Bifurcation of a Unique Stable Periodic Orbit from a Homoclinic Orbit in Infinite-Dimensional Systems Author(s): Shui-Nee Chow and Bo Deng Source: Transactions of the American Mathematical Society, Vol. 312, No. 2 (Apr., 1989), pp. 539587 Published by: American Mathematical Society Stable URL: http://www.jstor.org/stable/2001001 Accessed: 17-01-2016 17:45 UTC Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend access to Transactions of the American Mathematical Society. http://www.jstor.org This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions TRANSACTIONS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 312, Number 2, April 1989 BIFURCATION OF A UNIQUE STABLE PERIODIC ORBIT FROM A HOMOCLINIC ORBIT IN INFINITE-DIMENSIONAL SYSTEMS SHUI-NEE CHOW AND BO DENG we showhowa uniquestablepericonditions, ABSTRACT. Undersomegeneric orbitforsemilinear froma homoclinic parabolicequaodic orbitcan bifurcate of equations.Thisis a generalization functional differential tionsand retarded a resultof Sil'nikovforordinary differential equations. 0. INTRODUCTION Consideran autonomousdifferential equationin the plane witha real parametere: (0.1) x = f(x ,e), equilibriumforall small e wherex E R . Assumethat x = 0 is a hyperbolic and Dxf(O , e) = A. Supposethat A > 0 and u < 0 are theeigenvaluesof A and (0.2) A+ , < 0. Assumethatthereexistsa homoclinicorbitof (0.1) to theorigin0 at e = 0 . conditionson equation(0.1) It is wellknownthatundercertaintransversality will fromthehomoclinicorbit an exponentially stableperiodicorbit bifurcate as the parametere changes(see Andronov,Leontovich,Gordonand Maier [1] and Chow and Hale [3], forexample,formoredetails). In [9] Neimark and Sil'nikovgeneralizedthe above resultto R3 replacingcondition(0.2) by is nontrivial an appropriateconditionon the eigenvalues.This generalization toreducethestudy sincewearenotabletomakea smoothchangeofcoordinates of local behaviorof solutionsneara hyperbolic equilibriumto thatof a linear on thelongarerelatedto somefineestimates system.In fact,all thedifficulties ofa hyperbolic timebehaviorofsolutionsin a smallneighborhood equilibrium. In [10],Sil'nikovconsideredsimilarproblemsin higherdimensions.(For some interesting applications,see, forexample,Evans,Fenicheland Feroe [4] and Feroe [5].) ReceivedbytheeditorsJune5, 1986and,in revisedform,June8, 1987and October13, 1987. SubjectClassification (1985 Revision).Primary35K22, 34K99; Secondary 1980 Mathematics 35B33,35B10. The first supported byDARPA and NSF GrantDMS 8401719. authorwas partially ? 1989 American Mathematical Society 0002-9947/89 $1.00 + $.25 per page 539 This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 540 S.-N.CHOW AND BO DENG The purposeof thispaperis to showthatSil'nikov'stheoremmaybe genersystems.We considertwokindsof infinitealizedto someinfinite-dimensional equationsand dimensionalsystems,namely,semilinearparabolicdifferential bytheseequaequations.Sincetheflowsdefined differential retarded functional tionsare semiflows (solutionsmaynotbe extendedbackwardsin time),we are forourproblems.However,thebasic notableto use Sil'nikov'smethoddirectly is of idea Sil'nikov approach based on an [10] whichreducesthebifurcation familyofmaps(see offixedpointsfora one-parameter problemto continuation (iii) in ?2 forsemilinearparabolicequationsor Lemma3.11 in ?3 forretarded cases,themain functional differential equations).As in thefinite-dimensional in theproofis to obtainlocal estimatesof solutionsof our equations difficulty equilibria.We notethatin manyaspects,ourestimatesare new nearhyperbolic cases. Our estimatesare all relatedto linearvarievenin thefinite-dimensional of the nonlinearequations.By usingthese ationalequationsalongsemiorbits of theabovementioned properties we are able to obtainsmoothness estimates, mapsand to findthefixedpoints. methods.In [2], In [2 and 12],similarresultsare obtainedbyusingdifferent BlazqueggeneralizedSil'nikov'stheoremto semilinearparabolicequationsby systems.Thisworkseemstobe usingSil'nikov'swork[10] on finite-dimensional Sil'nikov'stheoremto retardedfunctional sketchy.In [12],Walthergeneralized (inclination differential equations.Walther'sapproachis basedon the A-lemma equations. differential lemmaJforretardedfunctional We also notethatwhilewe requirethevectorfieldsto be at least C3, Walther differential needsonlyC2 vectorfieldsand obtainssimilarresultsforfunctional equations. In ?1,we considersemilinear parabolicequations.We deriveall thenecessary theoneestimatesrequiredin theproofof ourmainresult.In ?2,we construct familyof mapsand proveourmainresult(Theorem2.3). Retarded parameter differential functional equationsaretreatedin ?3. An applicationis givenin ?4. In thispaper,we willgivedetailedproofsforsemilinearparabolicequations. For retardedfunctionaldifferential equations,we will onlygiveproofswhen fromthoseforsemilinearparabolicequations. theyare different We thankthe refereeof this paper formanyveryhelpful Acknowledgment. version. revisedfromitsfirst This paperhas been substantially improvements. are due to thecarefulreadingof thereferee. Mostof theimprovements 1. LOCAL ANALYSIS In thissection,we considerthebehaviorofsolutionsofa semilinear equation on estimates are based in a Banachspace neara hyperbolic equilibrium.Our theworkof Henry[8] and a modifiedGronwall'sinequality(Lemma 1.1). The main estimateis statedin Lemma 1.3 whichgivesthe exponentialboundsin finitetimeof solutionsof thevariationalequationalonga solutionin a small of theequilibrium.We also givetheupperand lowerboundfor neighborhood This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 541 ofthehyperbolic thetimeforwhicha solutionwillstayin a fixedneighborhood equilibrium.These estimatesseem to be new even forordinarydifferential equationsin Rd. The mainpurposeofthissectionis to definea map in a small whichis closelyrelatedto theusual equilibrium ofthehyperbolic neighborhood In Lemma 1.7,we showthatthismap Poincaremap butis somewhatdifferent. of the map is is Lipschitzwitha small Lipschitzconstant.The construction based on Sil'nikov'sideas [10] and thesmoothness is based on theestimates. Let X be a Banach space withnorm I - I and A: X -? X be a linearsecan analyticsemigroup toraloperatorwithdensedomain92(A) whichgenerates > < on Aa t X. For 0 a let be the < I, {e A, 0} a-fractional powerof A the spectrum Let a(A) be .2r(Aa) . of with domain [8, pp. 24-30]) (see Henry 1, A and assumethat Re av(A)> 0 . Defineforeach 0 < a < Ixla = lAaxl. Xa= O(Aa), It is well knownthat X' is a Banach space withnorm j Ij, and X0 = X. for a > fi> 0, Xa is a densesubspaceof Xfl withcontinuous Furthermore, inclusion.Let Y1 and Y2 be Banach spaces. We denoteby L(Y1 , Y2) the Banach space of all boundedlinearoperatorsfrom Yj to Y2 equippedwith theoperatornorminducedby thenormsof Y, and Y2. of the originand e6 > 0 be Let U0 c X' x R be an open neighborhood autonomousequationwitha parameter semilinear fixed.Considerthefollowing re[-o (11)~ ix=-Ax = Ay + +-f (X,y,g), g(x,y e), (X,Iy) EUo I8E (X ,y) Y E Uo X I g: Uo X [-%,eo]o where f: Uo x [-%oIe o [-6o'80], e e [-eo 0 ] R and A > 0 is fixed. hypotheses: Let a:(A) be thespectrum of A. Considerthefollowing (HI) Rea(A) >? > A> 0, where u is a fixedconstant; (H2) f E CN[Uox [-e0,e0],X], g E C3[Uox [-e0,e0] ,R] and f(0 ?0,') = 0, D(X Y)f(0 g(0,0,?) = 0, D(XY)g(0,0,0) ' ? 0)' = 0, ' 0, [= ? E ccE [-eo I60] I o S are takenin theBanachspace Xa x R. wherethederivatives B(J1)= {(x ,y): (HI) and (H2), thereexistsa neighborhood By hypotheses IXla< 5, IyI < J3} of (0,0) containedin U0 and 0 < eI < e0 such that the local stableand unstablemanifoldsJs'Jce)and WJ'ujge)existin B(J1) theyare fore E [-eu'e], where5, > 0 is a smallconstant.Furthermore C3manifoldsand are givenby (1.2) (1.3) WIC(e)= {(x ,y) e B(51): y = hs(x ,e), IxIa < 3}1 Wc() = {(x,y) e B(t51):x = hu(y,g),IyI < J1}, This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 542 satisfying wherehs and hu are C3 in theirarguments (1.4) (1.5) 9 E[-e1,1], hu(O,e)=0, hs(O,e)=0, Dxhs(O,0)=O, Dyhu(O,O)=0. (See Henry[8, pp. 112-116].) For e E [-1, 'e1] we definea diffeomorphismH(. ,* , e) by (X,y) = H(x,y,e) = (x - hu(ye),y) (1.6) map by (1.5). Its inversemap is whichis neartheidentity (x ,y) = H_ (x 5,y,e) = (x + hu(.y e) (1.7) 8 E [-el 5 l]. yj) Let (x(t) ,y(t)) be a classicalsolutionof (1.1) (see Henry[8, pp. 53-55]) and U I(e) = H(B(c1) e) . Since y(t) is finitedimensional,H maps the solution (x(t) ,y(t)) of (1.1) intothesolution(x(t) ,y(t)) = H(x(t) ,y(t) ,e) of , x = Ax + (1.8) f (.x,y (X ,yV)E U I(9) 5 8 e), Y = Ay+g(x,y ,8), (1.8) E [-81 81] E U1(e), e[-el1], (x,y) where f (x y 8) f (1.9) H-l (_ 5y 5e) e) + Ahu(y, c) e)[Ay -Dyhu(y g(x ,y ,e) = +g(H- (x y e) )] ,y ,e) ,e). g(H-(x let (xc(t),y(t)) be a solutionof (1.8). Note that y(t) is finite Conversely, dimensional.By (1.7), (x(t) ,y(t)) = H'l(x(t) ,y(t) ,e) is a solutionof (1.1). Thus,we have shownthatequation(1.1) is conjugateto equation(1.8) via the changeof variables(1.6). Since H(. 5, ,e) is near the identitymap for all E E [-e1 ,61], without loss of generality we assumethereis a neighborhoodUI of (O,0) suchthat U1 c UI(e) foreverye E [-ci ,e]I Let 52 > 0 be so smallthat B(52) = < 2, IYI< 62} C Ul {(x, y): j&VIa Note that(1.9) impliesthat f E C2[B(2) x [-elIe] ,X]. By thedefinition of H, {(x ,y): x = 0, (x gy)E U1} is the local invariantunstablemanifold of equation(1.8) forall e e [-e1 el] Hence f(O ,y ,e) = 0 for (O ,Y) E U1 we have and e e[-e81 e1]. Therefore, f (Xx e) = P(x 9 ,E) -x where P(X ,Y ,E) = J Dff(sx ,Y ,e)ds: B(52) x [-El ,el ] ) L(Xa ,X). 2 . Notethat -PP iS C C I1 SinCe f iS C. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 543 It follows thatbyan appropriate changeofvariables wemayassumethatthe localunstable manifold of equation(1.1) is givenbythey-axisin theneighborhoodU1 of (0,0)and (1. 