LQ-optimal control of infinite-dimensional systems by spectral factorization (Q1194943)

The standard LQ-optimal control problem \[ \dot x(t)=Ax(t)+Bu(t),\;x(0)=x_ 0\subset D(A),\;y(t)=Cx(t), \] \[ J(x_ 0,u)=\int^ \infty_ 0 \left(\langle Cx(t),Cx(t)\rangle+\langle u(t),u(t)\rangle\right)dt \] is considered, where the state space \(X\) is a real separable Hilbert space with inner product \(\langle\cdot,\cdot\rangle\), \(u(t)\in\mathbb{R}^ m\), \(y(t)\in\mathbb{R}^ p\), \(A: D(A)\subset X\to X\) is the infinitesimal generator of a \(C_ 0\)-semigroup \((e^{At})_{t\geq 0}\) of bounded linear operators in \(X\), \(B\) and \(C\) are bounded linear operators. An extension of a previous result of the authors [in: Signal Processing, Scattering and Operator Theory, and Numerical Methods, Prog. Syst. Control Theory 5, 323-329 (1990; Zbl 0723.93034)] is reported to show that if the pair \((A,B)\) is exponentially stabilizable and the pair \((C,A)\) is exponentially detectable then the reachable restriction \(K_ 0\pi\) (where \(\pi\) is the orthogonal projection on the reachable subspace) of the LQ-optimal state feedback operator \(K_ 0\) can be identified by solving a spectral factorization problem delivering a bistable spectral factor with entries in the distributed proper-stable transfer function algebra \(\widehat{\mathcal A}_ -\) and by determining the constant solution of an operator diophantine equation over \(\widehat{\mathcal A}_ -\). The properties of \(K_ 0\) are discussed and it is shown that \(K_ 0\pi\) and \(K_ 0\) are qualitatively similar except for the non- optimality of certain unreachable states; if the unreachable components are unobservable then \(K_ 0\pi=K_ 0\) and the closed-loop system is globally optimal. This framework is used to develop an approximation scheme by employing an indirect model reduction approach, i.e. a model reduction which is performed on the exact (infinite-dimensional) LQ-optimal feedback operator to obtain a finite-dimensional approximation. This scheme is applied to solve the LQ-optimal control problem for a simple model of heat diffusion in a finite rod with ``collocated'' sensor and actuator. It is shown that in this case the LQ-optimal feedback operator \(K_ 0\) and its reachable restriction coincide. The exponentially fast converging procedure is based on an approximate spectral factorization by symmetric extraction and on simple residue calculus.