HAL Id: hal-00354175
https://hal.archives-ouvertes.fr/hal-00354175
Submitted on 19 Jan 2009
HAL
is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or
L’archive ouverte pluridisciplinaire
HAL, estdestinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires
Bifurcation delay - the case of the sequence: stable focus - unstable focus - unstable node
Eric Benoît
To cite this version:
Eric Benoît. Bifurcation delay - the case of the sequence: stable focus - unstable focus - unstable
node. Discrete and Continuous Dynamical Systems - Series S, American Institute of Mathematical
Sciences, 2009, 2 (4), pp.911-929. �hal-00354175�
stable fous - unstable fous - unstable node
Eri Benoît
∗
January19,2009
Abstrat
Letusgiveatwodimensionalfamilyofrealvetorelds.Wesupposethatthereexistsastationarypoint
where the linearized vetor eld hassuessively a stable fous, anunstable fous and anunstable node.
Whenthe parametermovesslowly,a bifurationdelay appears dueto theHopf bifuration. Thestudied
questioninthisartileistheontinuationofthedelayafterthefous-nodebifuration.
AMSlassiation: 34D15,34E15,34E18,34E20,34M60
Keywords: Hopf-bifuration,bifuration-delay,slow-fast,anard,Airy,relief.
1 Introdution
"Singularperturbations"isastudieddomainfrommanyyearsago. Sine1980,manyontributionswerewritten
beause new toolswere applied to the subjet. The main studied objetsare the slow fast vetor elds also
known assystems with two time-sales. Wewill givethe problem herewith amorepartiular point of view:
thebifuration delay,asin artiles[8,2,9,7℄. Wewritethestudiedsystem: εX˙ =f(t, X, ε),whereεisareal
positiveparameterwhih tendstozero. Forabetterunderstandingoftheexpressiondynami bifurationit is
bettertowrite thesystemafteraresalingofthevariable:
X˙ = f(a, X, ε)
˙
a = ε
whereaisa"slowlyvarying"parameter.
The main objetsin this study are theeigenvaluesof the linearpartof equation X˙ =f(a, X,0) near the
quasi-stationarypoint. Indeed,theygiveaharaterizationofthestabilityoftheequilibriumofthefastvetor
eldatthispoint. Theaimofthisstudyistounderstandwhathappenswhenthestabilityofaquasi-stationary
pointhanges. Abifurationourswhenatleastoneoftheeigenvalueshasanullrealpart.
Inthisartilewerestritourstudytotwo-dimensionalrealsystems. Inthissituation,thegeneribifurations
are: thesaddle-nodebifuration,theHopfbifuration andthefous-nodebifuration.
Thesaddle-nodebifurationissolvedbytheturningpointtheory: whentherealpartofoneoftheeigenvalue
beomes positive, there is no delay and a trajetory of the systems leaves the neighborhood of the quasi-
stationary point when it reahes the bifuration. For this study, the study of one-dimensional systems is
suient: we have a deomposition of the phase spae where only the one-dimensionalfator is interesting.
Thereexistmanyartilesonthissubjet,wewillbeinterestedpartiularlyby[3℄wherethemethod ofrelief is
used. Theartile[6℄introduesthegeometrialmethodsofFenihel's manifold.
TheHopfdelayedbifurationiswellexplainedin[10℄, wewillupgradetheresultsinparagraph2below.
Inafous-nodebifuration, thestabilityofthequasistationary pointdoesnothange,then, loally,there
isnoproblemofanardsorbifurationdelays. Indeed,whenthereis abifurationdelayat aHopf-bifuration
point,itis possibleto evaluatethevalueofthedelay,and themainquestionisto understandtheinuene of
thefousnodebifurationtothisdelay.
Inparagraph2,theHopfbifurationaloneisstudied,aswellasthefous-nodebifurationfollowingaHopf
bifurationinparagraph3.
