Heike Fassbender

Heike Fassbender, Philip Saltenberger

In this paper, we introduce a new family of equations for matrix pencils that may be utilized for the construction of strong linearizations for any square or rectangluar matrix polynomial. We provide a comprehensive characterization of the resulting vector spaces and show that almost every matrix pencil therein is a strong lineariza- tion regardless whether the matrix polynomial under consideration is regular or sin- gular. These novel "ansatz spaces" cover all block-Kronecker pencils as a subset and therefore contain all Fiedler pencils modulo permutations. The important case of square matrix polynomials is examined in greater depth. We prove that the intersection of any number of block-Kronecker ansatz spaces is never empty and construct large subspaces of blocksymmetric matrix pencils among which still almost every pencil is a strong linearization. Moreover, we show that the original ansatz spaces L1 and L2 may essentially be recovered from block-Kronecker ansatz spaces via pre- and postmultiplication, respectively, of certain constant matrices.

Navneet Pratap Singh, Kapil Ahuja, Heike Fassbender

Recently a new algorithm for model reduction of second order linear dynamical systems with proportional damping, the Adaptive Iterative Rational Global Arnoldi (AIRGA) algorithm, has been proposed. The main computational cost of the AIRGA algorithm is in solving a sequence of linear systems. These linear systems do change only slightly from one iteration step to the next. Here we focus on efficiently solving these systems by iterative methods and the choice of an appropriate preconditioner. We propose the use of relevant iterative algorithm and the Sparse Approximate Inverse (SPAI) preconditioner. A technique to cheaply update the SPAI preconditioner in each iteration step of the model order reduction process is given. Moreover, it is shown that under certain conditions the AIRGA algorithm is stable with respect to the error introduced by iterative methods. Our theory is illustrated by experiments. It is demonstrated that SPAI preconditioned Conjugate Gradient (CG) works well for model reduction of a one dimensional beam model with AIRGA algorithm. Moreover, the computation time of preconditioner with update is on an average 2/3 rd of the computation time of preconditioner without update. With average timings running into hours for very large systems, such savings are substantial.