A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES VIA CONVEX SURFACE THEORY

arXiv:1102.5398v2 [math.GT] 13 May 2011

YANG HUANG A BSTRACT. In [2], Y. Eliashberg proved that two overtwisted contact structures on a closed oriented 3-manifold are isotopic if and only if they are homotopic as 2-plane elds. We provide an alternative proof of this theorem using the convex surface theory and bypasses.

C ONTENTS 1. Preliminaries 2. Outline of the proof 3. Local properties of bypass attachments 4. Isotoping contact structures up to the 2-skeleton 5. Bypass triangle attachments 6. Overtwisted contact structures on S 2 [0, 1] induced by isotopies. 7. Classication of overtwisted contact structures on S 2 [0, 1] 8. Proof of the main theorem References 2 4 4 7 9 13 23 26 28

A contact manifold (M, ) is a smooth manifold with a contact structure , i.e., a maximally non-integrable codimension 1 tangent distribution. In particular, if the dimension of the manifold is three, it was realized through the work of D. Bennequin and Y. Eliashberg in [1], [3] that contact structures fall into two classes: tight or overtwisted. Since then, dynamical systems and foliation theory of surfaces embedded in contact 3-manifolds have been studied extensively to analyze this dichotomy. Based on these developments, Eliashberg gave a classication of overtwisted contact structures in [2], which we now explain. Let M be a closed oriented manifold and M be an oriented embedded disk. Furthermore, we x a point p . We denote by Contot (M, ) the space of cooriented, positive, overtwisted contact structures on M which are overtwisted along , i.e., the contact distribution is tangent to along . Let Distr(M, ) be the space of cooriented 2-plane distributions on M which are tangent to at p. Both spaces are equipped with the C -topology. Theorem 0.1 (Eliashberg). Let M be a closed, oriented 3-manifold. Then the inclusion j : Contot (M, ) Distr(M, ) is a homotopy equivalence. In particular, we have: Theorem 0.2. Let M be a closed, oriented 3-manifold. If and are two positive overtwisted contact structures on M, then they are isotopic if and only if they are homotopic as 2-plane elds.
1

YANG HUANG

Consequently, overtwisted contact structures are completely determined by the homotopy classes of the underlying 2-plane elds. On the other hand, the classication of tight contact structures is much more subtle and contains more topological information about the ambient 3-manifold. The goal of this paper is to provide an alternative proof of Theorem 0.2 based on convex surface theory. Convex surface theory was introduced by E. Giroux in [7] building on the work of Eliashberg-Gromov [4]. Given a closed oriented surface , we consider contact structures on [0, 1] such that {0, 1} is convex. By studying the lm picture of the characteristic foliations on {t} as t goes from 0 to 1, Giroux showed in [8] that, up to an isotopy, there are only nitely many levels {ti }, 0 < t1 < < tn < 1, which are not convex. Moreover, for small > 0, the characteristic foliations on {ti } and {ti + }, i = 1, 2, , n, change by a bifurcation. In [9], K. Honda gave an alternative description of the bifurcation of characteristic foliations in terms of dividing sets. Namely, he dened an operation, called the bypass attachment, which combinatorially acts on the dividing set. It turns out that a bypass attachment is equivalent to a bifurcation on the level of characteristic foliations. Hence, in order to study contact structures on [0, 1] with convex boundary, it sufces to consider the isotopy classes of contact structures given by sequences of bypass attachments. In particular, we will study sequences of (overtwisted) bypass attachments on S 2 [0, 1], which is the main ingredient in our proof of Theorem 0.2. This paper is organized as follows. In Section 1 we recall some basic knowledge in contact geometry, in particular, convex surface theory and the denition of a bypass. Section 2 gives an outline of our approach to the classication problem. Section 3 is devoted to establishing some necessary local properties of the bypass attachment. Using techniques from previous sections, we show in Section 4 that how to isotop homotopic overtwisted contact structures so that they agree in a neighborhood of the 2-skeleton. Section 5, 6 and 7 are devoted to studying overtwisted contact structures on S 2 [0, 1] which is the technical part of this paper. We nally nish the proof of Theorem 0.2 in Section 8. 1. P RELIMINARIES Let M be a closed, oriented 3-manifold. Throughout this paper, we only consider cooriented, positive contact structures on M, i.e., those that satisfy the following conditions: (1) there exists a global 1-form such that = ker(). (2) d > 0, i.e., the orientation induced by the contact form agrees with the orientation on M. A contact structure is overtwisted if there exists an embedded disk D2 M such that is tangent to D2 on D2 . Otherwise, is said to be tight. We will focus on overtwisted contact structures for the rest of this paper. Let M be a closed, embedded, oriented surface in M. The characteristic foliation on is by denition the integral of the singular line eld (x) x T x . One way to describe the contact structure near is to look at its characteristic foliation. Proposition 1.1 (Giroux). Let 0 and 1 be two contact structures which induce the same characteristic foliation on . Then there exists an isotopy t , t [0, 1] relative to such that 0 = id and (1 ) 0 = 1. Possibly after a C -small perturbation, we can always assume that M is convex, i.e., there exists a vector eld v transverse to such that the ow of v preserves the contact structure. Using

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

this transverse contact vector eld v, we dene the dividing set on to be {x | vx x }. Note that the isotopy class of does not depend on the choice of v. We refer to [7] for a more detailed treatment of basic properties of convex surfaces. The signicance of dividing sets in contact geometry is made clear by Girouxs exibility theorem. Theorem 1.2 (Giroux). Assume is convex with characteristic foliation , contact vector eld v, and dividing set . Let F be another singular foliation on divided by . Then there exists an isotopy t , t [0, 1] in M such that (1) 0 = id and t | = id for all t. (2) v is transverse to t () for all t. (3) 1 () has characteristic foliation F . We now look at contact structures on [0, 1] with convex boundary. The rst important result relating to this problem is the following theorem due to Giroux. Theorem 1.3 (Giroux). Let be a contact structure on W = [0, 1] so that {0, 1} is convex. There exists an isotopy relative to the boundary s : W W, s [0, 1], such that the surfaces 1 ( {t}) are convex for all but nitely many t [0, 1] where the characteristic foliations satisfy the following properties: (1) The singularities and closed orbits are all non-degenerate. (2) The limit set of any half-orbit is either a singularity or a closed orbit. (3) There exists a single retrogradient saddle-saddle connection, i.e., an orbit from a negative hyperbolic point to a positive hyperbolic point. In the light of Girouxs exibility theorem, one should expect a corresponding lm picture of dividing sets on convex surfaces. It turns out that the correct notion corresponding to a bifurcation is the bypass attachment, which we now describe. Denition 1.4. Let be a convex surface and be a Legendrian arc in which intersects in three points, two of which are endpoints of . A bypass is a convex half-disk D with Legendrian boundary, where D = , D , and tb(D) = 1. We call an admissible arc, and D a bypass along on . Remark 1.5. The admissible arc in the above denition is also known as the arc of attachment for a bypass in literature. Remark 1.6. We do not distinguish isotopic admissible arcs 0 and 1 , i.e., if there exists a path of admissible arcs t , t [0, 1] connecting them. The following lemma shows how a bypass attachment combinatorially acts on the dividing set. Lemma 1.7 (Honda). Following the terminology from Denition 1.4, let D be a bypass along on . There exists a neighborhood of D M diffeomorphic to [0, 1], such that {0, 1} are convex, and {1} is obtained from {0} by performing the bypass attachment operation as depicted in Figure 1 in a neighborhood of . It is worthwhile to mention that there are two distinguished bypasses, namely, the trivial bypass and the overtwisted bypass as depicted in Figure 2. The effect of a trivial bypass attachment is isotopic to an I-invariant contact structure where no bypass is attached, while the overtwisted bypass

