Recovery type superconvergence and a posteriori error estimates for control problems governed by Stokes equations (Q2475341)

The authors deal with a recovery type superconvergence analysis and a posteriori error estimates for the finite element approximation of distributed optimal control problems governed by Stokes equations. Here the following optimal control problem: Find \((\mathbf{y},r,\mathbf{u}) \in \mathbf{Y} \times Q \times \mathbf{U}\) such that \[ \min_{\mathbf{u}\in K \subset U} \biggl\{g(\mathbf{y})+ \frac{\alpha}{2}| | \mathbf{u}| | ^2_{O, \Omega_U}\biggr\}, \quad -\triangle \mathbf{y} + \nabla r = \mathbf{f} + B\mathbf{u} \text{ in } \Omega \operatorname{div} \mathbf{y} = 0 \text{ in } \Omega, \quad \mathbf{y} = 0 \text{ on } \partial \Omega \tag{1} \] (where \(\mathbf{y}=(y_1, y_2)\) is the velocity of fluids, \(r\) is the pressure, \(\mathbf{f} \in \mathbf{H} = (L^2(\Omega))^2\), \(\alpha\) is a positive constant, \(\mathbf{Y}=(H^1_0(\Omega))^2, \mathbf{U}=(L^2(\Omega_U))^2, Q=L^2_0(\Omega) = \{ q \in L^2(\Omega), \int_\Omega q\,d\Omega = 0\}\)), is investigated. For the finite element approximation of (1) a weak formula of the following form \[ \begin{aligned} &\min\{g(y)+ \frac{\alpha}{2}| | \mathbf{u}| | ^2_{0, \Omega_U}\},\\ &a(\mathbf{y}(\mathbf{u}),\mathbf{w})-b(\mathbf{w}, r(\mathbf{u}))= (f+B\mathbf{u}, \mathbf{w}) \quad \text{for any } \mathbf{w}\in \mathbf{Y},\\ &b(y(\mathbf{u}), \Phi )=0 \quad \text{for any } \Phi \in Q, \end{aligned}\tag{2} \] is introduced \((a(\mathbf{y},\mathbf{w})=\int_\Omega \nabla \mathbf{y} . \nabla \mathbf{w} \,d\Omega\) for any \(\mathbf{y},\mathbf{w} \in \mathbf{Y}\), \(b(\mathbf{v},r)= \int_\Omega r \operatorname{div} \mathbf{v} \partial \Omega\) for any \((\mathbf{v},r) \in \mathbf{Y} \times Q, (\mathbf{f}+B\mathbf{u},\mathbf{w})= \int_\Omega(\mathbf{f}+B\mathbf{u}) . \mathbf{w} \,\partial \Omega\) for any \(\mathbf{u},\mathbf{w}, \in \mathbf{H} \times \mathbf{Y})\). For the state and co-state equations only conforming elements are considered. The polynomial space in the construction of \(\mathbf{Y}^h\) contains \(P_1\) for triangular elements and \(Q_1\) for quadrilateral element, and \(Q^h\) contain \(P_0\). The two finite spaces \(\mathbf{Y}^h\) and \(Q^h\) are assumed to satisfy the approximation properties. The superconvergence analysis and a posteriori error estimates require a certain regularity for the Stokes problem and stable interpolators from \(\mathbf{U}\) to \(\mathbf{U}^h\). In this paper, partitions \(\Im^{H_i}, i=1,2\) are constructed such that they are quasi-uniform (they are regular and satisfy the inverse assumption). The authors' recovery technique (the postprocessing technique of least-squares surface fitting) is to project the finite element solution to another finite element space of higher order with a different mesh. Here after recovery the error is improved to be \(O(H^2_1 + H^2_2 + h^2 H^{-1}_1+h^2 H^{-1}_2+h^2+h^{3/2}_U)\), which is better than the original error \(O (h+h_U\)), if \(H_1\) and \(H_2\) are chosen suitably \((H_i=h^{\alpha _{i}}, i= 1,2\), with \([\alpha_1, \alpha_2] \in (0,1))\). The parameters \(\alpha_i\) play an important role in achieving a superconvergence for the finite element approximation \((\mathbf{y}_h, r_h, \mathbf{p}_h, s_h)\). The authors define the recovery process (for the suitable \(H_i\)) which leads to the global superconvergence result: \[ \begin{multlined}\|\Re_h \mathbf{u}_h-\mathbf{u}\|_{O, \Omega_{U}} + \|\nabla \mathbf{y} - \wp_{H_{1}} \mathbf{y}_h \|_{O, \Omega} + \|\nabla \mathbf{p} - \wp_{H_1} \mathbf{p}_h \|_{O, \Omega}\\ + \|r - \wp_{H_2} r_h \|_{O, \Omega} + \|s - \wp_{H_2} s_h \|_{O, \Omega} \leq C(h^{4/3} + h^{3/2}_U),\end{multlined} \] which is better than a standard error estimate: \[ \|\mathbf{u}_h-\mathbf{u}\|_{O, \Omega_{U}} + \|\nabla (\mathbf{y} - \mathbf{y}_h) \|_{O, \Omega} + \|(\nabla \mathbf{p} - \mathbf{p}_h) \|_{O, \Omega} + \|r - r_h \|_{O, \Omega} + \|s - s_h \|_{O, \Omega} \leq C(h + h_U). \] (\(\Re_h\) is recover operator on triangly meshes, and \(\wp_{H_1}\), \(\wp_{H_2}\) are the \(L^2\) projectors from \(L^2 (\Omega)\) onto the finite element spaces \(\mathbf{Y}^{H_{1}}\) and \(Q^{H_{2}}\), respectively). Main result: A global superconvergence result for the control \(u\) is derived (by applying patch recovery operator and the superconvergence analysis technique). Recovery and superconvergence for state and co-state are also presented (by using a least-squares surface fitting method). Recovery type a posteriori error estimates are derived (the precise proofs of all Theorems and Lemmas are given). Finally a numerical example to demonstrate the superconvergence and a posteriori error estimates theory is proposed.