The convergence of matrices generated by rank-2 methods from the restricted β-class of Broyden

Summary It is shown that the matricesB _k generated by any method from the restricted -class of Broyden converge, if the method is applied to the unconstrained minimization of a functionfC ²(R ⁿ ) with Lipschitz continuous ² f(x) and if the method is such that it generates vectorsx _k converging sufficiently fast to a local minimumx ^* off with positive definite ² f(x ^*). This result not only holds for constant step sizes _k 1 in each iterationx _k x _k+1=x _k – _k B _k ^–1 f(x _k ) of these methods but also for step sizes determined by asymptotically exact line searches. The paper generalizes corresponding results of Ge Ren-Pu and Powell [6] for the DFP and BFGS methods used in conjunction with step sizes _k 1.Dedicated to Professor F.L. Bauer on the occasion of his 60th birthday     

On cubature formulae with a minimal number of knots

Summary In this paper an approach is outlined to the two-dimensional analogon of the Gaussian quadrature problem. The main results are necessary and sufficient conditions for the existence of cubature formulae which are exact for all polynomials of degree m and which have a minimal number of 1/2k(k+1) knots,k=[m/2]+1. Ifm is odd, similar results are due to I.P. Mysovskikh ([5, 6]) which will be derived in a new way as a special case of the general characterization given here. Furthermore, it will be shown how this characterization can be used to construct minimal formulae of even degree.     

A method of solving nonlinear variational problems by nonlinear transformation of the objective functional. Part I

Summary A general globally convergent iterative method for solving nonlinear variational problems is introduced. The method is applied to a temperature control problem and to the minimal surface problem. Several aspects of finite element implementation of the method are discussed.     

On the order of iterated defect correction

Summary In a recent article [2] Frank and Überhuber define and motivate the method of iterated defect correction for Runge-Kutta methods. They prove a theorem on the order of that method using the theory of asymptotic expansions.In this paper we give similar results using the theory of Butcher series (see [4]). Our proofs are purely algebraic. We don't restrict our considerations to Runge-Kutta methods, but we admit arbitrary linear one-step methods. At the same time we consider more general defect functions as in [2].     

On the design of high order exponentially fitted formulae for the numerical integration of stiff systems

Summary Single step exponentially fitted integration formulae of orders 4 and 6 are derived. The final approximation to the required solution is obtained via a linear combination of asymptotically less accurate solutions obtained using conventional implicit integration formulae. This linear combination is performed in such a way that the final integration formula, as well as having an increased order of accuracy, integrates a certain pre-determined ordinary differential equation exactly. The algorithms developed are illustrated by means of some numerical examples.     

On efficient implementations of Kogbetliantz's algorithm for computing the singular value decomposition

Summary In this paper we compare several implementations of Kogbetliantz's algorithm for computing the SVD on sequential as well as on parallel machines. Comparisons are based on timings and on operation counts. The numerical accuracy of the different methods is also analyzed.     

Zur effektiven Lösung von booleschen,quadratischen Optimierungsproblemen

Zusammenfassung Es wird die zum booleschen, quadratischen Optimierungsproblem duale Aufgabe untersucht. Ihr optimaler Zielfunktionswert liefert wesentlich schärfere Schranken für den branch and bound Prozeß als die bisher verwendeten.

---

On the efficient treatment of the boolean quadratic programming problem

Summary The dual of the boolean quadratic programming problem is considered. The optimal value of their objective function gives bounds for the branch and bound process, which seem to be better then known from other autors.

     

A generalized conjugate gradient algorithm for minimization

Summary A generalized conjugate gradient algorithm which is invariant to a nonlinear scaling of a strictly convex quadratic function is described, which terminates after at mostn steps when applied to scaled quadratic functionsf: R _n R₁ of the formf(x)=h(F(x)) withF(x) strictly convex quadratic andhC ¹ (R₁) an arbitrary strictly monotone functionh. The algorithm does not suppose the knowledge ofh orF but only off(x) and its gradientg(x).     

A stable interior penalty method for convex extremal problems

Summary For a nonlinearly constrained convex extremal problem a general interior penalty method is given, that is Hadamard-stable and needs no compactness conditions for convergence. The rate of convergence of the values iso(t) fort+0.     

A multilevel algorithm for the biharmonic problem

Summary A finite element discretization of the mixed variable formulation of the biharmonic problem is considered. A multilevel algorithm for the numerical solution of the discrete equations is described. Convergence is proved under the assumption ofH ³-regularity.     

Comparison of splittings used with the conjugate gradient algorithm

We consider several possible criteria for comparing splittings used with the conjugate gradient algorithm. We present sharp upper bounds for the error at each step of the algorithm and compare several widely used splittings with respect to their effect on these sharp upper bounds. We then consider a more stringent comparison test, and present necessary and sufficient conditions for the error at each step with one splitting to be smaller than that with another, for all pairs of corresponding initial guesses.     

