SIAM Journal on Control and Optimization · 2003 · 56 citations · 6 references
Natural SciencesDecision SciencesComputational GeometryDesign OptimizationOptimizationConstrained OptimizationConvex AnalysisConvex OptimizationNonlinear ProgrammingOptimality TheoryOptimization TheoryConditionsConstraintsExistence Of MinimizersOptimality ConditionsOptimization Under ConstraintsRegularity CriteriaRegularity Theory
In this paper we consider an optimization problem with equality constraints given in operator form as F(x)=0, where $F : X \to Y$ is an operator between Banach spaces. The paper addresses the case when the equality constraints are not regular in the sense that the Fréchet derivative F,'(x* ) is not onto. In the first part of the paper, we pursue an approach based on the construction of p-regularity. For p-regular constrained optimization problems, we formulate necessary conditions for optimality and derive sufficient conditions for optimality. In the second part of the paper, we consider a generalization of the concept of p-regularity and derive generalized necessary conditions for optimality for an optimization problem that is neither regular nor p-regular. For this problem, we show that the tangent cone to a level surface of F can consist of rays (rather than lines). This is in contrast to the regular and the p-regular cases, for which the tangent cone is always "two-sided." We state that if the gradient of the generalized p-regular problem is nonzero, it can belong to an open set, despite the fact that all constructions are usually closed. Both p-regular and generalized conditions for optimality reduce to classical conditions for regular cases, but they give new and nontrivial conditions for nonregular cases. The presented results can be considered as a part of the p-regularity theory. (A correction to the this article has been appended at the end of the pdf file.)
6