YANG HUANG

(a)

(b)

F IGURE 1. A bypass attachment along . (a) The dividing set on {0} before the bypass is attached. (b) The dividing set on {1} after the bypass is attached. attachment immediately introduces an overtwisted disk in the local model, hence, for example, is disallowed in tight contact manifolds.

(a)

(b)

F IGURE 2. (a) The trivial bypass attachment. (b) The overtwisted bypass attachment. 2. O UTLINE
OF THE PROOF

Let and be two overtwisted contact structures on M, homotopic to each other as 2-plane eld distributions. Our approach to Theorem 0.2 has three main steps. Step 1. Fix a triangulation T of M. Isotop and such that T becomes an overtwisted contact triangulation in the sense that the 1-skeleton T (1) is a Legendrian graph, the 2-skeleton T (2) is convex and each 3-cell is an overtwisted ball with respect to both contact structures. We rst show that if e() = e( ) H 2 (M; Z), then one can isotop and so that they agree in a neighborhood of T (2) . Step 2. We can assume that there exists a ball B3 M such that and agree on M \ B3 . Observe that |B3 and |B3 can both be realized as sequences of bypass attachments. In section 5, we will dene the notion of a stable isotopy. Then we show that both of sequences of bypass are stable isotopic to some power of the bypass triangle attachment. Moreover, the Hopf invariants of |B3 and |B3 are uniquely determined by the number of bypass triangles attached according to [10]. Step 3. By elementary obstruction theory, the Hopf invariants of |B3 and |B3 are not necessarily the same, but they can at most differ by an integral multiple of the divisibility of the Euler class of either or . We show that this ambiguity can be resolved by further isotoping the contact structures in a neighborhood of T (2) . This nishes the proof of Theorem 0.2. 3. L OCAL
PROPERTIES OF BYPASS ATTACHMENTS

Let M be an overtwisted contact 3-manifold. Let M be a closed convex surface with dividing set . For convenience, we choose a metric on M and denote M \ the metric closure of the open

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

manifold M . We may assume without loss of generality each connected component of M \ is overtwisted. In order to isotop convex surfaces through bypasses freely, we must show that there are enough bypasses. In fact, bypasses exist along any admissible Legendrian arc on provided that the contact structure is overtwisted. This is the content of the following lemma. Lemma 3.1. Suppose that M \ is overtwisted. For any admissible arc , there exists a bypass along in M \ . If separates M into two overtwisted components, then there exists such a bypass in each component. Proof. The technique for proving this lemma is essentially due to Etnyre and Honda [5], and inde pendently Torisu [11]. Let D M \ be an overtwisted disk. We push the interior of slightly into M \ with the endpoints of unmoved to obtain another Legendrian arc , i.e., tb() = 1 and = = . Take a Legendrian arc connecting to D, namely, the two endpoints of are contained in and D respectively and the interior of is disjoint from D. Then we form the Legendrian connected sum = #D via . By results in [5] and [11], tb( ) = 1 and it is easy to see that these two arcs together bound a desired bypass D along . We then show the triviality of the trivial bypass, i.e, attaching a trivial bypass does not change the isotopy class of the contact structure in a neighborhood of the convex surface. The proof essentially follows the lines of the proof of Proposition 4.9.7 in Geiges [6]. Here the contact structure may be either overtwisted or tight. Lemma 3.2. Let ( [0, 1], ) be a contact manifold with the contact structure obtained by attaching a trivial bypass on ( {0}, |{0}). Then there exists another contact structure , which is isotopic to relative to the boundary, such that {t} is convex with respect to for all t [0, 1]. Proof. Since this is a local problem, we may assume that [0, 1] is a neighborhood of the trivial bypass attachment. By Theorem 1.2, any Morse-Smale type characteristic foliation adapted to {0} can be realized as the characteristic foliation of a contact structure isotopic to in a neighborhood of {0}. In particular, we can assume that the characteristic foliation on {0} looks exactly the same as in Figure 3(a) such that e does not connect to any negative hyperbolic point other than h along the ow line. Look at the characteristic foliations on {t} as t goes from 0 to 1. Generically we can assume that the Morse-Smale condition fails at one single level, say, {1/2}, where an unstable saddlesaddle connection has to appear as shown in Figure 4(a). Let {1/2} be an open neighborhood of the ow line from h to e as depicted in Figure 4(a). Observe that the characteristic foliation inside is of Morse-Smale type, and therefore stable in the t-direction. According to the proof of Proposition 4.9.71 in Geiges [6], for a small > 0, there exists an isotopy s , s [0, 1], compactly supported in (2, 2) [0, 1] and 0 = id, such that = (1 ) satises the following: (1) The characteristic foliation on {t} with respect to is isotopic to the one in Figure 4(b) for t [1/2 , 1/2 + ]. In particular, it is nonsingular. (2) For t (1/2 2, 1/2 ) (1/2 + , 1/2 + 2), The characteristic foliation on {t} with respect to is almost Morse-Smale except that there may exist a half-elliptic-halfhyperbolic point.
1

This is a stronger version of the usual Elimination Lemma.

YANG HUANG

h+

h+

(a)

(b)

F IGURE 3. (a) The characteristic foliation on {0}. The trivial bypass is attached along the Legendrian arc in dash line. (b) The characteristic foliation on {1} after attaching the trivial bypass. Here e (resp. h ) denote the -elliptic (resp. -hyperbolic) singular points of the foliation.

e h h+

(a)

(b)

F IGURE 4. (a) The characteristic foliation on {1/2}, where a saddle-saddle connect from h to h+ exists. The region contains exactly two singular points {e , h} which are in elimination position. (b) The nonsingular characteristic foliation on after the elimination.

