine, M. (2016). Outlier eigenvalues for deformed i.i.d. random matrices. <i>Comm. Pure Appl. Math.<\/i> <b>69<\/b> 2131\u20132194.","DOI":"10.1002\/cpa.21629"},{"key":"10","doi-asserted-by":"publisher","unstructured":"Br\u00e9zin, E. and Zee, A. (1998). Non-Hermitean delocalization: Multiple scattering and bounds. <i>Nuclear Phys. B<\/i> <b>509<\/b> 599\u2013614.","DOI":"10.1016\/S0550-3213(97)00652-4"},{"key":"11","unstructured":"Bryc, W. and Silverstein, J. W. (2018). Singular values of large non-central random matrices. Preprint. Available at <a href=\"arXiv:1802.02960\">arXiv:1802.02960<\/a>."},{"key":"12","doi-asserted-by":"publisher","unstructured":"Cai, T., Han, X. and Pan, G. (2017). Limiting laws for divergent spiked eigenvalues and largest non-spiked eigenvalue of sample covariance matrices. Preprint. Available at <a href=\"arXiv:1711.00217\">arXiv:1711.00217<\/a>.","DOI":"10.1214\/18-AOS1798"},{"key":"13","doi-asserted-by":"publisher","unstructured":"Cai, T. T. and Zhang, A. (2018). Rate-optimal perturbation bounds for singular subspaces with applications to high-dimensional statistics. <i>Ann. Statist.<\/i> <b>46<\/b> 60\u201389.","DOI":"10.1214\/17-AOS1541"},{"key":"14","unstructured":"Cai, C., Li, G., Chi, Y., Poor, H. V. and Chen, Y. (2019). Subspace estimation from unbalanced and incomplete data matrices: $\\ell _{2,\\infty}$ statistical guarantees. Preprint. Available at <a href=\"arXiv:1910.04267\">arXiv:1910.04267<\/a>."},{"key":"16","doi-asserted-by":"publisher","unstructured":"Cape, J., Tang, M. and Priebe, C. E. (2019). Signal-plus-noise matrix models: Eigenvector deviations and fluctuations. <i>Biometrika<\/i> <b>106<\/b> 243\u2013250.","DOI":"10.1093\/biomet\/asy070"},{"key":"18","doi-asserted-by":"crossref","unstructured":"Chen, Y., Cheng, C. and Fan, J. (2020). Supplement to \u201cAsymmetry helps: Eigenvalue and eigenvector analyses of asymmetrically perturbed low-rank matrices.\u201d <a href=\"https:\/\/doi.org\/10.1214\/20-AOS1963SUPP\">https:\/\/doi.org\/10.1214\/20-AOS1963SUPP<\/a>","DOI":"10.1214\/20-AOS1963SUPP"},{"key":"19","doi-asserted-by":"publisher","unstructured":"Chen, Y., Chi, Y., Fan, J., Ma, C. and Yan, Y. (2019a). Noisy matrix completion: Understanding statistical guarantees for convex relaxation via nonconvex optimization. Available at <a href=\"arXiv:1902.07698\">arXiv:1902.07698<\/a>.","DOI":"10.1137\/19M1290000"},{"key":"20","doi-asserted-by":"publisher","unstructured":"Chen, Y., Fan, J., Ma, C. and Wang, K. (2019b). Spectral method and regularized MLE are both optimal for top-$K$ ranking. <i>Ann. Statist.<\/i> <b>47<\/b> 2204\u20132235.","DOI":"10.1214\/18-AOS1745"},{"key":"21","doi-asserted-by":"publisher","unstructured":"Chen, Y., Fan, J., Ma, C. and Yan, Y. (2019c). Inference and uncertainty quantification for noisy matrix completion. <i>Proc. Natl. Acad. Sci. USA<\/i> <b>116<\/b> 22931\u201322937.","DOI":"10.1073\/pnas.1910053116"},{"key":"22","unstructured":"Cheng, C., Wei, Y. and Chen, Y. (2020). Tackling small eigen-gaps: Fine-grained eigenvector estimation and inference under heteroscedastic noise. Preprint. Available at <a href=\"arXiv:2001.04620\">arXiv:2001.04620<\/a>."