10) f(x,y,C) = P(x,y,) *x. However, we willlose twoderivatives in thisprocess.Thissaysthatif we P is onlyC1. assumef hastheform(1.10),thenthefunction In thefollowing, we assumef takestheform(1.10), P is C and U1= B(32). By hypothesis (H2) we choose 62 > 0, 0 < 62 < 61 and a constant C1 > 0 so thatforevery(x,y ,) E B(62) X 1-62,621,wehave (1.11) IDXhs(x ,E)j < Cixlj,,Xj (1.12) IDyhu(y,)1j* < C1IyI, (1.13) (1.14) IP(x ,y ,6)1' < Ci(IXIa+ IYI) ID(XYP(x,y,)lIa < C1, Ig(x ,y ,)j1 < C1(IxIc+ jyj)2, |D(X,y)g(x,Y,E)| < Cj(IxI + IYI). (1.15) (1.16) , there Foreach (1 , 6) E B(32) X1-62 ,621 existsa uniqueclassicalsolution (X(t; s,?,6),y(t;4, 1,6)) E B(c2), 0 < t < T < +oo , forsome T which ofconstants dependson (?, ,6) satisfying thevariation formula x(-4 j' ) e4 At eAtj7+| y(t4 e-,e) 'eX'-)(( 4 eA('-S) g(x(s; P1e) ,ys e) ,Y( y(s; ?I,e) e) ds, e) ds. Furthermore, forfixedt, x(t; ., , ): B 1-EXA X and y(t; -,*,): (-2) = X R are and the derivatives C3 D(4,)x(t ; , , 6) and p(t) 21 B(Q2) [-62 q(t) = D(4 I)y(t;?I, , E) are theuniquesolutionsof theintegralequations - (1.18) p()= e-AtpO e-A(t-s)[Dxf (x(s;,,e I~~~~~~ +| p +Dyf(x(s;4, q(t) = eA'q(?) + t eA(t-S)[Dxg(x(s.;4 ys,,r,),)p) q,tE) ,y(s;4 , Q , e) ,y(s ; + Dy g(x(s; ;, ,q,e),c)q(s)]ds, ,'I , e) ,ep(s) ?I ,E) ,y(s; ; , I ,Ec),e)q(s)] ds for0 < t < T. Notethat(p(t), q(t)) E L(Xa x R, Xax R) and (p(O), q(O))= I: X' x R- Xa x R (seeHenry(8, pp. 62-65]). This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 544 Underthe assumptionon the operatorA and its spectruma(A) we may assumethenorm I on X = X has been chosenso that (1.19) le-Atxl (1.20) le t > 0, < e-MtIXI At -xl, <t a e-t > whereCa is a constantdependingon a (see Henry[8, p. 31]). To obtaingrowthestimatesforsolutionsin B(652)x [-62 '62Y we need the following versionof Gronwall'sinequalities: Lemma1.1. Let u , a: [O , ox) constant.If O< a < 1 and [O , o) be locallyintegrable,and b be a positive 7, u(s) ds, u(t) < a(t) + b | then u(t) < a(t) + [bF(1 - a)I1/1a O < t < xo, / E/((t - s)[bF(l - a)]l/1a )a(s) ds, where 00 > 0 [bF( 1 -)l/la E (t) 1 - a) + 1) Etn(l)/v(n - n=O and F is the usual Gamma function. Proof. See Henry[8, pp. 188-189]. Remarks. (1) E'(t) 5 a) is boundedas t -- 0+. Both E'(t) (1 - a)e t t areboundedas t -? +x,which impliesEa(t)(l - a)e t tcT(l - and Ea(t) (1- a)e is boundedon [0, + o) . then (2) If a(t) a, a constant, u(t) < a(1 + E(t[b]F(1 - a)] Lemma 1.2. Let T> 0 be fixed and e E (x(t ;4 51, 5) 5Y(t ;4 )). Suppose [-62 592I 51, 5)) Ei B(J 2 for all O < t < T. Then jx(t;4 , 1 58)10, < 1410,(l +E E(0(j23t))e Pt O < t < T, where 0(3) = [C C F(l - a)J/2]1/1'. Proof.Fromthevariationof constantsformula(1.17), This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 545 Since f(x(s;E, ,c),y(s;g,r/,c),c) eX for (x(s;E,r,c),y(s;g,t,c)) B(J2), 0 < s < T, e e[-e2 ,e2j, by(1.20), le A(t.S A PX(xS;g,r1/,) < ,8) ,y(s;g 'l7-x|f(X(S;, g e 1,) 1,8 ) Y(s; 4, 1, 8) , 1) By (1.12) and (1.13), we have lf(X(S; ;, , 8) ,Y(s;g, 1, 8) , 8)1 ' C1521X(S ;g 5,6 la.e) Thus, jx(t;g,j,g)1a <e Let u(t) = emtIx(t;g, (t 1utjgla+, t --,e -j(t S) c Jl(sgCe)a , 6)Ic,. Then U + +J 1;Ca(t u(t) < u(O) 1J2u(s) ds. -S)a By Lemma 1.1, thedesiredestimateis obtained. E5 growthof Next,we willderivean estimatefortheexponential (D( ,x (t; , 1 ,6),D(e j)y (t;g 1 5)) whichis theuniquesolutionofthevariationalequation(1. 18) withinitialvalue I, theidentity map on Xa x R. By the remarksafterLemma 1.1, [1 + E (t(O2))] *(1 - a)exp(-tO(J3)) is boundedon (0, + o) . Hence,thereexist 0 < 33 < 2, a constantC2 > 2 and a fixedy1> 0 suchthat (1.21) (1.22) (1.23) /= i - A)/4- 0(3) > y1 > 0(3) > 0 forevery0 < - A)/4< ,u, + A)/2> A5 (1.24) (1+E a(t6(3)))e at<C2 , f+ (1.25) 0 <t<+Xo 0< 3 <?3, <C T e(-HhiA)sds e(-A+A)sds < C2 (1.26) f+00 (1.27) J e(-As ds < C2. In the following,A E (0, ,) is a givenconstant,and constantsatisfying ( 1.28) <33, 64< 34 > minJ3,(A- A)/(2C,), 1/(2vC1 (A/2C+1C(8C2) C~C2) 11[2C eC,C2(3 0 is a small + + 2C2) + 8C, C2]1. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 546 S.-N.CHOW AND BO DENG I L and 1 To simplify thenotation,we willuse as theoperatornormsin II I spaces L(Xa x R, Xa) and L(Xa x R, R), respectively.Also, we will use as a normin Xa x R whichis definedby (X,Y) E XAx R. Il(X,Y)IIal= IXIa + IYI, Lemma 1.3. Let A,34 be as above, and i,) be as in (1.22) and (1.23). Let T > O. Suppose (x(t; , , ,e), y(t; c, r , e)) E B(64) for O< t < T . Then < t < T, OO<t< T, 9 E [ -8'6 92 ee[-e2,eJ] x(t ;4, I , )1a< 2e-jt,', ID(<D(;^Zx(tE,r/,1a2e ID(Q7y (t; t ~,)1 < 2e Proof. Note that (D(, Z)x(t; of (1.18) such that (D(;, )x(0; it O< Z)Y(t; ,),D(, , ,, t< e E T, ,)) , , e) D(;, )y(0; ,t, [ -82'6 is theuniquesolution ,e)) = I, the identityon X' x R (see Henry[8, pp. 64-65]). Definea metricspace V = {(p ,q): and q: [O,T] x R,Xa) . L(Xa p: [O,T] L(Xa -. x R,R) are continuousin theoperatornormsI *la, I *I, respectively} with-metric d((p1 ql) (p24,2q) = max -P2(t)lI O<t<T (IpI(t) ql(t) - q2(t)l), + I and I also denote the norms in L(Xa x R, Xa) and L(Xa x R, R), respectively. Let V be thesubsetin V definedby where - V ={(p, - q): q) e V, (p(O), (p, q(O)) = I, x the identity map on Xa R, and Ip(t)Ia< 2eAt, q(t)l < 2eit, O?<t Clearly,V is a closed subsetof V. Let 1: V definedby (1.29) 'p(t) = e-Atp(O) + V, (p,q) (p,q), = be t , r (x(s; e-A(t-s)[Dxf + Dyf(x(s;, q(t) = eAtq(O)+ | - T}. eA(t' e()-,s[Dxg(x(s ;, + Dyg(x(s; . I ,8) rl, e , y(s; ;, I) 4 , e), y(s; , e) ,1, 8), 8)q(s)]ds, ,y(s;4, r, ~,6 ) , y(s;4 e)p(s) , e)p(s) ,,1 ,6) ,6)q(s)] ds. Firstwe showthat ID maps V intoitself.Becauseof (1.12), (1.14), (1.16), as in the proofof (1.19) and (p(O),q(O)) = I, by usingsimilararguments This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 547 BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT Lemma 1.2,we have e < a e -u' + a~~JO 14 Ct 2Ce-#(t-5)lx( s;qg +t =1I sIP(s)I ds -)'3C34e (t -S)a +I2 )l lq(s)lds for(p,q) E V +I3 Hence,we have Il = e -' and 3C 35e (Ip(s)Iads '2f 0o(t.s)Q f) _s)a41 =2 6CC_4~ a ( 16410(t a I )e(-"+A)('-s) =6C C164e 8,A t ds. By (1.25) we have I2 < 6CaC1C264eMt. By Lemma 1.2 and (1.24), we have t, IX(t; (1.30) Hence, I3= | e)1a < KU'(1+ Ej(t6(34)))e ft Ca 2Ce_Y(t s)jxs~j81j()d Ix(s;4,t,e)IaIq(s)Ids S)1 (t _)a (t )ae < 4CaC1C234 < 4CaCIC234eMt By (1.25) we have Therefore, 'ut< C264e f I3 < 4Ca M(ts) e(-+A)s es ds. (t - S)ae(-"+)(t-s) C2Cd4e ds t. ? (1 + CaCI C2(6 + 4C2)94)e IIY(t)Ia < 2e At Mt since 34 < 1/[CaC CC2(6 + 4C2)] by (1.28). Similarly, Iq(t)l < e+ + feA(tS)2C5 e (()2CI4+q(s)l ds = II + I2 + I3 whereIl = eAl and I2= Joe ?4C,34 ds 2C )2C54Ip(s)Ia JleA(ts)e ds es This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 548 Becauseof (1.23), (1.27) and -(A+,) 4C1 C234e I2 < I3 =|eA( 52C,641q(s)l f < 4C134 = + = e (ts)eAs 4C1(4e f ds ds e( +A)(ts)ds. By (1.26) we have I3 < 4C,C2.4eAf. Hence, Iq(t)I< (1 + 8C1C254)e < 2e (1.31) since <? 1/[8CIC2]by (1.28). It followsthat V: V -- V. Next,we willequip V withanothermetricd suchthat(V, d) is completeand themap VP:V , V mapping. underthisnewtopology(V, d) is a contraction Let (p1, ql) (P2 q2) E V and define d((pl, q ), (P2 q 2)) = Omtax (e .4 IPI(t) - P2 (t)1,} + e Atlql (t) - q2(t)D. It is easy to verifythat (V,d) is complete.To show 4): (V,d) -- (V,d) is contractive,let (pi ,qi) = F(pi ,qi), (pi),qi) E V for i = 1,2. From (1.29), we have IIPI(t) - P2(t) la < + ( tt C saPI Jt ) (s) - P2(S)lads 2C1e -( S)Ix(s; (s)- q (s)Ids I,r1,e)IaIq =IIX+ I2' where I = 3C1Cp14 = (t S)a (tCS)e 3CC14j a 105 (t < 3COC,64e-8/ IP((s) -p2(s)Iads -A S)e es(efsIlp S 1 (-2() (t1),( S)a X8 (s)-p2(s)I)ds a d (p,Q),p,2)). By (1.25) we have I < 1-3Ca CC264e 124 d((p, 1 ,ql), (p2,Pq2)q This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 549 Because of (1.27) and -,u + 2/+)A = 0 we have e )l2C, fto( = I2 )u 4q 4c ) la lq, (s) 1t-S lx(s;(-)(-ij+)s- 2CaC1C254 (I < ?2C JoCa->|(+2-+A)s (-(t+-)(ts-) (t 2 aC C24e'1 / =2C,CC26~4e (e e s)le t S)'e( ++)ds S) e +iL(ds ~ e q2 (S) I dS q(s)-q2(s)D)ds (- d((pq)(p(pqq)P2)) d((p, ,ql),~(P2,~q2). By (1.25) we have I < 2CC1C254e td((p, ql) , P2 ( q Hence, (1.32) e 'lpg (t) -gp2(t) la< CC, C2(3 + 2C2)94d((p Moreover, from (1.26) we have Iq2(t) I2(t)I < - f e (s)2CI541pI j + (s) - p2(s) la ds (s) - q2(s)I ds 2C(tS)2j>41q 0 , ql) ,(p1 q2)). I+12 where I= jeA(ts-)2Ci54Ipi(s)-p2(s)Ids 1 (eusIp (s) - p2(s)) =2CIf4 fei(t)e 2C 64 At|e(-A+A)(t-S) e (1+A)s ds ds d((p, ql, (P2 q2)) By (1.26) we have d((p, ql)) ,(P2 IIl < 2C, C264e' fj '2= (54-lq (s)-q2(s)I 0 =2Ci54 < 2C54e j(e e (e ft j e( , q2)) ds Iq,(s)-q2(s)I)ds +A)(tS)dsd((pi q1 ),(p2 q9)P By (1.26) we have I2 < 2C C254ei d((p1 ql) , (P2 , q2)). This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 550 S.-N.CHOW AND BO DENG Hence, e Atlql(t) - ,ql), (P2 , q2)) q2(t)l< 4C1C264d((p1, Therefore, from(1.32) and theestimateabovewe have d((1 1q1) 1 (g2 42)) = max(e" IP1(t) - p2(t)Iar+ e 1t1(t) - 42(t)1) < [C CIC2(3 + 2C2) + 4C1C2]54d((p ,ql) '(P2 q2)) 2 d((p, ) ql) ) (P2 ) q2)) since ?4 < 1/[CaCiC2(6+4C2)+8C1C2] by (1.28). By the contractionmapping theorem,thereexistsa unique ( E,q) e V, such that (p, q) = b(f, q). By the uniquenessof thesolutionof (1. 18), we have p (t) = D( ^q)x(t;4, I1, 8) , (t) = Dg ^Z)y(t; 41^ 8 twoestimatesof thetheoremhold: Thus,thefirst JD(<:)x(t; I, ,e) I < 2et, 0 < t < T, e E [-62 ,'62J, Next,we show C) ' Iei, ID, Y(t;( 181 O < t < T. By (1.18), we have (1.33) D17y(t; ,18) = e + |e ( [Dxg(x(s;4, r18) ,y(s;4,q^,8), E) *D,,x(s;4,r/ ,e) +Dyg(x(s;4,r1,,e),y(s;4,C *Dty(s; ,q,8)= DIy(O, ,l), E) ,j,e)]ds, 1. Let T, = sup{t: O < t < T , Aly(t;, , E) > 0}. Since D y(O0,,I,e) = 1, T1 > 0. We will show T1 = T. Suppose the contrary,T, < T, D17y(T,, ,6) = 0. By differentiating (1.33) in t, we have tl dt = [R+ Dxg(x(t; ,I ,e) ,y(t; ,I ,e) ,e)]JD,Iy(t;4,t1,e) This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 551 Thus,by (1.15), (1.16), (1.28) and (1.30), we havefor 0 < t < T1 dD?y(t;~ I,t ,) 6 dt > DA C.4Dy(; e t t4]Dly(t)- 22ClCC5e(eI+Iit 2C, I,,) - 2CIC264 > ADIy (t; 2 e(+)' It followsfromtheabove inequalityand (1.28) that D, y(t;' , 1, 8) > I eAt, Thus, D y(T1;I, 0 < t < T1. e) > Ie" T, # 0. This is a contradiction. Hence, T1 = t, and thedesiredestimateholds. o T By (1.21) and (1.24), we have (1.34) < C2 fort > 0, 0 < ?<(4. 1[1+Ea(tO(6))]e'"tj Let (1.35) Q(64, p,e) = {(x,y): (1.36) pe,e) = {(x,y): Q+(34 4/2C2, Iy- hs(x, e)l < p}, Ixa< (x,y) Ei (64,p P8) p ( 1.37) Q (64, p, e) = {(x, y): (x, y) Ei 0(64, P), ,< y -hs(xe,e) < p}, < y - hs(x, e) < - p O}, where e e [-62, 2]. Note thatby (1.11) and (1.28), 41/462 > 0 and IhS(x,,e)j< C1(5Afor IXIa? (4 and e E [-62 ,62J. Thus,forevery(x,y) E < lY- hs(x, 8)1 + Ihs(x 8)1 < 6414 (34, p ,e) with 0 < p < 64/4 - CA52 IYI Hence, Q(54, p ,e) c B((4/2) for every e E [-62 ),62J Lemma 1.4. Let (x(t; 0 < < p exists 4/4 -C1 e) T(, > ,e) ,y(t; . Then for every (, I, 9); ( , I , 9) - Y(T(4, if (4, J) _ J412 - 4roof 1fis > Proof.If (~')is if (4 412 , E Q+(44, p, e) uQ ),I hs(4 E t)jE 4 ,p ,e) U Y(%4 ,p ,e) Furthermore, T: 5e) ) 0 such that Y(T(4, T(4 bea solution of(1.1) in B(54) and ,)) , 1 l_In __ 4 __ _ _ _ Q (64 ,P, e) Q (64),P , e) -* T(4,t _ -A4 in ); (, ((4, p, e), there , 8), e) R is C2 and satisfies 1 ie) < - In - ?4 ?5 8C1 4 This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 552 S.