Now we can make {t} convex for t [1/2 , 1/2 + ] because the only unstable saddlesaddle connection is eliminated and therefore the characteristic foliation becomes Morse-Smale. For t [1/2 , 1/2 + ], there may exist half-elliptic-half-hyperbolic singular points, but we can as well construct a contact structure realizing this type of singularity so that each {t} stays convex. Hence constructed above is as required. Remark 3.3. Let ( [0, 1], ) be a contact manifold such that |0 = |1 and {t} is convex for all t [0, 1]. If S 1 S 1 and is tight, then it is a standard fact that is isotopic to an I-invariant contact structure relative to the boundary. However, if either = S 1 S 1 or is overtwisted, then the above fact is not true anymore. We will study this phenomenon in detail in the case when = S 2 and is overtwisted in Section 6.

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

4. I SOTOPING

CONTACT STRUCTURES UP TO THE

2- SKELETON

We are now ready to take the rst main step towards the proof of Theorem 0.2. Since we will isotop contact structures skeleton by skeleton, we start with the following denition. Denition 4.1. Let (M, ) be an overtwisted contact manifold, and T be a triangulation of M. The triangulation T is called an overtwisted contact triangulation if the following conditions hold: (1) The 1-skeleton is a Legendrian graph. (2) Each 2-simplex is convex with Legendrian boundary. (3) Each 3-simplex is an overtwisted ball. Remark 4.2. The overtwisted contact triangulation dened above is different from the usual contact triangulation where the 3-simplexes are assumed to be tight. The goal for this section is to prove the following Proposition. Proposition 4.3. Let M be a closed, oriented 3-manifold with a xed triangulation T . Let and be homotopic overtwisted contact structures on M. Then they are isotopic up to the 2-skeleton, i.e., there exists an isotopy t : M M, t [0, 1], 0 = id such that (1 ) = in a neighborhood of T (2) . Proof. Before we go into details of the proof, observe that if t : M M, t [0, 1], 0 = id is an isotopy, then (M, 1(), T ) and (M, , 1(T )) carries the same contact information. In fact, we will 1 isotop the skeletons of the triangulation T and think of them as isotopies of contact structures. By a C 0 -small perturbation of the 1-skeleton T (1) , we can assume that T (1) is a Legendrian graph with respect to and . Performing stabilizations to edges of T (1) if necessary, we can further assume that = in a neighborhood of T (1) . For each 2-simplex 2 in T (2) , we can always stabilize the Legendrian unknot 2 sufciently many times so that tb(2) < 0. Therefore a C small perturbation of 2 relative to 2 makes it convex with respect to (resp. ) with dividing set 2 (resp. 2 ). Both 2 and 2 are proper 1-submanifolds of 2 and generically the endpoints are contained in the interior of the 1-simplexes. See Figure 5 for an example. In order to make T an overtwisted contact triangulation for and , we still need to make sure that all 3-simplexes are overtwisted. We do this for , and the same argument applies to . Take an overtwisted disc D in (M, ). We can assume that D is contained in a 3-simplex 3 . Let 3 be 1 2 another 3-simplex which shares a 2-face with 3 , i.e., 3 3 = 2 is a 2-simplex. We claim that 1 1 2 by isotoping 2 relative to 2 if necessary, we can make both 3 and 3 overtwisted. The fact that 1 2 M is closed immediately implies that a nite steps of such isotopies will make T an overtwisted contact triangulation. To prove the claim, we rst take a parallel copy of the overtwisted disk D in an I-invariant neighborhood of D, denoted by D . Pick an arc connecting D to 2 inside 3 . 1 Let 2 be another 2-simplex obtained by isotoping 2 across D along , i.e., 2 satisfying the following conditions: (1) 2 = 2 . (2) 2 2 bounds a neighborhood of D . 2 (3) is convex. By replacing 2 with 2 , we obtain two new 3-simplexes, each of which contains an overtwisted disk in the interior as claimed.

YANG HUANG

F IGURE 5. An example of the dividing set on a 2-simplex. Now by Girouxs exibility theorem, it sufces to isotop and so that they induce isotopic dividing sets on each 2-simplex relative to T (1) . To achieve this goal, we dene the difference 2-cocycle by assigning to each oriented 2-simplex 2 an integer (R+ ( 2 )) (R ( 2 )) (R+ ( 2 )) + (R ( 2 )). Since is homotopic to as 2-plane elds, [] = e() e( ) = 0 H 2 (M, Z). Hence there exists an integral 1-cocycle so that 2d = since the Euler class is always even.2 One should think of as an element in Hom(C1 (M), Z). Let 2 T (2) be an oriented convex 2-simplex and 1 2 be an oriented 1-simplex with the induced orientation. We study the effect of stabilizing the 1-simplex 1 to the overtwisted contact triangulation. If we positively stabilize 1 once and isotop 2 accordingly to obtain a new 2simplex 2 , then the dividing set 2 on 2 is obtained from 2 by adding a properly embedded arc contained in the negative region with both endpoints on the interior of 1 as depicted in Figure 6. Similarly, if we negatively stabilize 1 once and isotop 2 accordingly as before, then the dividing set on the isotoped 2 is obtained from 2 by adding a properly embedded arc contained in the positive region and with both endpoints on the interior of 1 .

+ +

+ + +

(a)

(b)

F IGURE 6. (a) The dividing set on 2 divides it into -regions. The bottom edge is 1 . (b) One possible dividing set on 2 after positively stabilizing 1 once. Note that in general, the new overtwisted contact triangulation obtained by -stabilizing a 1simplex 1 is not unique. In fact, different choices may give non-isotopic dividing sets on the isotoped 2 in the new triangulation. However, for our purpose, we only care about the quantity (R+ ) (R ) on each 2-simplex and it is easy to see that different choices give the same value to this quantity. Thus we will ignore this ambiguity by arbitrarily choosing an isotopy of the 2-simplex. We denote the overtwisted contact triangulation obtained by -stabilizing 1 once in (M, ) by S 1 (). As remarked at the beginning of the proof, one should think of S 1 () as isotopies of . It
precisely, if we x a trivialization of T M and consider the Gauss map associated to the contact distribution, then the Euler class of the contact distribution is exactly twice the Poincar dual of the Pontryagin submanifold of the e Gauss map.
2More

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

is easy to see that S 1 () changes (R+ ( 2 )) (R ( 2 )) by 1 for any 2-simplex 2 T (2) so that 1 2 as an oriented boundary edge. The same holds for as well. Now we argue that one can isotop and so that (R+ ( 2 )) (R ( 2 )) = (R+ ( 2 )) (R ( 2 )) on each 2-simplex 2 . This can be done as follows. For each oriented 1-simplex + 1 T (1) , the 1-cocycle sends it to an integer n = (1 ). We perform n times the isotopy S 1 () to and n times the isotopy S 1 ( ) to at the same time. If we perform such operation to every 1-simplex in T , it is easy to see that the following properties are satised: (1) = in a neighborhood of T (1) . (2) (R+ ( 2 )) (R ( 2 )) = (R+ ( 2 )) (R ( 2 )), 2 T (2) .

