ebook img

Mathematical Neuroscience PDF

186 Pages·2013·3.805 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Mathematical Neuroscience

Part I Methods of Nonlinear Analysis CHAPTER 1 Introduction to Part I PartIisdevotedexclusivelytothetheoryandmethodsofnonlinearanalysisofinfinite systems.Theseinfinitesystemsplayanimportantroleintheintegrativeaspectsof neurosciencemodellingwithfunctionalanalyticmethodsofdescribingnonlinearphenomena throughsettheoryandvectorspaces.Inmodernmathematicsvectorspacesusuallyarecarried overfromCartesianspaces,andknowledgeofthesespacesisofhelpinanalyzingmore generalvectorspaces(e.g.,Banachspaces,Hilbertspaces).However,whenitcomestothe brainanduncoveringitsdynamicalnature,avectororpointinordinarythree-dimensional spacenolongermakesanysense.Visualizingneuronaldynamicsbasedonordinary three-dimensionalCartesianspacebecomescounterproductive,especiallyintheprocessof integratingacrossscale.Thedynamicsofthebrainwithitsenormouscomplexitycannotbe understoodthroughtheuseofelementaryfunctionsofclassicalmathematics.Firstly,weneed abroaderdefinitionoffunctionsothatanoncommittalnotationandthenatureoftheruleor mappingbecomeirrelevant.Secondly,weneedtoconsiderspacesofcontinuousfunctions suchasinfinite-dimensionalEuclideanvectorspacesthatcarrynonlinearphenomenafrom whichnewideascanbeharnessedfromdynamicalcontinuity. Thenoncommittalapproachunderlyingthefoundationaltheoryofinfinitesystemsenables newconstructsofmorepowerfulmodels,especiallyintheprocessofintegratingacrossscale. Forinstance,thesetofstatesofbiologicalneuralnetworksisinfinite,topologically continuous.Thisisbecausethestateofthenetworkisspecifiednotbythefiringofdigital spikesinneurons,butintheelectricalpatternsbroughtaboutbytheelectricalfield.However, thesetofstatesofartificialneuralnetworksisfinite,topologicallydiscreteduetospiking beingthepredominantparadigmwhererefractorypropertiesofneuronsareincludedafterthey havefired. Themethodsofnonlinearanalysispresentedinthebookapplyonlytosecond-orderequations arisinginneuroscience.Giventheimportanceofmorerealisticneuronalmodelsatthecellular level,itcanbesaidthatneuronalmodelsarefoundationalunitsofthenervoussystemthatcan beinterconnectedtoformdendritictrees,networksofdendritictrees,systemsofnetworks, andsoon.Considerataperingcableofadendriticsegmentwithacontinuouslytaperingcable asshowninFigure1.1. MathematicalNeuroscience.http://dx.doi.org/10.1016/B978-0-12-411468-5.00001-6 ©2014ElsevierInc.Allrightsreserved. 3 4 Chapter1 Figure1.1 Ataperingcablerepresentingasinglebranchofadendritictree. TheequationdescribingthedepolarizationV = V(x,t)(inmillivolts)atthespacepointx and timet foracontinuousradiusofthecylinderr(x)denotedbythedashedlinesisgivenby ∂V 1 ∂V(cid:3) ∂V(cid:4) (cid:5) C = (cid:2) r2(x) − g (x,t;V)(V −V ). (1.1) m ∂t 2ρr(x) 1+(dr(x)/dx)2 ∂x ∂x i i Inessence,byassumingC = 1 µF/cm2, z(t,x) = V(x,t),andtheoperators m D = ∂V,D = ∂V,D2 = ∂2V,etc.,theaboveequationsturnouttobe t ∂t x ∂x xx ∂x2 r(x) (dr(x)/dx) D z(t,x)− (cid:2) D2 z− (cid:2) t 2ρ 1+(dr(x)/dx)2 xx ρ 1+(dr(x)/dx)2 (cid:5) D z − g (x,t;z)(z(t,x)−z ). x i i