On bounds for eigenvalues of real symmetric matrices

Let A=(a_ij) be a real symmetric matrix of order n. We characterize all nonnegative vectors x=(x₁,...,x_n) and y=(y₁,...,y_n) such that any real symmetric matrix B=(b_ij), with b_ij=a_ij, i≠jhas its eigenvalues in the union of the intervals [b_ij?y_i, b_ij+ x_i]. Moreover, given such a set of intervals, we derive better bounds for the eigenvalues of B using the 2n quantities {b_ii?y, b_ii+x_i}, i=1,..., n.     

Extreme eigenvalues of real symmetric Toeplitz matrices

We exploit the even and odd spectrum of real symmetric Toeplitz matrices for the computation of their extreme eigenvalues, which are obtained as the solutions of spectral, or secular, equations. We also present a concise convergence analysis for a method to solve these spectral equations, along with an efficient stopping rule, an error analysis, and extensive numerical results.

     

On real eigenvalues of real tridiagonal matrices

Lower bounds for the number of different real eigenvalues as well as for the number of real simple eigenvalues of a class of real irreducible tridiagonal matrices are given. Some numerical implications are discussed.     

Eigenvalues and eigenvectors of symmetric centrosymmetric matrices

It is proved that the eigenvectors of a symmetric centrosymmetric matrix of order N are either symmetric or skew symmetric, and that there are ?N/2? symmetric and ?N/2? skew symmetric eigenvectors. Some previously known but widely scattered facts about symmetric centrosymmetric matrices are presented for completeness. Special cases are considered, in particular tridiagonal matrices of both odd and even order, for which it is shown that the eigenvectors corresponding to the eigenvalues arranged in descending order are alternately symmetric and skew symmetric provided the eigenvalues are distinct.     

Extremal eigenvalues and eigenvectors of deformed Wigner matrices

We consider random matrices of the form \(H = W + \lambda V, \lambda \in {\mathbb {R}}^+\), where \(W\) is a real symmetric or complex Hermitian Wigner matrix of size \(N\) and \(V\) is a real bounded diagonal random matrix of size \(N\) with i.i.d. entries that are independent of \(W\). We assume subexponential decay of the distribution of the matrix entries of \(W\) and we choose \(\lambda \sim 1\), so that the eigenvalues of \(W\) and \(\lambda V\) are typically of the same order. Further, we assume that the density of the entries of \(V\) is supported on a single interval and is convex near the edges of its support. In this paper we prove that there is \(\lambda _+\in {\mathbb {R}}^+\) such that the largest eigenvalues of \(H\) are in the limit of large \(N\) determined by the order statistics of \(V\) for \(\lambda >\lambda _+\). In particular, the largest eigenvalue of \(H\) has a Weibull distribution in the limit \(N\rightarrow \infty \) if \(\lambda >\lambda _+\). Moreover, for \(N\) sufficiently large, we show that the eigenvectors associated to the largest eigenvalues are partially localized for \(\lambda >\lambda _+\), while they are completely delocalized for \(\lambda <\lambda _+\). Similar results hold for the lowest eigenvalues.     

On the intermediate eigenvalues of symmetric sparse matrices

An algorithm has been developed for finding the eigenvalues of a symmetric matrixA in a given interval [a, b] and the corresponding eigenvectors using a modification of the method of simultaneous iteration with the same favorable convergence properties. The technique is most suitable for large sparse matrices and can be effectively implemented on a parallel computer such as the ILLIAC IV.     

On relative perturbation theory for eigenvalues and eigenvectors of block operator matrices

We are concerned with singularly perturbed spectral problems which appear in engineering sciences. Typically under the influence of a singular perturbation the model can be approximated by a simpler, perturbation independent model. Such reduced model is usually better amenable to analytic or numeric analysis. However, the question of the quality of approximation has to be answered. Frequently, correctors which yield an improved solution–capturing important phenomena which the reduced model does not “see”–to the original problems are required. We tackle both question for self-adjoint eigenvalue/eigenvector problems posed in a general Hilbert space. Our technique is constructive and is based on methods (relative perturbation theory) of modern Numerical Linear Algebra. (© 2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim)     

On the convergence of the classical Jacobi method for real symmetric matrices with non-distinct eigenvalues

Summary It is proved that the classical Jacobi method for real symmetric matrices with multiple eigenvalues converges quadratically.While this paper was in the press, SCHöNHAGE gave a different proof of the ultimate quadratic convergence of the classical Jacobi method. [See Numer. Math. 6, 410–412 (1964).     

