Distributed control problems for the Burgers equation (Q5931036): Difference between revisions

The system is described by the one-dimensional forced Burgers equation \[ \begin{aligned} y_t - \nu y_{xx} + y y_x &= f + u, \quad 0 \leq t \leq T, \;0 \leq x \leq 1 , \\ y(t, 0) = y(t, 1) &= 0, \quad 0 \leq t \leq T, \\ y(0, x) &= \phi(x), \quad 0 \leq x \leq 1, \end{aligned} \] where \(f \in L^2((0, T) \times (0, 1)),\) \(\phi \in L^2(0, 1),\) and \(u \in L^2((0, T) \times (0, 1))\) is the control variable. The control problem consists of minimizing \[ J(y, u) = {1 \over 2}\|{\mathcal C}y - z\|^2_{\mathcal H} + {\sigma \over 2}\|u\|^2_{L^2((0, T) \times (0, 1))} \] where the (linear, bounded) operator \(\mathcal C\) maps the state space into the real Hilbert space \(\mathcal H.\) The problem is solved using the augmented Lagrangian-SQP method of Ito and Kunisch, where the equation becomes an equality constraint realized by a Lagrangian term together with a penalty functional. The paper contains numerical experiments with the resulting algorithm.