UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

For Louis and Peter with admiration and gratitude.

Abstract. It is well known that a Reeb vector eld on S 3 possesses a periodic solution. Sharpening this result we shall show in this note that every Reeb vector eld X on S 3 possesses a periodic orbit which is unknotted and has selflinking number equal to ?1. If the contact form is non-degenerate, then there is even a periodic orbit P which, in addition, has an index (P ) 2 f2; 3g, and which spans an embedded disc whose interior is transversal to X . The proofs are based on a theory for partial di erential equations of Cauchy-Riemann type for maps from punctured Riemann surfaces into R S 3 , equipped with special almost complex structures related to the contact form on S 3 .

Contents

1. Introduction and results 2. Dealing with the covering-number 3. Proof for overtwisted contact structures 4. Proof for tight contact structures 5. The degenerate case References 1. Introduction and results

1 6 12 16 20 22

We consider the three sphere (S 3 ; ) equipped with a contact form . Recall that, by de nition, a contact form is a one-form having the property that ^ d is a volume-form. The two dimensional kernel, = ker TS 3, constitutes the contact structure de ned by . The restriction of d onto is non-degenerate so that ( ; d ) is a symplectic plane eld. The contact form also determines the so called Reeb vector eld X = X . It is uniquely de ned by iX d = 0 and iX = 1: (1.1) The Reeb vector eld is transversal to the contact structure , so that the tangent bundle TS 3 naturally splits: TS 3 = RX : (1.2)
1

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

If 't is the ow of X , then ('t ) = . Consequently, the linearized ow T't : TS 3 ! TS 3 leaves the splitting (1.2) invariant T't (m): m ! 't (m); (1.3) moreover, the map is symplectic with respect to d . A nowhere vanishing vector eld on S 3 need not admit periodic solutions, see 16] and 17]. However, every Reeb vector eld on S 3 possesses a periodic solution, see 10]. The aim of this note is to prove more: every Reeb vector eld on S 3 possesses a periodic orbit which is unknotted and which has some additional properties. In order to formulate the results we rst introduce some notations. Assuming S 3 to be equipped with an orientation we shall consider only contact forms satisfying ^ d > 0. Let X be the associated Reeb vector eld and = kern( ) the associated contact structure. We rst recall the concept of a selflinking number sl(x) 2 Z of an admissible loop x : S 1 ! S 3 . The loop x is called admissible if it has the following properties x is transversal to and x = y , (1.4) where y : S 1 ! S 3 is an embedded loop and : S 1 ! S 1 is a smooth map. A typical example is a periodic orbit (x; T ) of the Reeb vector eld X . In this case the loop xT , de ned by xT (e2 it ) = x(Tt) is admissible. The integer sl(x) for a given admissible loop x is now de ned as follows. We choose a smooth map u : D ! S 3 , with the disc D = fz 2 C j jz j 1g, satisfying u j @D = x; @D = S 1 : Then we choose a nowhere vanishing section Z of u . It induces a section along x nowhere tangent to x. Now we push x into the direction of Z to obtain a new loop x0 disjoint of x. The loops x and x0 have natural orientations induced from the orientation of S 1 as the boundary of D. The oriented intersection number of x0 with u will be denoted by sl(x). Since we are on S 3 , this integer does not depend on the choices involved and is called the self-linking number of the loop x. If x : R ! S 3 is a T -periodic solution of x = X (x) for some T > 0, we de ne the _ loop xT : S 1 ! S 3 by xT (e2 it ) = x(tT ): (1.5) Since xT is transversal to we can de ne the self-linking number sl(x; T ) in the obvious way. The smallest positive number T0 satisfying x(T0 ) = x(0) is called the minimal period of x. The positive integer T=T0 is called the covering number of the periodic solution (x; T ) and abbreviated in the following by cov(x; T ). The image of a periodic solution (x; T ) will be denoted by Px ; it is an embedded circle in S 3 . Px has a natural orientation and we de ne the self-linking number sl(Px) by sl(Px) = sl(x; T0 ): (1.6) A periodic orbit (x; T ) is called unknotted if T = T0 and if Px is the boundary of an embedded disc. Since in this case we can identify Px and (x; T0 ) we will just say that Px is unknotted. It will also be convenient in the following to denote by P a periodic orbit whose period is minimal. With these notations our rst result is the following.

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

Theorem 1.1. If is any contact form on S 3, the associated Reeb vector eld X possesses a periodic orbit P which is unknotted and has self-linking number sl(P ) = ?1.
We would like to mention an application to Hamiltonian vector elds XH in (R4 ; !0 ), with the standard symplectic form !0 . On a regular energy surface S = fx 2 R4 j H (x) = constg the ow lines of XH are the integral curves of the canonical line bundle kernel (!0 j S ) TS . Assuming the hypersurface S to be of contact type and di eomorphic to S 3 we conclude from theorem 1.1 that S carries an unknotted periodic orbit. We also conclude from theorem 1.1 that there are spherelike hypersurfaces in R4 on which the Hamiltonian ow cannot be conjugated to a Reeb ow. Indeed, K. Cieliebak constructed in 3] spherelike hypersurfaces S (R4 ; !0 ) whose Hamiltonian ows have only knotted periodic orbits. In his construction, the hypersurface S is even of confoliation type, that is, there exists a one-form on S satisfying !0 j S = d and ^ d 0 (in contrast to the contact condition ^ d > 0). Since every compact three manifold admits contact forms, theorem 1.1 and also theorem 1.4 below give necessary conditions for a 3-manifold to be di eomorphic to S 3 . For a characterization of the tight S 3 we refer to 11]. We next restrict the class of contact forms under consideration. We call the contact form non-degenerate if all the periodic orbits (x; T ) of the associated Reeb vector elds are non-degenerate. This requires, taking a transversal section of Px , that the linearisations of the Poincare section map and all its iterates do not contain 1 in their spectra. There are many non-degenerate contact forms as the following result from 12] shows. Proposition 1.2. Fix a contact form on S 3. There exists a dense set R C 1 (S 3 ; (0; 1)) so that for every f 2 R the contact form f is non-degenerate. With a non-degenerate periodic orbit (x; T ) of a Reeb vector eld X on S 3 we can associate another integer-valued index as follows. Again we take a disc map u : D ! S 3 extending xT and choose a nowhere vanishing section Z of u . This section can be used to de ne a symplectic trivialization of u . This way we obtain a smooth family of symplectic maps (z ): xT (z) ! C , for z 2 S 1 = @D. If 't is the ow of X , the maps ^ L(t) := T't(x(0)) j x(0) : x(0) ! x(t) (1.7) are symplectic and we de ne ^ L(t) = (e2 it )L(tT ) (1)?1; 0 t 1: (1.8) This is an arc of linear symplectic maps in C starting at the Identity at time t = 0 and ending at time t = 1 at a symplectic map which does not have 1 in its spectrum. For such arcs L(t) one can de ne a Maslov-type index as follows. Denote by the collection of all arcs of linear symplectic maps : 0; 1] ! Sp(1) starting at Id and satisfying 1 62 ( (1)). Denote by G the set of smooth arcs starting and ending at Id. The homotopy classes in G represent 1 (Sp(1)). We observe that G operates on via G ! : ( ; ) 7! ;