-N.CHOW AND BO DENG then (x(t ;g, I,E) ,y(t ;g, e,E)) is not in W!jOj(e). Hence, the solution cannot stay in B(?54) for all t > 0. If 1g1< ?4/2C1, then by Lemma 1.2 forO<t< Ix(t;4, ?,e)I, <C264/2C2=?54/2 T, where T> 0 is such that the solution (x(t ; g , , e) , y(t ; ,q, e)) is in B(64) for all 0 < t < T. Hence, it has to leave BQ(4) througheither hyperplane y = ?54 or hyperplaney = -(4. Therefore, T = inf{t > 0: Myt;4 ,21 IE = 64} is well defined. Let ,I 1,E A(t ; , ,E-y(t c) y(t ;g 2 Note that A is C2. Since (g,)e JA(? Since (x(t; 1,E 0 < t < T. ;g, hs(g I,E) It) I Q(4,p,e), I = I c1h( ) I < p < 6414 - C 64 < 64/2. ,hS(4,E),E)) isin Wjsc(e),by (1.11) wehave ,hS(g,E),E),y(t; l^(T;4,^1,E)|~~~~~~~~~~~ |yT4?,)-yTgh gE,) > b-14 4/2. Hence, by the intermediatevalue theorem, T(4,,E (1.38) ) =inf{t:<t< =4/2} t<I(A(t;g,,,e)I is well definedforevery (g I1) E Q+(?4, p, E) U Q? (64I p E) . Next, we will prove that A(T(g,I, e);~ In (1.39) We only prove A(T(g,I -4/2 e)= if( othercase followsthe same argument. Suppose the contrary, Since (x(t;,hs ( ),e),E),y(t; (1.28) we have (1.40) y(T(g,t6,E);4,t6,E) ,) ,) ,hs(g,e),e)) + = E W(4 for (,)E for (0t)e~?4Q+J,p I,E); g, E) = 2/ ~~~~~~~~~~~~~/ 4 ) ) the ic h Ip I )) since = -64/2. isin WOc (e)Ithus,by(l.ll)and ,E) <-?414 Define Since (,) E Q+(4, p, E), we have d (1.;11 nE)( = But, by ( 1.40), - ha v(e > 0. ( 1.1 1 ) and ( 1.28) we have This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 553 Hence,bytheintermediate valuetheorem thereexistsTo, 0< ToT < T( ,1iE), suchthat (, d(TO; 0. = 1,e) This means (x(TO; , n,e),y(TO; ,J,8)) E Wj'(e) and,therefore, (xt; (, C, 8), Y(t; I,^1 8)) C wic(g). This contradicts(4, C) E n(64 ,P , ) - Woc(e). So (1.39) holds. Since A(t;4,t1,e) is C2 and ,0i(t ;, ?1,e) = A/\(t;,qe) ?1, + g(x(t ;, 11,e) ,y(t;, 11,e) ,e) aA by (1.15) and (1.28) we have (1.41) ~ - (t(;4'' ~ > 2 -4- C,6 >O forevery(, t) E Q+(4, p, E). By applyingtheimplicitfunctiontheoremto theequation A(T;, ,) we havethat T iS C 6412, 2 . ) E Q (64 'P'8) (4 ' 9 E [8 82 582]' Moreover,byLemma 1.3 we have 54/2= Iy(r(4, ,E);X, ,E) < 2eAT( ' I'6)l-hs( , j)I < 2e 'A Thus T(4, > ,) 1 _ _ Iln4 '6) p. _ _ 5( 1 Similarly, by usingLemma 1.3 we can show < - In T(4,ti,) I~hs(~ , e)1V Finally,let ( ), E) E (, p, e) U QY (d5p, e). Using the chain rule and differentiating (1.39) withrespectto (4, ), we obtain 0 = at (t; 4X1 e) DD(4' t=l(4,Q,6) rT(4, 15 ) + D )A(t; 45 5, e)| ,6 ) ~~~~~~~~~~~~t=t(Q, Thus,by (1.41) and Lemma 1.3 we obtain ZJ4 -8 Cl 24 The proofis complete. o The following maybe foundin Henry[8, p. 71]. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 554 S.-N.CHOWANDBO DENG Lemma1.5. Let (E ,,r)E Q(N4 Ip ,e), CE [-E2 ,C2I, T > 0 . Forequation ( 1. ) thereexistsa constantC3 independent of I,r, e2, T and 54 (see (1.28)) such thatifthesolution(x(t; , , e), y(t; g,rI,e)) is in B(54) for0 < t < T, then forevery0 < < 1I, d X(t;,i,e) E L(R,X4) and I d O<t< T, ee[-82e21 0 x(t;,I,e) Ifl? C3tal e", Remark.In particular, choose ,B= a. Then 4dX(t;* n e) <f3Le A Notethatx(.; ',ii,e) E C [(O,T),X] but x(*; ,,e) in generalif a #0. However,+x(t; boundedby C3eTP/t for 0 < t < T. , t ,e) is an element in Cl [(0, T), Xa] L(R , Xa) and Lemma1.6. Let where(,11) E Q+(4I4,pI), p < 64/4- C1(2 and E [-e3, E331.Then,there existsa constant C4((4) independent of,, I, e and &3, suchthatfor(4j I ni) E Q+ (64, p,I i1,2, IXI (4I ' nl I 8) - XI (42 ' n2 I 8) la + lyY1 (4I I 11 I 8) - y1(42 '12 < C4( 4)l(g I I'1, I2 21a + II ?12)a(Il- 8) I 21) -1 where a = min{j/PI, (u - A))/).} and l(4I I l '42''12)=max{'1i-hS(gje): Proof.For (4j I 17)E W (64 I p I E) I i = I I 2, let (1.42) g(s,e)=(l-S)l+S4,2 6(s, IE) = 71 - hs(gj I E) Then 0 < S< I I + ( I(l ,e) I ?(l IC)) = (42 ) - hs(g(s 01(s i= 1 ,2}. hs(g(s, IE) I ?) I - hs(Qj I1 I) 0 < S < . + hs (2 IE)) 5 C) I ) = n - hs(gj I,E)I 0 < S< 1 Thus, (g(s, ), I (s,I)) E Q (34 I p IE) for 0 < s < 1. Let ( 1.43)4(S, lE) = (S I ) 214 ( )1SAO( IE) + S?72- Then (s ,e) - h (s ,e)) (4 ?z(s,I) -hs(g(s I c), = ( s( h))=(1 I) ) 1)) E) 5 ?z(l IC()) = (42 I ?12)I E) =( -s)(1I, - h(S(g I e)) This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 555 Thus, 0 < 6(,)-h((,)8 0 < s < 1 , and (64, P, 9 ), (4(S, 9), ri(S,9)) E Q 0< S < Next, I 8)0 XI (42 ' 12 IXI(4, =x1X(4I' 11, a)1 ) - Xl (42,q( 1, 8, 0 + Xl (42,q( 1, 8, - Xl (42 '72 = IJ +-xT(,(s, e), q(s, e), e) ds + dX (,(s, e), q(s, e), e) ds < ||D( 5XI (4(s, + D( e), q(s, e), e) d-(4(s, e), q(s, e)) ds| (s,e) e) d-((, x ( ,(Se) ' 1 a a e), (s, e)) ds Il +12. By usingthe chainrule,Lemmas 1.3, 1.4 and 1.5 and p < 64/4 - C154, we have x JDtM (4,1,8)la _d = -X(T(4, -dt ,;8) ,,e)D(eT468 M + | C3 e!'r(~' e A54 D( ,)X(t; e C5+12 - ,17,)= , ,)| )| + l2e-jk(4sq ?,) -ln(l 18CI4 -4CI4) + 2 (4)4 h Furthermore, by (1.11), (1.42) and (1.43) we have (') , i(s e)) , = Il - 421a + IDxh (4(s, a~~~~~~~~~~~~~ e), e)l |-4(s, e) < (1 + C 54)I11 - 42Ia and |T;E(4(S, a), 6( j e) + hs172 1 )(I < 12 -III Ihs(17 t = 12 + < (1 + C {1640 _ 21 11 II _721)' This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions a S.-N.CHOW AND BO DENG 556 if we let Therefore, 4 2 1-4 )C(1+~+ ( 44 -Iln(l -4CI 64) *(AJ4- 8CI j)+2 142'12) = max{i -h (j ,e): i= 1,2} l(,,1 . C4(4) a = min{(,u- ,t/I A0 , thenwe have II <5 C4 (64) lo16(s, - e) hs(4(s,e 8, .),a ds C4(J4)I1 -hsQ(1 e .)Ia - 2 1a + I'll 11 ' 2 ' 112) (4l- <C4(054)1(4,l 721)s Similarly, '2 < C(4(34) j i6(s 8) - hs(4(s , e) , I)a ds , < C4 (04) 1 (4 ,I1 42 '12) 6l 1a + 17 (1-2 121)' Hence, '7I IXI(4Il - (42 8)XI 12 8) la s 9 11 Q42 9 112) (I4l - 2Xla +1 I'll 2C4(054)1(4l 121) Next,by ( 1.11) and ( 1.39) we have IYl(4l , 111,E)= IY(T(4l Y1 (42 412 , I ,e );41 - hs(X(T , 4/2 +Y (T(42 il I ,;e)1 e) - ,1 e )-Y(T(42 ,'12 ,e); '11,,E); 41 , hs(l = 164/2+ y(T(1 = 4C) ' ,E) ,e) 1126) 42 ,hs(42 ,hs(1 ,e) hs(X(T(2 2,112 ,e) )) e) e) n2 ,E); '2 hs(42 e)E), ,E)| <CI(41X(T(l( , n, e) ;l ,hs(4 se) se) - X (T (2 9 q2 ' );2 , h (42 E)E). we can show Similarly, IY,(4l,n 9, E- YI (42 9n2 9E) I < 2CI&4'C4((54)1(4l, 9 ' 2 '9 2)a (1, -21a + ?ln -2) Finally, if we let C4(54) = 2(1 + C154)C4(54), then we have This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 557 BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT we let (4 and p < (4/4 - C1(42 be fixedand let In thefollowing, po < min{[2a+2(1 + (1.44) B(po) = {(, S(64 E) = {(g 1): (1.45) / p/2}, CI64)C4(64)1a IJLj< p0o 1QI< po}l = h (g, E) + 412, lxle < 4121. t): 6 Since Wisoj(e)and Q(54 , p ,e) vary continuouslyin E E[-63 small 0 < E4 < 63 suchthatfor po as in (1.44) (E) 0 B(po) n WiOc (1.47) B(po) C Q(4, 'PpE), We definea map 3], thereexists 6 E [-E4 ,64],1 (1.46) 7C B(po) , 6 E [-64 ,64J. X [-64 ,641 -+ X x R by (1.48) Ir' g if (g,j) ,y1(g, ,E)), if (g,I)E{fi2(4,p,E)-Q (x1(g,6) ) (0,64/2), EfQ (4,p,E)nB(po), +(,p,e)}nB(po), where (x1(g ,1,6 ),y(g ,6 ,E)) is as in Lemma 1.6. Note thatforeach 6 E of ,* ,6) maps B(po) into S(64 ,6) as in (1.45). This definition [-64,64], ft'(4 ft' is based on Sil'nikov'sideas in [9 and 10]. Let Y beaBanach space, f: Y - Y'beamapand 1: YxY -R beareallemmaand sections,we say f is Lipschitz valued function.In thefollowing continuouswithLipschitzconstantl(y , ,Y2) if lf(YI) - PYA) < l(Y, ,YA)YI - Y21 forevery(yI, Y2) E Y x Y. We notethatthe"constant"1 is dependenton y willletus havemorepreciseestimateslater. and Y2. This definition in (g ,11,6) E B(po) x [-64 ,64] and is Lemma1.7. ft1 (g , ,e) is continuous in (g, t) E B(po) withLipschitzconstant Lipschitzcontinuous C5 (64)41( of 6 E independent [-64, 64], )a 42 ' 2 41 l where C5((4) = 2(1 + Cl (4)C4((4), l(l 'l '2 '12) = max{I i - hs (j ,e)I: i= 1 ,2}, and C4((4) and a are as in Lemma 1.6. of ft followsfromLemmas1.3 and 1.4 and Definition Proof.The continuity to showthat of ftI and Lemma 1.6,it suffices (1.48). By (1.44), thedefinition llft(gl Il ,6) - fk'(42 112 ,6)IIa < C5((54)1(4, Il ' 2 ' 12)a(,l, - 21a + Ill - This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 21) 558 S.-N.CHOW AND BO DENG - hs(4j ,) > 0 and holdsforthose ( ,i1) t and (42 ?12) satisfying AI = J. A2= 112-hS(42,) ?<0, where 11f is thenorm1a+l I in X' xR. Let 12(S) = s(hs(42 ,8) + A0) + (1 - S)n2, lift1 ( (~ 1'61) - (4'2 ( - e)j 112 )1<11 a jf (4 If 1(~'1 Because (4'21')1 ,12(l) - ft(~2 '61) + llt (422 =I+ O < s < 1. "6)Ila 8) - :f'(42 ' 2S8) 212 Il I2. ByLemma 1.6, h2(1)-hS(22,6)=A1 >0, (42 2M) E+ (64 pE) 1 < C4(t54)Q1(1 s? 1 '4 2 2) (1 - 421a+ I1l -12(1D)1) = lhSQ(4 Since 1117 112(1)l ,) - h(4he)I II by (1.11), so < C13411 -42la < ( 1 + C134)C4(34)1(1 ' 11 ' 42' 12)a _- 21a- Note that 2( )- hS(4, e) = A > 0, but 42(0) - hs(42 . e)= 12 hs(4,2 e) A2 < 0. So thereexists 0 < s < 1 such that 42(S) = hs(42 , ) and i2(s)hs(42, ) > O fors < s < I1,implying(42 62(S)) E Q+ (64 ,p P, ) fors < s <1. Hence,by Lemma 1.6, r f 72(1) 7(442 (1.49) f8) < C4(64)l(41 - f (4~21f72(S) ' 8) - 't 12(1) (R2 ,112(S) 1 ,4 }212)a1/2(1) = 1i2 ' ' - 172(S)l (, q,e) in (?7) in e E uniformly ,e)jj ? C4(54)4'1 C4(54)l1Q 8)1Ia ' 11 ' 2 , 12)a, 12(1) of for s < s < 1. By the continuity [-84 e84] and (1.46), we have IIkl('2 6 -2MI 1)a,1(,l-)[(h < (1 + C1.54)C4(34)(1 ( II'4b 212)a ( - s h(1 e)) + (ti - 121). 421a+ 1171 s(42 e) - - 112)11 By the definitionof f' in (1.45), it1(4,2 2(?), E) = fr1(42' 2, 8) = (0 . 