On the computation of the optimal \(\mathbf H_\infty\) norms for two feedback control problems (Q1301304)

The authors consider the linear time-invariant output feedback \(H_\infty\) control problem with \(G\) being the plant to be controlled and \(K\) being the output feedback controller. The realization of \(G\) is taken to be of the form \[ \begin{bmatrix} {\l} {}& {\l} & {\l}\\ A & B_1 & B_2\\ C_1 & 0 & D_{12}\\ C_2 & D_{21} & 0\end{bmatrix}, \] where the matrices \(A\), \(B_1\), \(B_2\), \(C_1\), \(C_2\), \(D_{12}\) and \(D_{21}\) are real and satisfy the assumptions: (A1) \((A,B_1)\) is stabilizable and \((C_1,A)\) is detectable; (A2) \((A,B_2)\) is stabilizable and \((C_2,A)\) is detectable; (A3) \(D^T_{12}[C_1 D_{12}]=[0\;I]\). (A4) \(D_{21}[B^T_1 D^T_{21}]=[0\;I]\). The signal \(w\) contains all external inputs; the output \(z\) is an error signal; \(y\) is the measured variable; \(u\) is the control input; and the closed-loop transfer function from \(w\) to \(z\) is denoted by \(T_{zw}\). If one defines \[ H_{\infty}(r):=\left[\begin{matrix} A & \frac{1}{r^2} B_1B_1^T-B_2B_2^T\\ -C^T_1 C_1 & -A^T\end{matrix}\right],\;J_\infty(r):=\left[\begin{matrix} A^T & \frac{1}{r^2} C^T_1 C_1-C_2^T C_2\\ -B_1B_1^T & -A\end{matrix}\right], \] then it is known from a previous work, that there exists an admissible controller \(K\) such that \(\|T_{zw}\|_\infty<r\) if and only if the following three conditions hold: (C1) \(H_\infty(r)\in \text{dom(Ric)}\) and \(X(r):=\text{Ric}(H_\infty(r))\geq 0\); (C2) \(J_\infty(r)\in\text{dom(Ric)}\) and \(Y(r):=\text{Ric}(J_\infty(r))\geq 0\); (C3) \(\rho(X(r)Y(r))<r^2\). The authors study the problem of finding the infimum of \(r\) such that conditions (C1)--(C3) hold simultaneously, under assumptions (A1)--(A4). This problem is divided into three subproblems. Firstly, one determines \(r_x\), the infimum of \(r\) such that (C1) holds, one characterizes \(r_x\) by the singularity of the upper-half of the matrix stacked up by the basis vectors of the stable invariant subspace of \(H_{\infty}(r)\), and thus \(r_x\) is obtained using the secant method. The same technique is applied to find \(r_y\), the infimum of \(r\) such that (C2) holds. Next, one uses the radius function \(\rho(X(r) Y(r))\) in order to find \(r^*\geq \max\{r_x,r_y\}\), the infimum of \(r\) such that conditions (C1)--(C3) hold simultaneously.