Zvonimir Bujanović

On the Failure of Rank-Revealing QR Factorization Software -- A Case Study

This article reports an unexpected and rather erratic behavior of the LAPACK software implementation of the QR factorization with Businger-Golub column pivoting. It is shown that, due to finite precision arithmetic, the software implementation of the factorization can catastrophically fail to produce a properly structured triangular factor, thus leading to a potentially severe underestimate of a matrixs numerical rank. The 30-year old problem, dating back to LINPACK, has (undetectedly) badly affected many computational routines and software packages, as well as the study of rank-revealing QR factorizations. We combine computer experiments and numerical analysis to isolate, analyze, and fix the problem. Our modification of the current LAPACK xGEQP3 routine is already included in the LAPACK 3.1.0 release. The modified routine is numerically more robust and with a negligible overhead. We also provide a new, equally efficient, and provably numerically safe partial-column norm-updating strategy.

A new framework for implicit restarting of the Krylov–Schur algorithm

Summary This paper introduces a new framework for implicit restarting of the Krylov–Schur algorithm. It is shown that restarting with arbitrary polynomial filter is possible by reassigning some of the eigenvalues of the Rayleigh quotient through a rank-one correction, implemented using only the elementary transformations (translation and similarity) of the Krylov decomposition. This framework includes the implicitly restarted Arnoldi (IRA) algorithm and the Krylov–Schur algorithm with implicit harmonic restart as special cases. Further, it reveals that the IRA algorithm can be turned into an eigenvalue assignment method. Copyright

A block algorithm for computing antitriangular factorizations of symmetric matrices

Any symmetric matrix can be reduced to antitriangular form in finitely many steps by orthogonal similarity transformations. This form reveals the inertia of the matrix and has found applications in, e.g., model predictive control and constraint preconditioning. Originally proposed by Mastronardi and Van Dooren, the existing algorithm for performing the reduction to antitriangular form is primarily based on Householder reflectors and Givens rotations. The poor memory access pattern of these operations implies that the performance of the algorithm is bound by the memory bandwidth. In this work, we develop a block algorithm that performs all operations almost entirely in terms of level 3 BLAS operations, which feature a more favorable memory access pattern and lead to better performance. These performance gains are confirmed by numerical experiments that cover a wide range of different inertia.

Efficient generalized Hessenberg form and applications

This article proposes an efficient algorithm for reducing matrices to generalized Hessenberg form by unitary similarity, and recommends using it as a preprocessor in a variety of applications. To illustrate its usefulness, two cases from control theory are analyzed in detail: a solution procedure for a sequence of shifted linear systems with multiple right hand sides (e.g. evaluating the transfer function of a MIMO LTI dynamical system at many points) and computation of the staircase form. The proposed algorithm for the generalized Hessenberg reduction uses two levels of aggregation of Householder reflectors, thus allowing efficient BLAS 3-based computation. Another level of aggregation is introduced when solving many shifted systems by processing the shifts in batches. Numerical experiments confirm that the proposed methods have superior efficiency.

RADI: a Low-Rank ADI-Type Algorithm for Large Scale Algebraic Riccati Equations

This paper introduces a new algorithm for solving large-scale continuous-time algebraic Riccati equations (CARE). The advantage of the new algorithm is in its immediate and efficient low-rank formulation, which is a generalization of the Cholesky-factored variant of the Lyapunov ADI method. We discuss important implementation aspects of the algorithm, such as reducing the use of complex arithmetic and shift selection strategies. We show that there is a very tight relation between the new algorithm and three other algorithms for CARE previously known in the literature—all of these seemingly different methods in fact produce exactly the same iterates when used with the same parameters: they are algorithmically different descriptions of the same approximation sequence to the Riccati solution.