On the eigenvalues and eigenvectors of certain finite,vertex-weighted,bipartite graphs

The established, spectral characterisation of bipartite graphs with unweighted vertices (which are here termed homogeneous graphs) is extended to those bipartite graphs (called heterogeneous) in which all of the vertices in one set are weighted h₁ , and each of those in the other set of the bigraph is weighted h₂. All the eigenvalues of a homogeneous bipartite graph occur in pairs, around zero, while some of the eigenvalues of an arbitrary, heterogeneous graph are paired around $\text{1}\text{2}$(h₁ + h₂), the remainder having the value h₂ (or h_l). The well-documented, explicit relations between the eigenvectors belonging to “paired” eigenvalues of homogeneous graphs are extended to relate the components of the eigenvectors associated with each couple of “paired” eigenvalues of the corresponding heterogeneous graph. Details are also given of the relationships between the eigenvectors of an arbitrary, homogeneous, bipartite graph and those of its heterogeneous analogue.     

On the concentration of eigenvalues of random symmetric matrices

It is shown that for every 1≤s≤n, the probability that thes-th largest eigenvalue of a random symmetricn-by-n matrix with independent random entries of absolute value at most 1 deviates from its median by more thant is at most 4e ⁻ t ²32 ^s2. The main ingredient in the proof is Talagrand’s Inequality for concentration of measure in product spaces. Research supported in part by a USA — Israel BSF grant, by a grant from the Israel Science Foundation and by the Hermann Minkowski Minerva Center for Geometry at Tel Aviv University. Research supported in part by a USA — Israel BSF grant and by a Bergmann Memorial Grant.     

A computational method for eigenvalues and eigenvectors of a matrix with real eigenvalues

This paper describes a new computational procedure for calculating eigenvalues and eigenvectors of a square matrix. The method is based on a matrix function, the sign of a matrix. Eigenvalues and eigenvectors of matrices with distinct eigenvalues and nondefective matrices with repeated roots can be determined in a straightforward manner. Defective matrices require additional calculations.     

Computation of eigenvalues and eigenvectors of a symmetric quindiagonal matrix

A recursive algorithm for the implicit derivation of the determinant of a symmetric quindiagonal matrix is developed in terms of its leading principal minors. The algorithm is shown to yield a Sturmian sequence of polynomials from which the eigenvalues can be obtained by use of the bisection process. Further modifications to the inverse iteration method using Wilkinson's technique (1962) yields the required eigenvectors.     

Relative perturbation results for eigenvalues and eigenvectors of diagonalisable matrices

Let and be a perturbed eigenpair of a diagonalisable matrixA. The problem is to bound the error in and . We present one absolute perturbation bound and two relative perturbation bounds. The absolute perturbation bound is an extension of Davis and Kahan's sin θ Theorem from Hermitian to diagonalisable matrices. The two relative perturbation bounds assume that and are an exact eigenpair of a perturbed matrixD ₁ AD ₂ , whereD ₁ andD ₂ are non-singular, butD ₁ AD ₂ is not necessarily diagonalisable. We derive a bound on the relative error in and a sin θ theorem based on a relative eigenvalue separation. The perturbation bounds contain both the deviation ofD ₁ andD ₂ from similarity and the deviation ofD ₂ from identity. This work was partially supported by NSF grant CCR-9400921.     

On graphs whose Laplacian matrices have distinct integer eigenvalues

In this paper, we investigate graphs for which the corresponding Laplacian matrix has distinct integer eigenvalues. We define the set S_i,n to be the set of all integers from 0 to n, excluding i. If there exists a graph whose Laplacian matrix has this set as its eigenvalues, we say that this set is Laplacian realizable. We investigate the sets S_i,n that are Laplacian realizable, and the structures of the graphs whose Laplacian matrix has such a set as its eigenvalues. We characterize those i < n such that S_i,n is Laplacian realizable, and show that for certain values of i, the set S_i,n is realized by a unique graph. Finally, we conjecture that S_n,n is not Laplacian realizable for n ≥ 2 and show that the conjecture holds for certain values of n. © 2005 Wiley Periodicals, Inc. J Graph Theory     

Variational equations for the eigenvalues and eigenvectors of nonsymmetric matrices

The tracking of eigenvalues and eigenvectors for parameterized matrices is of major importance in optimization and stability problems. In the present paper, we consider a one-parameter family of matrices with distinct eigenvalues. A complete system of differential equations is developed for both the eigenvalues and the right and left eigenvectors. The computational feasibility of the differential system is demonstrated by means of a numerical example.The work of R. Kalaba and L. Tesfatsion was partially supported by the National Science Foundation under Grant No. ENG-77-28432 and by the National Institutes of Health under Grant No. GM-23732-03.     

On Lyapunov scaling factors of real symmetric matrices

A necessary and sufficient condition for the identity matrix to be the unique Lyapunov scaling factor of a given real symmetric matrix A is given. This uniqueness is shown to be equivalent to the uniqueness of the identity matrix as a scaling D for which the kernels of A and AD are identical.     

On real matrices with positive definite symmetric component

Inequalities concerning real square matrices A with positive definite symmetric component A+A^*are derived from certain inertia relations which hold for any complex (not necessarily real) square matrices A with positive definite

A+A^*