简介:(x)?S.Ifg(x)=0,S*(x)=S*.MomnitS/0andnitS(x)2nitS,nitS(x)/0.Hellce(S(x))*=S*(x)hasacompactconvexbase,denotedbyA,andthecompactconvexbaseofQ*isdenotedbyB.{0}*=Re.LetC={xERe:IIxll=1}.ThenCisacompactbaseof{0}*.Letsupposethat(3dOER")(VPEV*\{0})(PTp)"(x;do)<0,fromLemma2.3thereedestsadER"togetherwithb>0suchthatInparticular,fixingTEBandA6A,settingp ̄(ET,ac, ̄)EV*withE>0andueC,weIndeed,ifthereexistsaalE(0,6)suchthath(x+a'd)/0,wecanchooseFIECsatisfyingU,"h(x+a'd)>0.In(4),lettingE-0,itfollowsthatU"h(x+rs'd)<0,wh?
简介:Inthispaper,wepresentasuccessivequadraticprogramming(SQP)methodforminimizingaclassofnonsmoothfunctions,whicharethesumofaconvexfunctionandanonsmoothcompositefunction.ThemethodgeneratesnewiterationsbyusingtheArmijo-typelinesearchtechniqueafterhavingfoundthesearchdirections.Globalconvergencepropertyisestablishedundermildassumptions.Numericalresultsarealsooffered.
简介:Akindofnondecreasingsubgradientalgorithmwithappropriatestoppingrulehasbeenproposedfornonsmoothconstrainedminimizationproblem.Thedualtheoryisinvokedindealingwiththestoppingruleandgeneralglobalminimiizingalgorithmisemployedasasubroutineofthealgorithm.Themethodisexpectedtotacklealargeclassofnonsmoothconstrainedminimizationproblem.
简介:Inthispaper,wepresentanonmonotonealgorithmforsolvingnonsmoothcompositeoptimizationproblems.Theobjectivefunctionoftheseproblemsiscompositedbyanonsmoothconvexfunctionandadifferentiablefunction.Themethodgeneratesthesearchdirectionsbysolvingquadraticprogrammingsuccessively,andmakesuseofthenonmonotonelinesearchinsteadoftheusualArmijo-typelinesearch.Globalconvergenceisprovedunderstandardassumptions.Numericalresultsaregiven.
简介:Thispaperpresentsacoordinategradientdescentapproachforminimizingthesumofasmoothfunctionandanonseparableconvexfunction.Wefindasearchdirectionbysolvingasubproblemobtainedbyasecond-orderapproximationofthesmoothfunctionandaddingaseparableconvexfunction.UnderalocalLipschitzianerrorboundassumption,weshowthatthealgorithmpossessesglobalandlocallinearconvergenceproperties.Wealsogivesomenumericaltests(includingimagerecoveryexamples)toillustratetheefficiencyoftheproposedmethod.
简介:Backlash-likehysteresisisoneofthenonsmoothandmulti-valuednonlinearitiesusuallyexistinginmechanicalsystems.Thetraditionalidentificationmethodisquitedifficulttobeusedtomodelthesystemsinvolvedwithsuchcomplexnonlinearities.Inthispaper,anonsmoothrecursiveidentificationalgorithmforthesystemswithbacklash-likehysteresisisproposed.Inthismethod,theconceptofClarkesubgradientisintroducedtoapproximatethegradientsatnonsmoothpointsandtheso-calledbundlemethodisusedtoobtaintheoptimizationsearchdirectioninnonsmoothcases.Then,arecursivealgorithmbasedontheideaofbundlemethodisdevelopedforparameterestimation.Afterthat,theconvergenceanalysisofthealgorithmisinvestigated.Finally,simulationresultstovalidatetheproposedmethodonasimulatedmechanicaltransmissionsystemarepresented.
简介:Inthispaper,nonsmoothunivex,nonsmoothquasiunivex,andnonsmoothpseudounivexfunctionsareintroduced.Byutilizingthesenewconcepts,sufficientoptimalityconditionsforaweaklyefficientsolutionofthenonsmoothmultiobjectiveprogrammingproblemareestablished.WeakandstrongdualitytheoremsarealsoderivedforMond-Weirtypemultiobjectivedualprograms.