Searcharxiv⌕ Search

arXiv subjects

Robbe Vermeiren

Publications and source records attributed to Robbe Vermeiren.

3 recordsLinked to original sources

Orthogonal Vector-Valued Rational Functions with Applications in Rational Approximation

We present two algorithms for constructing orthonormal bases of vector-valued rational functions with respect to a discrete inner product. \rev{Every pole is prescribed together with a tangential direction, which fixes the direction of the residue at that pole.} Building on the pencil formulation of the inverse eigenvalue problem for scalar rational functions by Van Buggenhout, Van Barel and Vandebril (2022), we show that, for functions with $k$ components, the recurrence relations are represented by a pencil of $k$-Hessenberg matrices, i.e., matrices with at most $k$ nonzero subdiagonals, \rev{and that the poles are the ratios of the entries on the $k$th subdiagonal. This extends the banded Hessenberg structure of orthogonal vector-valued polynomials to rational functions, and it replaces the semiseparable-plus-diagonal representation of Delvaux and Van Barel (2005), in which every pole belongs to a single component, by a pencil.} We derive an updating algorithm based on rotations and a Krylov-type algorithm \rev{that orthogonalizes a tangential rational Krylov subspace, and we compare their accuracy. Finally, we show that the lightning method and the AAA algorithm fit in our framework; for AAA, the barycentric weights follow from a singular value decomposition of a small block of the pencil.} The approximation of $\sqrt{z}$ on $[0,1]$ with exponentially clustered poles and sample points serves as a stress test for our methods.

math.NA↗

Global Lanczos with stochastic error correction for trace estimation of matrix functions

We propose a new method for computing the trace of a matrix function $f(A)$, where $A$ is a large, sparse, symmetric matrix. Our approach combines a deterministic global Lanczos approximation with a stochastic correction of its error. Starting the global Lanczos process with the identity matrix yields a direct approximation of the trace through a small tridiagonal matrix. The resulting basis matrices are polynomials in $A$, allowing the process to be implemented using sparse matrix-matrix multiplications when low-degree polynomials in $A$ have moderate fill-in. However, increased fill-in may limit the number of iterations that can be performed efficiently. To address this limitation, we derive an exact representation of the global Lanczos error as the trace of a matrix function and apply a stochastic estimator to the remainder. Even a moderately accurate estimate can substantially improve the accuracy, combining the deterministic method with the rapid initial convergence of stochastic trace estimators. We discuss error bounds and variance reduction, and present numerical experiments that compare the resulting method with existing ones.

math.NA↗

Krylov and core transformation algorithms for an inverse eigenvalue problem to compute recurrences of multiple orthogonal polynomials

In this paper, we develop algorithms for computing the recurrence coefficients corresponding to multiple orthogonal polynomials on the step-line. We reformulate the problem as an inverse eigenvalue problem, which can be solved using numerical linear algebra techniques. We consider two approaches: the first is based on the link with block Krylov subspaces and results in a biorthogonal Lanczos process with multiple starting vectors; the second consists of applying a sequence of Gaussian eliminations on a diagonal matrix to construct the banded Hessenberg matrix containing the recurrence coefficients. We analyze the accuracy and stability of the algorithms with numerical experiments on the ill-conditioned inverse eigenvalue problemshave related to Kravchuk and Hahn polynomials, as well as on other better conditioned examples.

math.NA↗