A posteriori error estimates for mixed finite element solutions of convex optimal control problems (Q2475380)

A posteriori error estimates for the mixed finite element approximation of a convex optimal control governed by elliptic equations are derived. The authors construct the mixed finite element approximation for the following optimal control problem: \[ \min_{u\in K\subset L^2(\Omega_U)} [g_1(\mathbf{p})+g_2(y)+h(u)],\tag{1} \] \[ \operatorname{div} \mathbf{p}=f+Bu \quad \text{in }\Omega \qquad \mathbf{p}=-A \operatorname{grad}y \quad\text{in }\Omega,\qquad y=0 \quad\text{on } \partial \Omega,\tag{2} \] where the bounded open set \(\Omega \subset\mathbb R^2\), is a convex polygon or has the smooth boundary \(\partial \Omega, \Omega_U\) is a bounded open set in \(\mathbb R^2\) with the Lipschitz boundary \(\partial \Omega_U\), and \(K\) is a closed convex set in \(L^2(\Omega_U)\). Using a weak formula for the state equality (2), the mixed finite element approximation of (1), (2), is introduced as follows: Find \([\mathbf{p}_h, y_h, u_h] \in \mathbf{V}_h \times W_h \times U_h\) such that \[ \min_{u_h\in K_h \subset U_h} \{g_1(\mathbf{p}_h)+g_2(y_h)+h(u_h)\},\tag{3} \] \[ \begin{aligned} (A^{-1} \mathbf{p}_h, \mathbf{v}_h)-(y_h, \operatorname{div} \mathbf{v}_h) = 0, &\quad \text{for any } \mathbf{v}_h \in \mathbf{V}_h \subset \mathbf{V}= H(\operatorname{div}, \Omega), \\ \operatorname{div}\mathbf{p}_h, w_h)=(f+Bu_h, w_h), &\quad \text{for any } w_h \in W_h \subset W = L^2 (\Omega), \end{aligned}\tag{4} \] where \(\mathbf{V}_h \times W_h \subset \mathbf{V} \times W\) denotes the RT, BDM or BDFM space of index \(k\) associated with triangulation or rectangulation \(\Im_h\) of \(\Omega\), where \(k \geq 0, K_h\) is a non-empty closed convex set in \(U_h\). For fixed \(h\) the control problem (3), (4) has a unique solution \([\mathbf{p}_h, y_h, u_h] \in \mathbf{V}_h \times W_h \times U_h\) if and only if there is a co-state \((\mathbf{q}_h, z_h)\in \mathbf{V}_h \times W_h\) such that \((\mathbf{p}_h, y_h, \mathbf{q}_h, z_h, u_h)\) satisfies the following optimality conditions \[ \begin{aligned} (A^{-1} \mathbf{p}_h, \mathbf{v}_h)-(y_h, \operatorname{div}\mathbf{v}_h) = 0, &\quad \text{for any } \mathbf{v}_h \in \mathbf{V}_h,\\ (\operatorname{div}\mathbf{p}_h, w_h)=(f+Bu_h, w_h), &\quad \text{for any } w_h \in W_h,\\ (A^{-1} \mathbf{q}_h, w_h)-(z_h, \operatorname{div}\mathbf{v}_h) = - (g_{1}' (\mathbf{p}_h), \mathbf{v}_h), &\quad \text{for any } \mathbf{v}_h \in \mathbf{V}_h,\\ (\operatorname{div}\mathbf{q}_h, w_h)= (g'_{2} (y_h), w_h), &\quad \text{for any } w_h \in W_h, \\ (h'(u_h)+B^*z_h, \tilde{u}_h-u_h) \geq 0, &\quad \text{for any }\tilde{u}_h \in K_h. \end{aligned}\tag{5} \] Main results: A posteriori error estimates for some intermediate errors for the RT, the BDM and the BDFM mixed method, are derived. The authors' analysis relies on a decomposition of the flux functions in the spirit of a generalized Helmholtz decomposition. Further, a posteriori error estimates for convex optimal control problems in general cases where the convex set \(K\) may not be the whole control space \(U\) and the discretized constraint set \(K_H\) may not be the subset of \(K\), are also presented. Finally equivalent a posteriori error estimates for a class of optimal control problems which are most frequently met in applications are established. Such estimates can be used to construct reliable adaptive mixed finite elements for the control problems.