简介：AccordingtoLorenz,chaoticdynamicsystemshavesensitivedependenceoninitialconditions(SDIC),i.e.,thebutterfly-effect:atinydifferenceoninitialconditionsmightleadtohugedifferenceofcomputer-generatedsimulationsafteralongtime.Thus,computer-generatedchaoticresultsgivenbytraditionalalgorithmsindoubleprecisionareakindofmixtureof'true'(convergent)solutionandnumericalnoisesatthesamelevel.Today,thisdefectcanbeovercomebymeansofthe'cleannumericalsimulation'(CNS)withnegligiblenumericalnoisesinalongenoughintervaloftime.TheCNSisbasedontheTaylorseriesmethodathighenoughorderanddatainthemultipleprecisionwithlargeenoughnumberofdigits,plusaconvergencecheckusinganadditionalsimulationwithevensmallernumericalnoises.Intheory,convergent(reliable)chaoticsolutionscanbeobtainedinanarbitrarylong(butfinite)intervaloftimebymeansoftheCNS.TheCNScanreducenumericalnoisestosuchalevelevenmuchsmallerthanmicro-leveluncertaintyofphysicalquantitiesthatpropagationofthesephysicalmicro-leveluncertaintiescanbepreciselyinvestigated.Inthispaper,webrieflyintroducethebasicideasoftheCNS,anditsapplicationsinlong-termreliablesimulationsofLorenzequation,three-bodyproblemandRayleigh-Bénardturbulentflows.UsingtheCNS,itisfoundthatachaoticthree-bodysystemwithsymmetrymightdisruptwithoutanyexternaldisturbance,say,itssymmetry-breakingandsystem-disruptionare'self-excited',i.e.,out-of-nothing.Inaddition,bymeansoftheCNS,wecanprovidearigoroustheoreticalevidencethatthemicro-levelthermalfluctuationistheoriginofmacroscopicrandomnessofturbulentflows.Naturally,muchmoreprecisethantraditionalalgorithmsindoubleprecision,theCNScanprovideusanewwaytomoreaccuratelyinvestigatechaoticdynamicsystems.