A perturbation method for optimizing matrix stability (Q616397)

A perturbation method is proposed for optimizing the stability of a matrix by minimizing its spectral abscissa. This approach is based on the perturbation theory for a matrix with simple eigenvalues. The active eigenvalue (i.e. the eigenvalue with the greatest positive real part) is shifted to the left (in the complex plane) by a small amount. The proposed procedure is repeated until the minimum of the spectral abscissa is reached. The affine matrix problem is considered: given \(m+1\) real \(n\times n\) matrices \(A_0,A_1,\dots,A_m\), find a real vector \({\mathbf t}=(t_1,t_2,\dots,t_m)\) such that the affine matrix family \(A({\mathbf t})=A_0+\sum_{i=1}^mt_iA_i\) is stabilized with a minimum spectral abscissa. A perturbation algorithm is provided for this problem, which uses the solution of a linear programming problem. A similar algorithm is derived for the output feedback control problem. This algorithm is applied to a model of a Boeing 747 aircraft jet at a flutter condition and to a nuclear powered turbo-generator. A necessary optimality condition is given for matrices with a unique active eigenvalue, as well as a sufficient condition for the case of several active eigenvalues. These conditions are used in the proposed algorithms.