54/2). Hence, we have '42' 112)(I, I2 < (1 + C1c54)C4(54)l(1 I1' -21a + 117- 1121). Finally, 7r ( 11l( ?1,) 8 - 7 (42 'M2 3,E)la < 2(1 + Cl(54)C4(54)l(41 ) n14294 2)(Kag- 421a+ 1ll- 1121)This proves the lemma. O 2. HoMOCLINIC BIFURCATIONS Consider the evolution equation in Xa x R with a parameter 6 E R, (2.1) *=-Ax+f(x,y,e), jI =).y+g(x,y,e). This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 559 In thissection,wewillconsiderthebehaviorofsolutionsneara homoclinic orbit of (2.1) whichis asymptotic to (x ,y) = (0,0) at e = 0. We notethatthe C3 local changeof variables(1.6) is valid locallyat theoriginforequation(2.1). Thissaysthattheestimatesobtainedin ?2 aretrueforthenewvariablesand the map ir is welldefinedin thenewvariables.In termsof originalvariables,it is notclearthatsimilarestimatesin Lemma 1.3 willhold. However,themap 7I is welldefinedin theoriginalvariablesand is Lipschitzwitha smallLipschitz constantsince 7rl is just a map and the local changeof variablesis C3 and is neartheidentity map. We also notethatthedomainof themap xl is also givenby theoriginalvariables. Considerthefollowing hypothesis: = to the (H3) At E 0, equation(2.1) has a homoclinicorbit FO asymptotic equilibrium(0,0). Remark.In fact,hypothesis (H3) impliesthat Fo belongsto Xa x R. In thissection,we considerthebehaviorof solutionsof equation(2.1) near Fo for IE? 1. To do this,we considercertainreturnmapsnear FO. (i) Construction ofthemap 7r1. Recall from?1 thatunderhypotheses (H1) and (H2), thelocal stableand unstablemanifoldsof theorigin(0, 0) existand theyare givenby ( 1.2), (1.3) in ?1: < 1I} y = hs(x ,E) l ItxIa Wju(e) = {(x ,y): x = hu(y, ?),IyI < a}' W.oc(E) = {(x,y): where sl > 0 is as in ?1 and hs and hu are C3 and satisfy(1.4), (1.5). Note thatthechangeof variablesH(. , E) of (1.6) is neartheidentity map. In the new coordinatesystemit has been shownin Lemma 1.7 thatthemap Zl: B(po) x [-64 ,E4] ) Xa x R is welldefinedby (1.46) wherepO ,64 are as in Lemma 1.7. Define (2.2) (2.3) e) = H 1(54,e) = H G(po ? E [-84 (B(po) ,), (S(54 ,e) 5E), ,84], e E [-e4 ,641, whereS(34, e) is as in (1.45). Since G(po , e) is open,thereexistsp1 satisfying (2.4) (2.5) p1 < po/(2+ C1c54), B(p,) c B(p,) c G(po00) where B(pl) is the closureof B(pl) in Xa x R. Since both ''oc(6)and in 6 E [-E64,64], thereexistsa small 0 < E5 < 64 G(po , 6) varycontinuously suchthatthecorresponding properties (1.46) and (1.47) in ?1 hold for B(p1), i.e., (2.6) (2.7) B(p1) n c(e) 5) 0 B(fp,) c G(po, E), EE [-E5 ,E5], EE [-E5,E5]. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 560 S.-N.CHOW AND BO DENG Note thatby ( 1.6) and ( 1.11), (2.8) IID(41)H( < 2+ C134, ,68)lla ( ,11 (2.9) IID(4,)H a (, ?e) < 2 + C1(54, Thus, H(., -, e) maps B(pl) into B((2 + C,54)pl) we have H(B(p) X [-65 ,E5], E B(pl) (1,8,E) E B(p) for 6 E , E) C B((2 + C,54)pl) c B(po) X[-5,eS] [-6 5, 5]. By (2.4) 6E [-E5 ,65]. DefineHxI: B(p)x[_5_,E5] Xa xR by (HxI)( , 1,?E) = (H(, ,6,E), __ and ft'x I: B(po) x [-E65, E5] Xa xR by (ft x I)(, 1, E) = ( (4 'g ) 6E) Then & : B(pl) X [-65 ,56]-1 x R is welldefined, where (2.10) 7r =H ((ft1 xI)o(HxI),.). By Lemma 1.7, (2.8) and (2.9), it is clearthat 7t is continuousand Lipschitz continuousin ( , ) witha Lipschitzconstant (2.11) (2+C134) 2C5(4)1(1 ,15 ,$2,?12) a whereC5(34), 1 and a are as in Lemma 1.7, and (4j 5qi) = H(j 5qi,'E) and (4j ' Ji) E B(pl ) i = 1 52, E E [-E65, E5] .6 Next,let p =( 0? 10) E B(pl) n rO be fixed.For every 0< p< dist(po ,OB(pj)), thedistancefromp0 to theboundaryof B(pj), thereexists0 < 66(p) < 65 suchthatfor B(p?0 P) = {(4,1): 1-~ 0 la< P,1 - 01< p} it satisfies (2.12) B(po ,p) Since H(.,,6) 6 n Wi.c(6) $ 0, 6 maps B(po ,p) into B(H( (41 ' 61 4 2 ' q2): (4i ' qi) forevery , q,o),(2+C134)p) E [-E66(p), E66(p)], sup{ E [-E6(p) ,66(p)]. =H(i 51i, ) (4i ' qi , E) E B(p ,p) x [-E6(p), E6(p)], i = 1 ,2} < 2V2(2 + C 14) p , wherefunction1 is definedin Lemma 1.7. of it' to B(p?0,p) X [-E6(p) , E6(p)] Finally,we define7r' to be therestriction where7r' is as in (2. 10). In otherwords, r' is definedby (2.14) (2.13) 1 - 1 ) 1 E 5 5E) E B(p 0(P), 6(P)] Define (2.15) p' (e) = ( '(e),6 '(E)) = H ((0,54/2) , E), 6 E [-E6(p) ,e6(p)]- It is clearthatp' (e) is theonlyintersection 6E). pointof Wj,(6) with (64/22, This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 561 Wemaysummarize theproperties of r1inthefollowing: Lemma2.1. Let r'I: B(po,p) x [-E6(p),E6(p)] Xa x R bedefined by(2.14), where0 < p < dist(p?,aB(p )) and B(po p) ,E6(p) satisfy (2.12). Then70 thefollowing: satisfies (a) Let ( ,tJ) E B(p0,p), E E [-E6(P),E6(p)J* If q > hs(@,E), then 70I(4,? ,E) is theintersection pointof L(54/2,E) withtheorbitof thesolutionof(2.1) withinitialvalue( , 1) andparameter E. On theotherhand,if rI< hS(, E), then 1'(E,i,E) = ( ((E),eq(E)), where( '(e), tjI(e)) is defined by(2.13), 1, ll,E) is Lipschitzian constant in (,,r) e B(po,p) withLipschitz (b) nr(4 in E e [-E6(p) ,E6(p)], where C76(j4)pauniformly C6 (94)= 23 /(2 + Cl 4) C5(a4) C5(34) and a areas inLemma1.7. Proof.(a) follows fromthedefinitions of ft1and nr. from(2.11) and (2.13). o (b) follows 2 01 ofthemap a . Let p and p (e) be as in Lemma2.1 (ii) Construction and(2.15)respectively. Sincep0 ,pI(0) areon J70and p1(0) E Wluo,(0), there existsT> 0 suchthat (x(T;4'(O), (0) ,0) ,y(T;l'(O) ,&(0) 0))= (4? ?,) ?p. on theinitialvaluesandparameter Bythecontinuous dependence property forsolutions ofequation(2.1) (seeHenry[8,pp. 62-70]),thereexistconstants r, > 0 and E7 > 0 suchthatthemapping (x(T; -~, ),) ,y(T;- ), ),)): B(pI(O) . r,) x [-E7 ,E87]- B(Pl) is C3, whereB(pl) is as in (2.6) and(2.7). Wedefine ,r1) X [-E7, E17]. B(pl ) by ir2=(x(T; ,,.),y(T;,-, -)). Since 2 iS C3 at (p1(0),0),thereexist 0 < r2< r1 and 0 < E8<E7 suchthat 7t2: IID forevery ( (2.16) B((p?() 7r2(1'I1 E) < ID(, qD ) 7r2(pI(0), 0)11a + 1 ) EB(pB(0),, r2) X [-e8, e8]. Let C7 = IID( e)72(PI(P(0), ,0)Il + 1. 2 maps Notethatforevery 0<r<r2 and 0<E<E8, B(pr(0),r)x[-c,e] intoB(p , C7r) and 72 is Lipschitzian withLipschitz constant C7. = 7. Construction the Let (iii) of map p' (E) (4I(e) ,6 ?I(E)) be as in (2.15). in 6 E [-E6(p) ,E6(p)] forevery Since p1(e) is continuous 0 < p < dist(po OB(p1)), This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 562 S.-N.CHOW AND BO DENG thereexists0 < e9(P) < e6(P) suchthat lXI(e) (2.17) (2.1) 1 (O)l < C6(.54)pap, - (e)-1 (O)I < C6( 4) PaP, e [-e9(p) ,89(p)] 9 E [-eg(p)e 9(p)] where86(P) , C6(54) and a are as in Lemma2.1. Let (2.18) p2 =mi 10= { [2C7C6(64)l min{68 - [2C (4)l /] ,dist(p? ,OBGoi))} e ,e9(P2)} wherer2, C7 and 68 are as in (2.16). By Lemma2.1 and (2.17), I maps B(p?,P2 x [-en ] into B(p'(0) ,2C6(54)pap9. By (1.19), 2C6(c54)pap2< r2 and e1O ?< 8 . Hence, by the constructionof r2, rc2maps B(p1 (0) ,2C6(54)p2P2) X [-e10 '610I into B(p0,2C7C6(34)ppa . By (2.18) we have 2C7C6(c54)pa< 1/2, thus,the composition (2.19) 7r(. - 8) = n2(7n(( ) ,e) it is clearfrom maps B(p0 , p2)X [-61o , 60] into B(pO,p2/2). Furthermore, Lemma 2.1 and (ii) that in is continuousin (4 ,7, ) and ir(. ,., e) is Lipschitzcontinuousfor everygiven 6 E [-e8 , 6I]. The Lipschitzconstant is C6C6(34)pa whichis less than 1/2 by (2.18) and independentof 6 E Therefore 7(, *,e) is a contractionmapping for every [-el,6 06]0. 6 E [-e,10 e 90] - We maysummarizetheproperties of in in thefollowing: Theorem 2.2. (a) in is continuous and in(., *,6) is a contraction mapping from in constantlessthan 1/2 uniformly B(p , P2) into B(p , P2) withcontraction 8 E [-610 ' 610]. (b) If (4, ) E B(pO,p2) and r > hS( , e), then r(, r,e) is on theorbit of equation(2.1) containing( , q) . If (E ,t) satisfies1 - hS( ,6 ) < 0, then 7c( ,,1 ,) isaconstant map with rq, 6, e) = 72(p (6) ,e),whereir and p1(6) are as in (ii) and (2.15). (c) B(pO P2) n B(pO, P2) n j( ,r: n) q - hs(4 ,8) > ?} 7&0 , ): q - hs(4 ,6) < O} 740 , E [-810 9 10] E [-,610 910]. of in as in (2.19). Proof. (a) followsfromtheconstruction (b) followsfromLemma2.1. (c) followsfromcondition(2.12). o This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 563 BIFURCATIONOF A UNIQUE STABLEPERIODIC ORBIT (iv) Homoclinicbifurcation theorem.Let pl(e) (2.15), and E (2-21) E [-E10,E10]. Let P2(e) (2.20) Wu(e) (e)) be as in = = =(2 (,,): (E) I 72(e))= ir2(p(e) ,e), (47,) E E [-E10 I,10] ,y(t;4OIt0 = (X (t ;4O?o~o10,) forsome (40,I170)E Wu-,,(g) I qo - hs(40 E > )) ? I t _>?} E E [-E10, E610]* is partof the globalunstablemanifoldof (0,0). Note that p2(e) E thatFo = HU(0) . B(po Ip2)n Wu(e) . We willassumewithoutlossofgenerality In thefollowing by a neighborhoodN(ro) of Fo we meanan open subset in X' x R whichcontainsFo u {(0, 0)} . W:(e) Theorem2.3. Considerthefollowing semilinearevolutionequationwitha real parameter E E [-EO egO]eE > 0: (2.22) ( (x,y,e)eX ix=-Ax+f(x,y,e), =Ay+ g(x,y,E), xRx[-eo0eo], E X X R x [-e ,E0]. (X,y,E) Suppose(2.22) satisfies: (HI) A is a sectorialoperatorwithRe (A) > ,u> A > 0; (H2) For some fixed 0 < a < I, f E C3[Xa x R x [-eo,Eo],X], 3a C [X x R x [-EO,O], R] and f(0,0,e)=0, g(0,0,e) D(X = 0, Y)f(00I0)=0I D(X y)g(050I0) ? = 0, Ee[-%o g E I80 I E E [-eo ,60I; to the orbitrO asymptotic (H3) At E = 0, equation(2.22) has a homoclinic equilibrium(0, 0). < go suchthat Thenthereexistsa neighborhood N(I0) of rO and 0 < = exists a periodic Wij(e)c N(FO) and W+(e)n WjsOj(e) 0 ifand onlyifthere orbitin N(Io) , whereW+(e) is theorbitof (2.22) through(E,, io) satisfying thisperiodic (4o I7) E Wu,U(e)and rO = W-uj(0)(see (2.21)). Furthermore, stable. orbitis uniquein N(Fo) and is exponentially orbitally asymptotically lemmas. To proveour mainresultabove,we needthefollowing Let Lemma2.4. Let p' (e) and T > 0 be as in (2.15) and (2.16), respectively. ro = {(x ,y): x = x(t;pI(0) 0) ,y = y(t;pI(0) ,0) ,0 < t < T}. Thenthere N1 and N2 of (0,0) E X" x R and FO, exist e81 > 0 and neighborhoods suchthat respectively, orbitro; (a) N(ro) = N, u N2 is a neighborhood ofthehomoclinic (b) if y isan orbitofequation(2.22) at E E [-el11 1] satisfying ynN1$ 0, y C N(FO), then l- hs(E,e) >O forevery(4,I) E yfnN1, This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 564 S.