No label defined (Q4544443)

This monograph is on the modeling of certain optimal control problems for partial differential equations with applications to combustion, field modeling, superconductivity and fluid dynamics. The model is \[ \text{minimize} \;J(x) \qquad \text{subject to} \;e(x) = 0 \] where \(X, Y\) are real Hilbert spaces and \(J : X \to R\) and \(e : X \to Y\) are sufficiently smooth functions. The augmented Lagrangian \(L_c : X \times Y \to R\) of this system is \[ L_c(x, \lambda) = J(x) + \langle e(x), \lambda \rangle_Y + {c \over 2}\|e(x)\|^2_Y \] and the augmented Lagrangian-SQP method goes after the solution of the system \(L'_c(x^*, \lambda^*) = 0,\) \(e(x^*) = 0\) by means of an iteration process involving at each step the solution of a linear-quadratic minimization problem. Quadratic convergence is insured by second order optimality conditions. Implementation involves projection of the dynamical system onto subspaces generated by basis elements that contain characteristics of the expected solution rather than straight finite element discretization. There are results on numerical implementation (e.g., mesh independence) and various actual computations.