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

where ( )(t) = (t) (t). Given 2 we de ne the arc ?1 2 by ?1 (t) = (t)?1 . We recall that there is a natural isomorphism M : 1 (Sp(1)) ! Z, called Maslov isomorphism, mapping the loop t ! e2 it Id] onto 1. This induces a homotopy invariant map M : G ! Z. The following proposition is a special case of a result in 14]. Proposition 1.3. There exists a unique map : ! Z having the following properties: is homotopy invariant. For 2 and 2 G we have ( ) = ( ) + 2 M ( ). ( ) + ( ?1 ) = 0. ( ) = 1 for the arc (t) = e it Id; t 2 0; 1]. The index (x; T ) 2 Z of a non-degenerate periodic orbit (x; T ) of the Reeb vector eld X associated with on S 3 is de ned by (x; T ) = (L), where L 2 is the symplectic arc introduced in (1.8) above. Since we are on S 3 , the de nition does not depend on the choices involved in the construction. Recall that if T = T0 is the minimal period, we can identify (x; T0 ) and Px . Our second result is as follows. Theorem 1.4. Assume the contact form on S 3 is non-degenerate. Then the associated Reeb vector eld X possesses an unknotted periodic orbit P having the following properties. The orbit P spans an embedded disc whose interior is transversal to X . Moreover, P has self-linking number sl(P ) = ?1 and index (P ) 2 f2; 3g. If the contact form is overtwisted we nd an unknotted periodic orbit P with selflinking number ?1 and index (P ) = 2. It is an open question whether for a non-degenerate tight contact form there always exists an unknotted periodic orbit P having self-linking number ?1 and index (P ) = 3. We point out that if (P ) = 3 in theorem 1.4 and if, moreover, the period T0 is minimal among all periodic solutions of X then the periodic orbit P is the binding orbit of an open book decomposition of S 3 into special embedded planes. Every plane of this decomposition is a global surface of section for the Reeb ow on S 3 n P . It follows that the contact structure considered is tight. This is a special case of results in 11] and 12]. In order to prove theorem 1.1 and theorem 1.4 we need some results on pseudoholomorphic curves described next. We consider a compact three-manifold equipped with the contact form , associated contact structure and Reeb vector eld X . We now choose an almost complex structure J : ! compatible with the symplectic structure d on in the sense that (h; k) ! d (h; Jk) de nes an inner product on . On the four-manifold R S 3 we next de ne a special, R-invariant almost complex structure associated with and J as follows. ~ J (a; m)(h; k) = (? (m)(k); J (m) k + hX (m)); (1.9) where : TM = RX ! is the projection along X and (h; k) 2 T(a;m)(R S 3 ). A nite energy plane is a smooth map u = (a; u): C ! R M satisfying, in the ~ coordinates z = s + it, the following partial di erential equation of Cauchy-Riemann type: us + J (~)~t = 0; ~ ~u u (1.10)

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

and the energy requirement 0 < E (~) < 1. The energy is de ned by u ~ E (~) = sup u d ' : u
C

(1.11)

The one-form ' on R M is de ned by (1.12) ' (a; m)(h; k ) = '(a) (m)(k ); where ' : R ! 0; 1] is smooth and satis es '0 (s) 0 for all s 2 R. The supremum in (1.11) is taken over all such ''s. The asymptotic behaviour of a nite energy plane u(z ) as jz j ! 1 is determined by periodic solutions of the Reeb vector eld ~ X and we recall the following crucial result from 10] and 13]. Proposition 1.5. Assume u = (a; u): C ! R M is a nite energy plane. Then ~ there exists a sequence Rk ! 1 such that a(Rk e2 it ) ! 1 and lim u(Rk e2 it ) = x(Tt) as k ! 1 in C 1 (S 1 ), for a T -periodic solution x of the Reeb vector eld X on M . The period is equal to T = E (~). Moreover, if this solution is nonu degenerate, then a(Re2 it ) ! 1 and lim u(Re2 it ) = x(Tt) R!1
in C 1 (S 1 ), and the convergence is of exponential nature.

If the periodic solution x in proposition 1.5 is non-degenerate, we call it the asymptotic limit of the nite energy plane u = (a; u). In this case we can extend ~ u : C ! M over a circle at in nity to obtain a map u : D = C S 1 ! M satisfying u(e2 it ) = x(tT ) over S 1 . The map u de nes a homology class in H2 (M; Px ; Z) denoted by u]. Using a symplectic trivialization of u we can de ne the index (~) 2 Z by taking the symplectic arc associated to the periodic solution (x; T ) at u the boundary via this trivialization. This integer does not depend on the choice of the trivialization. It only depends on the homotopy class of disc maps u : D ! M satisfying u(e2 it ) = x(Tt). Therefore, on M = S 3 the integer (~) = (x; T ) does u not depend on the choice of the disc map. We should recall from 14], theorem 4.4, that (~) 2 u (1.13) for every nite energy plane u : C ! R M having a non-degenerate asymptotic ~ limit (x; T ). The next results show how the index (~) in uences the geometry of the plane u u(C ) S 3 . The results extend proposition 4.12 in 13] to multicovered asymptotic limits. Theorem 1.6. Assume u = (a; u): C ! R S 3 is an embedded nite energy plane ~ with non-degenerate asymptotic limit (x; T ). If (~) 3 then u(C ) \ Px = ; and u u : C ! M n Px is an embedding transversal to the Reeb vector eld X . Theorem 1.6 describes a situation in which the next theorem is applicable. It does not require the asymptotic limits to be non-degenerate nor simply covered. Theorem 1.7. Let u = (a; u): C ! R S 3 be a non constant nite energy plane. ~ Assume u : C ! S3

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

is an injective immersion. Then every asymptotic limit (x; T ) of u is simply covered, ~ cov(x; T ) = 1. The self-linking number of (x; T ) is sl(x; T ) = ?1. Moreover, Px = x(R) is unknotted and Px \ u(C ) = ;.

Our main result (Theorem 1.1) will follow immediately from Theorem 1.7 and Proposition 1.5 in view of the following existence result. Theorem 1.8. Let be a contact form on S 3 and J a compatible almost complex structure on the associated contact structure . Then there exists a non-constant nite energy plane u := (a; u): C ! R S 3 such that u is an injective immersion. ~ The proofs of the theorems will be based on the techniques of pseudoholomorphic curves in symplectizations of contact manifolds developped in 10]{ 15]. 2. Dealing with the covering-number The aim of this section is to prove Theorem 1.6 and Theorem 1.7 of the introduction. Theorem 2.1. Assume u := (a; u): C ! R S 3 is an embedded nite energy plane ~ with non-degenerate asymptotic limit (x; T ). If (~) 3 then u(C ) \ Px = ; and u u : C ! M n Px is an embedding transversal to the Reeb vector eld X . We begin the proof with Lemma 2.2. Assume the nite energy plane u := (a; u): C ! ~ degenerate and satis es (~) 3. If R > 0 is su ciently large, u u : C n B R ! S 3 n Px and is an embedding.
R