-N.CHOW AND BO DENG wherehs is themapwhosegraphis thelocalstablemanifold (see (1.2)); (c) if (e ,) EN2 and e E [-,el, then(x(t;E,j,,e), y(t;E,J,e)) E B(p , P2) forsomet > O, whereB(p , P2) is as in Theorem 2.2. Proof.Let 34 , p and Q(34 , p, e) be as in Lemma 1.4,and 64 and S(34, e) as in Lemma 1.7. Define ~N(8) = ( )23 (2.23) for8 E [-E4 , 4] . j,t): {( = , 141, < 34, 054/1 < Q = < i) IIa ?54 64/4 Then D N(E (2.24) (4P8, eE[ By Lemma 1.7, forevery(, that (x(t;4, I,6),y(t;4,J,9)) (1.6) and N, =H ) 84,8] e E [-64 N(e) D S(34, e), (2.25) hs(EIXe)< 34} - hs( ,e)} E N(e) n Ea N7(6) n 84]. ((4 , p , e) for 8 thereexistst > 0 such [-e4,84]. Let H be as in E aN7 = H' (N0 (), 0). (N(O), 0), Since H is continuousin e, thereexists0 < I< le (see Theorem2.2) such that (2.26) f~~~~~~~~~~~~~~~~~~~~ N, D H_ ()(64, p P,E) ,8) D H (B(po),8) = G(po,8) D B(p ,P2), jN, E [-ell I 911 e~~~~~~~~~~~~~~~~ < D H ((564 ,6) ,e) = 1(34 ,e), , 8 E [-el ,ell 6 [ whereB(p0), G(p0, 6), Z(34 ,6) and B(p0, P2) are as in (2.2), (2.3) and Theorem2.2. This impliesthefollowing properties: (2.27) if ( ,1t) EN, nHM'(- then (x(t; and e [-e611, 1], willleave N2 throughN7 ; ( p4, p, e),e)- Wlc(e) ,t6,6)) ,t6,e),y(t; fN, ON, is closedwithTo n (2.28) = 0. By continuousdependenceon initialvalues and parametersand property (2.28), foreveryf = (x(t ;pl (0), 0),y(t ;p' (0), 0)) E ro thereexistr = r(t) > O and E = E(t) > 0 suchthat B(p-,r) n aN, = 0 and if (4, )E B(-, F) and 6 E [-t-,iJ, then (x(T - t;4 I6 ,E) ,y(T - t;4 ,6I,E)) E B(po ,p2). Note that UO<t<T B(pf F) is an open coverof ro. Since ro is compact,there existsa finiteopen subcoverN2 = Ul7 B(fi , P,),where , fi = (X(ti;p2(0) 0) ,Y(ti;p2(0), 0)) This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT r 565 P and 0 < to < t1 < < t, < T. Without loss of generalitywe r(ti) may assume to = 0 and fp(0) = p I(0). Define eI = min0<i<n{e(t1)} and N(I0) = N1 U N2. We need to showthat N(ro) and eI satisfytherequired = properties (a), (b) and (c). It is obviousthat N(F0) is an open neighborhood of Fo U {0} . This yields (a). Since N2 nON1= 0 and ONJ c ON1c ON(FO),theboundaryof N(FO) . This and property (2.27) imply(b). (c) followsfromthedefinition of B(p-,r) . 0 Lemma2.5. Let p' (E), N2 and T be as in Lemma 2.4. Thereexist 0 < fo < and F > 0 suchthatthefollowing hold: properties 6le (a) if (4,rl,E) E B(p (0) ,)x[-Uo N2 for all 0 < t < T; fo], then(x(t; ,?I,e),y(t; ,i ,E)) E (b) if 6 E [-10, %], thenIpI(E) -p (0)ll,< F/4; (c) ifp*(E) = (e*)(, q* (E)) is theuniquefixed pointofthemap ir in Theorem 2.2 and E E [-0Eo, 50], IpI(e) - then (E() ,hs(* (E) ,e))Ila < min{C7r/4 '.P2} wheretheconstantsC7 and P2 are givenin (2.17) and (2.19), respectively. Proof.Let r(t) be as in theproofof Lemma2.4. By Lemma2.4, if ( , i?) E B(pl (O) r(O)) and E E [-11,1 1J],then (x(t;, I, E), y(t; , , E)) is defined for all 0 < t < T. We claim that thereexist 0 < 612 < I and 0 < F < r(0) suchthatforevery(, 7,t,E) E B(p (0) , F) x [-e12 ,612], (x(t ; X, r1 E ) ,y(t ; X, q, E)) E N2 forevery0 < t < T. Suppose the contrary.Then,thereexistsa sequence 1(4k 5 nk 5ek, k = 1,525 .. * . such that t WI (X (tk ; 4k 5 ?1k 5ek) 5Y (tk ; Xk 5 1k 5Ek)) EON2 , (4k ,1k ,ek) (4(0), 52(0) ,0) as k- +oo and O< tk < T forevery k = 1,2, .... Since [0, T] is compact,we mayassume tk - + to E [0, T] as k +oo. By passingto thelimits,we have r03(x(to;4 - (0) ,q (0) ,O),yt0;4 lim (x(tk;k k--*oo ,k 5 (0) 6ek), y(tk; 51 (0) 50) k 5r,k 5Ek)) E aN2. This contradictsro c N2. This yields(a). Next,since p*(E) and p1(6) are continuousin 6, thereexists 0 < e13 < 610 such thatproperties(b) and (c) hold true for E E [-e13 ,E13]. Define 0o= min{610,ell 12 '6 13}. This yieldsthedesiredresults. 0 Lemma 2.6. Let to and p*(E) be as in Lemma 2.5 and N(Fo) be the neighborhoodin Lemma 2.4. If y is a periodicor homoclinicorbitof equation (2.22) at E E [-E-,fo%]and y C N(I0), then p*(6) E y. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 566 S.-N.CHOW AND BO DENG Proof. Let y be eithera homoclinic,or periodic orbitof equation (2.22) at e E [-,o ,o] . We will show thatthe inclusion y c N(FO) implies p* (e) E y, where p* (e) is the unique fixedpoint of the map r(. , *,e) definedin Theorem 2.2. If y is a homoclinic orbit,then p I(e) E y and i(e) = 7l2(,P 8 O)E B(po, where P2 is as in (2.18) and e e [-fo, 0 Since y c N(FO), then y f N ln H-1(IQ (54, p , E) , E) = 0 by Lemma 2.4(b). By Theorem 2.2(b), (7)k (p (e) , E) is on y forevery k = O, 1, 2,... ,where (7r)k(fi(e), C)= 7((7t)kl1(p(e), E), E) is the kth iterate of 7r(. ,. , e). Since y n Wj3S(e) #&0, there exists K > 0 such that (7K) (p(E) , ) E Wlso(e) . By Theorem 2.2, i((7f) K(p(E),)C) = ( Thus, 15(e) is a fixedpoint 8) = (8) ie p(7) K+1 (p() e)= 7r2(p(e) of (7)K+1 (K *, e) . Since every fixed point of any given iteration of a contraction mappingis a fixedpoint of itself,we have p5(e) = p*(e) . Similarly,if y is a periodic orbitin N(FO), by Lemma 2.4(b), ynH1 (Q7 (454,p , e) , c) = 0. By the hyperbolicity of (O , 0) E N1, y ? N1 , i.e., ynN2 $ 0. By Lemma 2.4(c), there Since ynH 1(Q (64,p,e),e)=0,by mustbeapoint q0 EynB(p0,po). Theorem 2.2(b), qk = (7r)k+1 (q? e) is on y for every k =0, 1, 2, .... Since there K is exists > 0 with qK = q , i.e. q = (t)K(qI, y periodic, E) . This implies that ql = q*(e) by the uniqueness of the fixedpoint for (r)k( , , ) k = 1, ~2, ... .a Proof of Theorem 2.3. We show that the condition in the theorem is necessary. Suppose W<V (e) c N(I70) for some e [-to, 90], then by Lemma 2.4(c) WCu(e)n H l(Q ((54, p,8), E) =0 . We claim that p* ( ) H-'(n-(64 p5e) ). Suppose the contrary.By Theorem 2.2(b), p (e) = t(p*(e) e) = 72(p1(e), E) E Wu(e). This is a contradiction.Since p*(e) 0 H 1(Q- (4 ,p ,p) ,), by Theorem 2.2(b), 7t(p*(e) ,) = p*(e) is on an orbit y of a solution of equation (2.22). It followsthat y is a periodic orbit. Next, we need to show y c N(IFO). Let the real valued function z be as in Lemma 1.4 and T*() = T(H(p* (e) ,e) ,e). Then, by Lemma 1.4, (x(t;p*(e) ,e),y(t;p*(e),8)) E N, for 0 By (2.18), Lemma 2.1 and Lemma 2.5(b), (c) we have (2.29) < t < T*(8). l1(p*(e) ,e) -P1(O)II< 7IIr(P(*(8) ,e) --P (C)11 + Ilp1(e) -P'(O)jIl < C6(64)p min{C7r/4, P2} + t/4 < r/4 + r/4 = F/2, where C6(64) a, P2 and C7 are as in Lemma 2.1 and Theorem 2.2; r- is as in Lemma 2.5. Thus, by Lemma 2.4(c), (X(t + T* () ;p* (e) ,) ,y(t + T* () ;p* (e) ,)) E N2 This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 567 BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT for 0 < t < T. Thereforey c N(70). We notethattheperiod wo*(e)of y is o () + T. = T*( Let y We now show thatthe conditionin the theoremis also sufficient. be a periodicorbit. By Lemma 2.6, p*(C) E y. Let f(C) = it2(p (C) ,) E Wu(e)nB(po P2) be as in theproofof Lemma2.6 and T be as in Lemma1.4. Define zk p (C) z for k= 1, 2 . (2.30) (C) ~k = (ir) = T(ft(C) E [-o ) ,) ,C) E]' CE 'C), - ] [_Ce We willshowthat pf() E H1(Q+ (J45P2 ,8) '8), ee [-o ,Jo], k = 1,2,... and U O<t<T (2.31) {(X,y):X=X(t;p(C)C8)),y=y(t;p U (C) ,)} c N(FO), {(x ,y): x = x(t;fk (e) Ce)y=y(t;fk (C) ,)}cN(F), O<t<TA(e)+T 8E -oeoi]k= 152.... Note thatif (2.30) and (2.31) hold true,then,givena p E W(C), p is in either 8)),y=y(t;p (e) ,C)} c N(ro) U {(X,y):X=X(t;p1(C O<t<T or U {(X O<t<tk (e)+T ,y): x = x(t;fk () ,C) ,y = )} y(t;fk(C) -theconclusion W:(C) c N(FO) holds forsome k = 1 ,,2.. Therefore, true. Thus it suffices to show (2.30) and (2.31). Suppose (2.30) is not true, -K thenthereexists K such that f (e) E H-1(Q7 (34, p2 ,E) ,E ) U WI'c(E) and t (C) H (Q 4,4 P2, ), E) U Wis(C(), k = 1, 2,.. ,K-1 . By Theorem -r(iK((C) 5)= (7 K(e 6 ,C) = p(C). ,C) = 7((7) It followsthatp(E) is the 2.2(b), r(f (p(C) ,C))58 K+ i fixedpointof (7r) . Thus,it is thefixedpointof 7r and p(C) = p*(C) E y. -kc and proves(2.30). By Theorem2.2(b), ft(e) is on This is a contradiction Wu(e) foreveryk = 1, 2, .... By Lemma2.5, we have U {(X, y): x = x(t;p O<t<T (C), C),y = y(t;p (E), E)} c N(ro). By Theorem2.2(a) and Lemma2.5(b),(c) we have (2.32) jjtl () -P*() jj_ < (1) IIjf(E) * ()jjl This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 568 S.-N.CHOW AND BO DENG whereft*(e)= (5*(e) ,hs(s*(e),e)) E Wjsc(e).ByLemma2.1, (2.18) and (2.30) we have (2.33) 1171(-k(8) 8) e)I (f e)e (p*(g ? 2 C6(64)P211P ()-k (ek)1 *81c < C6 (34) C7 P2(1/2)ky/4< F/8 By (2.29) and (2.33) we have lit (fk (e) e)- p'(0)ll < t/2+r/8 < F. It followsthat,by Lemma2.5(a), U {(x ,y): x =x(t;k (e) e) y = y(t;pk(e),)} -k2 T (e)<-t<jk(e)+T By thedefinition of N, and Lemma 1.4 we have U {(x,y):x=x(t;pz 0<t<T (6) (e) e),y=y(t;f k(e) ,)}CNJ. Therefore (2.31) is proved.Finally,theexponentially of y asymptotic stability followsfromTheorem2.2(a). o Corollary2.7. Supposethesame hypotheses of Theorem2.3 holdforequation (2.22Y.Thenthereexistsa neighborhood N(I7o) of 1Tou {(o , 0)} in Xa x R and 0 < tf < e0 suchthatifthereis a homoclinic orbitin N(I7o) forequation(2.22) < at 1e1 u., thenthereexistno periodicorbitsin N(I0). 3. HoMOCLINIC BIFURCATION FOR RETARDED FUNCTIONAL DIFFERENTIAL EQUATIONS In thissection,we willderiveanalogousresultsof theprevioussectionsfor retardedfunctional differential ofretardedfuncequations.Forthebasistheory tionaldifferential of Hale [7]. equations,we followtheusual treatment Let Rn be the n-dimensional Euclideanrealspace,let r be a fixedpositive constantand let C be thespace of continuous from[-r , 0] into Rn functions withtheusual sup-norm1/1= sup-r<0<O 10(0) | Considerthelinearautonomousdelayequation x(t) = L(xt), 0 E C -r where I is an n x n matrixfunctionof bounded variation and x, e C with autonomousequation xt() = x(t + 0) for -r < 0 < 0 . Considertheperturbed witha realparametere E [-e0 e,0] (3.1) (3.2) L(0) = f[d?1(0)]$(0), x(t) = L(xt) + f(xt 8) , where f: Cx[-e0, e0] Rn is continuousand takesbounded setsintobounded sets, eo > 0 is a fixedconstant. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 569 Let x (q) and x (q , e) be thereal-valued solutionsof (3. 1) and (3.2) respectivelywithinitialvalue 0 E C at t = 0. Let T(t) be thesemigroup generated by the solutionxt(q) and let A be its infinitesimal withdomain generator 9(A) = { : 0 E C1 ,qS(O)= L(0q)}. The spectruma(A) of A containsonly pointspectrum and A E a(A) if and onlyif A satisfies thecharacteristic equation detA(A) = 0, where A(A) AI - |ei dtl(0). Our discussionwillbe based on thefollowing hypotheses: (H4) The characteristic equationdetA(A)= 0 has a uniquerootA > 0 which is simpleand the real partsof all othersolutionsof detA(A) = 0 are strictly less than -A; (H5) f(q, e) is C3 in f(O,e)=O, k and e and satisfies 6E[-e60,e01, D?f(O,e8)=O, e8E [-e0 ,60]. By hypothesis (H4), C is decomposedby the eigenvalueA of A as C = S E U, whereU is theone-dimensional eigenspaceof {A}A.Let qp = (p;(0)e to theeigenbe theeigenvector of A and v/' be a roweigenvector corresponding A* with thebilinear value A of the formaladjointoperator of A associated form (3.) (, q$)= (O)q$(O)- ffoc4 r?r di(()q$(c4)dc -) on [0, r] and 0 E C. It is known wherea is a 1 x n continuousrowfunction that U = {0: q e C, S= Forevery $E C, = 0$ = { 0: q EC,(yi,(A$) + ai, forsomea e R}, =0}. U with qU=(y, Note that U and S are closedsubspacesof C. definedby Let X0(0) be thematrixfunction 0, I ) E U and 0 $= 0 -u E S - r < 0 < 0, 60=0. of T(t) to includeitsactionon X0 (see [7]). X0 We mayextendthedefinition has thefollowing decomposition: This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 570 S.