A geometric characterization of the reachable and the controllable subspaces of descriptor systems (Q1085119)

This paper gives algorithms to compute the reachable subspace and a new proof of the feedback invariance of it in a descriptor system. Consider the linear, time-invariant descriptor system \(D\Sigma\) :\(E\dot x(t)=Ax(t)+Bu(t)\), where \(E,A:C^ n\to C^ n\) and \(B:C^ m\to C^ n\). Let \({\mathfrak R}(B)\) be the range space of B. We define a (A,E,\({\mathfrak R}(B))\) invariant subspace of \({\mathfrak K}\) as \({\mathcal F}_{A,E,B}({\mathfrak K})=\{{\mathfrak S\subset K}:A{\mathfrak S}\subset E{\mathfrak S}+R(B)\}\) and a restricted (E,A,R(B)) invariant subspace as \({\mathcal F}_{E,A,B}({\mathfrak K})=\{{\mathfrak S}\subset C^ n:{\mathfrak S}=K\cap E^{-1}(A{\mathfrak S}+{\mathfrak R}(B)\}\). \({\mathcal F}_{A,E,B}(K)\) contains a unique supremal element \({\mathfrak L}^*(K)\). \({\mathcal F}_{E,A,B}({\mathfrak K})\) contains a unique infimum \({\mathfrak S}_*({\mathfrak K})\). Algorithms to compute \({\mathfrak L}^*({\mathfrak K})\) and \({\mathfrak S}_*({\mathfrak K})\) are derived. In case \({\mathfrak K}=C^ n\), we shall simply write \({\mathfrak L}^*,{\mathfrak S}_*\) rather than \({\mathfrak L}^*({\mathfrak K}),{\mathfrak S}_*({\mathfrak K})\). Let \({\mathfrak R}\) and \({\mathfrak C}\) be the reachable subspace and the controllable subspace respectively. Main results of this paper are given by the following theorems: Theorem 3.1. \({\mathfrak R}={\mathfrak L}^*\cap {\mathfrak S}_*\), \({\mathfrak R}={\mathfrak L}^*({\mathfrak S}_*).\) Theorem 3.2. \({\mathfrak C}={\mathfrak S}_*\cap ({\mathfrak L}^*+{\mathfrak N}(E))\), \({\mathfrak C}={\mathfrak L}^*({\mathfrak S}_*)+{\mathfrak N}(E)\), where \({\mathfrak N}(E)\) denotes the null space of E. Let a feedback \(u(t)=Fx(t)+v(t)\) be applied to \(D\Sigma\) to yield the closed-loop system CLD\(\Sigma\) :\(E\dot x(t)=(A+BF)x(t)+Bv(t)\). Let the subscript/superscript CL denote closed-loop quantities. Then, we have \({\mathfrak S}_*^{CL}={\mathfrak S}_*\), \({\mathfrak L}^*_{CL}={\mathfrak L}_*\). Accordingly, in virtue of Theorem 3.1 and 3.2, the following theorem follows: Theorem 4.1. \({\mathfrak R}_{CL}={\mathfrak R}\).