Perturbation theory for matrix equations (Q1396759)

The main object of study in this book are equations of the type \[ F(A,X)=0, \tag{1} \] and their properties with respect to perturbations. Here \(X\) is the unknown matrix which is assumed to belong to a known set \({\mathcal X}\in {\mathbf F}^{n\times m}\); we denote by \({\mathbf F}^{n\times m}\) the vector space of \(n\times m\) matrices with entries in \({\mathbf F}\), either the real field or the complex field. The element \(A\in{\mathcal V}\) is a parameter, where \({\mathcal V}\) is a Cartesian product of matrix spaces. The function \(F:{\mathcal D}\to {\mathbf F}^{p \times q}\) is a given continuous function defined on an open and connected set \({\mathcal D}\subseteq {\mathcal V}\times {\mathcal X}\). Denote by \(\Xi(A)\) the set of solutions \(X\) of (1), with a particular value of the parameter \(A\). A solution \(X\in \Xi(A)\) is called well-posed if \(X\) is contained in a family of solutions (supporting family) which depends continuously on the parameter in a neighborhood of \(A\). A solution \(X\in\Xi(A)\) is called proper if it is well-posed and isolated in \(\Xi(A)\). A solution \(X\in\Xi(A)\) is called regular if it is proper and at least one of its supporting families is Lipschitz continuous; otherwise a solution \(X\in\Xi(A)\) is called singular. Equation (1) is considered together with a perturbed equation \[ F(A+\delta A,X+\delta X)=0,\tag{2} \] where \(\delta A\) and \(\delta X\) are perturbations of \(A\) and of \(X\), respectively. If \(X\in\Xi(A)\), and the partial Fréchet derivative of \(F\) with respect to \(X\) is invertible, whereas the partial Fréchet derivative of \(F\) with respect to \(A\) exists, then \(X\) is guaranteed to be regular. For regular solutions, using an operator equation \(\delta X=\Pi (\delta A,\delta X)\) which is equivalent to (2), a non-local perturbation bound on \(\|\delta X\|\) (with respect to a specified matrix norm) is obtained via fixed point theorems for the operator \(\Pi\). The first seven chapters (pp. 1-120) contain extensive development of general notions and concepts of perturbation analysis, illustrated with examples. In particular, local first order and non-local perturbation bounds, and Lyapunov majorants are introduced and studied there. These notions and concepts are further fleshed out in detail in Chapters 8-16 (pp. 121-326) for many particular classes of matrix equations. The classes include matrix equations of continuous and discrete control theory (in particular, descriptor systems), such as Sylvester equations, Lyapunov equations, quadratic equations (in particular, continuous time and coupled Riccati equations), and general and symmetric fractional affine equations. The latter are equations of the form \[ F_1+F_2F_3^{-1}F_4=0, \] where \(F_i\), \(i=1,2, 3,4\), are affine matrix expressions in the unknown matrix \(X\). The book also contains eight appendices (pp. 327-405), with background material mainly on more advanced matrix theory, but also on normed spaces, functions, and fixed point principles. The bibliography is extensive (241 items), and includes many recent works. The book is a valuable addition to the literature. It offers a unique perspective, concise and systematic development of perturbation analysis of matrix equations, and a wealth of material on the topic. Large part of the material appeared first in papers authored by various subsets of the four authors of the book, oftentimes with other co-authors. Although sometimes neglected by practitioners of computation, perturbation analysis as presented in the book serves as a necessary theoretical basis for development of numerical methods, primarily concerning matrix equations in control, and is indispensable for estimating the accuracy of computed results.