},{"key":"23","doi-asserted-by":"publisher","unstructured":"Chi, Y., Lu, Y. M. and Chen, Y. (2019). Nonconvex optimization meets low-rank matrix factorization: An overview. <i>IEEE Trans. Signal Process.<\/i> <b>67<\/b> 5239\u20135269.","DOI":"10.1109\/TSP.2019.2937282"},{"key":"25","unstructured":"Eldridge, J., Belkin, M. and Wang, Y. (2018). Unperturbed: Spectral analysis beyond Davis\u2013Kahan. In <i>Algorithmic Learning Theory 2018<\/i>. <i>Proc. Mach. Learn. Res. (PMLR)<\/i> <b>83<\/b> 38. Proceedings of Machine Learning Research PMLR."},{"key":"26","doi-asserted-by":"publisher","unstructured":"Erdos, L., Knowles, A., Yau, H.-T. and Yin, J. (2013). Spectral statistics of Erdos\u2013R\u00e9nyi graphs I: Local semicircle law. <i>Ann. Probab.<\/i> <b>41<\/b> 2279\u20132375.","DOI":"10.1214\/11-AOP734"},{"key":"27","unstructured":"Fan, J., Wang, W. and Zhong, Y. (2017). An $\\ell_{\\infty}$ eigenvector perturbation bound and its application to robust covariance estimation. <i>J. Mach. Learn. Res.<\/i> <b>18<\/b> Paper No. 207, 42."},{"key":"28","doi-asserted-by":"publisher","unstructured":"Feinberg, J. and Zee, A. (1997). Non-Hermitian random matrix theory: Method of Hermitian reduction. <i>Nuclear Phys. B<\/i> <b>504<\/b> 579\u2013608.","DOI":"10.1016\/S0550-3213(97)00502-6"},{"key":"34","doi-asserted-by":"publisher","unstructured":"Keshavan, R. H., Montanari, A. and Oh, S. (2010). Matrix completion from a few entries. <i>IEEE Trans. Inf. Theory<\/i> <b>56<\/b> 2980\u20132998.","DOI":"10.1109\/TIT.2010.2046205"},{"key":"35","doi-asserted-by":"publisher","unstructured":"Khoruzhenko, B. (1996). Large-$N$ eigenvalue distribution of randomly perturbed asymmetric matrices. <i>J. Phys. A<\/i> <b>29<\/b> L165\u2013L169.","DOI":"10.1088\/0305-4470\/29\/7\/003"},{"key":"36","doi-asserted-by":"publisher","unstructured":"Knowles, A. and Yin, J. (2013). The isotropic semicircle law and deformation of Wigner matrices. <i>Comm. Pure Appl. Math.<\/i> <b>66<\/b> 1663\u20131750.","DOI":"10.1002\/cpa.21450"},{"key":"38","doi-asserted-by":"publisher","unstructured":"Li, X., Ling, S., Strohmer, T. and Wei, K. (2019). Rapid, robust, and reliable blind deconvolution via nonconvex optimization. <i>Appl. Comput. Harmon. Anal.<\/i> <b>47<\/b> 893\u2013934.","DOI":"10.1016\/j.acha.2018.01.001"},{"key":"39","doi-asserted-by":"publisher","unstructured":"Lytova, A. and Tikhomirov, K. (2018). On delocalization of eigenvectors of random non-Hermitian matrices. Preprint. Available at <a href=\"arXiv:1810.01590\">arXiv:1810.01590<\/a>.","DOI":"10.1007\/s00440-019-00956-8"},{"key":"40","doi-asserted-by":"publisher","unstructured":"Ma, C., Wang, K., Chi, Y. and Chen, Y. (2020). Implicit regularization in nonconvex statistical estimation: Gradient descent converges linearly for phase retrieval, matrix completion and blind deconvolution. <i>Found. Comput. Math.<\/i> <b>20<\/b> 451\u2013632.","DOI":"10.1007\/s10208-019-09429-9"},{"key":"41","doi-asserted-by":"crossref","unstructured":"Mehlig, B. and Chalker, J. T. (1998). Eigenvector correlations in non-Hermitian random matrix ensembles. <i>Ann. Phys.