Variance bounds, with an application to norm bounds for commutators (Q847200): Difference between revisions
From MaRDI portal
Latest revision as of 11:42, 2 July 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Variance bounds, with an application to norm bounds for commutators |
scientific article |
Statements
Variance bounds, with an application to norm bounds for commutators (English)
0 references
12 February 2010
0 references
Recall that the variance \(\text{Var}(X)\) of a random variable \(X\) that assumes real values \(x_{i}\) with probabilities \(p_{i}\) is defined as \(\text{Var}(X)=\sum_{i} p_{i}x_{i}^{2}-(\sum_{i} p_{i}x_{i})^2.\) It is of interest in mathematical statistics to have upper bounds for this quantity. Murthy and Sethi gave a sharp upper bound in terms of the range of values of the variable. This bound is generalized to complex case. Formally one can define the variance of a complex random variable \(X\) as \(\text{Var}(X)=\sum_{i} p_{i}|x_{i}|^{2}-|\sum_{i} p_{i}x_{i}|^2\) and then it follows from the first result of the paper (Theorem 2) that \(\text{Var}(X)\leq \min_{y\in \mathbb{C}} \max_{i}|x_{i}-y|^{2}.\) The notion of the variance of a random variable is generalized to the matrix case as well. For a complex matrix \(T\), let \(\| T\|_2\) be its Frobenius norm. \textit{A. Böttcher} and \textit{D. Wenzel} [Linear Algebra Appl. 429, 1864--1885 (2008; Zbl 1173.15010)] have proved that \( \| [T,S]\|_{2} \leq \sqrt{2}\| T\|_{2}\| S\|_{2}\), where \([T,S]=TS-ST\) is the commutator of two complex matrices. A new and elegant proof of this inequality gives a good motivation for introducing a quantity which can be seen as a generalization of the variance of a random variable to the matrix case. More precisely, for a complex matrix \(T\), let \(|T|_{L}=(T^* T)^{1/2}\), \(|T|_{R}=(T T^*)^{1/2}\), and \(|T|_{C}=\sqrt{(|T|_{L}^{2}+|T|_{R}^{2})/2}\). The \(\ast\)-variance of a \(d\)-by-\(d\) complex matrix \(T\) w.r.t.\ a density matrix \(\rho\), i.e., a positive semidefinite matrix with trace 1, is defined as \(\text{Var}_{\ast}(T)=Tr[\rho|T- Tr[\rho T] I_{d}|_{\ast}^{2}]\), where \(\ast\) stands for \(L\), \(R\) or \(C\) and \(I_{d}\) is the \(d\times d\) identity matrix. Some upper bounds for \(\text{Var}_{\ast}(T)\) are given. For instance, if \(T\) is a non-normal complex \(d\)-by-\(d\) matrix, then \(\text{Var}_{\ast}(T)\leq \min_{y\in \mathbb{C}}\| |T-yI_{d}|_{\ast}^{2}\|_{\infty}\). These results alow to prove the sharp bound on the Frobenius norm of a commutator. Namely, if \(T\) and \(S\) are complex matrices and \(p\geq 1\), then \(\| [T,S]\|_{2} \leq 2^{\max(1/2, 1-1/p)}\| T\|_{2} \|S\|_{p}.\)
0 references
norm inequalities
0 references
commutators
0 references
variance
0 references
Frobenius norm
0 references
numerical range
0 references