∗
Laboratoire deMathématiques et Appliations, Université dela Rohelle, avenueMihelCrépeau, 17042 LA ROCHELLE
and/orprojetCOMORE,INRIA,2004routedesLuioles06902SOPHIA-ANTIPOLISourriel:ebenoituniv-lr.fr
steadystateinthewhole domain,sothistrajetoryhasaninnitedelay. Theusedmethods arereal, andthe
systemhastobesmooth(atuallyonlyC2). Inparagraphs2.2et3.2,weavoidthisveryspeialhypothesis. It isheresupposedthatthesystemisanalyti,andwestudythesolutionsonomplexdomains. Unfortunately,I
havenotaproofforthemainresultofthisartile. Butitseemstomethattheproblemisinteresting,andthe
resultsareargumented.
WeuseNelson'snonstandardterminology(seeforexample[5℄). Indeed,almostallsentenesanbetranslated
in lassialterms,where ε isonsidered asavariableand notasaparameter. Often, thetranslationis given onfootnotes.
2 The delayed Hopf bifuration
Theproblemis studiedand essentiallyresolvedin [10℄. Wegiveheretheproofsto improvetheresultsandto
xtheideasforthemainparagraphoftheartile. Themaintoolistherelief'stheoryofJ.L.Callot,explained
in[4℄.
Thestudiedequationis
εX˙ =f(t, X, ε) (1)
wheref isanalytionadomainDofC×C2×C.
Hypothesis and notations
H1 Thefuntionf isanalyti. Ittakesrealvalueswhentheargumentsarereal.
H2 Theparameterεisreal, positive,innitesimal 1
.
H3 Thereexists ananalytifuntion φ,dened onaomplexdomainDt sothat f(t, φ(t),0) = 0. Theurve X =φ(t) is alled the slow urve of equation(1). We assumethat the intersetion ofDt with the real
axisis aninterval]tm, tM[.
H4 Letusdenoteλ(t)andµ(t)fortheeigenvaluesofthejaobianmatrixDXf,omputedatpoint(t, φ(t),0).
Weassumethat ,fortreal,thesignsoftherealandimaginarypartsaregivenbythetablebelow:
t tm a tM
ℜ(λ(t)) - 0 +
ℜ(µ(t)) - 0 +
ℑ(λ(t)) - - -
ℑ(µ(t)) + + +
Then,whentinreasesfromtm totM,thequasi-steadystateisrstanattrativefous,thenarepulsive fous,with aHopfbifurationat t=a.
2.1 Input-output funtion when there exists a big anard
Inthissetion,weassumethatthereexistsabiganardX˜(t)i.e. asolutionofequation(1)suhthat2X˜(t)≃φ(t)
foralltintheS-interiorof]tm, tM[. Wenowwanttostudytheotherssolutionsofequation(1)byomparison
withX˜.
Themaintoolforthatisasequeneofhangeofunknowm: rst,weperformatranslationonX,depending
ont toputthebiganardontheaxis:
X = ˜X(t) + Y
Itgivesthesystem
εY˙ = g(t, Y, ε) with g(t, Y, ε) =f(t,X˜(t) +Y, ε)−f(t,X˜(t), ε)
1
Inlassialterms,weassumethatεleavesinasmallomplexsetor: |ε|boundedandarg(ε)∈]−δ, δ[.
2
Withoutnonstandardterminology,abiganardisasolutionofequation(1)dependingontheparameterεsuhthat
∀t∈]tm, tM[, lim
ε>0,ε→0
X˜(t, ε) =φ(t)
ThematrixDXf(t, φ(t),0)hastwoomplexonjugatedistinteigenvalues(seehypothesisH4),thenthereexists alineartransformation P(t)whih transformsthejaobianmatrixinaanonialform. We denethehange
ofunknown
Y =P(t)Z
Thenewsystem,hasthefollowingform (wewroteonlytheinterestingterms):
εZ˙ = h(t, Z, ε) with h(t, Z, ε) =
α(t) −ω(t) ω(t) α(t)
Z+O(ε)Z+O(Z2) , λ(t) =α(t)−iω(t)
Thenexthange isgivenbythepolaroordinates:
Z =
rcosθ rsinθ
εr˙ = r(α(t) +O(ε) +O(r)) εθ˙ = ω(t) +O(ε) +O(r)
Thelastoneisanexponentialmirosope 3
:
r = expρ ε
ρ˙ = α(t) +O(ε) +eρεk1(r, θ, ε)
εθ˙ = ω(t) +O(ε) +eρεk2(r, θ, ε) (2)
While ρ is non positive and non innitesimal, r is exponentially small and the equation (2) gives a good approximationof ρ with ρ˙ =α. When ρ beomes innitesimal, with a moresubtle argument (see [1℄)using dierentialinequations,weanprovethatrbeomesnoninnitesimal. Thisgivesthepropositionbelow:
Proposition 1 Let us assumehypothesis H1 toH4(Hopf bifuration) for equation(1). However, we assume
that there exists aanard X(t)˜ going along4 the slow urve at least on ]tm, tM[. Then if X(t) goes along the
slow urveexatly 5
on ]te, ts[ with[te, ts]⊂]tm, tM[,then Z ts
te
ℜ(λ(τ))dτ = 0
Theinput-output relation(betweente andts)is dened byRts
te ℜ(λ(τ))dτ = 0. It is desribedby itsgraph
(seegure1). Inthisase,thisrelationisafuntion.