S 3 is non-

Proof. Setting z = e2 (s+it) we know from proposition 1.5 that u(s; t) := u(e2 (s+it) ) ! x(Tt) as s ! 1 and we can study the map u in the convenient local coordinates S 1 R2 of a tubular neighborhood of P = Px S 3 introduced in 13]. In these coordinates S 1 f0g corresponds to P and f0g R2 R R2 corresponds to the contact planes along P . For the period we have T = kT0 where k = cov(x; T ). We set the minimal period T0 = 1 for notational convenience. Working in the covering space R of S 1 = R=Z the map u : R2 ! R R R2 has the following presentation ~ in these local coordinates, where s s0 is large: u(s; t) = (a(s; t); #(s; t); z (s; t)) ~ (2.1) and a(s; t) = s + a0 + (s; t) #(s; t) = kt + (s; t) (2.2) Rs ( )d e(t) + r(s; t) : z (s; t) = e s0 The functions ; and r are periodic in t of period 1. Moreover, as s ! 1, r(s; t) ! 0 with all its derivatives, uniformly in t. The functions and converge exponentially fast to zero together with all their derivatives. In addition, (s) ! , where is a negative eigenvalue with normalized eigenfunction e(t) = e(t +1) 2 R2 _ of a linear selfadjoint operator. It is of the form h 7! ?J (t)h ? A(t)h in L2 (S 1 ; R2 )

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

on the domain H 1;2 (S 1 ; R2 ). Clearly e(t) 6= 0. The matrix functions A(t) and 1 J (t) are periodic in t with period k . In abstract terms one may consider e as a distinguished section of xT so that e(t) 2 x(kt) . For details we refer to 13]. Considering the eigenfunction e(t) we assume that 1 j k is the smallest integer satisfying and show that for an integer l and
?j e(0) = e k

for some > 0

(2.3) (2.4)

j 1 k=l e(t) = e t + 1 ; l
?

(2.5) Since e(t) is an eigenfunction of a rst order di erential equation whose coe cients j j 1 are k -periodic we conclude form (2.3) that e(t) = e( k + t) so that e(t) = m e(m k + t) for every integer m and t 2 R. Setting m = k and recalling e(t + 1) = e(t) we conclude from e(t) = k e(t) that = 1. It remains to verify that k is an integer. j j r Arguing by contradiction we assume k = l + r and 1 r j ? 1. Then l k = 1 ? k j j and r r j e(t) = e(l k + t) = e(1 ? k + t) = e(? k + t): r r Setting t = k we obtain e(0) = e( k ) contradicting the choice of the integer j , which is the smallest such integer. Hence (2.4) and (2.5) are proved. We next show that the assumption (~) 3 implies that l = 1. We recall the u de nition of the winding number wind1 (~) 2 Z associated with a non-degenerate u nite energy plane u, introduced in 14]. Recalling that TS 3 = RX we take the ~ tubular neighborhood V = (U ) of the periodic orbit P , where is a di eomorphism :U j P ! V = (U ) S 3 (2.6) of a neighborhood U of the zero section of j P satisfying, at every p 2 P , that (0p ) = p. Moreover, the brewise derivative of at 0p is the inclusion of p into TpS 3 . For large s we can represent u(s; t) uniquely as u(s; t) = (x(k (s; t)); w(s; t)); (2.7) with w(s; t) 2 x(k (s;t)) . The function satis es (s; t + 1) = (s; t) + 1 and (s; t) ! t as s ! 1. We obtain the distinguished nowhere vanishing section v of xT by w(s; t) v(t) = slim jw(s; t)j 2 x(kt) : (2.8) !1 Let D = fz j jz j 1g and choose a disc map ' : D ! S 3 satisfying '(e2 it ) = x(kt). Then choose a nowhere vanishing section Z of ' so that 0 6= Z (z ) 2 '(z) . The winding number of u is de ned as ~ ? wind1 (~) = wind v j 0; 1]; Z j 0; 1] = wind(f ): u (2.9) Here, wind(f ) 2 Z is the winding number of the function f : S 1 ! C n f0g de ned by v(t) = f (t) Z (t) 2 x(kt) , where 0 t 1. The complex multiplication in is

t 2 R:

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

de ned by the almost complex structure J . In view of its homotopy invariance, the integer wind1 (~) does not depend on the choice of the non-vanishing section Z of u ' , and since we are on S 3 it is also independent of the choice of the disc map ' satisfying '(e2 it ) = x(kt). In particular, we can homotope ' to a special disc map given by (z ) = 0 (z k ), where 0 : D ! S 3 satis es 0 (e2 it ) = x(t). This way 1 we nd a section Z of satisfying, in addition, Z (t) = Z (t + k ). Observe that, in the local coordinates (2.2), we have v(t) = (t)e(t) with 0 6= (t) 2 R. Since 1 wind(f ) = wind( f ) we conclude from e(t + 1 ) = e(t) and 1 = j k that l l ? wind(v j 0; 1]; Z j 0; 1]) = l wind v j 0; 1 ]; Z j 0; 1 ] : (2.10) The invariants wind1 (~) and (~) can be compared with each other in a symplectic u u 1 u trivialisation of ' . One nds 0 wind1 (~) ? 1 2 (~) ? 1, see 14]. Conseu quently, the assumption (~) 3 implies wind1 (~) = 1 and we conclude that both u u integers on the right hand side of (2.10) are equal to 1. In particular, l = 1 and we have proved that, if (~) 3, then u

j e(t + k ) 6= e(t) for all t 2 R, > 0 and 1 j k ? 1.

(2.11)

From the asymptotic formula (2.2) one easily deduces for large R that u(C n BR ) \ P = ; and u : C n BR ! S 3 n P is an immersion, we refer to 13] for a proof. We shall show that u j C n BR is injective. Arguing by contradiction we assume u(sj ; tj ) = u(s0j ; t0j ) (2.12)

j Since z (sj ; tj ) = z (s0j ; t0j ) we conclude from (2.2), as j ! 1, that e(t ) = e(t + k ), for some > 0. Consequently, by (2.11), we nd j = 0 and = 1 so that t = t0 . Taking norms, kz (sj ; tj )k = kz (s0j ; t0j )k, we deduce from (2.2) that sj ? s0j ! 0. Summarizing, sj ? s0j ! 0 and tj ? t0j ! 0: (2.14) Since u is an immersion one concludes from (2.2), (2.14) and u(sj ; tj ) = u(s0j ; t0j ) that (sj ; tj ) = (s0j ; t0j ), if j is large, contradicting (2.13). We have proved that u : C n BR ! S 3 n P is an injective immersion for R su ciently large. In view of the asymptotic behaviour (2.2) it must be an embedding. The proof of Lemma 2.2 is complete. Lemma 2.3. Assume u := (a; u): C ! R S 3 is an injective, non-degenerate ~ nite energy plane satisfying (~) 3. Then u uc(C ) \ u(C ) = ;; ~ ~ c 6= 0; where uc(z ) = (a(z ) + c; u(z )). ~

for sequences satisfying (sj ; tj ) 6= (s0j ; t0j ) and sj ! 1; s0j ! 1: (2.13) In view of the periodicity in t we may assume tj ! t 2 0; 1) and t0j ! t0 2 0; 1). Since #(sj ; tj ) = #(s0j ; t0j ) mod 1 we deduce from (2.2), as j ! 1, that k(t ? t0 ) = 0 mod 1 so that j t ? t0 = k and j 2 f0; 1; : : : ; k ? 1g:

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

Proof. Given 0 < r0 < r1 there exists an R1 > 0 such that for all R R1 the following holds true. If z; z 0 2 C satisfy jz j = R, jz 0j R, then uc(z ) 6= u(z 0) for ~ ~ every c 2 R in r0 jcj r1 . Indeed, arguing indirectly we nd a sequence cj in r0 jcj j r1 , sequences zj ; zj0 with jzj0 j jzj j and jzj j ! 1 satsifying ~ (2.15) ucj (zj ) = u(zj0 ): ~ Outside a large ball BR0 the map u is an embedding, in view of Lemma 2.2. Therefore, we may assume that jzj0 j R0 . Recall a(z ) ! 1 as jz j ! 1. Since jzj j ! 1 and cj is bounded we conclude from (2.15) that a(zj0 ) = a(zj ) + cj ! 1. This, 0 however, is not possible since the sequence zj is bounded, and the claim is proved. As a consequence, there is a well de ned intersection number int(~c ; u) if c 6= 0. u ~ It is de ned by restricting the maps to su ciently large balls. By the excision and the homotopy property of the intersection number, the integer int(~c ; u) is u ~ independent of c for c 6= 0. By the positivity of the local intersection numbers of pseudoholomorphic curves, int(~c ; u) 0. Moreover, uc(C ) \ u(C ) = ; for c 6= 0 u ~ ~ ~ if and only if int(~c; u) = 0. We shall prove that int(~c ; u) = 0. Arguing by u ~ u ~ contradiction we nd a sequence cj ! 0, cj 6= 0 and sequences zj ; zj0 such that (2.16) ~ ucj (zj ) = u(zj0 ): ~ Since outside a ball u is an embedding (Lemma 2.2) we may assume taking a 0 u suitable subsequence, that zj0 ! z0 . Since (~) 3, the map u : C ! S 3 is an 0 immersion in view of Corollary 4.2 in 13]. Therefore, jz0 ? zj j " if j is large, for a suitable " > 0. Since a(zj ) = a(zj ) ? cj is bounded, also the sequence zj must be bounded. Hence, for a subsequence, zj ! z0 , so that by (2.16) u(z0 ) = u(z0 ). Since ~ ~ 0 0 this contradicts our assumption that u is injective. The proof of Lemma z0 6= z0 ~ 2.3 is complete.

In order to complete the proof of theorem 2.1 we observe that Lemma 2.3 implies that the map u : C ! S 3 is injective. We have seen that the assumption (~) 3 implies wind1 (~) = 1 and conclude, using Corollary 4.2 in 14], that u u dimR Tu(z )(C ) = 2 for every z 2 C . Recalling TS 3 = RX with the projection : TS 3 ! along X , the map u is an immersion transversal to the Reeb vector eld X . If follows that u(C ) \ P = ;. Indeed arguing indirectly we assume u(z0) 2 P . Then u(C ) intersects the solution P of X transversally in u(z0). De ning the loops SR by SR (t) = u(Re2 it) we know that SR (t) ! x(Tt) as R ! 1 in C 1 (S 1 ). Hence there is an open neighborhood B" (z0 ) satisfying u(B" (z0 )) \ SR 6= ; if R > 0 is large. This contradicts the injectivity of u. In view of the asymptotic behaviour of u as jz j ! 1 it follows that the injective immersion u : C ! S 3 n P is an embedding. This nishes the proof of theorem 2.1. 2 Theorem 2.4. Let u = (a; u): C ! R S 3 be a non-constant nite energy plane. ~ Assume u : C ! S3 is an injective immersion. Then every asymptotic limit (x; T ) of u is simply covered, ~ cov(x; T ) = 1. In addition, Px = x(R) is unknotted and has self-linking number sl(x; T ) = ?1. We point out that the theorem does not require the contact form to be nondegenerate.

10

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

Proof. Recalling the projection : TS 3 = RX ! we conclude from the di erential equation (1.10) for u that ~ Tu i = J (u) Tu: (2.17) Hence Tu(z ): C ! u(z) is complex linear and since u is an immersion we have Tu(z ) 6= 0 for every z 2 C : (2.18) We shall use this to show that u does not intersect any of its asymptotic limits. Assume (x; T ) is an asymptotic limit of u. Then there exists a sequence Rk ! 1, ~ so that u(Rk e2 it ) ! xT (t) = x(Tt) (2.19) in C 1 (R) as k ! 1. Setting Px = x(R) we claim that u(C ) \ Px = ;: (2.20) Arguing indirectly we assume that u(z0 ) 2 Px . Then u intersects Px at u(z0 ) transversally in view of (2.18). If k is large we therefore conclude from (2.19) that Sk (R) \ u(B" (z0 )) 6= ;, with Sk (t) = u(Rk e2 it ). This contradicts the postulated injectivity of u and proves the claim (2.20). Let D = fz j jz j 1g. Given a disc map v : D ! S 3 satisfying v(e2 it ) = xT (t); the self-linking number of (x; T ) is de ned by sl(x; T ) = int(v; yT ); (2.21) where the loop yT is obtained by pushing xT into the direction of a nowhere vanishing section Z of v ! D, so that xT (R) \ yT (R) = ;. Introducing DRk = fz j jz j Rk g we de ne the maps wk : DRk ! S 3 by wk (z ) = v( Rzk ). Clearly int(wk ; yT ) = int(v; yT ) so that sl(x; T ) = int(wk ; yT ) for every k. For large k the maps wk and uk := u j DRk : DRk ! S 3 are homotopic through maps not intersecting yT on their boundaries. Consequently, by the homotopy property of the intersection number we obtain for k large enough sl(x; T ) = int(uk ; yT ); uk = u j DRk : (2.22) Consider now the nowhere vanishing section Zk of uk ! DRk de ned by

@ Zk (z ) = Tu(z )( @s ); z = s + it; and the section Wk along @DRk de ned by @ Wk (z ) = Tu(z )( @r ); jz j = Rk where r is the radial coordinate z = rei' 2 C . For k large enough Wk is homotopic, as vector eld not tangential to uk (@DRk ), to the outward pointing tangent vector to the embedded disc uk (DRk ) at its boundary. Along @DRk we can write, using the complex multiplication in de ned by the almost complex structure J , Wk (z ) = fk (z ) Zk (z ); jz j = Rk ; (2.23) with a function fk : @DRk ! C n f0g having winding number wind(fk ) = 1: (2.24)

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

11

We claim that for k large enough there exists a C 1 -small deformation of uk = u j DRk : DRk ! S 3 through immersions into the deformed map vk : DRk ! S 3 satisfying the following properties: vk (@DRk ) = Px vk : @DRk ! Px has degree cov(x; T ) (2.25) vk is an immersion _ vk (DRk ) \ Px = ;: To prove this claim we identify a closed tubular neighborhood U of Px = P with P D and de ne by , : S 1 D ! P D; (t; z ) 7! (x(Tt); z ) (2.26) the cov(x; T )-fold covering map. De ning the subset Ak = fz 2 C j jz j Rk and uk (z ) 2 U g we can lift the map uk j Ak to the map uk : Ak ! S 1 D ^ so that uk = uk . In view of (2.20) ^ uk (Ak ) \ (S 1 f0g) = ;: ^ (2.27) For k large enough, the loops k : t 7! uk (Rk e2 it ) and : t 7! (t; 0); t 2 0; 1] are ^ C 1 -close. Hence we nd a family ; 2 0; 1], of di eomorphisms of S 1 D, which are C 1 -close to the identity map, compactly supported in the interior of S 1 D and satisfying 1 ( k ) = . Note that corresponds to xT in our tubular neighborhood. Since 1 is a di eomorphism we conclude (S 1 f0g) \ 1 uk (Ak n @DRk ) = ;: ^ We now de ne a C 1 -small isotopy of immersions ~ : DRk ! S 3 , 2 0; 1] by ~ (z ) = uk (z ); z 2 DRk n Ak (2.28) ~ (z ) = uk (z ); z 2 Ak : ^ Then vk (z ) = ~ 1 (z ); z 2 DRk has the desired properties. Along the above deformation ~ of uk , the sections Zk resp. Wk can be deformed as nowhere vanishing sections of along vk resp. its boundary. We shall denote these new sections by the same letters. We may assume that Wk is an outward pointing tangent vector eld to the immersed disc vk at the boundary vk (@DRk ). In view of the continuous deformation we still have Wk = fk Zk and wind(fk ) = 1. Since Zk is a nowhere vanishing section of vk , the loop yT is homotopic in the complement of Px to the loop obtained from xT by pushing it into the direction of Zk . Therefore, we may assume that yT is obtained from xT by pushing slightly into the direction of Zk , so that, by de nition of the self-linking number, sl(x; T ) = int(vk ; yT ): (2.29) _ Since, by (2.25), vk (DRk ) \ Px = ;, the curve ck := vk j @DRk ?"k for some small "k > 0 is homotopic in S 3 n Px to a constant map. The curve ck may be viewed as obtained from xT by pushing it into the direction of ?Wk . For the homology classes yT ] and ck ] in H1 (S 3 ; Px ; Z) we have the relation ck ] = yT ] + wind(fk ) @ k ]; (2.30)