Definea(x) = √ r(x) , b(x) = √−(dr(x)/dx) ,and 2ρ 1+(cid:6)(dr(x)/dx)2 ρ 1+(dr(x)/dx)2 f(t,x,z(t,x), z) = g (t,x,z)(z(t,x)−z )sothattheaboveequationbecomes i i D z(t,x)−a(x)D2 z(t,x)+b(x)D z(x,t) = f(t,x,z(t,x),z). (1.2) t xx x Weconsidercoupledinfinitesystemsofequationswherethecouplingarisesfromboththe connectivitybetweenbranchesofadendritictreeofasingleneuronandthesynapticcoupling betweenindividualneuronsinnetworks.Wethereforeintroduceasuperscript“i”todenote, forexample,themembranedepolarizationoftheithneuronwhere“i”rangesfrom1to∞. Thecoefficientsa(x)andb(x)haveintroducedsubscripts“j”and“k”todenotethe electrogeometricalinhomogeneityofthebranchingprocessesandweassumethereare“m” branchesinaparticulardendritictree,sothat(1.2)becomes (cid:5)m (cid:5)m D zi(t,x)− a (x)D2 zi(t,x)+ b (x)D zi(x,t) = fi(t,x,z(t,x),z). (1.3) t jk xjxk j xj j,k=1 j=1 IntroductiontoPartI 5 Insettheorynotation(1.3)isconsideredaweaklycoupledinfinitesystemofequationsofthe form (cid:5)m Fi[zi](t,x) := D zi(t,x)− ai (t,x)D2 zi(t,x) t jk xjxk j,k=1 (cid:5)m + bi(t,x)D zi(t,x) = fi(t,x,z(t,x),z) fori ∈ S (1.4) j xj j=1 for(t,x) = (t,x ,...,x ) ∈ D,where D isaboundedcylindricaldomain, 1 m D := (0,T]×G, 0 < T < ∞, G ⊂ Rm isanopenandboundeddomainwhoseboundary ∂G isan(m −1)-dimensionalsufficientlysmoothsurfacem = 1,2,or3, S isanarbitraryset ofindices,Fi, i ∈ S,arediagonaloperators1 whichareuniformlyparabolicin D,and fi, i ∈ S,aregivenfunctions fi: D×B(S)×C (D) → R, (t,x,y,s) (cid:7)→ fi(t,x,y,s), i ∈ S, S where S := {(t,x) : t = 0, x ∈ G}, σ := (0,T]×∂G isthelateralsurfaceand(cid:5) := S ∪σ 0 0 istheparabolicboundaryofthedomain D and D := D∪(cid:5) = [0,T]×G.Theright-hand sides fi ofthesystem(1.4)arefunctionalswithrespecttothelastvariable.Theydescribethe localnonlinearityduetothemembraneelectricalproperties,andthediffusiveinteractiondue toelectricalcurrentflowinneurons. ThefollowingexamplesofVolterrafunctionalsareconsidered: (cid:7) t f (t,x,z) = m(t −τ)K(z(τ,x))dτ, (1.5) 1 (cid:7)0 t f (t,x,z) = K(t,τ,x,z(τ,x))dτ, (1.6) 2 (cid:7)0 (cid:7) t f (t,x,z) = K(t,τ,x,ξ,z(τ,ξ))dτdξ. (1.7) 3 0 G Equationsincludingsuchfunctionalsneedtobeconsideredinappropriatelychosendomains. Otherexamplesinclude: (cid:5)∞ (cid:5)∞ f1(t,x,z)=−z1(t,x) a1zk(t,x)+ b1z1+k(t,x), (1.8) 4 k k k=1 k=1 1 OperatorsFi,i ∈S,arediagonalifFi dependsonzi onlyforalli ∈S. 6 Chapter1 (cid:5)j−1 (cid:5)∞ f j(t,x,z)= 1 aj−kzj−k(t,x)zk(t,x)−zj(t,x) ajzk(t,x) 5 2 k k k=1 k=1 (cid:5)∞ (cid:5)j−1 + bjzj+k(t,x)− 1zj(t,x) bj−k for j = 2,3..., (1.9) k 2 k k=1 k=1 and (cid:7) (cid:5)∞ f1(t,x,z)=−z1(t,x) a1(x,ξ)zk(t,ξ)dξ 6 k k=1 G (cid:7) (cid:5)∞ + B1(x,ξ)z1+k(t,ξ)dξ, (1.10) k k=1 G (cid:7) (cid:5)j−1 f j(t,x,z)= 1 Aj−k(x,ξ,η)zj−k(t,ξ)zk(t,η)dξdη 7 2 k=1 G×G k (cid:7) (cid:7) (cid:5)∞ (cid:5)∞ −zj(t,x) aj(x,ξ)zk(t,ξ)dξ + Bj(x,ξ)zj+k(t,ξ)dξ k k k=1 G k=1 G (cid:5)j−1 − 1zj(t,x) bj−k(x) for j = 2,3,..., (1.11) 2 k k=1 (cid:8) (cid:8) where Aj(x,ξ,η)dx = aj(ξ,η)and Bj(x,ξ)dx = bj(ξ)arethenonnegative G k k G k k coefficientsofratesaj andbj,while f j, f j, f j,and f j areVolterrafunctionals. k k 4 5 6 7 Systemswithright-handsidesofthistypeareconsideredasthediscrete