简介:由Ginchev在我们证明足够、必要的optimality条件的那一些的这篇文章获得了,Guerraggio,Luc[Appl。数学,51,5鈥?6(2006)]概括(严格地)那些由Guerraggio介绍了,Luc[J。Optim。理论Appl,109,615鈥?29(2001)]。当以前的纸显示出在那里给的条件是有效的为例子时但是从后者纸的失败,它不通常证明那它建议的条件更强壮。在现在的笔记,我们与缺乏的证明完成这比较。关键词C1,1功能-概括秒顺序方向性的衍生物-Dini衍生物-弱有效的minimizer-秒顺序先生(2000)题目分类49K10的孤立的minimizer-49J52-49J50-90C29-90C30由捷克的管理(MSM6198959214)的委员会支持了
简介:Thispaperpresentsatrustregiontwo-phasemodelalgorithmforsolvingtheequalityandboundconstratinednonlinearoptimizationproblem.Aconceptofsubstationarypointisgiven.Undersutableassumptions.theglobalconvergenceofthisalgorithmisprovedwithoutassumingthelinearindependenceofthegradientofactiveconstraints.Anumericalexampleisalsopresented.
简介:在这篇论文,变光滑QP免费的不能实行的方法为非线性的不平等被建议抑制优化问题。这个反复的方法基于被获得由的非线性的方程的答案多钳子;为KKT一阶的optimality条件的变光滑的Fisher-Burmeister功能。与另外的QP免费的方法作比较,这个方法不请求重复的严格的可行性。特别地,这个方法是工具能;没有假定严格的补充条件全球性会聚;累积的孤立的海角指。而且,活跃限制的坡度没被请求线性地独立。初步的数字结果显示那这变光滑QP免费的不能实行的方法是相当有希望的。
简介:Somepropertiesofconvexconesareobtainedandareusedtoderiveseveralequivalentconditionsaswellasanotherimportantpropertyfornearlycone-subconvexlikeset-valuedfunctions.Undertheassumptionsofnearlycone-subconvexlikeness,aLagrangianmultipliertheoremonBensonproperefficiencyispresented.Relatedresultsaregeneralized.
简介:Thealternatingdirectionmethodofmultipliers(ADMMforshort)isefficientforlinearlyconstrainedconvexoptimizationproblem.ThepracticalcomputationalcostofADMMdependsonthesub-problemsolvers.Theproximalpointalgorithmisacommonsub-problem-solver.However,theproximalparameterissensitiveintheproximalADMM.Inthispaper,weproposeahomotopy-basedproximallinearizedADMM,inwhichahomotopymethodisusedtosolvethesub-problemsateachiteration.Undersomesuitableconditions,theglobalconvergenceandtheconvergencerateofO(1/k)intheworstcaseoftheproposedmethodareproven.Somepreliminarynumericalresultsindicatethevalidityoftheproposedmethod.
简介:Basedonanewefficientidentificationtechniqueofactiveconstraintsintroducedinthispaper,anewsequentialsystemsoflinearequations(SSLE)algorithmgeneratingfeasibleiteratesisproposedforsolvingnonlinearoptimizationproblemswithinequalityconstraints.Inthispaper,weintroduceanewtechniqueforconstructingthesystemoflinearequations,whichrecurstoaperturbationforthegradientsoftheconstraintfunctions.Ateachiterationofthenewalgorithm,afeasibledescentdirectionisobtainedbysolvingonlyonesystemoflinearequationswithoutdoingconvexcombination.ToensuretheglobalconvergenceandavoidtheMaratoseffect,thealgorithmneedstosolvetwoadditionalreducedsystemsoflinearequationswiththesamecoefficientmatrixafterfiniteiterations.Theproposedalgorithmisprovedtobegloballyandsuperlinearlyconvergentundersomemildconditions.WhatdistinguishesthisalgorithmfromthepreviousfeasibleSSLEalgorithmsisthatanimprovingdirectionisobtainedeasilyandthecomputationcostofgeneratinganewiterateisreduced.Finally,apreliminaryimplementationhasbeentested.
简介:Thispaperstudiesthetwo-dimensionallayoutoptimizationproblem.Anoptimizationmodelwithperformanceconstraintsispresented.Thelayoutproblemispartitionedintofinitesubproblemsintermsofgraphtheory,insuchawayofthateachsubproblemovercomesitson-offnatureoptimalvariable.Aminimaxproblemisconstructedthatislocallyequivalenttoeachsubproblem.Byusingthisminimaxproblem,wepresenttheoptimalityfunctionforeverysubproblemandprovethatthefirstordernecessaryoptimalityconditionissatisfiedatapointifandonlyifthispointisazeroofoptimalityfunction.
简介:Inthispaper,westudyHenigefficiencyinvectoroptimizationwithnearlycone-subconvexlikeset-valuedfunction.TheexistenceofHenigefficientpointisprovedandcharacterizationofHenigefficiencyisestablishedusingthemethodofLagrangianmultiplier.Asaninterestingapplicationoftheresultsinthispaper,weestablishaLagrangemultipliertheoremforsuperefficiencyinvectoroptimizationwithnearlycone-subconvexlikeset-valuedfunction.
简介:InthispaperwereportasparsetruncatedNewtonalgorithmforhandlinglarge-scalesimpleboundnonlinearconstrainedminimixationproblem.ThetruncatedNewtonmethodisusedtoupdatethevariableswithindicesoutsideoftheactiveset,whiletheprojectedgradientmethodisusedtoupdatetheactivevariables.Ateachiterativelevel,thesearchdirectionconsistsofthreeparts,oneofwhichisasubspacetruncatedNewtondirection,theothertwoaresubspacegradientandmodifiedgradientdirections.ThesubspacetruncatedNewtondirectionisobtainedbysolvingasparsesystemoflinearequations.Theglobalconvergenceandquadraticconvergencerateofthealgorithmareprovedandsomenumericaltestsaregiven.
简介:Aclassoftrustregionmethodstorsolvinglinearinequalityconstrainedproblemsispropo6edinthispaper.Itisshownthatthealgorithmisofglobalconvergence.Thealgorithmusesaversionofthetwo-sldedprojectionandthestrategyoftheunconstrainedtrustregionmethods.Itkeepsthegoodconvergencepropertiesoftheunconstrainedcaseandhasthemeritsoftheprojectionmethod.Insomesense,ouralgorithmcanberegardedasanextensionandimprovementoftheprojectedtypealgorithm.