-N. CHOW AND BO DENG Note thatif v E Rn, then T(t)XYOuv E U forall t E R and T(t)Xsv e S for t > r. Furthermore, thereexistconstantsK1 and u > A> O suchthat ' Kle JUt losl < K e 5t IT(t)X81l t >O O,5E S, IT(t)osl t > T(t)0(0) = T(t + 0)q$(0), T(t)0(0) = O(t + 0), T(t)qU =qUeAt, -x < qu(6) =qu(O)eAO Let the solutionx(q ,) O, t + 0 > O, r < t+ 0 < t < +x, E [-r O],qu 0 O, u E U E U of (3.2) be withinitialvalue q. Then x t(q5 ,) c), with xs(q, e) E S and can be decomposed as xt(q, e) = xts(o, e) + xu(q, xtU(0,) E U. Then we have thevariationof constantsformula(see [7, pp. 143-147, 185-188]): xt (q, e) (3.6) = f T(t - a)X8 f(x,,(k, e), e) daa, T(t)/s + t xt(?> e) = e t/Y+I e(tfa)X f(xa(O ,e) , e) da. S and Pu: C -- U be the projection operators, i.e. if 0 = ,then Ps5q=qOs and Puq=$$u Since f is C3, xt(q$,e) is C3 Os+quC in (4 , e) forall t > 0 in the maximalintervalof existence.Let ys4(q,) = D?,Xts(+e) and ytu(q e) = D x u(q , e) . It followsfrom(3.6) that Let Ps: C -* , (3.7) yts(q,e) = T(t)Ps + Yu(q, e) = T(t)Pu + rt j T(t - a)X8sD,f(XQ (:, e), e) . Ya(q ,e) ea, t T(t - a)XOuDqf(xa(k , e) , e) *ya(q, e) dea. Notethatif q E C, then T(t)Pu$= eAtqU te R. By hypotheses (H4) and (H5), thereexist 6, > 0 and 0 <el < e0 such thatthe local stableand unstablemanifoldsWlsoc(e) , W"uc(e)existin B(31) for e E [-el ,el], whereB(61) = {q E C: kk51 < 15 1 < 61} (see [7, pp. 210-214]). Moreover, Wloc(e)= {O = Os+ u (3.8) )Wluoc() = {q = Os5 +?U u= hs(zs5 e) $s = h (,qu e) 1 sI < ,I,uI ?1} whereh5 and hu are C3 . hU is givenby (3.9) n This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 571 BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT e), t < 0, is theuniqueboundedsolutionof (3.2) with (q wherext4 t < 0, lx*(qU, ,e)I < K21buIe.t, for some constant K2 independentof (ou, e), and hu(OU e) = (xOu(/, ))s qu = (xoQu,u))u. Because of hypothesis(H5), we may assume for constant < k1 ?< 6 and K2 > 0 that the followingestimates also hold for kfOsI ,I e E [-el1 '61]: Ihs(s K210s 1 ~,P)l' Dphs(s ,e)- Is <K2kksIlIvsI, ID shs(OS,e) *(,s Y/seS E S, ' ) < K2IIIKI i= 1,2, y K21,u12U Ihu(hU 5e) ' lDuhu(o , 8) . /ul < K210ul ly/ul /u E U 2 v 4 EU, ID h i=1,2, operatorwithrespectto 0. whereD denotesthe kthpartialdifferentiation We willuse D1 for D 1. We havethefollowing: Proposition3.1. For every lIul 0 E [-r ,0]. < 61 and e E [-e1 ,e1], hu(q a,e)(0) is C4 in Since qU + Proof. Since oU(O) = aq#(O)eA for some a E R, qu is C'. forevery = q in manifold unstable local the element is an e) , WluJ'(e), hu(u = by the This implies q. suchthat xt(vI,e) t > 0 thereexists Y/E Wluoc(e) of thesolutionoperatorthat q(O) is C4 in 0E [-r,O]. smoothing property o Hence, hu(ou,8)(0)= 0(09(o) is C4 in 0E[-r,0]. Let lIUl< s and lei< 1. thenotation,we let x* = x (qU,e) To simplify (3.9) withrespect and hu = hu(qU , e) . Note that x*, hu e C. By differentiating to 0 E [-r, 0], we have (3.11) hu(G) = XO(O)f(x*) + L - J| d (I T(-a)Xof(x4+0 (O - a)v(0)f(x , ,e) da) e) da. Since dhu(0) is a functionin (ou ,e) withvalues in Rn, D4u dhu(0) is a continuouslinearmap fromU to Rn. On theotherhand, d h is a function in (u ,e) with values inC and D., d h is a continuous map from U into the orderof C. The followingpropositionshowsthatwe may interchange differentiations. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 572 Proposition3.2. Let qu E U, IQu < 6 (a) d E [-r,O] and d / dO u)(o) (D'b,hu , u l, (o) ) Then < < e?e1 -e E U; (b) d(D2 h2(A u (DOqh ,dU\(6) ON' (0) D;u (DTod u) hu(6))5 (du EU; /,qu (c) * 5) (0) = dO (Dfu d69hu Proof. Let fi (t) = Deu [f(x (Du h (qu * Xu=O/fT(-a)XosfL $u E U. ,e)]. Then fi (t): U-) f2(t) = D,u [f(x and bilinear. f2(t): U x U -- Rn is continuous 0 2 01 and ,e)] and linearand Rn is continuous By (3.9) we have DDuhh )( *, do2 hu(6)) (a) * (q =X|O-a)Xof2(a) u f +uE daX U u u ,q2)da u E U. -00 By direct computation, we have (3.12) H(Duhu u* and ~ ~ (0) =Xo ~ (f9) (Jo *+ L T(-a)Xosf2 (a + 0) u da) ~ - t)oi(6 )y()f(a)* -lda~,q5)d 10fA0- h H D== (dQ uE8) (X 0 q$h"())',q$ (DQ (6) =a (D0) f, (a) hu(D,/,,,2U)d(}*u XOu(0) =(D$u 7h()2uu (Dou dO 2 U X,+ -u * This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions eU BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 573 to showby To showthat (D,I,,de *XU)(0) is C1 in 0 E [-r, 0], it suffices < t Recall in 0. (3.12) that f1(t) is differentiable . f1(t) = Doff(x* , e) *Doux. Since f is C3, we onlyneed to show x* and D x * are at least C1 in t. Since x*(qu ,e) is a solutionof (3.2) and D ,ux*(ku,8) is thesolutionof the variationalequationalong x*, L(yt)+D, f(xt ,e) *ytS = whichis definedfor t < 0, we have ax at *(ru/ ,) = (X* k ,8))t - D x**Que) = (D0JC*(0U,8))t' where (x*(Ou,e))t(O) = S*(qU,e)(t + 6). Hence f(t) is C' in t < 0. By (3.12) differentiating (3.11) withrespectto 0 and qu and thendifferentiating withrespectto 0, we obtain(c). ol of Dciu,uh(6),D,uh(6) and D5 dh(0) in Remark.The smoothness properties 6 E [-r, O] are notoptimalin theabove proposition. 3.2 3.3. Underthehypotheses ofProposition Corollary d[(Do2uhu) (u 1. u)] and d[(D;u 5 hu) 0u] . K3 there existsa constant areelements in C forall 9u,qu E U. Furthermore, on ,, l, K1 and K2 suchthat depending d [(D2 hu)*(ouq02u)]< K3jql TO i qu u,qE U; u)*I1]|31OU11011, (DQdO 1,E Proof.This followsfrom(b) and (c) of Proposition3.2, (3.5) and (3.10). o u above. Notethat I I appearson theright-hand sideofthesecondinequality This is becausetheunstablemanifoldis tangentto thelinearspace U at the origin. Note thatby (3.9) we have -Idh (O) = u L(hu) + X0(O)f(hu + q$ ~) This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 574 if qu $ 0 . However,we have This saysthat hu(qU, e) 0 -(A) 3.4. Let qu E U, I uI < 3, 0 E [-r ,0] and JeI< e . Thenthere Proposition on b,, e and K1- K3 suchthat existsa constantK4 depending d (a) T(t)hUIt=+= hu- Xof(hu+ u 8); (b) IT(t)dT(T)h 2o+I < K4Iq| e-A,tt > 0 d is taken in L?. where thederivative Proof. (a) By (3.5) and (3.9) we have (3.13) T(t)h^(6) = f [xoo+ t+6 L(T(f6)Xo)dfl6f (x f [XO(O)+ f , e) da ,t +< O fo (P(t0-)NJ()(O,)a -|) T(t)hu(0) = rt+Oa_ t > 0. L(T(fl)Xo)d6] f (xo, e) da, (3.13) we have If 0 < 0, thenby (3.11) and bydifferentiating d = d hu(0) = d hu(0) - Xo()f (hu + Od T(t)hU(0) dt ~~t=O+ u e). If 0 = 0, thenby (3.19), (3.11) and (3.13) we have d T(t)hU(0) = rlin [T(t)hu(0) - hu(o)] t-O~~~~ t- = lim - | t+0+ ta = L (J [ f(xJ L(T(J)Xo)dflJ ,)da T(-a)XOf(xa e) d) dOhu(O) - XO(O)f(hu+ q$U ,e). This proves(a). (b) By Proposition3.1 and (a) above, T(t) dr T(T)hu __ is well defined. Moreover,since T(t)hUE S is continuousfor t for t > r and S is a closedsubspace, differentiable T(t)dr h dt (t)h eS, t>r. It followsfrom(a) of Proposition3.2 that T(t),d T(T)h , t > 0, This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions > 0 and BIFURCATIONOF A UNIQUE STABLEPERIODIC ORBIT 575 is C3 in (qu , c). Moreover,since D -dhU(0,e)=0?, dO E [-C1,C811' we have DollT(t) d t > 0, C E [-C1 ,C1]. = 0, T(T)hU( ,C) Therefore,by (3.10) we have Kand sup K depends on l, elC K1, and K3 . Thus, by (3.5) we have K2 < K1Ke T(t) d+ T(T)hu < +00, T(T)hu e2lr D?UT(2r)d lu t > 2r. 2 Again by (3.10) we have K= sup e28r DoT(t)TT(T)hu < +00 le1<6, 0<t<2r and K depends on T(t)d 1, el, K1, K2 < T(T)hu and K3. Thus, Ke' Define K4 = max{KK I, K}. This proves (b). Define a change of variables 0 = H(0 , s + = (3.14) 0 tlq$uj2 < t < 2r. o ) , 0 E B(31) , II < - hu(ou ,C), 1 by ?, =0 ?, It is clear from (3.10) that H is near the identityand in the new variable 1 WiU(C) = {0: E H(B(.) , qs = 0} . The inverse H of H is givenby S q U =qU = s+hUW C) J Note that since xu(q,e) is C3 in (t, ,C), t > 0, 1?k1 < 3,e < e , the composition map h (xt(q C) , ) is C3 in t and ( ,C). By this change of variables, (3.6) becomes tsW e) = T(t) (3.15) + T(t - a)Xsf(xFa, e) da + T(t)hu(0u ,C) xu($ )=eAtu +j - eA(t-a)Xu T(O)hu(Fu ,C), (x )d& This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 576 where X = XI(,E) =H(x,(q, e), e) and f is defined by f (Xa, 8) = f (H I (xaz ,e), 8). (3.16) Note thatin the new variables, xt(5, e) may not satisfya functionaldifferential discuss the behavior of equation. However,we can make such a transformation, xt(0,e) as a solution of the integralequation and then returnto xt to obtain informationabout the originalequation. By using Propositions 3.2-3.4, (3.15) can be writtenas rt -ts(+, T(t)+s + IJT(t e) == T() +1a It(k3,8) (3.17) -u xt (, e) = At u + etU 0 ft (t-a) JeA - a)g(xa e) da, , ,)a u~, XJxf(x, c) da where g is given by g(+, e) = (3.18) ~d S Xof( - - ,Pe)- -tT(t)hu(0U,e) * + X4f ( Diuh"(QU 8) [)Au =X""(0 e) + X0[f ( dhu (o o- e)D,O,,hu C) + f(hu(Ou , C) + , 8)] (o + xof( ,)[Ao ,u , 8)]. Note that X0 appears in (3.18). This says that g(q , e) has a discontinuityat = L??([-r ,0] , Rn). Since H is near the identity 0 = 0, hence g(q ) E map, by (3.10) thereexist J2> 0 and 0 < C2 < C1 such that ,E B(62) C H(B(3,),)a (3.19) ICI< C2 Therefore, g: B(62) X f-C2,621-* L? is C2. By (H5), (3.10) and (3.11), ?2 and IEI< C2. ByProposition3.2, g(q, C)(0) and D(E))g(O,C) = 0 for VkI< in 0 E -r , 0) and have a discontinuity differentiable are actually Dtg(q, c)(0) at 0 = 0 in general. However, the followinglemma says that if q E WIU0C(E), < 62 and 1C1<? C2 theng(k, e) = 0, thatis, the discontinuitywill disappear 101 on the unstable manifold. 0 < 33 < 32 and 0 < 3.5. Thereexistconstants Proposition F: B(33) x [-83 , g(,c) -* L(S , L?) C3 <C62 suchthat = F(q5,C) .4, E B(53), I anda map < E3, we have whereOs = Ps5. Furthermore, (i) F is C'; s (ii) if E S, then(F(Q,e) q s)(O) is C1 in OE [-r,0); on 33, C3 and K,- K4 suchthat Ks depending (iii) there existsa constant estimates hold: foreverykE B(3), ICl < C3 and @ E S, thefollowing This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 577 OF A UNIQUESTABLEPERIODICORBIT BIFURCATION IF(Q,e) S IL? l, KI L55IsII d < KI sup |j(F(,6). l)(6)?0Ks) 1 thatthereexist 0 < J3< J2 and 0 < 63 < 62 such Proof. We willshowfirst thatif I0I< 3,qE U and 1I1<?3, then (3.20) Recall that q 0 E [-r,0]. )(0) = 0, 9,( 6) is definedby = H(q, s - hU(oU U= eg), Assumethat II < J2 and 1el < 62. Let to= to(o ,) = sup{t > 0: Xt(O 8) E H (B(J2) g)}. thatif 32> 0 is sufficiently It is nothardto see fromthesaddlepointproperty small,then to > 2r forall 1q$< 32 and 1el<612. Since H is neartheidentity, thereexist 0 < 83 < 62 and 0 < J3< J2 suchthat (3.21) B(33) C H(B(32) I1I ? 