zbMATH — the first resource for mathematics

Une extension de la programmation quadratique successive. (French) Zbl 0559.90081
Analysis and optimization of systems, Proc. 6th int. Conf., Nice 1984, Part 2, Lect. Notes Control Inf. Sci. 63, 16-31 (1984).
[For the entire collection see Zbl 0538.00036.]
Nonlinear programming codes based on successive quadatic programming enjoy a large popularity. They may encounter some difficulties, however, if the quadratic subproblem is non-well posed at some iterations. We give a mean to compute a descent direction of a penalized functional (differentiable or not), using a linear quadratic approximation of the criterion and constraints. At each iteration, the subproblem to be solved is well-posed, even if the linearized constraints are not consistent. If the penalization is quadratic, we get the formulae of M. C. Bartholomew-Biggs [J. Inst. Math. Appl. 21, 67-81 (1978; Zbl 0373.90056)]. In the case of the \(L^{\infty}\) (resp. \(L^ 1)\) penalization, we extend the result of B. N. Pshenichnyj and Yu. M. Danilin [”Numerical methods in extremal problems” (1978; Zbl 0384.65028)] (resp. S. P. Han [J. Optimization Theory Appl. 22, 297- 309 (1977; Zbl 0336.90046)]. We also propose a new exact penalty function based on the \(L^ 2\) norm.

90C20 Quadratic programming
65K05 Numerical mathematical programming methods
90C30 Nonlinear programming
49M30 Other numerical methods in calculus of variations (MSC2010)