A software package for the solution of generalized algebraic Riccati equations

Abstract

The generalized eigenvalue problem provides a powerful framework for the solution of quite general forms of algebraic Riccati equations arising in both continuous-and discrete-time applications. This general form is derived from control and filtering problems for systems in generalized (or implicit or descriptor) state space form. A software package called RICPACK has been developed to solve such Riccati equations by means of deflating subspaces for certain associated Hamiltonian or symplectic generalized eigenvalue problems. Utilizing an embedding technique, the package also calculates a solution even in cases where all cost or covariance matrices are singular or ill-conditioned with respect to inversion. Cross-weighting or correlated noise is handled directly. Both system-theoretic balancing and Ward's balancing for the generalized eigenvalue problem are available to improve condition and accuracy. Condition estimates for the solution are also calculated. An iterative improvement calculation via Sylvester equations is available and can be used to generate new solutions for small changes in the model. An interactive driver with numerous convenient default options has also been written. A numerical example is shown.