12

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

where k is any su ciently small disc transversal to Px with orientation induced by . Since ck ] = 0 we deduce from (2.30) int(vk ; yT ) = ? wind(fk ) int(vk ; @ k ]): (2.31) Recall that wind(fk ) = 1 and observe that int(vk ; @ k ]) = cov(x; T ) for a sufciently small disc k . Since sl(x; T ) = int(vk ; yT ), in view of (2.29), we have proved sl(x; T ) = ? cov(x; T ): (2.32) If T0 is the minimal period of (x; T ) such that T = cov(x; T ) T0 we have, since we are on S 3, the formula sl(x; T ) = cov(x; T )2 sl(x; T0): (2.33) Consequently, in view of (2.32), ? cov(x; T ) = cov(x; T )2 sl(x; T0); implying that cov(x; T ) = 1 and sl(x; T ) = ?1. In particular (x; T ) = (x; T0 ) is simply covered. Since (x; T0 ) is the C 1 -limit of unknotted loops, it must be unknotted as well. This completes the proof of theorem 2.4. 3. Proof for overtwisted contact structures We consider S 3 equipped with the contact form which determines the contact structure and the Reeb vector eld X . We have to distinguish two classes of contact forms according to their induced structures which can be either tight or overtwisted, see 5, 6, 8]. Recall that a contact structure is called overtwisted, if there exists an embedded disc D S 3 such that Tm D 6 m for all m 2 @ D and T@ D j @ D. If no such disc exists, the contact structure is called tight. We mention that the contact structures on S 3 have been classi ed by Ya. Eliashberg in 4, 6]. In this section we shall prove theorem 1.4 for overtwisted contact structures. By assumption, is a non-degenerate contact form inducing an overtwisted contact structure . Hence, arguing as in 10] we nd an overtwisted disc D S 3 whose characteristic foliation T D \ contains precisely one singularity, e 2 interior (D), which is positively elliptic. Moreover, the boundary @ D is the only limit cycle for the characteristic foliation of D. In addition, we may assume that D does not contain a periodic orbit for X . This can be achieved by a C 1 -small perturbation disjoint from @ D. We choose an almost complex structure J0 on compatible with d and take the associated R-invariant almost complex structure ~ J0 on R S 3 de ned by (1.9). It is determined by and J0 . We study disc maps u = (a; u): D ! R S 3 , with the closed disc D := fz 2 C j ~ jz j 1g, satisfying, in the coordinates z = s + it, the partial di erential equation us + J0 (~)~t = 0 on D ~ ~ uu (3.1) and the boundary conditions u(@D) f0g (D n feg) R S 3 ~ The winding number of u j @D : @D ! D n feg (3.2) is equal to 1.

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

13

The disc D is oriented in such a way that the unique singularity e of the characteristic foliation is positive. For a suitable choice of J0 near e we can achieve ~ that the associated structure J0 is integrable near (0; e) 2 R S 3 , hence a complex structure. Therefore, by a classical result due to Bishop, there is a 1-parameter family of embedded holomorphic discs satisfying (3.1) and (3.2) emerging from the singularity (0; e), which have the property that the discs u(D) are mutually disjoint ~ and the boundaries u(@D) = (0; u(@D)) foliate a punctured neighborhood of e in ~ D, see 1, 2, 5, 10].

Still following 10], this local family can be continued by means of an implicit function theorem to a (essentially unique) maximal 1-parameter family B of embedded, mutually disjoint disc maps satisfying (3.1) and (3.1). The boundary u(@D) D for every u = (a; u) 2 B is transversal to the singular foliation of D ~ by the maximum principle. B, however, is not a compact family, since otherwise there would be a disc in B touching the boundary @ D which is a leaf of the singular foliation on D. This would contradict the maximum principle. Analysing the failure of compactness one concludes that even a reparametrisation of the discs in B by Moebius transformations does not prevent the gradients jruj of u 2 B from ~ ~ exploding. In view of the gradient bounds near the boundary @D, the gradients blow up at nitely many points ? interior (D). A bubbling o analysis of these singularites based on rescaling methods, and carried out in detail in 10] and 11], establishes the existence of a complicated structure of various kinds of nite energy surfaces u = (a; u) into R S 3 all having not identically vanishing ~ Tu. The

14

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

geometric realisation in S 3 is illustrated by the following picture together with its associated graph.

The top part of the graph represents a map u := (a; u): D n ? ! R S 3 , where ~ ? 6= ; is a nite set of points in interior (D), which solves the following problem: us + J0 (~)~t = 0 on D n ? ~ ~ uu a j @D = 0 (3.3) u : @D ! D n feg has winding number 1 0 < E (~) < 1: u The energy E (~) is obtained by maximising the integrals u
Z

over the set of smooth ' : R ! 0; 1] satisfying '0 (s) 0. The points ? are negative punctures of u near which u tends asymptotically to periodic solutions of ~ the Reeb vector eld X . Recall that, by assumption, the periodic solutions are non-degenerate. It will be important later on that u is an embedding. Near @D this is a conse~ quence of the maximum principle, as shown in 10]. Near the non-degenerate periodic orbits u is either multiple-covered or embed~ ded in view of the arguments in 14] and 13]. The rst alternative cannot occur, since using the similarity principle u would have to be multiple-covered also near ~ @D where it is an embedding. By the asymptotic behaviour u must be an embed~ ding near the boundary in the sense of D. McDu . Moreover, by the bubbling o construction u is a limit of embeddings obtained by properly rescaled disc maps. ~ Consequently, u has to be an embedding, in view of the results of D. McDu in ~ 19, 20]. The bottom parts of the graph represent nite energy planes u := (a; u): C = ~ S 2 n f1g ! R S 3. At the positive puncture f1g, the plane is asymptotic to a periodic solution of X . Every middle piece of the graph represents a punctured _ nite energy sphere u : S ! R S 3 , where S = S 2 n ? is the sphere with a non-empty ~ _ nite set ? of points removed, namely one positive puncture and a positive number of negative punctures where the map tends asymptotically to periodic solutions of X.