<\/i> <b>7<\/b> 427\u2013436.","DOI":"10.1002\/andp.199851005-610"},{"key":"42","doi-asserted-by":"publisher","unstructured":"O\u2019Rourke, S., Vu, V. and Wang, K. (2016). Eigenvectors of random matrices: A survey. <i>J. Combin. Theory Ser. A<\/i> <b>144<\/b> 361\u2013442.","DOI":"10.1016\/j.jcta.2016.06.008"},{"key":"45","unstructured":"Rajagopalan, A. B. (2015). Outlier eigenvalue fluctuations of perturbed iid matrices. Preprint. Available at <a href=\"arXiv:1507.01441\">arXiv:1507.01441<\/a>."},{"key":"47","doi-asserted-by":"publisher","unstructured":"Silverstein, J. W. (1994). The spectral radii and norms of large-dimensional non-central random matrices. <i>Comm. Statist. Stochastic Models<\/i> <b>10<\/b> 525\u2013532.","DOI":"10.1080\/15326349408807308"},{"key":"48","doi-asserted-by":"crossref","unstructured":"Sommers, H.-J., Crisanti, A., Sompolinsky, H. and Stein, Y. (1988). Spectrum of large random asymmetric matrices. <i>Phys. Rev. Lett.<\/i> <b>60<\/b> 1895\u20131898.","DOI":"10.1103\/PhysRevLett.60.1895"},{"key":"49","doi-asserted-by":"crossref","unstructured":"Tao, T. (2012). <i>Topics in Random Matrix Theory<\/i>. <i>Graduate Studies in Mathematics<\/i> <b>132<\/b>. Amer. Math. Soc., Providence, RI.","DOI":"10.1090\/gsm\/132"},{"key":"50","doi-asserted-by":"publisher","unstructured":"Tao, T. (2013). Outliers in the spectrum of iid matrices with bounded rank perturbations. <i>Probab. Theory Related Fields<\/i> <b>155<\/b> 231\u2013263.","DOI":"10.1007\/s00440-011-0397-9"},{"key":"53","doi-asserted-by":"publisher","unstructured":"Vu, V. and Wang, K. (2015). Random weighted projections, random quadratic forms and random eigenvectors. <i>Random Structures Algorithms<\/i> <b>47<\/b> 792\u2013821.","DOI":"10.1002\/rsa.20561"},{"key":"57","doi-asserted-by":"crossref","unstructured":"Xia, D. (2019). Confidence region of singular subspaces for low-rank matrix regression. <i>IEEE Trans. Inf. Theory<\/i> <b>65<\/b> 7437\u20137459.","DOI":"10.1109\/TIT.2019.2924900"},{"key":"59","doi-asserted-by":"publisher","unstructured":"Zhang, A. and Xia, D. (2018). Tensor SVD: Statistical and computational limits. <i>IEEE Trans. Inf. Theory<\/i> <b>64<\/b> 7311\u20137338.","DOI":"10.1109\/TIT.2018.2841377"}],"container-title":["The Annals of Statistics"],"original-title":[],"link":[{"URL":"https:\/\/projecteuclid.org\/download\/pdfview_1\/euclid.aos\/1611889234","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T15:07:01Z","timestamp":1791472021000},"score":1,"resource":{"primary":{"URL":"https:\/\/projecteuclid.org\/journals\/annals-of-statistics\/volume-49\/issue-1\/Asymmetry-helps--Eigenvalue-and-eigenvector-analyses-of-asymmetrically-perturbed\/10.1214\/20-AOS1963.full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2021,2,1]]},"references-count":59,"journal-issue":{"issue":"1","published-online":{"date-parts":[[2021,2,1]]}},"URL":"https:\/\/doi.org\/10.1214\/20-aos1963","relation":{},"ISSN":["0090-5364","2168-8966"],"issn-type":[{"value":"0090-5364","type":"print"},{"value":"2168-8966","type":"electronic"}],"subject":[],"published":{"date-parts":[[2021,2,1]]}}}