An algorithm for segment approximation

Summary An algorithm for computing a set of knots which is optimal for the segment approximation problem is developed. The method yields a sequence of real numbers which converges to the minimal deviation and a corresponding sequence of knot sets. This sequence splits into at most two subsequences which converge to leveled sets of knots. Such knot sets are optimal. Numerical results concerning piecewise polynomial approximation are given.     

A parallel algorithm for the eigenvalues and eigenvectors of a general complex matrix

Summary A new parallel Jacobi-like algorithm is developed for computing the eigenvalues of a general complex matrix. Most parallel methods for this problem typically display only linear convergence, Sequential norm-reducing algorithms also exist and they display quadratic convergence in most cases. The new algorithm is a parallel form of the norm-reducing algorithm due to Eberlein. It is proven that the asymptotic convergence rate of this algorithm is quadratic. Numerical experiments are presented which demonstrate the quadratic convergence of the algorithm and certain situations where the convergence is slow are also identified. The algorithm promises to be very competitive on a variety of parallel architectures. In particular, the algorithm can be implemented usingn ²/4 processors, takingO(n log² n) time for random matrices.This research was supported by the Office of Naval Research under Contract N00014-86-k-0610 and by the U.S. Army Research Office under Contract DAAL 03-86-K-0112. A portion of this research was carried out while the author was visiting RIACS, Nasa Ames Research Center     

On best cubature formulas and spline interpolation

Summary A general cubature formula with an arbitrary preassigned weight function is derived using monosplines and integration by parts. The problem of determining the best cubature is formulated in terms of monosplines of least deviation and a solution to the problem is given by Theorem 3 below. This theorem may also be viewed as an optimal property of a new kind of two-dimensional spline interpolation.This work was done while the author was working at CERN, Geneva, Switzerland     

On the numerical treatment of a small divisor problem

Summary Quasiperiodic solutions of perturbed integrable Hamiltonian equations such as weakly coupled harmonic oscillators can be found by constructing an appropriate coordinate transformation which leads to a small divisor problem. However the numerical difficulties are not merely caused by the small divisors but rather by the appearence of ghost solutions, which appear in any reasonable discretization of the problem. Our numerical treatment, based on a Newton-type iteration, guarantees an approximation of the relevant solution of the nonlinear problem. Numerical solutions are found up to a critical value of the coupling constant, which is much larger than the coupling constants allowed by the existence theory available so far.     

A note on a one-sided Jacobi algorithm

Summary We propose a one-sided or implicit variant of the Jacobi diagonalization algorithm for positive definite matrices. The variant is based on a previous Cholesky decomposition and currently uses essentially one square array which, on output, contains the matrix of eigenvectors thus reaching the storage economy of the symmetric QL algorithm. The current array is accessed only columnwise which makes the algorithm attractive for various parallelized and/or vectorized implementations. Even on a serial computer our algorithm shows improved efficiency, in particular if the Cholesky step is made with diagonal pivoting. On matrices of ordern=25–200 our algorithm is about 2–3.5 times slower than QL thus being almost on the halfway between the standard Jacobi and QL algorithms. The previous Cholesky decomposition can be performed with higher precision without extra time or storage costs thus offering considerable gains in accuracy with highly conditioned input matrices.This work was partly done during the first author's visit to the Department of Mathematics, The University of Tennessee-Knoxville, while participating in the Special Year on Numerical Linear Algebra sponsored by the UTK Departments of Computer Science and Mathematics, and the ORNL Mathematical Sciences Section, Engineering Physics and Mathematics Division as well as during a second author visit to the Fernuniversität Hagen. Both authors gratefully acknowledge the support of the respective institutions     

Convergence analysis of the general Gauss-Newton algorithm

Summary The convergence of the Gauss-Newton algorithm for solving discrete nonlinear approximation problems is analyzed for general norms and families of functions. Aquantitative global convergence theorem and several theorems on the rate of local convergence are derived. A general stepsize control procedure and two regularization principles are incorporated. Examples indicate the limits of the convergence theorems.     

A modified bisection algorithm for the determination of the eigenvalues of a symmetric tridiagonal matrix

Summary A modification to the well known bisection algorithm [1] when used to determine the eigenvalues of a real symmetric matrix is presented. In the new strategy the terms in the Sturm sequence are computed only as long as relevant information on the required eigenvalues is obtained. The resulting algorithm usingincomplete Sturm sequences can be shown to minimise the computational work required especially when only a few eigenvalues are required.The technique is also applicable to other computational methods which use the bisection process.     

On the convergence and divergence of Bairstow's method

Summary Necessary and sufficient condition of algebraic character is given for the invertibility of the Jacobian matrix in Bairstow's method. This leads to a sufficient condition for local quadratic convergence. Results also yield the rank of the Jacobian, when it is singular. In the second part of the paper we give several examples for two step cyclization and more complicated kind of divergence. Some of the polynomials in divergence examples are proved to be irreducible over the field of rational numbers.     

A new method for the solution ofAx=b

Summary An alternative to Gauss elimination is developed to solveAx=b. The method enables one to exploit zeros in the lower triangle ofA, so that the number of multiplications needed to perform the algorithm can be substantially reduced.