New algorithms for polynomial \(J\)-spectral factorization (Q1293696): Difference between revisions

Polynomial \(J\)-spectral factorizations are factorizations of real \(q\times q\) matrix polynomials of a real variable \(\xi\) of the following type: \(Z(\xi)= (F(-\xi))^TJF(\xi)\), where \(J\) is a fixed real signature matrix (i.e., symmetric and unitary). The real matrix polynomial \(F(\xi)\) is often required to have additional properties, for example, stability. Such factorizations arise in many areas of control systems, notably in \(H_\infty\) control problems. It is known that a regular real matrix polynomial \(Z(\xi)\) admits a polynomial \(J\)-spectral factorization if and only if \(Z(\xi)\) is para-Hermitian, i.e., \((Z(-\xi))^T= Z(\xi)\), and the signature of the Hermitian matrix \(Z(i\xi)\) is constant for almost all real \(\xi\). In the reviewed paper several new algorithms for computing the factor \(F(\xi)\), often with additional properties, are developed. The first algorithm is based on the calculus of two-variable matrix polynomials and associated quadratic forms; see \textit{J. C. Willems} and \textit{H. L. Trentelman} [SIAM J. Control Optimization 36, No. 5, 1703-1749 (1998; Zbl 0912.93006)] for a development of this calculus. Briefly, one associates a real matrix polynomial of two variables \(\Phi(\xi,\eta)\) with a given para-Hermitian matrix polynomial \(Z(\xi)\) such that \(\Phi(-\xi,\xi)= Z(\xi)\). Then polynomial \(J\)-spectral factorizations of \(Z(\xi)\), if they exist, can be constructed using the coefficient matrices \(\Phi_{i,j}\) of the matrix polynomial \(\Phi(\xi,\eta)= \sum^\infty_{i,j= 0}\Phi_{i,j}\xi_i\eta_j\) (only finitely many of the \(\Phi_{i,j}\) are different from zero). Another algorithm is based on existence of a symmetric solution to an algebraic Riccati equation the coefficients of which are explicitly given in terms of the \(\Phi_{i,j}\). It turns out that a polynomial \(J\)-spectral factorization with a Hurwitz \(F(\xi)\) exists if and only if \(\Phi_{0,0}\) has the signature determined by \(J\) and the Riccati equation has a symmetric ``almost'' stabilizing solution. A third algorithm is based on a Pick matrix approach. Starting with a polynomial \(J\)-spectral factorization \[ Z(\xi)= (F(-\xi))^T\left[\begin{matrix} I & 0\\ 0 & -I\end{matrix}\right] F(\xi) \] with a possibly rectangular factor \(F(\xi)\), the algorithm produces another \(J\)-spectral factorization (if it exists) \[ Z(\xi)= (H(-\xi))^T\left[\begin{matrix} I & 0\\ 0 & -I\end{matrix}\right] H(\xi), \] where the \(q\times q\) matrix polynomial \(H(\xi)\) has the properties that \(MN^{-1}\) is proper (finite at infinity) and that the \(q\times q\) matrix polynomial \(\left[\begin{smallmatrix} M_-\\ H_+\end{smallmatrix}\right]\) is Hurwitz; here \(M_-\) and \(H_+\) are the parts of \(M\) and \(H\) corresponding to \(I\) and to \(-I\), respectively. Necessary and sufficient conditions for the existence of such an \(H\) are obtained by \textit{H. L. Trentelman} and \textit{J. C. Willems} [IEEE Trans. Autom. Control 44, No. 3, 521-536 (1999)]. Examples illustrating each of these algorithms are presented.