The second property implies that 2 can be obtained from 2 by attaching a sequence of by passes for each 2-simplex 2 . Recall that T is an overtwisted contact triangulation and in particular each 3-simplex is an overtwisted ball. Hence bypasses exist along any admissible arc in 2 inside any 3-simplex with 2 as a 2-face by Lemma 3.1. Therefore by isotoping 2-simplexes through bypasses, we can assume that and induce isotopic dividing sets on each 2-simplex relative to its boundary. The conclusion now follows immediately from Girouxs exibility theorem. 5. B YPASS
TRIANGLE ATTACHMENTS

In this section we study the effect of attaching a bypass triangle to the contact structure, in particular, we give an alternative denition of the bypass triangle attachment. We start with the denition of the bypass triangle attachment. Notation: Let be a convex surface and be an admissible arc. We denote the bypass attachment along on by . Let be another admissible arc on the convex surface obtained by attaching the bypass along on . We denote the composition of bypass attachments by , where the composition rule is to attach the bypass along rst, then attach the bypass along . If (M, ) is a contact manifold with convex boundary, then denotes the contact structure obtained by attaching a bypass along to (M, ). Remark 5.1. In general, bypass attachments are not commutative unless the attaching arcs are disjoint. Denition 5.2. Let be a convex surface and be an admissible arc. A bypass triangle attachment along is the composition of three bypass attachments along admissible arcs , and in a neighborhood of as depicted in Figure 7. We denote the bypass triangle attachment along by = . Remark 5.3. The second admissible arc in the bypass bypass triangle is also known as the arc of anti-bypass attachment to . Warning: When we dene a bypass attachment along on (, ), there are several choices involved. Namely, we need to choose a multicurve, i.e., a 1-submanifold of , representing the isotopy class of , an admissible arc representing the isotopy class of , a neighborhood of where is supported. Since the space of choices of and its neighborhood is contractible according to Theorem 1.2, we can neglect this ambiguity. However the space of choices of multicurves representing is not necessarily contractible. This point will be made clear in the next section. For the

10

YANG HUANG

(a)

(b)

(c) F IGURE 7. (a) A neighborhood of on , along which the rst bypass is attached. (b) The second bypass is attached along the dotted arc . (c) The third bypass is attached along the dotted arc and nishes the bypass triangle. rest of this paper, always means a multicurve on rather than its isotopy class. Remark 5.4. If = S 2 and = S 1 , then the space of choices of multicurve is also contractible since there is a unique tight contact structure in a neighborhood of S 2 up to isotopy. Observe that, up to an isotopy supported in a neighborhood of the admissible arc , the bypass triangle attachment does not change . In what follows we look at bypass triangle attachments along different admissible arcs, which leads to our alternative denition of the bypass triangle attachment. Lemma 5.5. Let and be two (overtwisted) contact structures on S 2 [0, 1], where and are admissible arcs on S 2 {0}, such that (1) S 2 {0, 1} is convex with respect to both and . (2) = in a neighborhood of S 2 {0} and #2 {0} = #S2 {0} = 1. S (3) is obtained by attaching a bypass triangle to |S 2 {0} , and is obtained by attaching a bypass triangle to |S 2 {0} . Then is isotopic to relative to the boundary. Proof. Up to isotopy, there are only two different admissible arcs on (S 2 {0}, |S 2 {0} ) (or, (S 2 {0}, |S 2 {0} )). Namely, one gives the trivial bypass and the other gives the overtwisted bypass. We may assume without loss of generality that is not isotopic to , and is the trivial bypass and is the overtwisted bypass. We complete the bypass triangles and as depicted in Figure 8. Observe that is isotopic to , is isotopic to and bypass attachments along and are trivial according to Lemma 3.2, we have the following isotopies: = = .

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

11

F IGURE 8. Since S 2 {0, 1} are convex, we can make sure that the isotopies above are supported in the interior of S 2 [0, 1]. Denition 5.6. A minimal overtwisted ball (B3 , ot ) is an overtwisted ball where B3 has a tight neighborhood, and the contact structure ot is obtained by attaching a bypass triangle to the standard tight ball (B3 , std ). Remark 5.7. By Lemma 5.5, the minimal overtwisted ball is well-dened even if we do not specify the admissible arc along which the bypass triangle is attached. With the above preparation, we can now redene the bypass triangle attachment which is more convenient for our purpose. Let (M, ) be a contact 3-manifold with convex boundary M = . Identify a collar neighborhood of M with [1, 0] such that M = {0} and the contact vector eld transverse to M is identied with the [1, 0]-direction. Let M be an admissible arc along which the bypass triangle is attached. Push into the interior of M to obtain another admissible arc, parallel to , contained in {1/2}, which we still denote by . Let N be a neighborhood of in {1/2}. Consider the ball with corners D [2/3, 1/3] M. By rounding the corners, we get a smoothly embedded tight ball (B3, |B3 ) (M, ), in particular, B3 has a tight 1 1 1 neighborhood in (M, ). Let (B3 , ot ) be a minimal overtwisted ball. We construct a new contact 2 manifold (M, ) = (M \ B3 , ) (B3 , ot ), where is an orientation-reversing diffeomorphism 1 2 identifying the standard tight neighborhoods of B3 and B3 . It is easy to see that is isotopic to 1 2 the contact structure obtained by attaching a bypass triangle to (M, ) along . Remark 5.8. The uniqueness of the tight contact structure on 3-ball, due to Eliashberg, guarantees that the bypass triangle attachment described above is well-dened. Using the above alternative description of the bypass triangle attachment, we prove the following generalization of Lemma 5.5. Lemma 5.9. Let (M, ) be a contact 3-manifold with convex boundary, and let , be two admissible arcs on M. Let (resp. ) be the contact structure on M obtained by attaching a bypass triangle (resp. ) along (resp. ) to (M, ). Then is isotopic to relative to the boundary.

