Natural SciencesComputer ScienceMathematicsAlgorithmsOptimizationTheory Of ComputationConstrained OptimizationConvex OptimizationGradient-based OptimizationIterative SolversNonlinear ProgrammingNumerical AlgorithmsFinite-dimensional Hilbert SpacesInfinite-dimensional Hilbert Spaces
The gradient-projection algorithm (GPA) is a powerful method for solving constrainedminimization problems in finite (and even infinite) dimensional Hilbert spaces. Weconsider GPA with variable stepsizes and show that if GPA generates a bounded sequence,then under certain assumptions, every accumulation point of the sequence is a solution of theminimization problem. We also look into the issue where the sequence of stepsizes is allowedto be the limiting case (e.g., approaching to zero).
5