Figure1: Theinput-outputrelationforequation(3)whenthereexistsabiganard.
3
Allthepreeedingtransformationswereregularwithrespettoε.Thislastoneissingularatε= 0.
4
AsolutionX˜(t, ε)goesalongtheslowurveatleaston]t1, t2[if
∀t∈]t1, t2[, lim
ε>0,ε→0
X(t, ε) =˜ φ(t)
5
AsolutionX˜(t, ε)goesalongtheslowurveexatly on]te, ts[ifit goesalongtheslowurveatleast on ]te, ts[,andifthe
interval]te, ts[ismaximalforthisproperty.
Inthisparagraph, tbeomesomplex,in the domainDt. Weassumethat forallt in Dt, thetwoeigenvalues
λ(t)andµ(t)are distint. Itis aneessaryonditiontoapplyCallot'stheoryofreliefs.
WedenethereliefsRλ andRµ by:
Fλ(t) = Z t
a
λ(t)dt , Rλ(t) =ℜ(Fλ(t)) Fµ(t) =
Z t a
µ(t)dt , Rµ(t) =ℜ(Fµ(t))
It iseasyto seethat λ(t) =µ(t),and Fλ(t) =Fµ(t), thenRλ(t) = Rµ(t). ThetwofuntionsRλ andRµ
oinideonthereal axis. WewilldenoteR(t).
Denition1 Wesay thatapath γ:s∈[0,1]7→ Dt goesdown the relief Rλ if andonly if d
dsRλ(γ(s))<0 for
allsin [0,1].
Denition2 Letus give apoint te suh that (te, φ(te),0) ∈ D. We say that Dt isa domain below te if and
only if for all t in the S-interior of Dt, there existtwo paths in Dt, from te tot,the rst one goes down the
relief Rλ andthe seondonedown Rµ.
Theorem2(Callot) Let us assume that Dt is a domain below te. A solution X(t) of equation (1) with
an initial ondition X(te) innitesimally lose to φ(te) is dened at least on the S-interior of Dt where it is
innitesimally losetoφ(t).
Letusapplythistheoremtothefollowingexample,hosenasthetypialexamplesatisfyinghypothesisH1
toH4(Hopfbifuration).
εx˙ = tx+y+εc1
εy˙ = −x+ty+εc2
(3)
Theeigenvaluesareλ=t−ietµ=t+i. ThelevelurvesofthetworeliefsRλ(t) = 12(t−i)2andRµ(t) = 12(t+i)2
aredrawnongure2.
Figure2: Thelevelurvesofthetworeliefsof equation(3),andadomainbelowte
Generially 1
thereis nosurstabilityat pointt =i(see [3℄forthedenition ofsurstability). Consequently, wehavethefollowingresultsforallequationssuhthatthereliefsareonthesametypeasthose ofgure2:
Denition3 Letusgive tc apoint2 where the eigenvaluevanishes: λ(tc) = 0. Thevalue of therelief atpoint tc isaritial valueof therelief Rλ. The bump3 istherealnumber t∗ biggerthana,minimalsuhthatRλ(t∗)
isaritial value. Theanti-bumpisthe real numbert∗∗ smallerthana,maximalsuhthatRλ(t∗∗)isaritial
value.
