No label defined (Q4514934)

In this dissertation the author studies the discrete-time system \(x_{j+1}= Ax_j+ Bu_j\), \(y_j = Cx_j+ Du_j\) with the associated algebraic Riccati equation (DARE). \[ A^* PA- P+ C^*JC= (D^*JC+ B^*PA)^* (D^*JD+ B^*PB)^{-1} (D^*JC+ B^*PA), \] where \(J\) is the cost operator. All these symbols denote bounded linear operators on suitable Hilbert spaces. Different representations of the discrete linear system are introduced in the first chapter, i.e., as the difference equation such as given above, and as a quadruple of operators working on the initial state and the input signal, giving the state- and the output signal. Furthermore, in this chapter various notions that are important for the rest of this dissertation are introduced. These notions, such as state feedback and stability, are explained for both representations. For \(J\) a bounded and selfadjoint operator, the following cost functional is defined: \(J(x_0, u)= \sum^\infty_{j=0} \langle y_j, Jy_j\rangle_Y\). Note that the author uses the same symbol for the operator and the associated cost functional. The rest of the dissertation discusses various aspects of the discrete linear system with this cost functional. In chapter 2 the relation between critical control, spectral factorization, inner-outer factorization, and the DARE (in weak form) is investigated. Critical controls are those control sequences \(u\) for which the Fréchet derivative of \(J(x_0, u)\) is zero. All the results in this chapter hold under the assumption that the set of initial conditions such that, with zero control, \(y\in\ell^2(\mathbb{N})\) forms a dense set. From Chapter 3 on, it is assumed that this dense set equals the whole state space, and that the system is input-output stable. Under these assumptions the DARE is called the \(H^\infty\) DARE. In chapter 3 and 4 a more detailed study of the solutions to the DARE is made. In chapter 3 the emphasis is on solutions \(P\) to the DARE for which \(A^{*j} PA^j\) converges strongly to zero. It is shown that these solutions correspond to stable spectral factorizations. In chapter 4 this class is further restricted by the extra assumption that \(P\geq 0\). Furthermore, it is assumed that the cost operator is nonnegative. The main result of this chapter is that under these conditions the input-output map \({\mathcal D}\) admits for every \(P\) a factorization \({\mathcal D}={\mathcal D}_{\phi^P}{\mathcal D}_{\phi_P}\). Under some additional assumptions this provides a factorization of the transfer function on the unit circle. In the last chapter the author studies the question if the natural ordering in the solutions to the DARE can be encoded into invariant subspaces of some linear bounded operator.