Algorithms for the computation of the pseudospectral radius and the numerical radius of a matrix

^** Email: mengi{at}cs.nyu.edu^*** Email: overton{at}cs.nyu.edu Two useful measures of the robust stability of the discrete-timedynamical system x_k+1 = Ax_k are the -pseudospectral radius andthe numerical radius of A. The -pseudospectral radius of A isthe largest of the moduli of the points in the -pseudospectrumof A, while the numerical radius is the largest of the moduliof the points in the field of values. We present globally convergentalgorithms for computing the -pseudospectral radius and thenumerical radius. For the former algorithm, we discuss conditionsunder which it is quadratically convergent and provide a detailedaccuracy analysis giving conditions under which the algorithmis backward stable. The algorithms are inspired by methods ofByers, Boyd–Balakrishnan, He–Watson and Burke–Lewis–Overtonfor related problems and depend on computing eigenvalues ofsymplectic pencils and Hamiltonian matrices.     

Isometries of a generalized numerical radius

For 0<q<1, the q-numerical range is defined on the algebra M_n of all n×n complex matrices by

W_q(A)={x^∗Ay:x,y∈Cⁿ,∥x∥=∥y∥=1,〈y,x〉=q}.     

Convex numerical radius

In this work we introduce the concept of convex numerical radius for a continuous and linear operator in a Banach space, which generalizes that of the classical numerical radius. Besides studying some of its properties, we give a version of James's sup theorem in terms of convex numerical radius attaining operators.     

On the spectral radius of a nonnegative centrosymmetric matrix

In this paper, we discuss the spectral radius of nonnegative centrosymmetric matrices. By using the centrosymmetric structure, we establish some estimations of the spectral radius.     

Matrix polynomials with spectral radius equal to the numerical radius

In this paper we study a class of matrix polynomials with the property that spectral radius and numerical radius coincide. Special attention is paid to the spectrum on the boundary of the numerical range.     

Sharp bounds on the spectral radius of a nonnegative matrix

We give upper and lower bounds for the spectral radius of a nonnegative matrix using its row sums and characterize the equality cases if the matrix is irreducible. Then we apply these bounds to various matrices associated with a graph, including the adjacency matrix, the signless Laplacian matrix, the distance matrix, the distance signless Laplacian matrix, and the reciprocal distance matrix. Some known results in the literature are generalized and improved.     

On the numerical radius and its applications

We give a brief account of the numerical radius of a linear bounded operator on a Hilbert space and some of its better-known properties. Both finite- and infinite- dimensional aspects are discussed, as well as applications to stability theory of finite-difference approximations for hyperbolic initial-value problems.     

A recursive computation of the determinant of a pentadiagonal matrix

A new algorithm for treating numerically a pentadiagonal matrix is given. As an application the evaluation of the eigenvalues is performed. Also the stability of the solution of some difference equations is examined.     

Linear operators preserving the decomposable numerical radius

The characterization of all linear operators on matrices which preserve the decomposable numerical radius is obtained. This result refines those of Tam. Marcus and Filippenko on the topic. The proof of the main theorem depends on a characterization of scalar multiples of unitary matrices in terms of decomposable numerical radius that is of independent interest.     

An algorithm for computing the numerical radius

Received on 23 October 1995. Revised on 15 July 1996. This paper is concerned with the calculation of the numericalradius of a matrix, an important quantity in the analysis ofconvergence of iterative processes. An algorithm is developedwhich enables the numerical radius to be obtained to a givenprecision, using a process which successively refines lowerand upper bounds. It uses an iteration procedure analogous tothe power method for computing the largest modulus eigenvalueof a Hermitian matrix. In contrast to that method, convergenceis possible here to a local maximum of the underlying optimizationproblem which is not global, so that only a lower bound is provided.This is used in conjunction with a technique based on the solutionof a generalized cigenvalue problem to provide an upper bound.Numerical results illustrate the performance of the method.     

On the effective computation of the inertia of a non-hermitian matrix

Summary given a complex lower Hessenberg matrixA with unit codiagonal, a hermitian matrixH is constructed such that, ifH is non-singular InA= InH. IfA is real,H is real symmetric. Classical results of Fujiwara on the root-separation problem and of Schwarz on the eigenvalue-separation problem are included as special cases.The authors' research was conducted at the Universidade Estadual de Campinas and supported by the Fundação de Amparo à Pesquisa do Estado de São Paulo, Brasil, under grant n⁰ 78/0490.     

On the computation of the generalized inverse of a polynomial matrix

The main purpose of this paper is to present a quicker and less memory-expensive algorithm for the generalized inversionof polynomial matrices than those presented earlier (Karampetakis,1997a Computation of the generalized inverse of a polynomialmatrix and applications. Linear Algebr. Appl. 252, 35–60 and Karampetakis, 1997b Generalized inverses of two variable polynomial matrices and applications. Circuit Syst. & Signal Process. 16, 439–453). Received 24 January, 1999. ⁺ karampetakis@ccf.auth.gr     

A compound matrix algorithm for the computation of the Smith form of a polynomial matrix

In the present paper is presented a numerical method for the exact reduction of a singlevariable polynomial matrix to its Smith form without finding roots and without applying unimodular transformations. Using the notion of compound matrices, the Smith canonical form of a polynomial matrixM(s)^nxn[s] is calculated directly from its definition, requiring only the construction of all thep-compound matricesC _p (M(s)) ofM(s), 1<pn. This technique produces a stable and accurate numerical algorithm working satisfactorily for any polynomial matrix of any degree.