3. ,6) e U and 161< 63. Now, we will show (3.20) holds true for all 101< 33, Suppose the contrary;thenthereexist q$ E U, J$oj< d33 I6oI? 63 and 00e [-r , 0] such that g(q$ ,o)(0o) (3.22) # 0. IXo Then 00 E Wju0c(6o)and Let 00 E H1'(qug,o). < 32. By the choice of 32' to = to(00,0o) > 2r. Hence, [-00,to - 00] is nonempty. Let Zt = H(xt(q$0,go) ,6o), - 0 < t < to- 0h.Since WOC(60)c { : 0S = Ps = 0} and 0 for -0 < t < to-6 0, x0 = H(q$ ,60) =qu. By (3.17), we have xFC 0= 100 T(t Xt+00 - ,o) a)g(xca+60 da. Let t = -00 + a and 0 < a < to with a < -00 if 00 $ 0. Since the above integralis an elementin C, by (3.5) we have 0 (3.23) -00+a = 1 = [T(- 0 + a- -I00+a 0 f g g(Fca+o T(a- d) o0)](00) ca)g(xa+o0, I e0) (a- a) da , a)[g(xF ,6o)(0)]daQ, - 00 > a> O, 00 = 0. Dividing (3.23) by a and lettinga -) 0+, we have 0= lim I -00+a [T(-0o + a - a)g(xFc0 ,60)](60) da = g(q$g ,6o)(0o) This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 578 forany -r (3.22) and proves(3.20). Therefore, 00 < 0. This contradicts < e) = g(q$S +q, g(0qe) - (I - g(0U, e) = j +ouq Lg9(aqS e)da C e) da). ~D<sg(aO4 + Define F(b, e) =f (3.24) D,sg(aqs + u , 8) da. Since D - (DOhu) . (Xo ft) + X011 whereh=hu (ku, e) and fi e , a) = Ddsf (a)s + 4u , c) , F($, e): B((3) x L[S ,L??] is C1 and [F(f,e) *y-S](Q() is C1 in 0 E [-r ,0) for [-83 ,83] (3.25) + (as =Xu ,u) - it is not hardto see from all 5 E S. This proves(i) and (ii). Furthermore, (3.25) that F(O ,e) = 0 forall IeI < e3. Thus,thereexistsK5 dependingon 62' e3 and K,-K4 suchthat JF(>, 8) - sl K-5 1?>1< 63, 1> I.s, 181 < Moreover,by Corollary3.3 thereexistsK5 dependingon suchthat - r< yo(F(+ ,e) *Vs)(0)| < K51sl , (53, E3 1eI <e3 and K,- K4 < 0, 0 forall q$I< (3, proves(iii). o SEES. '635 and Vs E S. Choose K5 = max{K5,K5}. This By (3.18), we have T(t)g(q ,e) = T(t)Xof- T(t)d T(T)hU (3.26) - T(t)D hu * ())WU + Xou ), (qu ,e). Since S is a closed subspace, (Dp hu) and hu = hu + Xouf) e S . We also have that T(t)Xosf e S for t > r. Therefore, by where f=f(,e) (,?u Proposition3.4(b), (3.5) and (3.26) thereexistsa constantK6 dependingon e and K1- K suchthat (3, for all q t> 0 jT(t)g( ,e8)j< K6e'HtkkjjlSj (3.27) E B((3) and 1Ie<I?3 Now,we rewrite(3.17) as S( ) . rt = T(t)qs + (3.28) T(t - a)F(x(qc , e), e) t Xu+ E) =eiU + e (t-a)Xuj(Xa( , e), e) * , e) da , da. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 579 It followsby differentiating (3.28) with respectto q in L?? thatthe derivatives DgxY(t,e) and D,x1't(q, e) satisfy the variational equations along U= f rt T(t - a)[D F(x(,ce),e) T(t)Ps + (ua,xS ,e)) F(x(c ,) 5c) - us ]da, ~~~~~~~~~~~~+ (3.29) 0< At =u e P +]e u u D)XJ f (Xa (Pe),) (ta )ua da, t < to 0 < t< to where ut = D,?xt(q, e) and [O , to] is a definitionintervalfor x(q+, ) . Note that even though xt, X,u and Xs are C3, we cannot differentiate(3.28) more than once because F is only Cl. This says that by using the change of variables (3.13) we have lost some differentiabilityproperties. However, the special forms in (3.28) and (3.29) will be sufficientto derive similar local results as in ?1. By using (3.28), (3.29), (3.5), (3.10), (3.27), and Propositions 3.1-3.5, the proofs of the following lemmas can be adapted from those of Lemmas 1.21.6. We will therefore omit the proofs of Lemmas 3.6-3.8 and 3.10. We will present the ptoof of Lemma 3.9 since it is differentfrom the case of semilinear parabolic equations in ? 1. on (3, e3 Lemma3.6. For equation(3.2) thereexistsa constantK7 depending and K1-K6 such thatif X E(b,e) E B(63) forall 0 < t < to whereto > 0 is then anyconstant, lt (o 5 )l < K710 le A, 0 < t < to. -~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Lemma3.7. Forequation(3.2) thereexist0 < 34 < 33 andconstantsj A, , A > O then suchthatif WXt(b,E) E B(64) for 0< t < to whereto is anyconstant, ID xt + ,E)| < 2e ID+Xu,E)1 ( ID X O0< , < 2eA [-E3,E3] 0 < t < to, E E [-E , E)<!e t < to, E E < t < to, E E [-E3,E3]. 0, Note that by the change of variables H in (3.13), the local stable and unstable manifolds Wloc(e) and Wju(Ce) are given by JIoc(e) = {1 WJ1j(e) = { 4 : qs hs(W0,E), WI < 64}, =0?I a I <34} = This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 580 whereh3 is C3, h (0,C)= 0, e - [-C3,e3] wedefine forevery ee[-E3,83] (3.30) and Dhs(O,0) = 0. As in ?1, ISi < 64/K7,I I(i I u_ o-: hs(0S,))I < p} C B(64) Q(64I P Ie) = {q Q+04 ,P I 8) ={: ,e)) < PI IXu- hs(0s5 Q(6S4,P I C) I? < (vA1 E Q2(04,P, )={ '20S4,P I8) ,-P < (,V,Xu- hs(0s,C))< 01, 0E has where (*,*) is the bilinearformgivenby (3.3). Note thatsince WlsJc(e) u U U Wjsoc(e) where Q = codimensionone and Q+n Q- = 0, Q 2(64,p,e) and &f = Q (S4 , e) (3.28) in B(34) and 0 < p < 4/4. Lemma3.8. Let (XYs(q,C)xu(f,Ce))satisfy If E Q (44, p, e) u Q- (34, p, e), thenthereexistsT = T(,C) > 0 suchthat Xu(? ,e) C) + hs(S E) - Xu (s 6412, if -b4/2, if+ E Q2 (54p,g,) I ,In 4l (5~4 -A u _ ?z S 1l In < T < #5e) I1A C3 181< 83' and II ?<3, then E)U Q-(34,p,I) if Q+$(34,p, E Furthermore, 1l 4I 181< ,P ,IC) I E fl+(64 354 1t?u-h(/5e) 32 34 -8K 24 whereK2 is givenin (3.10). o Lemma39. Let 'b E (34 ,p,I), ICI<?e3 and to> 2r be as in (3.21). There on 33, 83 and K1-K7 suchthatifa solution existsa constantK8 depending is in B(34) for 0 < t < t , thenxts(q,e) (3.28) equation e)) of (,s (4IC) I,X1u(q$ I is differentiable in t E (r , to) and satisfies d xS(k,e) < K81bje yt, dt is takenin L?' . wherethederivativedt and x = Proof. Let t > r, X1=x (,) T(t + t~~~~~~~~ xts(0) = +/ rt Q)q$s (0) + f t+O t < to, . By (3.28) we have x (,e) T(t - a + 0)[F(xca, e) *xs](0) da [F(.x,e) *xS](0+t-a)da, Since [F( C,e) *@s](0) is discontinuous only at 6 and (3.25) we have tt r< d tf( 0 e [-r, 0]. = 0 for all ViS E S, +_d This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions by(3.24) BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 581 where F(a, 6) * 6 [F(xYc,e) V-s= @ s]() -Xo0(6)j f1(x e )) d,B * s and f (q 5,e, fl) is as in (3.25). Note thatby (3.24) and (3.25), F(a , 6) is C in 0 E [-r , 0] and by Proposition 3.5(d) we have (3.31) <K TOF(a,6)*@S ?Ks IF(a,60).* I 'KsIsII f9E[-r <a?<to, I, O<a<to, ,0], 6E[-r,0], for all Vs E S. Hence d-Y (6) = L(T(t + 6)qs) + [F(Xt+05) * ot+0~~~~~~t + L[T(t+6, 0). - F(t J(0) a + 6)F(xc, 6) *xs] da + F(t, 6)*xs t+O -- / t+0 00 F(a , 6 + t - a) Xs da. a Therefore,the desired estimate follows simplyfrom (3.5), (3.31), Proposition 3.2 and Lemma 3.6. 0 Define 1: B(54) x [-c3, 3]3 (3.32) 1( , 2) , R by = max{(q' ,1u - hs(i ,))} , i= 1 ,2 where ( ,.) is as in (3.3). The followinglemma is analogous to Lemma 1.5. Lemma 3.10. Let and 6 E [-83 ,83]. ) = X,(?> ) X 1 (? Q+(54,p, ), p < 54/2 58), where qE ThereexistconstantsKg and a > 0 depending on64, 63 and K1-K8 suchthatif E1' K2e Q+(54 ,p,6 ) and ]e [-63,e3], then lX1 ($1 ,6) -X (k2 6)I < K9[l(?1 5 ?>2 ,6)]a1j, - >21 ' where1 is givenby(3.32). o Define (3.33) Wu(e) where = {q: thereexist t > 0 and V E Wuoc(6) u with (V. , hs(V,SE)) > 0 and k = nX;(54/2 1(8) = WuJ(6) X( 5,e)={q (YA $ - hs(q ,6)}, 161 ?6< 3S < ,e), ,e)) xt(q, 11?6 = ,5$ = H($, 6)}. Let 0 < pO < 64/2 be fixed. Since H is near the identitymap, there exist 0 < pi < pO and 0 < 64 < 63 such that This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 582 S.-N.CHOW AND BO DENG Let q0 E Wloc (O) withk01I of Wlsc() property I < p1 be fixed.Bythecontinuity in , for every 0 < p < dist(OB(p)I, 00), thereexists 0 < e5(p) < 64 such that n W 6c(g)7$0, 1:1_ B(Oo,p) <5(p). By usingtheabove lemmas,it is nothardto see thattheproofsof thefollowingresultsare exactlythesame as Lemma2.1, Theorem2.3, and Corollary 2.4, respectively. Lemma3.11 willbe neededin thenextsection. Lemma 3.11. Thereexistsa continuousmap satisfying r1: B(0, p)X[-65(P) 65(p)] 5, -* C 0 - Xo < p and (V.A, ou - hs(Os, e)) > 0, then ,1 (0, E) 18I< e5(P) is the intersection point of X((54/2, 6) and the orbitof thesolutionof (3.2) with initial value q and parameter 6 . On theotherhand, if (YA, Os - hs(Os ,6)) < 0, then 7rl (q) =l(e); (b) 7r ( ,6) is continuousin ( ,6) and is Lipschitzian in X for each fixed 6. Furthermore,thereexist constants a and Klo dependingon 654, 64 and K1-Kg such that (a) If 17I (ob1 5,) - 7r1(02 ,68)1 < K,opalq$ - 021 E] In thefollowing theorem, we mayassumewithoutloss of generality thatthe homoclinicorbitIF = WJu'(O). Theorem 3.12. Consider the nonlinearautonomous delay equation witha real parameter (3.34) 6 E [-o0 Xgo] 60 > x*(t) 0: = L(xt) + f(xt , 8), where L: C -* Rn is a bounded linear operatorand f: C x [-60o go1 Rn in C3. Suppose (3.34) satisfiesthefollowing: (H4) thecharacteristic equation detA(A)= 0, whereA(A) = AI-JfreAO[d(6)], has a unique positivesolution A > 0 whichis simple and the real parts of the othersolutionsof A(i) = 0 are smaller than -i; (H5) f (q, 6) satisfies f (0, 6) = 0 and D, f (0, 6) = 0, where 6 E [-6o ,g] (H6) at 6 = 0, equation (3.34) has a homoclinicorbit FO asymptoticto the equilibrium 0. Then thereexist a neighborhoodN(Io) of FO U {0} in C and 0 < 6f < 6o such that WuZ(6)c N(FO) and WuJ(6)n Wlsc() = 0 ifand onlyif thereexistsa periodicorbitin N(Io), where WuJ(6)is theorbitof(3.34) through00 satisfying 0 E Wlu4c'()and (v. , Ou - hs(Os0,6)) > 0 . Furthermore,thisperiodic orbitis stable. o unique and exponentiallyasymptotically We willonlyoutlinetheproofofthistheorem becausetheproofis exactlythe same as thatof Theorem 2.3 afterwe have establishedLemma 3.1 1. Let to > 0 be thetimesuchthat xto(I(0) 5,0)= 00 and fo = {q0:0 = xt(qI(0) ,0) , 0 < t < to} C FO. We definea familyof maps 7r2(.,) from B((0) This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 5,p) into BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 583 small e a neighborhood of 00 by 7t2(q$,e) = xt0(q, e). Then forsufficiently witha and po, thecompositionmaps 7r(.,,e)= 72 (7r1(,e) ,e) are contractive a neighborhood uniform contractive constantlessthan 1/2. Then,we construct thefollowing < co satisfying properties N(17) c C of FOU {O} and 0 < (see Lemmas2.4 and 2.5): of {0} containing (1) N(FO) = N1 U N2, where N1 is a neighborhood of FO; B(q0 ,po) and B(q$I(0) ,p1), and N2 is a neighborhood (2) if y is an orbitin C of equation (3.34) at e E [-U ,Uo] satisfying yn N1: 0 and y c N(FO), then (WA, ou - h5(ou,e)) > O forevery0 E yn N ; (3) if 0 E N2 and Iel < U0, thenthereexists t > 0 such that xt(q$,e) E B(0o Po); (4) if 0 E B((0) ,p1) and eI?