12

YANG HUANG

Proof. Without loss of generality, we can assume that and are disjoint. If not, we take another admissible arc which is disjoint from and . We then show that and , which implies . As before, since M is convex, we can push and slightly into the manifold M, which we still denote by and . Now let B3 M and B3 M be smoothly embedded tight balls containing and respectively. Take a Legendrian arc connecting B3 and B3 , i.e., the endpoints of are contained in B3 and B3 , respectively, and the interior of is disjoint from B3 and B3 . Moreover, we can assume that B3 B3 and B3 B3 . Let N() be a closed tubular neighborhood of . By rounding the corners of B3 B3 N(), we get a smoothly embedded ball B3 M with tight convex boundary. Using our cut-and-paste denition of the bypass triangle attachment, it is easy to see that (B3 , |B3 ) and (B3 , |B3 ) are isotopic, relative to the boundary, to the contact boundary sums (B3 , ot )#b (B3 , std ) and (B3 , std )#b (B3 , ot ), respectively. Hence both are isotopic to the minimal overtwisted ball. One simply extends the isotopy by identity to the rest of M to conclude that on M. According to Lemma 5.9, the isotopy class of the contact structure obtained by attaching a bypass triangle does not depend on the choice of the attaching arcs. We shall write for a bypass triangle attachment along an arbitrary admissible arc. An immediate consequence of this fact is that the bypass triangle attachment commutes with any bypass attachment. This is the content of the following corollary: Corollary 5.10. Let (M, ) be contact 3-manifold with convex boundary, and be an admissible arc on M. Then . Proof. By Lemma 5.9, we can arbitrarily choose an admissible arc M along with the bypass triangle is attached. In particular, we require that is disjoint from . Hence a neighborhood of where is supported in is also disjoint from . Thus we have the following isotopies: . which proves the commutativity. Corollary 5.11. Let (S 2 [0, 1], ) be a contact manifold with convex boundary, where is isotopic to a sequence of bypass attachments 1 2 n , i.e., there exists 0 = t0 < t1 < < tn = 1 such that S 2 {ti } are convex for 0 i n and S 2 [ti1 , ti ] with the restricted contact structure is isotopic to the bypass attachment i . Then is isotopic to k for 0 k n, where k is the contact structure isotopic to a sequence of bypass attachments 1 k k+1 n . Proof. This is an iterated application of Corollary 5.10. However, observe that subtracting a bypass triangle is in general not well-dened. So we need the following denition. Denition 5.12. Two contact structures and on S 2 [0, 1] are stably isotopic, denoted by , if they become isotopic after attaching nitely many bypass triangles to S 2 {1} simultaneously, i.e., n n for some n N.

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

13

6. OVERTWISTED

CONTACT STRUCTURES ON

S 2 [0, 1] INDUCED

BY ISOTOPIES .

Let be an overtwisted contact structure on S 2 [0, 1] such that S 2 {0} and S 2 {1} are convex spheres. In general, any such can be represented by a sequence of bypass attachments. More precisely, by Theorem 1.3, there exists an increasing sequence 0 = t0 < t1 < < tn = 1 such that S 2 {ti } is convex and |S 2 [ti1 ,ti ] is isotopic to a bypass attachment i for i = 1, , n. In this section, we consider a special class of overtwisted contact structures on S 2 [0, 1] such that S 2 {t} is convex for t [0, 1], in other words, there is no bypass attached. Let 0 be an I-invariant contact structure on S 2 [0, 1]. Let t : S 2 S 2 , t [0, 1], be an isotopy such that 0 = id. We dene a new contact structure = (0 ) on S 2 [0, 1], where : S 2 [0, 1] S 2 [0, 1] is dened by (x, t) (t (x), t). Observe that S 2 {t} is convex with respect to for all t [0, 1] by construction. Hence we get a smooth family of dividing sets S 2 {t} for t [0, 1]. Conversely, a smooth family of dividing sets S 2 {t} , t [0, 1] denes a unique contact structure on S 2 [0, 1], which is isotopic to constructed above for some isotopy t , t [0, 1]. In practice, it is usually easier to keep track of the dividing sets rather than the isotopy. Denition 6.1. A contact structure on S 2 [0, 1] is induced by an isotopy if S 2 {t} is convex for all t [0, 1], or, equivalently, there exists an isotopy : S 2 [0, 1] S 2 [0, 1] such that is isotopic to as constructed above. It is convenient to have the following lemma. Lemma 6.2. Let , be two contact structures on S 2 [0, 1] induced by isotopies and let t , t be dividing sets on S 2 {t}, 0 t 1, with respect to , respectively. If 0 = , 1 = and 0 1 there exists a path of smooth families of multicurves ts , 0 s 1 satisfying the following: (1) ts is a multicurve, i.e., a nite disjoint union of simple closed curves, contained in S 2 {t} for 0 s 1, 0 t 1. (2) 0 = t , 1 = for 0 t 1, t t t s s (3) 0 = 0 , 1 = 1 for 0 s 1. then is isotopic to relative to the boundary. Proof. By Girouxs exibility theorem, the path ts , 0 s 1 of multicurves determines a path of contact structures s on S 2 [0, 1] such that 0 = , 1 = . Hence is isotopic to relative to the boundary by Grays stability theorem. We rst consider a bypass attachment to the contact structures on S 2 [0, 1] induced by an isotopy. Lemma 6.3. Let be a contact structure on S 2 [0, 1/2] induced by an isotopy t , t [0, 1/2], and (S 2 [1/2, 1], ) be a bypass attachment along an admissible arc S 2 {1/2}. Then there exists an admissible arc S 2 {0} such that (S 2 [0, 1], ) is isotopic, relative to the 2 boundary, to (S [0, 1], ). Proof. We basically re-foliate the contact manifold (S 2 [0, 1], ). Recall that attaches a bypass D on S 2 {1/2} so that D = is the union of two Legendrian arcs, where tb() = 1, tb() = 0. We extend D to a new bypass D on S 2 {0} through the isotopy t , t [0, 1/2], by dening D = D ( [0, 1/2]), where = 1 () S 2 {0} is the new admissible arc along 1/2 2 which D is attached, and : S [0, 1/2] S 2 [0, 1/2] is dened by (x, t) (t (x), t). By

14

YANG HUANG

attaching the new bypass D on S 2 {0}, observe that the rest of S 2 [0, 1] can be foliated by convex surfaces, and the contact structure is also induced by the isotopy . Hence is isotopic to as desired. Denition 6.4. The admissible arc constructed in Lemma 6.3 is called a push-down of . Con versely, we call a pull-up of . The rest of this section is rather technical and can be skipped at the rst time reading. The only result needed for our proof of Theorem 0.2 is Proposition 6.15. We consider a subclass of the contact structures on S 2 [0, 1] induced by isotopies which we will be mainly interested in. Fix a metric on S 2 . Without loss of generality, we assume that there exists a small disk D2 (y) S 2 centered at y of radius and a codimension 0 submanifold S 2 {0} 2 2 2 of S 2 {0} such that S 2 {0} D (y) and D (y) S 2 {0} = S 2 {0} . Let (s) S {0}, s [0, 1] be an embedded oriented loop such that (0) = (1) = y. Let A() be an annulus neighborhood of containing D2 (y) and disjoint from other components of the dividing set as depicted in Figure 9. We dene an isotopy t , t [0, 1], supported in A() which parallel transports D2 (y) along in A(). More precisely, by applying the stereographic projection map, we can identify A() with an annulus in R2 . Then the parallel transportation is given by an afne map t : x x + (t) (0) for any x D2 (y) and t [0, 1].
A() \ \

F IGURE 9. Denition 6.5. With the small disk D2 (y) S 2 {0} , the annulus A() and the isotopy t chosen as above, we say that the contact structure on S 2 [0, 1] is induced by a pure braid of the dividing set, where : S 2 [0, 1] S 2 [0, 1] is induced by t as before. We denote such contact structures by ,(,D2 (y),) . When there is no confusion, we also abbreviate it by ,D2 , . Remark 6.6. For any simply connected region D S 2 {0} containing S 2 {0} , one can isotop so that D becomes a round disk with small radius as required in Denition 6.5. The isotopy class of the contact structure on S 2 [0, 1] induced by a pure braid of the dividing set only depends on the choice of D S 2 {0} and the isotopy class of . Remark 6.7. If is a contact structure on S 2 [0, 1] induced by a pure braid of the dividing set, then S 2 {0} = S 2 {1} . Before we give a complete classication of contact structures on S 2 [0, 1] induced by pure braids of the dividing set, we make a digression into the study of its homotopy classes using the Pontryagin-Thom construction. We can always assume that the isotopy (, D2 (y), ) is supported in a disk D2 S 2 . Trivialize 2 the tangent bundle of D [0, 1] by embedding it into R3 so that D2 is contained in the xy-plane.