Dn?

u d( ' ); ~

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

15

We shall prove now that at least one of the bottom nite energy planes u has an ~ asymptotic limit (x; T ) with index (~) = 2. From (1.13) we recall that (~) 2. u u Arguing indirectly we shall therefore assume that (~) 3 for all bottom nite u energy planes. We cancel now in the graph all elements at the bottom representing a nite energy plane and write at the lower ends of the remaining graph the index of the corresponding asymptotic limit which, by assumption is at least 3. We now study the bottom elements of the new graph. These elements are trees with one positive puncture having an index + and a positive number of negative punctures having corresponding indices 3. To this situation we can apply the following special case of Theorem 5.8 in 14] formulated as Lemma 3.1. Let u : S ! R S 3 be a punctured nite energy sphere with not ~ _ identically vanishing Tu. Then (~) = + ? ? wind (~) + 4 ? 2#?0 ? #?1 : u u (3.4) Here + is the sum of all -indices of the positive punctures in ? while ? is the sum over all negative punctures. Furthermore, wind (~) 0 is a non-negative u integer introduced in 14]. Finally, ?0 is the set of punctures in ? having even -index, while ?1 is the set having odd -index, ? = ?0 ?1 . Applying Lemma 3.1 to our situation, observing that ? 3(#? ? 1), we estimate the -index of the positive puncture by + ? + 4 ? 2#? 3(#? ? 1) + 4 ? 2#? #? + 1 2+1 = 3: We now iterate the procedure starting in the next round again with negative punctures having all -indices 3.

After a nite number of steps only the top piece of the graph remains, namely the piece representing the solution u : D n ? ! R S 3 of the problem (3.3). We ~ recall that it is an embedding. By the inductive construction, its punctures ? are negative and have -indices at least 3. Note that our construction is independent of the choice of the almost complex structure J0 on . Choosing now a generic structure J we have arrived at a contradiction with the following result from 15],

16

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

Theorem 1.15, proved by means of Fredholm theory. Denote by J the set of all almost complex structures on compatible with d . Lemma 3.2. Given an almost complex structure J0 2 J and an embedded disc D as above. Then there exists a J 2 J in every C 1 neighborhood of J0 , whose ~ associated almost complex structure J on R S 3 de ned by (1.9) has the following property. There exists no embedded solution u = (a; u): D n ? ! R S 3 ~ of the mixed boundary value problem (3.3) satisfying Tu(z ) 6= 0 for some z , if ? 6= ; and if all the corresponding asymptotic limits (xj ; Tj ) are non-degenerate and have indices (xj ; Tj ) 3:

u := (a; u): C ! R S 3 having an asymptotic limit with index (x; T ) = 2. Now, a ~ nite energy plane u is either somewhere injective or u = v P , with a somewhere ~ ~ ~ injective nite energy plane v and a higher degree complex polynomial P , see 14]. ~ If (~) = 2, the latter is not possible since the asymptotic limit of v would have an u ~ index at least 2 in view of (1.13), and consequently u would have an index at least ~ 2 deg(P ) 4. Hence u is somewhere injective and we can apply Lemma 3.1 to the ~ nite energy plane u : C = S 2 n f1g ! R S 3 . We have ?1 = ; and #?0 = 1 so ~
that (~) = 2 wind (~) + 2: u u (3.5) Hence, in view of wind 0, we conclude wind (~) = 0. This implies, in view of u the properties of wind in 14] that u : C ! S 3 is an immersion. Consequently u ~ is an immersion. Near f1g it is an embedding by Lemma 2.2. Assume that u has ~ sel ntersections. Then they are necessarily isolated and moreover have positive intersection indices, since we are dealing with pseudoholomorphic curves, see 18, 20]. By the bubbling o analysis, u can be approximated by a rescaled family of disc ~ maps and we conclude that also the discs in this family possess sel ntersections if they are su ciently close to u, by the positivity of the intersection index. This, ~ however, contradicts the fact, that, by our construction based on the Bishop family B, the discs are all embedded. We have proved that u is an injective immersion. In ~ view of its asymptotic behaviour described in Lemma 2.2 the map u is an embed~ ding. Since (~) = 2, theorem 1.4 is now an immediate consequence of theorem 1.6 u and theorem 1.7. This nishes the proof of theorem 1.4 in the overtwisted case. For later use we point out an a-priori bound for the period T of the distinguished periodic solution of X found above, namely Z 1Z (3.6) T = E (~) = u d 2 jd j: u
C

With this contradiction we have established the existence of a nite energy plane

The bound follows from the bubbling o analysis, the proof is contained in 10]. 4. Proof for tight contact structures

On S 3 the tight contact structures are well understood due to the work of Ya. Eliashberg and E. Giroux see 5, 6, 7, 8, 9]. Consider the round sphere

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

17

S 3 = fz 2 C 2 j jz j2 = 1g equipped with its standard contact form 0 = ~0 j S 3 , where the Liouville{form ~0 on C 2 is, in the coordinates zj = aj + ipj given by
If is any tight contact form on S 3 such that the volume form ^ d is compatible with the given orientation S 3 , then there exists an orientation preserving di eomorphism : S 3 ! S 3 such that = f 0 for a smooth function f : S 3 ! (0; 1). This is proved in Eliashberg 6]. In order to prove theorem 1.4 in the tight case, we can therefore assume that our tight contact form on S 3 is given by = f 0; (4.2) 3 . We assume , in addition, to be nonfor a smooth positive function f on S degenerate and denote by X the associated Reeb vector eld. Using a modi cation of a construction in 12] we shall establish an embedded nite energy plane u whose ~ asymptotic limit has an index (~) 2 f2; 3g. u We choose a function fE on S 3 , de ned by ~0 = 1 X(qj dpj ? pj dqj ): 2 j=1
2

(4.1)

fE (z ) =

2 ?X

2 2 with positive real numbers 0 < r1 < r2 such that r1 =r2 is irrational and f (z ) < fE (z ); z 2 S 3 ; (4.4) where f is given in (4.2). We denote by XE the Reeb vector eld de ned by the contact form (4.5) E = fE 0 : The vector eld XE has precisely 2 non-degenerate, in fact elliptic, periodic orbits P0 and P1 having minimal periods T0 < T1 and indices (P0 ) = 3 and (P1 ) = 2k + 1 5; (4.6) 2 < k + 1. This is where the integer k 2 is determined by k < 1 + (r2 =r1 ) easily veri ed by a direct computation carried out in 11]. We now interpolate the contact forms = f 0 and E = fE 0 in a monotonic way by choosing a function h : R S 3 ! (0; 1) satisfying f (z ) h(a; z ) fE (z ), (a; z ) 2 R S 3 , and, moreover, h(a; z ) = f (z ) if a ?2 h(a; z ) = fE (z ) if a 2 @h (a; z ) 0 (4.7)

2 j =1 rj

jzj j2

?1 ; 2

(4.3)

> 0 if jaj 1: For every a 2 R, the contact form a := h(a; ) 0 on S 3 determines the associated Reeb vector eld Xa . By construction, a = ; Xa = X if a ?2 and a = E ; Xa = XE if a 2. The contact structure is, of course, independent of a. We next choose a smooth family a 7! Ja of almost complex structures of , compatible

