Analyse fonctionnelle / Functional Analysis Smallest singular value of random matrices with independent columns
Analyse fonctionnelle / Functional Analysis Smallest singular value of random matrices with independent columns
复制标题
分析 fonctionnelle / 泛函分析 具有独立列的随机矩阵的最小奇异值
DOI:
--
复制
发表时间:
2009
期刊:
影响因子:
--
通讯作者:
N. Tomczak
中科院分区:
文献类型:
--
作者:
Rados law;O. Guédon;A. Litvak;A. Pajor;N. Tomczak
We study the smallest singular value of a square random matrix with i.i.d. columns drawn from an isotropic symmetric log-concave distribution. We prove a deviation inequality in terms of the isotropic constant of the distribution. Sur la plus petite valeur singulière de matrices aléatoires avec des colonnes indépendantes Résumé. On étudie la plus petite valeur singulière d’une matrice carrée aléatoire dont les colonnes sont des vecteurs aléatoires i.i.d. suivant une loi à densité log-concave isotrope. On démontre une inégalité de déviation en fonction de la constante d’isotropie. The behaviour of the smallest singular value of random matrices with i.i.d. random entries attracted a lot of attention over the years. Major results were recently obtained in [5, 8, 9, 10]. In asymptotic geometry one is interested in sampling vectors uniformly distributed in a convex body. In particular the entries are not necessarily independent. In this note, we study the more general case when the columns are i.i.d. random vectors with a symmetric isotropic log-concave distribution. We prove a deviation inequality for the smallest singular value in terms of a parameter Lμ which, in the case of sampling from a convex body, corresponds to the isotropic constant of the body. Recall that a non-negative function f on R is called log-concave if for all x, y ∈ R and all θ ∈ (0, 1), f((1 − θ)x + θy) ≥ f(x)1−θf(y)θ. In this paper a symmetric probability measure μ on R is said to be log-concave if its density f is symmetric log-concave and it is called isotropic if its covariance matrix is the identity. We will also set Lμ = f(0). Let us observe that if μ is an isotropic probability measure uniformly distributed on a symmetric convex body K then Lμ is the 1This work was done when this author held a postdoctoral position at the Department of Mathematical and Statistical Sciences, University of Alberta in Edmonton, Alberta. The position was sponsored by the Pacific Institute for the Mathematical Sciences. 2This author holds the Canada Research Chair in Geometric Analysis.