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

15

Consider the Gauss map G : (D2 [0, 1], ,D2 , ) S 2 . By Lemma 6.2, we can assume with out loss of generality that the dividing set is a disjoint union of round circles in D2 {t} for all 0 t 1, and p = (1, 0, 0) S 2 R3 is a regular value. Suppose the number of connected components #D2 {0} = m, then the Pontryagin submanifold B = G1 (p) is an oriented framed monotone braid in the sense that B transversely intersects D2 {t} in m points for any 0 t 1, and each connected component of the dividing set contains exactly one point. It is easy to check that the pull-back framing is the blackboard framing, and consequently the self-linking number of B is exactly writhe(B). Example 6.8. If D2 {0} is the disjoint union of two isolated circles, and D2 {0} = S 1 D2 (y) is the circle on the left as depicted in Figure 10. The isotopy t parallel transports D2 (y) along the oriented loop . We compute the homotopy class of the contact structure ,D2 , .
p1 + p1 p2 p2

+ p1 p2 p1 p2

D2 [0, 1]

(a) (b) 2 F IGURE 10. (a) The contact structure on S [0, 1] induced by a full twist of the dividing circles, where {p1 , p2 } are pre-images of the regular value p = (1, 0, 0) S 2 . (b) The oriented braid with the blackboard framing B as the Pontryagin submanifold. According to the Pontryagin-Thom construction, since writhe(B) = 2, the homotopy class of ,D2 , is in general different from the I-invariant contact structure, and the difference is measured by a decreasing the Hopf invariant by 2.3 Example 6.9. If D2 {0} is the disjoint union of three circles, and D2 {0} = S 1 D2 (y) is the circle on the left as depicted in Figure 11. The isotopy t parallel transports D2 (y) along the oriented loop . We compute the homotopy class of the contact structure ,D2 , . In this case, one computes that writhe(B) = 0, hence ,D2 , is homotopic to the I-invariant contact structure. Now we are ready to classify the contact structures induced by pure braids of the dividing set up to stable isotopy in the sense of Denition 6.5. One goal is to establish an isotopy equivalence relation between a pure braid of the dividing set and the bypass triangle attachment. To start with, we consider the contact structures induced by two special pure braid of the dividing set as depicted in Figure 12. In Figure 12(a), the dividing set D2 (y) is a single circle, and the dividing set
However, if the divisibility of the Euler class is 2, then t gives a contact structure which is homotopic to the I-invariant contact structure. We will discuss the divisibility of the Euler class in detail in Section 6.
3

16

YANG HUANG
p1 + p1 p2 + p3 p2 p3

+ p1 p2 + p3 D2 [0, 1] p1 p2 p3

(a) (b) F IGURE 11. (a) A braiding by a full twist of the left-hand side dividing circle along , where {p1 , p2 , p3 } = G1 (p) is the pre-image of the regular value p = (1, 0, 0) S 2 . (b) The oriented framed braid B as the Pontryagin submanifold. contained in the disk bounded by and disjoint from is also a single circle. In Figure 12(b), D2 (y) consists of m isolated circles nested in another circle, and the dividing the dividing set set contained in the disk bounded by and disjoint from consists of n isolated circles nested in another circle. We also assume that either m or n is not zero. For technical reasons, it is convenient to have the following denitions. Denition 6.10. Given two disjoint embedded circles , D2 , < if and only if is contained in the disk bounded by . Denition 6.11. Let D2 be a nite disjoint union of embedded circles. The depth of is the maximum length of chains 1 < 2 < < r , where i is a single circle for any i {1, 2, , r}. Observe that the depth of the dividing set in Figure 12(a) is 1, and the depth of the dividing set in Figure 12(b) is 2. It turns out that to study the contact structure induced by an arbitrary pure braid of the dividing set, it sufces to consider a nite composition of these two special cases.

m n

(a) F IGURE 12.

(b)

Lemma 6.12. If (S 2 [0, 1], ,D2 , ) is a contact manifold with contact structure induced by a pure braid of the dividing set where D2 and are chosen as in Figure 12(a), then (S 2 [0, 1], ,D2 , ) is isotopic relative to the boundary to (S 2 [0, 1], 2), where 2 denotes the contact structure obtained by attaching two bypass triangles on (S 2 {0}, ,D2 , |S 2 {0} ). Proof. Let be an admissible arc as depicted in Figure 13(b). Suppose that both bypass triangles are attached along . Observe that = , where , and are all trivial bypass attachments. Hence the contact manifold (S 2 [0, 1], 2 ) can be foliated by convex surfaces by Lemma 3.2.

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

17

,D2 ,

(a)

(b)

F IGURE 13. (a) The contact structure is induced by parallel transporting D2 along . (b) Attaching two bypass triangles along the admissible arc . In other words, it is induced by an isotopy. By Theorem 0.1 in [10], we know that attaching two bypass triangles 2 decreases the Hopf invariant by 2. In Example 6.8, we checked by Pontryagin Thom construction that ,D2 , also decreases the Hopf invariant by 2. Observe that the isotopy class relative to the boundary of a 2-strand oriented monotone braid with blackboard framing is uniquely determined by its self-linking number, which is equal to the Hopf invariant. Hence 2 is isotopic ,D2 , in the region where both operations are supported. By extending the isotopy by identity to the rest of S 2 , we conclude that (S 2 [0, 1], ,D2 , ) is isotopic relative to the boundary to (S 2 [0, 1], 2 ). Lemma 6.13. If (S 2 [0, 1], ,D2 , ) is a contact manifold with contact structure induced by a pure braid of the dividing set where D2 and are chosen as in Figure 12(b), then (S 2 [0, 1], ,D2 , ) 2 2(m1)(n1) is stably isotopic to (S [0, 1], ). Proof. Let S 2 {1} be an admissible arc as depicted in the left-hand side of Figure 14(a). By Lemma 6.3, if is the push-down of , then ,(,D2 ,) ,(D2 ,) , where is obtained from by attaching a bypass along . We remark here that ,(,D2 ,) and ,(D2 ,) are contact structures induced by the same isotopy, but are push-forward of different contact structures on S 2 [0, 1]. Choose D2 to be the m isolated circles on the left and be an oriented loop as depicted in the right-hand side of Figure 14(a). Let ,D2 , be the contact structure induced by an isotopy which parallel transports D2 along . Then Lemma 6.2 implies that ,(D2 ,) is isotopic, relative to the boundary, to ,D2 , , where is an isotopy rounding the outmost dividing circle. An iterated application of Lemma 6.12 implies that ,D2 , 2m(n1) . We next isotop the contact structure . Consider the n isolated circles nested in a larger cir cle. Let D2 be the leftmost circle among the n circles and be an oriented loop as depicted in the right-hand side of Figure 14(b). We pull up through an isotopy which parallel transports D2 along , and observe that the pull-up of is isotopic to . By using Lemma 6.3 one more time, we get the isotopy of contact structures ,D2 , . It is left to determine the isotopy class of the contact structure ,D2 , . Since is oriented counterclockwise, by ap plying Lemma 6.12 n 1 times, we get a stable isotopy ,D2 , 2(1n) , i.e., ,D2 , 2(n1) is isotopic to the I-invariant contact structure. To summarize what we have done so far, we have the following (stable) isotopies of contact structures:

18

YANG HUANG

,D2 ,

,D2 ,

(a)

,D2 ,

(b) F IGURE 14. (a) Pushing down the bypass attachment . (b) Pulling up the bypass attachment .

,D2 , = ,(,D2 ,) ,(D2 ,) ,D2 , 2m(n1) ,D2 , 2m(n1) 2(1n) 2m(n1) 2(m1)(n1) = 2(m1)(n1) .

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

19

Hence by denition, ,D2 , is stably isotopic to 2(m1)(n1) as desired. We now completely classify contact structures on S 2 [0, 1] induced by pure braids of the dividing set. Proposition 6.14. If (S 2 [0, 1], ,D2 , ) is a contact manifold with contact structure induced by a pure braid of the dividing set, then ,D2 , is stably isotopic to (S 2 [0, 1], l ) for some l N. Proof. Recall that D2 is a union of components of S 2 {0} , and is an oriented loop in the complement of S 2 {0} as in Denition 6.5. Let be the union of components of S 2 {0} contained in a disk bounded by and outside of A(). We may choose the disk so that is the oriented boundary. Since (, D2 , ) is a pure braid of the dividing set, we have S 2 {0} = S 2 {1} . Hence we also view and as dividing sets on S 2 {1}. Choose pairwise disjoint admissible arcs 1 , 2 , , r , r+1 , , k on S 2 {1} such that the following conditions hold: (1) 1 , 2 , , r1 are admissible arcs contained in D2 such that by attaching bypasses along these arcs, the depth of becomes at most 2. (2) r , r+1, , k are admissible arcs contained in the disk bounded by and outside of A() such that by attaching bypasses along these arcs, the depth of becomes at most 2. Observe that we choose 1 , 2 , , k so that the isotopy class of each i is invariant under the time-1 map 1 which is supported in A() \ D2 . Hence, by abuse of notation, we do not distinguish i and its push-down through (, D2 , ). By Lemma 6.3, we have the isotopy of contact structures (,D2 ,) 1 k 1 k , where is the contact structure induced by a nite composition of special pure braids of the dividing set considered in Lemma 6.12 and Lemma 6.13, Therefore is stable isotopic to a power of the bypass triangle attachment, say l for some l N. To summarize, we have the following (stable) isotopies of contact structures, relative to the boundary. ,D2 , k (,D2 ,) 1 k = (,D2 ,) (1 1 ) (k k ) 1 k ) ( ) ((,D2 ,) 1 k ) (1 1 k k (1 k ) (1 ) (k ) 1 k (1 k l ) (1 ) (k ) 1 k l (1 1 ) (k k ) 1 k = l k . Hence ,D2 , is stably isotopic to l by denition. To conclude this section, we prove the following technical result which asserts that under certain assumptions and up to possible bypass triangle attachments, one can separate two bypasses. Proposition 6.15. Let (S 2 , ) be a convex sphere with dividing set and (S 2 , ) be an admissible arc such that the bypass attachment increases # by 1. Suppose that (S 2 , ) is the new convex sphere obtained by attaching to (S 2 , ) and suppose (S 2 , ) is another admissible arc such that the bypass attachment decreases # by 1. Then there exists an admissible arc

20

YANG HUANG

(S 2 , ) disjoint from , a map : S 2 [0, 1] S 2 [0, 1] induced by an isotopy, and an integer l N such that l relative to the boundary. Proof. Let be the arc of anti-bypass attachment to contained in (S 2 , ) as discussed in Remark 5.3. Then intersects in three points {p1 , p2 , p3 } as depicted in Figure 15(b). Let 1 and 2 be subarcs of from p1 to p2 and from p2 to p3 respectively. Observe that, in order to nd an admissible arc (S 2 , ) which is disjoint from and satisfy all the conditions in the lemma, it sufces to nd an admissible arc on (S 2 , ), which we still denote by , and which is disjoint from and also satises the conditions in the lemma. In fact, by symmetry, we only need to be disjoint from 1 . Without loss of generality, we can assume that intersects transversely and the intersection points are different from p1 , p2 and p3 .

p1 p2 p3

(a)

(b)

F IGURE 15. (1) The convex sphere (S 2 , ) with an admissible arc . (2) The convex sphere (S 2 , ) obtained by attaching a bypass along , where is the arc of the anti-bypass attachment. Claim: Up to isotopy and possibly a nite number of bypass triangle attachments, one can arrange so that and 1 do not cobound a bigon B on S 2 as depicted in Figure 16(1).

(1) (2) (3) F IGURE 16. (1) The admissible arc together with 1 bound a minimal bigon, which contains other components of the dividing set in the interior. (2) Choose a disk D2 containing all the dividing sets in the bigon and an oriented loop so that it intersects in exactly one point. (3) The pull-up of through the contact structure ,D2 , bounds a trivial bigon with 1 . To verify the claim, note that if B is a trivial bigon, i.e., it contains no component of the dividing set in the interior, then we can easily isotop to eliminate B. If otherwise, we consider a minimal

