On the cost of controlling unstable systems: The case of boundary controls (Q1385413)

The control system is \[ {\partial y \over \partial t} = - \Delta^2 y - k\Delta y \quad (x \in \Omega) \tag{1} \] in a domain \(\Omega \subset\mathbb{R}^n\) with boundary \(\Gamma\). The boundary conditions are \[ y(t, x) = \begin{cases} u(t, x) & (x \in \Gamma_0 \subseteq \Gamma)\\ 0 & (x \in \Gamma \setminus \Gamma_0)\end{cases}, \qquad {\partial y \over \partial n} = 0 \quad (x \in \Gamma). \tag{2} \] and the problem is that of approximate controllability in time \(T\), that is, driving from the zero initial state to a neighborhood of a target \(y^T\) by means of a control \(u(t, x) \in L^2((0, T) \times \Gamma_0)\). In general, one can do this by means of different controls, and the infimum of their norm is the \textit{cost of control.} As \(k \to +\infty\), the semigroup \(S_k(t)\) generated by \(- \Delta^2 - k\Delta\) and the homogeneous boundary conditions in \(L^2(\Omega)\) becomes ``less stable'' in the sense that \[ \lim_{t \to +\infty} {1 \over t} \log \| S_k(t)\| \to + \infty \] and the authors conjecture that the cost of control tends to zero as \(k \to +\infty\). This paper deals with results of this type for certain Galerkin approximations of (1)--(2).