简介:提出一个带线性搜索的非单调信赖域算法.算法将非单调wolfe线搜索与非单调信赖域方法相结合,使算法不需要重新求解子问题.在适当条件下,分析了算法的全局收敛性,并通过数值实验说明了算法的可行性.
简介:AinteriorpointscalingprojectedreducedHessianmethodwithcombinationofnonmonotonicbacktrackingtechniqueandtrustregionstrategyfornonlinearequalityconstrainedoptimizationwithnonegativeconstraintonvariablesisproposed.Inordertodealwithlargeproblems,apairoftrustregionsubproblemsinhorizontalandverticalsubspacesisusedtoreplacethegeneralfulltrustregionsubproblem.Thehorizontaltrustregionsubprobleminthealgorithmisonlyageneraltrustregionsubproblemwhiletheverticaltrustregionsubproblemisdefinedbyaparametersizeoftheverticaldirectionsubjectonlytoanellipsoidalconstraint.Bothtrustregionstrategyandlinesearchtechniqueateachiterationswitchtoobtainingabacktrackingstepgeneratedbythetwotrustregionsubproblems.Byadoptingthel1penaltyfunctionasthemeritfunction,theglobalconvergenceandfastlocalconvergencerateoftheproposedalgorithmareestablishedundersomereasonableconditions.AnonmonotoniccriterionandthesecondordercorrectionstepareusedtoovercomeMaratoseffectandspeeduptheconvergenceprogressinsomeill-conditionedcases.
简介:借鉴无约束优化问题的BFGS信赖域算法,建立了非线性一般约束优化问题的BFGS信赖域算法,并证明了算法的全局收敛性.数值实验表明,算法是有效的.