A new inequality about matrix products and a Berger-Wang formula

DOI10.5802/JEP.114MaRDI QIDQ2288148zbMATH OpenOpenAlexWikidataFDO

Authors Eduardo Oregón-Reyes

Publication date 17 January 2020

Published in Journal de l’École polytechnique — Mathématiques (Search for Journal in Brave)

Copyright license Creative Commons Attribution 4.0 International

Full work available at URL https://arxiv.org/abs/1710.00639

zbMATH Keywords

Berger-Wang formula joint spectral radius linear cocycle Lyapunov exponent product of nilpotent matrices

Mathematics Subject Classification ID

Miscellaneous inequalities involving matrices (15A45) Norms of matrices, numerical range, applications of functional analysis to matrix theory (15A60) Random dynamical systems aspects of multiplicative ergodic theory, Lyapunov exponents (37H15) Norms (inequalities, more than one norm, etc.) of linear operators (47A30)

Abstract: We prove an inequality relating the norm of a product of matrices

A_{n} c d o t s A_{1}

with the spectral radii of subproducts

A_{j} c d o t s A_{i}

with

1 l e q i l e q j l e q n

. Among the consequences of this inequality, we obtain the classical Berger-Wang formula as an immediate corollary, and give an easier proof of a characterization of the upper Lyapunov exponent due to I. Morris. As main ingredient for the proof of this result, we prove that for a large enough

n

, the product

A_{n} c d o t s A_{1}

is zero under the hypothesis that

A_{j} c d o t s A_{i}

are nilpotent for all

1 l e q i l e q j l e q n

.

Review text

The Beruling-Gelfand spectral radius formula asserts that for each \(A\in M_d(\mathbb{C})\), \[ \lim_{m\to \infty} \|A^m\|^{1/m} = \rho(A),\] where \(\rho(A)\) is the spectral radius of \(A\) and \(\|\cdot\|\) is the spectral norm. The formula is also valid for any matrix norm, i.e., a norm with submultiplicative property \(\||AB\||\le \||A\|\||B\||\) for all \(A, B \in M_d(\mathbb{C})\), Let \(\|\cdot\|\) be the operator norm \(M_d(\mathbb{C})\) induced by a vector norm \(\|\cdot\|\) on \(\mathbb{C}^n\). The Berger-Wang formula (see [\textit{M. A. Berger} and \textit{Y. Wang}, Linear Algebra Appl. 166, 21--27 (1992; Zbl 0818.15006)]) is an extension of Beruling-Gelfand formula and asserts that if \({\mathcal M}\subset M_d(\mathbb{C})\) is a bounded set, then \[\mathrm R({\mathcal M}) = {\lim\sup}_{n\to \infty} (\sup \{\rho(A_1\cdots A_n): A_i\in {\mathcal M}\})^{1/n}\] where \[\mathrm R({\mathcal M}):= \lim_{n\to \infty} (\sup\{\|A_1 \cdots A_n\|: A_i\in {\mathcal M}\})^{1/n}\] is called the joint spectral radius of \({\mathcal M}\), introduced by \textit{G.-C. Rota} and \textit{W. G. Strang} [Nederl. Akad. Wet., Proc., Ser. A 63, 379--381 (1960; Zbl 0095.09701)]. It represents the maximal exponential growth rate of the partial sequence of products \((A_1\cdots A_n)_n\) of a sequence of matrices \(A_1, A_2, \dots, \) with \(A_i\in \mathcal M\). The quantity \(\sup \{\rho(A_1\cdots A_n): A_i\in {\mathcal M}\}\) is called the generalized spectral radius. Along with several results, the author proves the following: \vspace{.2cm} Theorem. Let \(d\in \mathbb{N}\), \(k\) be a local field, and \(\|\cdot\|\) be matrix norm on \(M_d(k)\). There exist constants \(N=N(d)\le \prod_{i=1}^d \binom di\), \(r=r(d,N)\le (Nd+1)^{Nd^2+2}\) and \(C =C(d,\|\cdot\|)>1\) such that for all \(n\ge N\) and \(A_1, \dots, A_n\in M_d(k)\): \[ \|A_n\cdots A_1\|\le C \left(\prod_{1\le i\le n} \|A_i\| \right) \max_{1\le \alpha \le \beta \le n} \left( \frac {\rho(A_\beta\cdots A_\alpha)}{\prod_{\alpha\le i\le \beta} \|A_i\|}\right)^{1/r},\tag{1} \] where the right side is treated as zero if one of the \(A_i\) is zero. Thus, for large \(n\), if \(\|A_n\cdots A_1\|\) is comparable to \(\prod_{1\le i\le n} \|A_i\|\), then there exist s sub-product \(A_\beta, \dots, A_\alpha\) whose spectral radius is comparable to \(\prod_{\alpha\le i\le \beta} \|A_i\|\). The Beruling-Gelfand formula is used in the proof of Theorem 1.2 to get a uniform bound for the multiplicative constant in the general case of matrix norms, and is not needed for the case of operator norms. The proof of the inequality is based on the non trivial case of equality, where the right hand side is zero but the matrices \(A_i\) are nonzero. This particular case can be stated in the context of nilpotent matrices in Theorem 1.3 which is related to some nice theorems such as Levitzki's theorem or Burnside-Schur theorem. An interesting polynomial identity is obtained in the process. Remark: The author proves (see Proposition 4.2) that when \(k=\mathbb{C}\), there is \(C =C(d)>1\) such that (1) holds for all operator norms \(\|\cdot\|\) on \(M_d(\mathbb{C})\) and \(A_1, \dots, A_n\in M_d(\mathbb{C})\). The reviewer would like to point out that it is true for all matrix norms as Lemma 4.3 is also true for matrix norms \(\||\cdot\||\) because \(\||SAS^{-1}\||\) (see Lemma 4.3) is also a matrix norm for each nonsingular \(S\in M_d(\mathbb{C})\). It is worth to observe that operator norms are induced by vector norms and are matrix norms (see, e.g.,Chapter 5 of [\textit{R. A. Horn} and \textit{C. R. Johnson}, Matrix analysis. 2nd ed. Cambridge: Cambridge University Press (2013; Zbl 1267.15001)]).

Recommendations

Cites work

Cited in

(9)

This page was built for publication: A new inequality about matrix products and a Berger-Wang formula

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2288148)