1
Idonotknowtheexatgenerihypothesis. Wehavetoombinetheonstraintsgivenbythesurstabilitytheoryof[3℄andthe
fatthattheequation(1)isreal
2
Insomeases,itispossiblethattcisinnite. ForεX˙ =
„ sint cost
−cost sint
«
X+O(X2) +O(ε)wehavetc= +i∞.
3
Thename"bump"isatranslationofthefrenhname"butée"
Forequation(3),thebumpist∗= 1andtheanti-bumpt∗∗=−1.
Theorem3 Atrajetoryof equation(1)an goalong the slowurve X=φ(t)exatly on]te, ts[ ifandonlyif
oneofthe following isveried:
te< t∗∗ and ts=t∗ te=t∗∗ and ts> t∗
t∗∗ < te< a and a < ts< t∗ and R(te) =R(ts)
Thistheoremisillustratedbythegraphoftheinput-outputrelation,drawnongure3.
Figure3: Theinput-outputrelationforequation(3)
3 Delayed Hopf bifuration followed by a fous-node bifuration
Thestudiedequationis
εX˙ =f(t, X, ε) (4)
wheref isanalytionadomainDofC×C2×C,andsatisesthefollowinghypothesis:
Hypothesis and notations
HFN1 Theanalytifuntionf takesrealvalueswhentheargumentsarereal.
HFN2 Theparameterεisreal, positive,innitesimal.
HFN3 There exists an analyti funtion φ, dened on aomplex domain Dt suh that f(t, φ(t),0) = 0. The
urveX =φ(t)isalledtheslowurveofequation1. WeassumethattheintersetionofDtwiththereal
axisis aninterval]tm, tM[.
HFN4 Letusdenoteλ(t)andµ(t)fortheeigenvaluesofthejaobianmatrixDXf,omputedatpoint(t, φ(t),0).
Weassumethat ,fortreal,thesignsoftherealandimaginarypartsaregivenbythetablebelow:
t tm a b tM
ℜ(λ(t)) - 0 + + +
ℜ(µ(t)) - 0 + + +
ℑ(λ(t)) - - - 0 0
ℑ(µ(t)) + + + 0 0
Then, when t inreases on the real interval ]tm, tM[, we have suesively an attrative fous, a Hopf
bifuration att =a, arepulsivefous, afous-node bifurationat t=b andarepulsivenode. At point t=b,thetwoeigenvaluesoinide. Weassumethatλ(t) =µ(t)onlyat pointb. Atually,intheomplex
plane, the two eigenvalues are the two determinations of a multiform funtion dened on a Riemann
surfaewithasquarerootsingularityatpointb.
However, there is asymmetry: if the funtion
√
is dened with a ut-o on the positivereal axis, it
satises
√s=−√
sandwethenhave
µ(t) = λ(t)
Rλ(t) =ℜ Z t
a
λ(t)dt
and Rµ(t) =ℜ Z t
a
µ(t)dt
arethetwodeterminationsofamultiformfuntionwithasquarerootsingularityatpointt=b. However,
there is a symmetry: if the funtion
√
is dened with a ut-o on the positive real axis, it satises
√s= −√
s and we havethen: Rµ(t) = Rλ(t)exept on the ut-ohalf line [b,+∞[. For real t > b,
wehoosedeterminationsofsquareroot suhthat λ(t)< µ(t). WeassumethatRλ hasauniqueritial
pointwithritialvalueRc. Weassumethat Rλ(b)< Rc. Anexampleisgivenandstudiedinparagraph
3.2.1.
3.1 Input-output funtion when there exists a big anard
Weassume now that there exists abig anard X(t)˜ i.e. a solutionof equation (1) suh that X˜(t) ≃φ(t) for
all t in the S-interior of ]tm, tM[. The study below is similar to paragraph 2.1. The added diulty is the
oinideneofthetwoeigenvaluesatpointb whihdonotallowtodiagonalizethelinearpart.
