Inexactly constrained discrete adjoint approach for steepest descent-based optimization algorithms (Q1652805)

Convergence properties of the inexactly constrained gradient-based discrete adjoint steepest descent optimization algorithm are studied. A relationship was derived between the error in the solution to the constraint equations and the resulting error in the gradient, and it is established analytically that it is possible to maintain the Q-convergence rate of an exact gradient-based algorithm by adapting the constraint equation error based on the norm of the gradient and the convergence rate.