简介:Inthispaper,wepresentasmoothingNewton-likemethodforsolvingnonlinearsystemsofequalitiesandinequalities.Byusingtheso-calledmaxfunction,wetransfertheinequalitiesintoasystemofsemismoothequalities.ThenasmoothingNewton-likemethodisproposedforsolvingthereformulatedsystem,whichonlyneedstosolveonesystemoflinearequationsandtoperformonelinesearchateachiteration.Theglobalandlocalquadraticconvergencearestudiedunderappropriateassumptions.Numericalexamplesshowthatthenewapproachiseffective.
简介:InthispaperwediscusstheconvergenceofamodifiedNewton’smethodpresentedbyA.Ostrowski[1]andJ.F.Traub[2],whichhasquadraticconvergenceorderbutreducesoneevaluationofthederivativeateverytwostepscomparedwithNewton’smethod.Aconvergencetheoremisestablishedbyusingaweakconditiona≤3-2(21/2)andasharperrorestimateisgivenabouttheiterativesequence.
简介:AconicNewtonmethodisattractivebecauseitconvergestoalocalminimizzerrapidlyfromanysufficientlygoodinitialguess.However,itmaybeexpensivetosolvetheconicNewtonequationateachiterate.InthispaperweconsideraninexactconicNewtonmethod,whichsolvesthecouicNewtonequationoldyapproximatelyandinsonmunspecifiedmanner.Furthermore,weshowthatsuchmethodislocallyconvergentandcharacterizestheorderofconvergenceintermsoftherateofconvergenceoftherelativeresiduals.
简介:Recentexperiencehasshownthatinterior-pointmethodsusingalogbarrierapproacharefarsuperiortoclassicalsimplexmethodsforcomputingsolutionstolargeparametricquantileregressionproblems.Inmanylargeempiricalapplications,thedesignmatrixhasaverysparsestructure.Atypicalexampleistheclassicalfixed-effectmodelforpaneldatawheretheparametricdimensionofthemodelcanbequitelarge,butthenumberofnon-zeroelementsisquitesmall.AdoptingrecentdevelopmentsinsparselinearalgebraweintroduceamodifiedversionoftheFrisch-NewtonalgorithmforquantileregressiondescribedinPortnoyandKoenker[28].Thenewalgorithmsubstantiallyreducesthestorage(memory)requirementsandincreasescomputationalspeed.Themodifiedalgorithmalsofacilitatesthedevelopmentofnonparametricquantileregressionmethods.Thepseudodesignmatricesemployedinnonparametricquantileregressionsmoothingareinherentlysparseinboththefidelityandroughnesspenaltycomponents.ExploitingthesparsestructureoftheseproblemsopensupawholerangeofnewpossibilitiesformultivariatesmoothingonlargedatasetsviaANOVA-typedecompositionandpartiallinearmodels.
简介:Inthispaper,aswitchingmethodforunconstrainedminimizationisproposed.ThemethodisbasedonthemodifiedBFGSmethodandthemodifiedSR1method.Theeigenvaluesandconditionnumbersofboththemodifiedupdatesareevaluatedandusedintheswitchingrule.WhentheconditionnumberofthemodifiedSR1updateissuperiortothemodifiedBFGSupdate,thestepintheproposedquasi-NewtonmethodisthemodifiedSR1step.OtherwisethestepisthemodifiedBFGSstep.Theefficiencyoftheproposedmethodistestedbynumericalexperimentsonsmall,mediumandlargescaleoptimization.Thenumericalresultsarereportedandanalyzedtoshowthesuperiorityoftheproposedmethod.
简介:Thispaperconsiderstheexistenceandasymptoticestimatesofglobalsolutionsandfinitetimeblowupoflocalsolutionofnon-Newtonfiltrationequationwithspecialmediumvoidofthefollowingform:{ut/|x|^2-△pu=u^q,(x,t)∈Ω×(0,T),u(x,t)=0,(x,t)∈ЭΩ×(0,T),u(x,0)=u0(x),u0(x)≥0,u0(x)全不等于0,where△pu=div(|△↓u|^p-2△↓u),ΩisasmoothboundeddomaininR^N(N≥3),0∈Ω,2
简介:WeprovideconvergenceresultsanderrorestimatesforNewton-likemethodsingeneralizedBanachspaces.TheideaofageneralizednormisusedwhichisdefinedtobeamapfromalinearspaceintoapartiallyorderedBanachspace.Convergenceresultsanderrorestimatesareimprovedcomparedwiththerealnormtheory.
简介:首先用微分中值定理推出了Newton-Leibniz公式,同时也用Newton-Leibniz公式推出了三个微分中值定理,从而证明了微分中值定理与Newton-Leibniz公式可互相证明.
简介:Inthisstudy,weuseinexactnewtonmethodstofindsolutionsofnonlinear,nondifferenti-ableoperatorequationsonBanachspaceswithaconvergencestructure.ThistechniqueinvolvestheintroductionofageneralizednormasanoperatorfromalinearspaceintoapartiallyorderedBanachspace.Inthiswaythemetricpropertiesoftheexaminedproblemcanbeanalyzedmoreprecisely.Moreover,thisapproachallmvsustoderivefromthesametheorem,ontheonehand,semi-localresultsofKantorovich-type,andontheotherhand,globalresultsbasedonmono-tonicityconsiderations.Furthermore,iveshowthatspecialcasesofourresultsreducetothecorrespondingonesalreadyintheliterature.Finally>ourresultsareusedtosolveintegralequationsthatcannotbesolvedwithexistingmethods.
简介:Inthispaper,thesupercouvergenceofaclassofWilson-likeelementsisconsidered,andasuperconvergentestimateofWilson-likeelementsisobtainedforstrongregularmeshes.
简介:Inthispaperweimprovethetwoversionsofthetwo-sidedprojectedquasi-Newtonmethod-onewasproposedbyNocedal&Overtonin[1]andtheotherwasdiscussedinourpreviouspaper,byintroducingthreedifferentmeritfunctionstomakeinexactone-dimensionalsearches.Itisshownthattheseimprovedquasi-Newtonalgorithmshavegainedglobalconvergencepropertywhichisnotpossessedbytheoriginaltwoalgorithms.
简介:InthispaperwereportasparsetruncatedNewtonalgorithmforhandlinglarge-scalesimpleboundnonlinearconstrainedminimixationproblem.ThetruncatedNewtonmethodisusedtoupdatethevariableswithindicesoutsideoftheactiveset,whiletheprojectedgradientmethodisusedtoupdatetheactivevariables.Ateachiterativelevel,thesearchdirectionconsistsofthreeparts,oneofwhichisasubspacetruncatedNewtondirection,theothertwoaresubspacegradientandmodifiedgradientdirections.ThesubspacetruncatedNewtondirectionisobtainedbysolvingasparsesystemoflinearequations.Theglobalconvergenceandquadraticconvergencerateofthealgorithmareprovedandsomenumericaltestsaregiven.
简介:Fortheimprovedtwo-sidedprojectedquasi-Newtonalgorithms,whichwerepresentedinPartI,weproveinthispaperthattheyarelocallyone-steportwo-stepsuperlinearlyconvergent.Numericaltestsarereportedthereafter.ResultsbysolvingasetoftypicalproblemsselectedfromliteraturehavedemonstratedtheextremeimportanceofthesemodificationsinmakingNocedal&Overton’soriginalmethonpractical.Furthermore,theseresultsshowthattheimprovedalgoritnmsareverycompetitiveincomparisonwithsomehighlypraisedsequentialquadraticprogrammingmethods.