@a @h (a; z ) @a

18

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

^ with d a and such that Ja does not depend on a in jaj 2. With J we denote the 3 de ned by associated almost complex structure on R S ^ J (a; z )( ; k) = (? a (z )k; Ja (z ) a k + Xa (z )): (4.8) ~ Finally, we de ne the special almost complex structure J on R S 3 as follows. We ~(a; z ) = J (a; z ) if jaj 1. If jaj < 1 we require that J is compatible with ^ ~ set J 3 . We denote by S 2 = S 2 n ? the _ the symplectic form = d(h 0 ) on ?1; 1] S 2 is a nite subset of points to wich we refer as punctured sphere, where ? S punctures. We shall study the equation _ u = (a; u) : S 2 ! R S 3 ~ (4.9) ~ ~ J T u = T u i; ~ where i is a complex multiplication on S 2 . In complex coordinates, z = s + it, the equation (4.9) becomes the familiar Cauchy-Riemann equation us + J (~)~t = 0 for ~ ~u u the map u. We are only interested in solutions having nite energy E (~) de ned ~ u as follows. We denote by the set of smooth functions ' : R ! 0; 1] satisfying 1 '0 (s) 0 and ' = 2 on ?1; 1]. Given ' 2 we de ne the one-forms ' and on 3 by R S ' (a; x)( ; k ) = '(a) a (x)(k ) (a; x)( ; k) = a (x)(k); 3 ). Then the energy E (~) for a solution u of of the above where ( ; k) 2 T(a;x)(R S u ~ equation is now de ned as the supremum over all ' 2 of the numbers
Z

Any non-constant solution of (4.9) having nite energy will be called a generalized nite energy sphere. Next we de ne positive and negative punctures as follows. A neighborhood of a puncture looks like the complement of a compact set in C and one shows that near the puncture the limit always exists in R. If T > 0, the puncture is called positive, in this case the Rcomponent of u is unbounded, but bounded from below. If T < 0, the puncture ~ is called negative, the R-component of u is unbounded, but bounded from above. ~ If T = 0, the solution u can be extended smoothly over the puncture by Gro~ mov's removable singularity theorem. In case of a non constant nite energy plane, the puncture is always positive. There are no non constant nite energy surfaces without positive punctures, in particular no closed surfaces. We refer to 14] for a ~ detailed discussion. By de nition of J , a solution u(z ) = (a(z ); u(z )) of (4.9) solves ~ at points z where a(z ) 2 the equation associated with E while at points where a(z ) ?2 it is a solution of the desired equation associated with our given contact form . Following 11] one establishes an embedded generalized nite energy plane u = ~ _ = C ! R S 3 with inf(a) 2 which is asymptotic to the distinguished (a; u): S periodic solution P0 of XE having index (P0 ) = 3. (It is again the result of an explosion of a non-compact Bishop family B.) As described in 12], this energy plane lies in a 2-parameter family F of mutually disjoint and embedded generalized nite

u d ': ~

T = Rlim !1

? u Re2 it 1

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

19

A bubbling o analysis of the singularities of F , carried out in 12], gives rise to a tree of generalized energy spheres which is illustrated in the following picture:

energy planes which are, as jz j ! 1, all asymptotic to the same periodic orbit P0 of XE . The family F is non-compact and contains elements u = (a; u): C ! R S 3 ~ having arbitrary negative R-components a: ? sup maxjru(z )j = 1; u2B(min a(z )) = ?1: ~ inf z (4.10) z ~
u2B ~

The top part of the graph represents an embedded (since P0 is simply covered) generalized energy sphere u = (a; u): S 2 n ? ! R S 3 ~ (4.11) with one positive puncture, 1, asymptotic to the distinguished periodic solution P0 of XE , and a positive number of negative punctures which are all asymptotic to non-degenerate periodic solutions of X . The bottom parts of the graph represent nite energy planes (not necessarily embedded) u : (a; u): C = S 2 nf1g ! R S 3 , ~ for the -equation, hence asymptotic to periodic solutions of X . All intermediate parts represent nite energy spheres u : S 2 n ? ! R S 3 for the -equation, having ~ one positive puncture and a positive number of negative punctures, all asymptotic to periodic solutions of X . We shall prove now that there is at least one nite energy plane at the bottom of the graph whose -index is either 2 or 3. Arguing indirectly we can assume that all the nite energy planes at the bottom have indices 4. Starting at the bottom we work our way up in the graph as we already did in the overtwisted case and conclude inductively that all the negative punctures of the energy spheres have indices at least 4. Here we use Lemma 3.1 in order to estimate the positive puncture of the middle parts in the graph as follows: + 4(#? ? 1) + 4 ? 2#? = 2(#?) 4: After nitely many steps we arrive at the top piece of the graph. It represents the generalized nite energy sphere u in (4.11) having one positive puncture with index ~ (P0 ) = 3 and (as we just have proved) a positive number of negative punctures, having all an index at least 4. This will lead to a contradiction. We can choose the ~ almost complex structure J to be, in addition, generic, see 15]. Since u : S 2 n ? ! ~ 3 is an embedding, the linearized map along u is a Fredholm map having a R S ~

20

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

non-negative Fredholm index Fred(~) 0. Denoting by ? (~) the sum over all u u -indices associated with the negative punctures, and by (P0 ) the index of the positive puncture, the Fredholm index is given by Fred(~) = (P0 ) ? ? (~) ? (S 2 ) + #? 0: u u (4.12) The formula is proved in 15]. Consequently, in view of (S 2 ) = 2, (P0 ) = 3 and ? (~) 4(#? ? 1) we arrive at the contradiction u 3 ? (~) ? (#? ? 1) + 1 u 4(#? ? 1) ? (#? ? 1) + 1 = 3(#? ? 1) + 1 4: In view of this contradiction, there exists a nite energy plane u for with asymp~ totic index (x; T ) 2 f2; 3g. Arguing now as in the overtwisted case one shows, using Lemma 3.1, that u is an immersion. Again, in view of the asymptotic be~ haviour in the non-degenerate case, u has to be an embedding. Indeed, otherwise ~ the isolated sel ntersections have positive intersection indices. But, by the bubbling o analysis in 12], u can be approximated by a sequence of rescaled generalized ~ nite energy planes, which would have sel ntersections too. This would contradict the fact, that by their construction based on elements of F , they are embeddings. To sum up, we have established the existence of an embedded nite energy plane u : C ! R S 3 for the -equation having a non-degenerate asymptotic limit ~ with index (x; T ) 2 f2; 3g. Hence theorem 1.4 is consequence of theorem 1.6 and theorem 1.7. The proof of theorem 1.4 in the tight case is complete. For later use we should mention that there is an a-priori bound for the minimal period T of the special asymptotic orbit found above in terms of the parameter r1 occuring in the de nition (4.3) of the function fE , see 12]. 5. The degenerate case The results in the previous sections for non-degenerate contact forms on S 3 will now be used in order to prove theorem 1.8 by an approximation argument. Theorem 5.1. Let be a contact form on S 3 and J a compatible almost complex structure on the associated contact structure . Then there exists a nite energy plane u = (a; u): C ! R S 3 such that u is an injective immersion. ~ The main result (theorem 1.1) follows immediately from theorem 5.1 in view of proposition 1.5 and theorem 1.7.
Proof. Let be a contact form on S 3 . By proposition 1.2 we nd a sequence fk : S 3 ! (0; 1) of smooth functions converging in C 1 to the constant function f0 1 so that the contact forms k = fk are non-degenerate. The associated contact structures k = coincide. Choose a J on compatible with and choose a sequence Jk compatible with k so that for the associated almost complex ~ structure Jk on R S 3 the constructions in the sections 3 and 4 can be carried out. Then there exists a sequence of embedded non-degenerate nite energy planes uk = (ak ; uk ): C ! R S 3 with indices (~k ) 3. In view of theorem 1.7, and the ~ u

