A sequential quadratic programming method for potentially infeasible mathematical programs (Q2640450): Difference between revisions

The author presents a modified sequential quadratic programming (SQP) technique that is applicable to the nonlinear programming problem \[ (NLP):\;\min f(x)\text{ subject to } g(x)\in C,\quad x\in X, \] where f: \({\mathbb{R}}^ n\to {\mathbb{R}}\) and g: \({\mathbb{R}}^ n\to {\mathbb{R}}^ m\) are \(C^ 1\) and \(C\subseteq {\mathbb{R}}^ m\) and \(X\subseteq {\mathbb{R}}^ n\) are nonempty closed convex sets. The usual application of the SQP technique requires one to iteratively solve a direction-finding subproblem of the form: \[ \min \nabla f(x)^ Td+d^ THd\text{ subject to } g(x)+g'(x)d\in C\text{ and } x+d\in X. \] The procedure discussed in this paper is a modification of the above constraints in the form: \(g(x)+g'(x)d\in C+k{\mathbb{B}}^ n\), \(x+d\in X\), \(d\in \beta {\mathbb{B}}^ n\), where \({\mathbb{B}}^ n\) and \({\mathbb{B}}^ m\) are closed unit balls. The author claims that this modification makes it possible to overcome the difficulty of the subproblem being potentially infeasible. The global convergence properties of the method are also discussed. For related and recent work the readers' attention is directed to a paper by \textit{M. Sahba} [J. Optimization Theory Appl. 52, No.2, 291-309 (1987; Zbl 0585.90076)].