Model reduction techniques are often required in computationally tractable algorithms for the solution of simulation, optimization, or control problems for large scale dynamical systems. For linear systems, essentially all reduced models can be produced using projection onto subspaces determined by the approximation constraints of the problem. For example, rational interpolation, e.g., Rational Krylov methods, and its generalization to tangential interpolation require projection onto so-called generalized Krylov subspaces whose bases solve a particular family of Sylvester equations. In this paper, we derive a numerically reliable way to compute an orthogonal basis of these generalized Krylov subspaces. The residual error of the large linear systems of equations that are solved in order to produce the bases are controlled so as to yield a small backward error in the associated Sylvester equations and in the model reduction problem. The efficiency and effectiveness of the algorithm is demonstrated for single and multipoint tangential interpolation examples.
[1]
W. Kahan,et al.
The Rotation of Eigenvectors by a Perturbation. III
,
1970
.
[2]
Eric James Grimme,et al.
Krylov Projection Methods for Model Reduction
,
1997
.
[3]
Paul Van Dooren,et al.
A collection of benchmark examples for model reduction of linear time invariant dynamical systems.
,
2002
.
[4]
Antoine Vandendorpe,et al.
Model reduction of linear systems : an interpolation point of view/
,
2004
.
[5]
Paul Van Dooren,et al.
Model Reduction of MIMO Systems via Tangential Interpolation
,
2005,
SIAM J. Matrix Anal. Appl..
[6]
Åke Björck,et al.
Numerical methods for least square problems
,
1996
.
[7]
Robert Skelton,et al.
Model reductions using a projection formulation
,
1987,
26th IEEE Conference on Decision and Control.