< U0, thenxt( ,e) E N2 forall 0 < t < to; closeto 0) (5) 01(e) and thefixedpoint 0*(e) of 7r(-,e) are sufficiently and 00, respectively(see Lemma 2.5(b), (c)). AfterhavingdefinedN(FO) and %0,we can showthatthe conditionthat ,e)) > (e) = 0 is equivalentto (V. , *u-h W+u(e)c N(FO) and W+u(e)nWloc 0, where = (e) is thefixedpointof 7r(-,e). However,thelastcondition is equivalentto theexistenceof a uniqueperiodicorbitof equation(3.34) in property oftheperiodicorbitfollowsfromthecontraction N(Jo) . The stability property of themap 7t. ofTheorem3.12 holdforequation Corollary 3.13. Supposethesame hypotheses (3.34). Thenthereexistsa neighborhood N(FO) of FO U {0} in C and 0 < orbitin N(Fo) forequation(3.34) at fo<,e0 suchthatifthereis a homoclinic IeI< o, thenthereexistnoperiodicorbitsin N(FO) . 4. APPLICATIONS As an application,we considerthe followingfunctionaldifferential delay equation(see [6 and 11]): (4.1) x*(t)= af(x(t - 1)), x E R, where a is a real parameter.We assumethat f: R -* R is periodicwith minimalperiod -A + B, A < 0 < B, f (A) = f(B) = 0 and f'(A) = 1. Note thatif x is a solutionof (4.1), then x + n(-A + B) is a solutionof (4.1) for everyintegern. x _ A and x _ B are equilibriaof (4.1). A solutionx of solutionfromA to B if (4.1) is calleda heteroclinic lim x(t) =A t- -oo and lim x(t) = B. t-*E+oo A solutionx of (4.1) is called a periodicsolutionof thesecondkindif there existsp > 0 suchthat foralltER, x(t)=x(t+p)-(-A+B) forall0<q<pandsometER. x(t)7x(t+q)-(-A+B) This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions S.-N.CHOW AND BO DENG 584 Note thata periodicsolutionof the second kind correspondsto a periodic solutionfrom rotationof the statevariableon the circleand a heteroclinic to a homocliniccurveof thestatevariableon the circle A to B corresponds r(t) ). We are (bya homocliniccurve17(t)we mean limt-00 1(t) = limt-++00 ofperiodicsolutionsofthesecondkind interested in theproblemofbifurcation solutionfromA to B. It is obviousthatour mainresult froma heteroclinic will to thisproblem.However,a slightmodification cannotbe applieddirectly be sufficient. delayequationwitha realpadifferential Theorem4.1. Considerthefunctional rametera e R: x(t) = af(x(t - 1)). (4.2) Suppose(4.2) satisfies: (H7) f is C3, periodicwithminimalperiod -A + B, A < 0 < B, solutionxao fromA to (H8) at a = aO > 0 equation(4.1) has a heteroclinic B; equation0 = A- a e-A has onlyonepositivesolution (H9) thecharacteristic Re A < -AO solutions and all other A. satisfying AO Then thereexists a neighborhoodN(Xao) in C of U {xao} U{A} U{B} -oo<t<oo and a+ > a0 such thatif equation(4.1) has a periodicsolutionof thesecond kind x at a E (aO, a+) satisfying (4.3) XtE U{N(xao) + n(-A + B)}, nEZ t E R, stable. asymptotically thenit is uniqueand exponentially Proof.Since solutionsof 0 = A- a e i are theeigenvaluesof the infinitesicontinuoussemigroupdefinedby the linearized of the strongly mal generator equation xc(t) = aox(t - 1), by (H8) and (H9), local resultsderivedin ?3 are applicableto equation(4.2) nearthe equilibriaA and B. Let B(3) be a sufficiently of A suchthatthelocal stableand unstable smallneighborhood and Wuc(a) of A aredefinedfora > ao closeto ao. Since manifoldsWlsoc(a) has codimension one,we mayassume B(6) is dividedby Wlsoi(a)into (a) WlsJ'' two disjointparts Q+(a) and Q (a) (see (3.30)). Withoutloss of generality, a we assume x c Q+ as t -* -oo. Since lim X ao _(_A +B) = A, t-*+oo we can choose 00 E Xax- (-A + B) with40 E B(5). By (3.33) we can define q (a) E Wluoj(A) for a > aO closeto ao suchthat - 1(a)E Q+(a) is continuous in a with q1(ao) = xta forsome t < 0. By Lemma 3.11, foreverysmall p > 0, thereexist a(p) > 0 and a continuousmap 7r: B($0 , p) x [ao ,a(p)] This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 585 BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 10 -qoI the orbitof the solutionx(q , a) satisfying(a) if a E [ao,a(p)], is on (- (a), then 7r1(q , a) = q 1(a); > p and q E Q +(a), then 7r (,a) of (4.2). On the otherhand,if q E (b) 7r1(q, a) is Lipschitzian in q for each fixeda E[ao, a(p)] witha LipschitzconstantKp6, whereconstantsK and p are independentof a E [ao ,a(p)]. Let to > 0 be the time such that x (0, (ao), ao) - (-A + B) = 00 . Then, we definea map '2: B(0q (ao), pl) x small number,by B(3), where p1 and a1 are some sufficiently [ao ,a] aI small,thenthe if p0 is sufficiently 7r2(q , a) = x,( , a) - (-A + B). Therefore, composition7r = 7r(Xr (2 ,), ,) B(q0, po) x [ao , a(po)] B((0 , po) satisfies ( , a) is a contractionmap with a unithat for each fixed a E [aO , a(po)], a neighborhood constantless than 1/2. Next,we construct formcontractive {A B} satisfying and a+ N(x,a) c C of {xao: t E R} U , of A containing (1) N(xa) = N1U N1U N2, whereN1 is a neighborhood - (-A + B) EN1} B(00, p0) and B(1 (ao), p), N1 = N1+ (-A + B) ={: of {x,(1 I(ao) ,ao): of B and N2 is a neighborhood whichis a neighborhood 0 < t < to); (2) if y is an orbitin C satisfying y c U {N(ro) + n(-A - B)} uEZ and yn {N, + n(-A + B)} forsome n, then 0 - n(-A + B) E V+(a) UWjo (a) foreveryq e yn 1NI+ n(-A + B)}; (3) if 0 E N2 and a E [ao , a+], thenthereexistst > 0 suchthatx,(q, a) (-A + B) E B(qo Ipo); (4) if 0 E B(01(0) ,pl) and a E [ao,a+], thenx,(q5,a) E N2 for0 < t < to; closeto 0q1 (5) ,1(a) and thefixedpointq5 (a) of 7r(,a) aresufficiently (ao) (see Lemma2.5(b),(c)). and 00, respectively anyorbitin C of a periodic By (a) and (2) and (3) of theabove properties, (4.2) mustcontainthe fixedpoint solutionof the second kind x satisfying q5 (a) of 7r(.,a) and q5 (a) E Q+ (a). This provestheuniqueness.By (b) and theperiodicsolutionof thesecondkindis (4) and (5) of theabove properties, stable. 0 asymptotically exponentially is takenfromWalther[ 1]. The following Suppose r > 0 and 1 I2 E R satisfy A + 2r < 4j < 42 < -r < r < B - r. Let g: R -R be C' andsatisfy (i) g is periodicwithminimalperiod -A + B, (ii) g(A)=0, 0<g in (A,0), g(0)=0, g<0 in (0,B), (iii) IgI < r/2 in (A - r, A + r) U (-r,B - r), (iv) thereexistsq E (O, 1) with Ig(4) I < q - AI forall 4 E (A - r ,A + r), This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions 586 S.-N.CHOW AND BO DENG (v) a+ = g'(A) is positive with loga+ < -A(a+), where A(a+) satisfies A(a+)= a+e-A(a+) (vi) for qE C, A ?< < 1 and q(O) = 41 imply c1 -+ f01 g q(s) ds <?2 and for y E C, Consider 41 < < V ?2 and ,(O) = 2 imply B + r< * 42 +f1 gy(s) * ds. E R. (4.4) fP() = g(4)/g (A), Theorem4.2. Let f: R -* R be defined conditions by(4.4). Supposef satisfies (i)-(vi). Then (a) thereexists a nonemptyinterval [ao ,a) with a, > 0 such that if > ao a = ao, thenequation(4.1) has a heteroclinic solutionxao fromA to B, and if a E (ao , a1) thenequation(4.1) has a periodicsolutionofthesecondkind Xa. foreverya E [aO, a1 thereexistsOa E ya (theorbitin C of xa) Furthermore, suchthatOa `(ao as a- a0; (b) thecharacteristic equation0 = A- aoe has exactlyonepositivesolution AOand all othersolutionsA satisfytheinequalityRe)A< o By Theorems 4.1 and 4.2 above, we have the following: 4.3. Suppose f: R -- R is C3 and satisfies Corollary in Lemma theconditions 4.2. Thenthereexists0 < a < aO+ < a and a neighborhood N(xao) c C of {xao: t E R} U {A} U {B} such thatif a E (ao a+), thenequation(4.1) has a uniqueand exponentially asymptotically stableperiodicsolutionofthesecond kindin G= U {N(xao) + n(-A + B)}. neZ Proof. Let N(xa,) and a+ be defined as in Theorem 4.1. Without loss of generality,we may assume Oa - (-A + B) = +*(a), the unique fixedpoint of the map 7r(. , a) definedin Theorem 4.1 for every a E [ao a+). Therefore, the orbit ya of a periodic solution of the second kind xa must be in G for all a E (ao , a+). Hence, by Theorem 4.1 Xa is unique and exponentially asymptoticallystable. o , Example. Let p E (7r/2 7r) be given. Let g.(4) = p(sinw - sin(4 + w)) for 4 E R and w E (0, 7r/2). Then thereexists wo E (O , 7r/2) such that for every w E (wo 7r/2) the real numbers A = -7r - 2w, B = 7r- 2w, r = B/3, Ej = A + w, E2= -w and g.(4) satisfyhypotheses(i)-(vi) (see [6]). The correspondingequation xk(t)= af(x(t - 1)) with a > 0 and f = gWlg(A) models phase-lockedloops for the controlof high frequencygenerators(see [6 and 11]). REFERENCES 1. A. A. Andronov, E. A. Leontovich, J. I. Gordonand A. G. Maier,Theoryofbifurcations of dynamical systems on a plane,Wiley,New York,1973. 2. C. M. J.Blazqueg,Bifurcation fromhomoclinic orbitsinparabolicequations, preprint. This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions BIFURCATIONOF A UNIQUE STABLE PERIODIC ORBIT 587 3. S.-N.Chowand J.K. Hale,Methodsofbifurcation New York,1982. theory, Springer-Verlag, 4. J.W. Evans,N. Fenicheland J.A. Feroe,Doubleimpulsesolutionsin nerveaxon equations, SIAM J. Appl. Math. 42 (1982), 219-234. 5. J.A. Feroe,Temporal stability ofsolitary impulsesolutions ofa nerveequation,Biophys.J.21 (1978), 102-110. 6. T. Furomochi, Existenceofperiodicsolutionsofdimensional-delay equations, T6hokuMath. J. 30 (1978), 13-35. 7. J.K. Hale, Functionaldifferential equations,Springer-Verlag, New York,1977. 8. D. Henry,Geometric theoryofsemilinear parabolicequations,Springer-Verlag, New York, 1981. 9. Ju.I. Neimarkand L. P. Sil'nikov,A case ofgeneration ofperiodicmotions,SovietMath. Dokl. 6 (1965), 1261-1264. 10. L. P. Sil' nikov,On thegeneration ofperiodicmotion fromtrajectories doublyasymptotic toan equilibrium stateofsaddletype,Mat. Sb. 77 (1968), 427-438. 11. H.-O. Walther,Bifurcation froma heteroclinic solutionin delaydifferential equations, Trans. Amer. Math. Soc. 290 (1985), 213-233. 12. , Bifurcation fromsaddleconnection infunctional differential equations:an approachwith inclination lemmas,preprint. DEPARTMENT OF MATHEMATICS, MICHIGAN STATE UNIVERSITY, EAST LANSING, MICHIGAN 48824 Currentaddress (Shui-Nee Chow): Center for Dynamical Systems,Georgia Instituteof Technology,Atlanta,Georgia 30332 Currentaddress (Bo Deng): Departmentof Mathematicsand Statistics,Universityof NebraskaLincoln, Lincoln, Nebraska 68588 This content downloaded from 129.93.180.108 on Sun, 17 Jan 2016 17:45:33 UTC All use subject to JSTOR Terms and Conditions