Maximum-norm a posteriori error estimates for an optimal control problem (Q2419577)

Let $\Omega$ be an open, bounded polytope in $\mathbb{R}^2$ or $\mathbb{R}^3$. The optimal control problem is that of minimizing a cost functional \[ J(y, u) = \frac12 \|y - y_\Omega \|^2_{L^2(\Omega)} + \frac{\lambda}{2}\|u\|^2_{L^2(\Omega)} \] among the functions $y$ satisfying \[ - \Delta y = f + u \quad (x \in \Omega) \qquad y = 0 \quad (x \in \partial \Omega) \tag{1} \] where $y_\Omega, f \in L^2(\Omega)$. The control $u$ satisfies the constraint $a \le u(x) \le b$ in $\Omega$. The authors set up a weak version of (1) and of the equation $-\Delta p = y - y_\Omega$ defining the adjoint vector $p,$ and obtain the optimal control from the variational inequality $\langle \bar p + \lambda \bar u. u - \bar u\rangle_{L^2(\Omega)} \ge 0$. By introducing a mesh $\mathcal{T}$ satisfying suitable conditions and approximating the functions using piecewise linear finite elements, a finite dimensional approximation of the problem is obtained. The subject of this paper is to provide a posteriori error bounds in the $L^\infty$ norm, where ``a posteriori'' means error estimators that depend only on the approximate solution and the data of the problem. The approximation is illustrated in two examples where $\Omega$ is a square or union of squares in $\mathbb{R}^2$. The initial mesh is constructed from sides and diagonals and refined adaptively in the same way.