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

21

a-priori estimates mentioned at the end of section 3 and 4 we nd a constant c > 0 such that ~ Jk T uk = T uk i ~ ~ ?1 E (~k ) c (5.1) c u 3 is an embedding. uk : C ! S Moreover, if Pk is the non-degenerate asymptotic limit of uk , we have ~ uk (C ) \ Pk = ;: (5.2) Denote by 0 > 0 a constant smaller than the smallest period of the periodic solutions of X and also smaller then c?1 . Reparameterizing the solutions uk we ~ may assume, in addition to (5.1), inf ak (z ) = ak (0) = 0 z Z Z (5.3) uk d k = uk d k ? 0 ; where D = fz j jz j 1g. We observe that
Z

(5.4) uk d k " c?1 ? 0 : Assume (at rst) that the gradients of uk remain uniformly bounded on compact ~ 1 subsets of C . Then, possibly passing to a subsequence, uk ! u in Cloc(C ) using ~ ~
D

the standard compactness results, see f.e. in 11]. Clearly, in view of (5.4), the map ~ u = (a; u) is a non constant nite energy plane for the structure J having d -energy ~ at least " and hence at least 0 . We shall prove that u is an injective immersion. In order to prove that u is an immersion we consider the section Tu of the bundle HomC (T C ; u ) ! C : (5.5) By the similarity principle, this section either vanishes identically or has isolated zeroes having all a positive index. For non constant nite energy planes the section does not vanish identically, see 14]. Let R > 0 and assume that Tu does not vanish on @DR. Since uk j DR converges in C 1 to u j DR it follows that the winding numbers wk of the nowhere vanishing sections ( k Tuk ) j @DR coincide with the winding number w of ( Tu) j @DR . Since k Tuk (z ) 6= 0 for every z 2 C we conclude wk = 0 and hence w = 0. Consequently Tu(z ) 6= 0 for z 2 DR , see 14]. This argument shows that Tu(z ) 6= 0 for every z 2 C . Consequently, u : C ! S 3 is an immersion. To prove that u is injective we argue by contradiction and assume u(z ) = u(z 0) for z 6= z 0. Hence we nd a number c 2 R satisfying uc (z ) = u(z 0 ); z 6= z 0: ~ ~ In view of the standard results in the intersection theory of pseudoholomorphic curves (see 18], 19], 20]), the intersection is isolated and has, moreover, a posi1 ~ tive intersection index. Since uk;c converges in Cloc to uc, the pseudoholomorphic ~ curves uk;c have to intersect uk if k is su ciently large. This, however, contradicts ~ ~ the injectivity of the maps uk . We have proved that u : C ! S 3 is an injective immersion, under the assumption that the gradients of uk are bounded. ~ It remains to consider the case in which the gradients of uk are not uniformly ~ bounded. In view of the normalization (5.3) it follows from a standard bubbling o

22

H. HOFER, K. WYSOCKI, AND E. ZEHNDER

The gradients of wk are uniformly bounded on every compact subset of C and, ~ moreover, jrwk (0)j = 1: ~ (5.8) Being a reparametrization of uk , the map wk : C ! S 3 is an injective immersion. 1 Passing now, as in the rst part of the proof, to the limit in Cloc , we obtain a 3 . Arguing again as in the non-constant nite energy plane w = (a; w): C ! R S ~ rst part we deduce that the map w : C ! S 3 is an injective immersion. The proof of theorem 5.1 is complete.
References

analysis (see 11]) that the bubbling o points must be contained in the unit disc D. For a subsequence we nd a nite set ? of points in D so that uk converges ~ 1 in Cloc(C n ?) to some nite energy curve v . Carrying out a bubbling o analysis ~ near one of the punctures, say z0 2 ?, we nd a sequence zk ! z0 and a sequence "k ! 0 satisfying j(ruk )(zk )j "k ! 1 ~ (5.6) j(ruk )(z )j 2j(ruk )(zk )j for all jz ? zk j "k : ~ ~ Set rk = j(ruk )(zk )j. Rescaling we de ne a new sequence wk of nite energy planes ~ ~ by (5.7) wk (z ) = (ak (zk + rz ) ? ak (zk ); uk (zk + rz )): ~
k k

1] C. Abbas and H. Hofer. Holomorphic curves and global questions in contact geometry. To appear in Birkhauser. 2] E. Bishop. Di erentiable manifolds in complex Euclidean space. Duke Math. J., 32:1{21, 1965. 3] K. Cieliebak. Symplectic boundaries: Creating and destroying closed characteristics. ETH Zurich, 1996. Preprint. 4] Y. Eliashberg. Classi cation of overtwisted contact structures on three manifolds. Inv. Math., pages 623{637, 1989. 5] Y. Eliashberg. Filling by holomorphic discs and its applications. London Math. Society Lecture Notes, pages 45{67, 1991. Series 151. 6] Y. Eliashberg. Contact 3{manifolds, twenty year since J. Martinet's work. Ann. Inst. Fourier, 42:165{192, 1992. 7] Y. Eliashberg. Classi cation of contact structures on R3 . Inter. Math. Res. Notices, 3:87{91, 1993. 8] Y Eliashberg. Legendrian and transversal knots in tight contact manifolds. In Topological methods in modern mathematics. Publish or Perish, 1993. 9] E. Giroux. Convexite en topologie de contact. Comm. Math. Helvetici, 66:637{677, 1991. 10] H. Hofer. Pseudoholomorphic curves in symplectisations with application to the Weinstein conjecture in dimension three. Invent. Math., 114:515{563, 1993. 11] H. Hofer, K. Wysocki, and E. Zehnder. A characterization of the tight three-sphere. Duke Math. J., 81:159{226, 1995. 12] H. Hofer, K. Wysocki, and E. Zehnder. The dynamics on a strictly convex energy surface in R4 . ETH Zurich, 1996. Preprint. 13] H. Hofer, K. Wysocki, and E. Zehnder. Properties of pseudoholomorphic curves in symplectisations I: Asymptotics. Ann. I.H.P. Analyse Non Lineaire, 13:337{379, 1996. 14] H. Hofer, K. Wysocki, and E. Zehnder. Properties of pseudoholomorphic curves in symplectisations II: Embedding controls and algebraic invariants. GAFA, Vol. 5, No. 2:270-328, 1995 15] H. Hofer, K. Wysocki, and E. Zehnder. Properties of pseudoholomorphic curves in symplectisations III: Fredholm theory. ETH Zurich, 1996. Preprint.

UNKNOTTED PERIODIC ORBITS FOR REEB FLOWS ON THE THREE-SPHERE

23

16] G. Kuperberg. A volume-preserving counterexample to the Seifert conjecture. Preprint 1994. 17] K. Kuperberg. A smooth counterexample to the Seifert conjecture. Annals of Mathematics, 140:723{732, 1994. 18] D. McDu . The local behavior of holomorphic curves in almost complex 4{manifolds. J. Di . Geom., 34:143{164, 1991. 19] D. McDu . The local behavior of J {holomorphic curves in almost complex 4{manifolds. J. Di . Geom., 34:143{164, 1991. 20] M. Micallef and B. White. The structure of branch points in minimal surfaces and in pseudoholomorphic curves. Annals of Mathematics, 114:35-85, 1994.
Department of Mathematics, ETH Zentrum, 8092 Zurich, Switzerland