ThersthangeofunknownisX = ˜X(t) +Z whihmovesthebiganard ontheaxisX = 0: εZ˙ = A(t)Z+O(ε)Z+O(Z2) , A(t) =DXf(t, φ(t),0)
Letusdenote
α(t) β(t) γ(t) δ(t)
theoeientsofthematrixA(t). Asinparagraph2.1,thehangeofunknowns Z =
rcosθ rsinθ
, r = expρ ε
givesthenewsystem:
ρ˙ = α(t) cos2θ+ (β(t) +γ(t)) cosθsinθ+δ(t) sin2θ+O(ε) +eρεk1(r, θ, ε)
εθ˙ = γ(t) cos2θ+ (δ(t)−α(t)) cosθsinθ−β(t) sin2θ+O(ε) +eρεk2(r, θ, ε) (5)
Fornonpositiveρ(morepreisely,forinnitesimalr),theseondequationisaslow-fastequation. Itsslowurve
isgivenby
θ = arctan
δ(t)−α(t)± q
α(t)2−2α(t)δ(t) +δ(t)2+ 4β(t)γ(t) 2β(t)
Ithastwobranheswhenλandµarereals, oneisattrative,theotherisrepulsive: seegure4.
When θ goes alonga branh of theslowurve, (and when r is innitesimal),an easy omputation shows that ρ˙ is innitely lose to one of the eigenvaluesλ or µ. The repulsive branh orresponds to the smallest
eigenvalue(whih isreal positive). Whent < b, theangleθ movesinnitelyfast,and anaveragingproedure
isneededtoevaluatethevariationofρ:
hρ˙i =
Rθ1+2π θ1
˙ ρ θ˙dθ Rθ1+2π
θ1 1 θ˙dθ
Aneasyomputationshowsnowthat,intheS-interiorofthedomaint < b,ρ <0,wehave hρ˙i ≃ α(t) +δ(t)
2 = ℜ(λ(t)) =ℜ(µ(t))
Let us give an initial ondition (t, θ) between the two branhes of the slow urve and ρ negative non
innitesimal(intheexample,weantaket= 0.8,θ= 0,ρ=−0.03). Forinreasingt,theurve(t, θ(t))goes
alongtheattrativebranhoftheslowurve,whileρbelievesnegativenon innitesimal. Fordereasingt,the
solutiongoesalongtherepulsivebranh,thenθmovesinnitelyfastwhileρbelievesnegativenoninnitesimal.
Consequently, we know the variation of ρ(t) (see gure 4). As in paragraph 2.1, amore subtle argument is
neededtoprovethatwhenρbeomesinnitesimal,thevariablerbeomesnoninnitesimalandthetrajetory
X leavestheneighborhoodoftheslowurve.
Fromthisstudy, allthebehavioursof ρ(t)areknown,dependingontheinitial ondition. Theyaredrawn
ongure5.
Figure4: Oneofthetrajetoriesofsystem0.002 ˙X =
t 1 t−0.3 t
X drawn withthe variables(θ, ρ). The
slowurveis alsodrawn
Figure 5: Thepossiblebehavioursofρ(t).
Proposition 4 Let us give an equation of type (6) with hypothesis HFN1 to HFN5. Assume also that there
exists a big anard X(t)˜ going along the slow urve on the whole interval ]tm, tM[. If a trajetory X(t) goes
alongthe slow urveexatly onan interval ]te, ts[with [te, ts]⊂]tm, tM[,then Z ts
te
ℜ(λ(τ))dτ ≤ 0 ≤ Z ts
te
ℜ(µ(τ))dτ
Conversely, iftheinequalities abovearesatised,thereexistsatrajetorygoing alongthe slowurveexatly
on]te, ts[.
Theinput-outputrelationisdesribedbyitsgraph,drawnongure6.
We ouldgivemorepreise results ifweonsider the two variablesr and θ forthe input-output relation.
Indeed,whenthepoint(te, ts)isintheinteriorofthegraphoftheinput-outputrelation,weknowthat,attime ofoutput,θ isgoingalongtheattrativeslowurvewhihorrespondstotheuniquefasttrajetorytangentto
theeigenspae ofthebiggesteienvalueµ.
3.2 The fous-node bifuration is a bump
Hereisthemainpartofthisartile. Today,Iamnotabletoprovetheexpetingresults,butIhavepropositions
inthisdiretion. Toexplaintheproblem,I willgiveonjetures.
Letusdenetheanti-bump t∗∗ andthetwobumpst∗λandt∗µ asin denition3:
Rλ(tc) = Rµ(tc) = Rλ(t∗∗) = Rµ(t∗∗) = Rλ(t∗λ) = Rµ(t∗µ)