coagulation-fragmentationmodelswithdiffusion.Therearetwodifferent(discreteand continuous)modelswhichdescribechangesintheconcentrationofclusters(proteins) undergoingcoagulation(composition)orfragmentation(decomposition).Ifavariable zj = zj(t,x)describestheconcentrationof j clusterscomprising j identicalbasicelements knownasmonoclusters,thenthevariableisadiscreteone,takingnonnegativeintegervalues. Themodelthusobtainedisadiscretecoagulation-fragmentationmodelbuiltofinfinite countablesystemsofequations. Thesemodelsareexpressedintermsoftheinfinitecountablesystemsofequationsoftheform ∂zj(t,x) (cid:5)m ∂2zj(t,x) −d = f j (t,x,z), j ∈ N, (1.12) ∂t j ∂x2 5,6 l=1 l for(t,x) ∈ (0,T]×G := D,whereG ⊂ Rm isaboundeddomainwithsufficientlysmooth boundary∂G,withtheinitialcondition z(0,x) =U (x) ≥ 0 for(t,x) ∈ G, (1.13) 0 IntroductiontoPartI 7 andthesealed-endboundarycondition dzj(t,x) = 0 for(t,x) ∈ σ, j ∈ N, (1.14) dν wherefunctions f j, f j aregivenby(1.8),(1.9),respectively,anddiffusioncoefficientsd are 4 5 j positive. Thismodelisexpressedintermsoftheinfinitecountablesystemsofequationsoftheform ∂zj(t,x) −Aj[zj](t,x) = f j (t,x,z) for(t,x) ∈ D j ∈ N (1.15) ∂t 6,7 withinitialandboundaryconditions(1.13)and(1.14),wherethediffusionoperators,having thedivergenceform (cid:9) (cid:10) (cid:5)m ∂ ∂zj(t,x) Aj[zj](t,x) := dj(t,x) , j ∈ N ∂x lk ∂x l,k=1 j k areuniformlyellipticin D,diffusioncoefficientsdi arepositiveandfunctions fi, fi are jk 6 7 givenby(1.10),(1.11),respectively,integraloperators. Unfortunately,themonotoneiterativemethoddoesnotcoverthesystems(1.12)and(1.15). Therefore,tosolveproblems(1.12)and(1.15),thetruncationmethodisapplied. Infinitecountablesystemsofordinarydifferentialequationscanbeusedtosolvesome problemsfornonlinearequationswiththeuseofthefinite-differencemethod.Inthisvery importantmethodwhichiscalledthenumericalmethodoflines,spatialderivativesonlyare discretizedbydifferenceexpressions,leadingtoaninfinitecountablesystemofequations. Ontheotherhand,whenacontinuousvariable(e.g.,concentrationperunitvolume)taking nonnegativerealvaluesisusedinthedescription,wearriveatacontinuousmodelof coagulation-fragmentation,amodelbuiltofinfiniteuncountablesystemsofequations. Adoptinganimportantassumptionthatthenumberofcreatedclustersisnotbounded,we concludethatthenumbersofbothequationsandvariablesmaybeinfinite. Continuousmodelsmaybeexpressedintermsofinfiniteuncountablesystemsofequationsof thereaction-diffusiontype.Studyinginfiniteuncountablesystemsofequationsisfarfrom simple.Inparticular,ifwehaveaninfinitesystemofequationsoftheform(1.4) (nonstationaryreaction-diffusionequations),thenthestudymaybebroughtdowntoan evolutionequationoftheform ∂u − A(t)u = F(t,u) fort > 0, x ∈ Rm ∂t 8 Chapter1 withtheinitialcondition u(0,x) = f(0,x) for x ∈ G, where−Aisalinearoperatorand F isanonlinearreactionterm,consideredinan appropriatelydefinedvectorspace. Forsystem(1.4),wewillconsidertheso-calledFourierfirstinitial-boundaryvalueproblem: findtheregular(classical)solutionz ofsystem(1.4)in D fulfillingtheinitial-boundary condition z(t,x) = φ(t,x) for(t,x) ∈ (cid:5). (1.16) Weremarkthatitisnotpossibletosolveinfinitesystemsofequations.Therefore,anatural approachtothestudyoftheexistenceanduniquenessofsolutionsofinitialandboundary valueproblemsforinfinitesystemsofsuchequationsistostartwithafinitenumberofsuch equations(thatis,studyingfinitesystemsofsuchequationsandthenextendingtheobtained resultstotheinfinitesystems).However,seriousbasicdifficultiesemergehere.Thedifficulty referredtoabovemaybesurmountedbyconsideringinfinitesystemsintheformofsystem (1.4)andinthediagonalformoftheoperatorsFi, i ∈ S.Themethodwillconsistinthe constructionofcertainapproximatingsequencestendingtosolutionsofinfinitesystemsin suitablychosenvectorspaces. Themethodofsuccessiveapproximationisoneofthemostbasicandsimplestmethodsfor provingexistencetheoremsforcertaintypesofequations.Numerousvariantsofthismethod arewellknown,usingdifferentconstructionsofasequenceofsuccessiveapproximationsand differentwaysofprovingtheconvergence. Toprovetheexistenceanduniquenessoftheregular(classical)andglobalinatimesolution (i.e.,thesolutionisdefinedonthewholeinterval[0,∞])ofthisFourierfirstinitialvalue problem,weapplyvariousmonotoneiterativemethodsinvectorspaces,partiallyorderedby positivecones.Themethodsofupperandlowersolutionsarewell-knownmethodsofproving theexistenceofsolutionsfornumerousclassesofinitial-boundaryvalueproblems.Applying thesemethodsrequiresassumingthemonotonicityoftheright-handsidesoftheequations, i.e.,thereactionfunctions fi(t,x,y,s), i ∈ S,withrespecttothefunctionargument y and thefunctionalarguments.Wealsoassumetheexistenceofanorderedpairofaloweru and 0 anupperv solutionstotheproblemconsidered.Moreover,abasicassumptionon 0 fi(t,x,y,s), i ∈ S,isalsotheleft-handsideLipschitzcondition.Theseassumptionsarenot typical,butweobtainconstructiveexistencetheoremsandaprioriinformationonasector (cid:10)u ,v (cid:11)inwhichthosesolutionsmustremain.Weremarkthattheright-handsideLipschitz 0 0 conditionisusedtoensuretheuniquenessofasolution. Wewillassumethatthereactionfunctions fi(t,x,y,s), i ∈ S,areVolterrafunctionals,i.e., theysatisfytheVolterraconditionwithrespecttothelastvariables.Thismeansthatthe valuesofthesefunctionsdependonlyonthepasthistoryoftheprocess.Anexceptionally IntroductiontoPartI 9 importantroleinthetheoryofnonlinearequationsplaysthenonlinearsuperpositionoperator, whosepropertiesarestudiedincertainimportantlemmas. Ineachcase,themonotoneiterativetechniquesasthemethodofprovingtheexistenceof solutionsconsistofthreesteps:(i)constructinganondecreasingandanonincreasingsequence ofloweranduppersolutions;(ii)showingtheuniformoralmostuniformconvergenceofthe constructedsequencesofapproximatesolutions;and(iii)provingthatthelimitfunctionsare solutions. Toexaminetheexistenceanduniquenessofasolutionofinfinitesystems,sixmonotone iterativemethodsareemployedinchapter4.Inthemethodofdirectiteration,Chaplygin methodanditsmodifications,thesuccessivetermsoftheapproximationsequencesaredefined assolutionsoflinearequations.Wealsopresentadifferentvariantofamonotoneiterative method,inwhichweapplytheimportantideaofapseudo-linearizationofnonlinearproblems (thesuccessivetermsofapproximationsequencesaredefinedassolutionsofequations). ThemaindifferencebetweentheChaplyginmethodandothermethodsofmonotoneiterations liesinthedefinitionsoftheapproximatingsequences.Thismethodrequiresastronger assumptiononthefunctions fi,namelytheconvexityassumption,butyieldssequencesof successiveapproximationsconvergingtothesolutionoftheproblemunderconsideration morequicklythantheiterativesequences{u }and{v }constructedothermethods,underthe n n obviousassumptionthatallmonotoneiterativemethodsstartfromthesamepairofalower solutionu andanuppersolutionv ,whoseexistenceisassumed.Thesesequencesare 0 0 convergedquadraticallytothesearchedsolution. UsingChaplygin’sideawewillusethelinearizationoffunctions fi(t,x,y,s),i ∈ S,with respecttobotharguments: y ands,simultaneously.Acharacteristicfeatureofthismethodis constructionofthemonotonesequencesconvergingtothesolutionsoughtwithinthesector (cid:10)u ,v (cid:11),andwhoseconvergentspeedisquadratic.Wewillusethismethodforinfinite 0 0 systemsofequationswithfunctionals. Nextwepresentamonotonemethodofdirectiterationsinanunboundeddomain,when consideredfunctionssatisfysomegrowthcondition. Twoideasareworthmentioningamongtheproposedapproacheswhichenableustodeletethe monotonicityconditionfortheright-handsideswithrespecttothefunctionandfunctional arguments(oratleastthelatterone).Forsimplicity,wewillconsiderthecasewherethe right-handsidedependsonlyonthefunctionalargument. Thefirstideaconsistsofthemodificationofeachoftheequationsbyadding,tobothsidesof theequations,thesamelineartermincludingtheunknownfunction,torenderthenew right-handsidesoftheequationsmonotone,shouldthisbepossible.Thismeansthatwewill bereplacingthemonotonicityconditionwithaweakerone,namely,asemi-monotonicity condition. 10 Chapter1 Theotherideaconsistsofdividingthefunctionargumentsintotwogroups,dependingon whethertheright-handsideofanequationisincreasingordecreasingwithrespecttoagiven argument.Wheneverthisisthecase,wespeakaboutmixedmonotonicity.Thesplittingofthe reactionfunctionleadstofourdifferenttypesoforderedpairsofloweranduppersolutions. Thesituationisbestreflectedinthechoiceofthepairoffunctionswhichstartthe approximationprocess. Itshouldbenotedherethatbothideasweredevelopedtoanswertheneedtostudyspecific finitesystemsofequationsandtheyadmitdirectextensiontotheinfinitesystemsofequations. TheproofsoftheoremsarebasedonSzarski’sresultsconcerningthedifferentialinequalities forweaklycoupledinfinitesystemsofequationsinwhichVolterrafuntionalsarise.The theoremsonweakandstrongdifferentialinequalities,thecomparisontheoremanda uniquenesscriterionaregiveninchapter3.Themaximumprinciple,whichplaysa fundamentalrole(seeChapter3)isreflectedinwhatisknownasapositivitylemma,which directlyfollowsfromthemaximumprinciple. Theuniquenessofsolutionstotheproblemsconsideredinourworkisguaranteedbythe LipschitzconditionandfollowsfromSzarski’suniquenesscriterion. Themonotoneiterativemethodsarequiteusefulforcomputationofnumericalsolutionsof finitesystemsofequations. Chapter4givesmonotoneiterativemethodsforthecaseofinfinitesystemsofequations whoseright-handsidedependsonz(t,·),i.e., fi(t,x,z(t,x),z(t,·)). InChapter5,wegivesomeremarksonthemonotoneiterativemethodsandonthe constructionsofupperandlowersolutions,forfinitesystemsofequations,andsomeremarks concerningthepossibilityofextendingthesemethodstomoregeneralinfinitesystemsof equations. InChapter6,wedescribethetruncationmethodforinfinitecountableanduncountable systemsofequationswhichplayaveryimportantroleamongapproximationmethods.The applicationofthesemethodsleadstosequencesofapproximatesolutions{zNN,ψ}N=1,2,..., which—underappropriateconditions—tendtotheexactsolutionofagivenproblemas N → ∞.Letusobservethatwedonotneedtoknowthepreviousapproximationsto determinethe Nthapproximationasasolutionoffinitesystemsofthefirst N equationswith N unknownfunctionswhicharecalledtruncatedsystems.Thesefinitesystemsofequations arediscretizedbythefinite-differencemethod.Therearethreebasicmonotoneiteration schemesforfinite-differencesystems:aPicarditeration,amodifiedversionofaJacobi iteration,andtheGauss-Seidelmethod.Theseiterationsalsoprovidenumerical approximationofsolutionswhichisimportantinpracticalapplications.Iftheinitialiteration isalwaysapairofknowncoupledloweranduppersolutionsoftheconsideredproblem,then IntroductiontoPartI 11 thesequenceofPicarditerationconvergesfasterthanthesequenceofaGauss-Seideliteration whichinturnconvergesfasterthanthesequenceoftheJacobiiteration.Thesemonotone iterativeschemesprovidethesequencesoflineariterationswhichconvergemonotonicallytoa uniquesolutionoftheconsideredtruncatedproblem.Thereforethismethodisveryusefuland commonlyusedinpracticalcomputationofapproximatesolutions. Tostudytheexistenceanduniquenessofasolutionofinfinitesystems,inChapter7weuse thetopologicalfixedpointmethod.ConsideringpartiallyorderedBanachspaceofcontinuous mappings,wegivesomenaturalsufficientconditionsfortheexistenceanduniquenessofthe solution.TheaprioriestimateswhichappearinapplicationsoftheBanachandSchauder fixedpointtheoremsareparalleltotheabove-mentionedassumptionsinthetheoryof monotoneiterativemethods.Finally,toprovetheexistenceanduniquenessofglobalsolutions inatime,i.e.,thesolutiondefinedonthewholeinterval[0,∞),weapplytheLeray-Schauder fixedpointtheorembutfirstweextendsomeaprioriestimatesoftheFriedmantype. InChapter8thetheoremonexistenceofasolutionoftheDirichletproblemforinfinite systemsofequationsisprovedwithuseofacertainvariantoftheChaplyginmethod.The resultsobtainedareappliedtostudytheasymptoticbehavioroftheglobaltime-dependent solutionsinrelationtothesolutionsofthetime-independentproblem.Ithasbeenprovedthat thelimitofthesolutionast → ∞isasolutionofthetime-independentproblem,obtainedby applyingthemonotonemethodofassociatedloweranduppersolutions. Thetoolsofmodernmathematicsinthedevelopmentofaframeworkthatunderpinsthe understandingofinfinitesystemsingeneralisdeveloped(PartI).Infinitesystemsofequations ariseinavarietyofphysicalcontexts,especiallywhenmodelingintheneurosciences.New insightscanbeachievedthatwillleadtoabetterunderstandingofthebrainandnervous systemfromamathematicalperspective(seePartII).

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.