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

21

bigon bounded by and 1 in the sense that the interior of the bigon does not intersect with . Take a disk D2 B containing all components of the dividing set in B, namely, D2 = 2 and (B \ D ) = . By our assumption, the bypass attachment decreases # by 1, so must intersect in three points which are contained in three different connected components of respectively. One can nd an oriented loop : [0, 1] S 2 \ with (0) = (1) D2 such that intersects in one point. Orient in such a way that it goes from to (1) in the interior of B as depicted in Figure 16(b). Suppose that : S 2 [0, 1] S 2 [0, 1] is induced by an isotopy t which parallel transports D2 along . By pulling up the the bypass attachment through , we get the following isotopy of contact structures (cf. proof of Lemma 6.13): ,(D2 ,) ,(,D2 ,) where is obtained from by attaching a bypass along , and is the pull-up of which is isotopic to the one depicted in Figure 16(c). Since and 1 cobound a trivial bigon, a further isotopy of will eliminate the bigon so that does not intersect 1 in this local picture. By Proposition 6.14, the contact structure ,(,D2 ,) is stably isotopic to n for some n N. Dene 1 : S 2 [0, 1] S 2 [0, 1] by (x, t) (1 (x), t), t then it is easy to see that ,(D2 ,) ,1 (D2 ,) is isotopic, relative to the boundary, to an I-invariant contact structure. Since we will use this trick many times, we simply write 1 for ,1 (D2 ,) when there is no confusion. To summarize, we have ,(,D2 ,) ,1 (D2 ,) n ,1 (D2 ,) n ,1 (D2 ,) By applying the above argument nitely many times, we can eliminate all bigons bounded by and 1 . Hence the claim is proved. Let us assume that intersects 1 nontrivially, and and 1 do not cobound any bigon on S 2 . We consider the following two cases separately. Case 1. Suppose does not intersect any of the three components of the dividing set generated by the bypass attachment . Let 1 , 2 and 3 be the three dividing circles which intersect with . If intersects 1 in exactly one point as depicted in Figure 17(a), then we choose a disk D2 1 and an oriented loop in the complement of the dividing set as depicted in Figure 17(b) such that ,(1 ,D2 ,) 1 m 1 by arguments as before for some m N, where 1 intersects in exactly two points and cobound a trivial bigon as depicted in Figure 17(c). Hence an obvious further isotopy of makes it disjoint from 1 as desired. If intersects 1 in more than one point, we orient so that it starts from the point q = 1 as depicted in Figure 18(a). Let q1 and q2 be the rst and the second intersection points of with 1 respectively. Note that since we assume and 1 do not cobound any bigon, there is no q more intersection point 1 between q1 and q2 . Let 1 , 1 q and 1 be oriented subarcs of qq q q 2 q and 2 be an oriented subarc of 1 . We obtain a closed, oriented (but not embedded) loop q 1 q q q = 1 1 2 1 q by gluing the arcs together. To apply Proposition 6.14 in this case, qq q 1 q 2 we take an embedded loop close to as depicted in Figure 18(2), which we still denote by . Let

22
3

YANG HUANG

(b) (c) 2 F IGURE 17. (1) The convex sphere (S , ) with an admissible arc intersecting 1 in exactly one point. (2) Choose a disk D2 containing 1 and an oriented loop , along which we apply the isotopy. (3) The pull-up of through the contact structure 1 ,D2 , bounds a trivial bigon with 1 . D2 be a small disk containing 1 as usual. Again by pulling up the bypass attachment through ,(1 ,D2 ,) , we have (stable) isotopies of contact structures ,(1 ,D2 ,) 1 r 1 and 1 bound a trivial bigon. Hence an obvious further isotopy eliminates for some r N, where the trivial bigon and decreases #( 1 ) by 1. By applying the above argument nitely many times, we can reduce to the case where intersects 1 in exactly one point, but we have already solved the problem in this case. We conclude that under the hypothesis at the beginning of this case, there ex ists a disjoint with 1 such that l for some isotopy and an integer l N.

(a)

q2 q1

(b) (c) 2 F IGURE 18. (1) The convex sphere (S , ) with an admissible arc intersecting 1 in at least two points, say, q1 and q2 . (2) The embedded, oriented loop approximating the broken loop qq1 q1 q2 q2 q1 q1 q. (3) The pull-up of through the contact structure 1 ,D2 , bounds a trivial bigon with 1 . Case 2. Suppose nontrivially intersects the union of the three components of the dividing set generated by the bypass attachment . Without loss of generality, we pick an intersection point r as depicted in Figure 19(a). Orient so that it starts from r. Let r1 be the rst intersection point of and 1 . Then , 1 and bound a triangle rr1 p1 . By the assumption that there exists no bigon bounded by and 1 , the interior of the triangle rr1 p1 does not intersect with . If the interior of the triangle rr1 p1 contains no components of the dividing set, then it is easy to isotop so that #( 1 ) decreases by 1. If otherwise, take a small disk D2 rr1 p1 containing all components of the dividing set in rr1 p1 , i.e., rr1 p1 \ D2 does not intersect with the dividing

(a)

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

r p1 r1

23

(a) (b) (c) F IGURE 19. (1) The admissible arc , the dividing set and 1 cobound a topological triangle rr1 p1 , which may contain other components of the dividing set in the interior. (2) Choose the disk D2 to contain all the components of the dividing set in the topological triangle rr1 p1 , and an oriented loop which intersects in exactly one point. (3) By applying the isotopy along , the admissible arc becomes which bounds a trivial triangle with the dividing set and 1 . set . Let be an oriented loop based at a point in D2 intersecting exactly once. By pulling up the bypass attachment through (,D2 ,) , we have (stable) isotopies of contact structures ,(,D2 ,) 1 n 1 so that , 1 and bound a trivial triangle in the sense that the interior of the triangle does not intersect with the dividing set. Hence we can further isotop to eliminate the trivial triangle and hence decrease #( 1 ) by 1. By applying such isotopies such that #( 1 ) = 0 and satisfy all the conditions nitely many times, we get an admissible arc of the proposition. 7. C LASSIFICATION
OF OVERTWISTED CONTACT STRUCTURES ON

S 2 [0, 1]

We have established enough techniques to classify overtwisted contact structures on S 2 [0, 1]. Proposition 7.1. Let be an overtwisted contact structure on S 2 [0, 1] such that S 2 {0, 1} is convex with S 2 {0} = S 2 {1} = S 1 . Then n for some n N, where n denotes the contact structure on S 2 [0, 1] obtained by attaching n bypass triangles to S 2 {0} with the standard tight neighborhood. Proof. By Girouxs criterion of tightness, both S 2 {0} and S 2 {1} have neighborhoods which are tight. Take an increasing sequence 0 = t0 < t1 < < tn = 1 such that is isotopic to a sequence of bypass attachments 0 1 n1 , where i S 2 {ti } are admissible arcs along which a bypass is attached. Dene the complexity of a bypass sequence to be c = max0in #S 2 {ti } . The idea is to show that if c > 3, then we can always decrease c by 2 by isotoping the bypass sequence and suitably attaching bypass triangles. To achieve this goal, we divide the admissible arcs on (S 2 , ) into four types (I), (II), (III) and (IV), according to the number of components of intersecting the admissible arc as depicted in Figure 20, where we only draw the dividing set which intersects the admissible arc. Observe that bypass attachment of type (I) increases # by 2, bypass attachment of type (II) and (III) do not change #, and bypass attachment of type (IV) decreases # by 2. Hence the complexity of a sequence of bypass attachments changes only if the types of bypasses in the sequence change. By repeated application of Lemma 6.3, we may assume that contact structures induced by isotopies

24

YANG HUANG

are contained in a neighborhood of S 2 {1}. By assumption, S 2 {1} has a tight neighborhood. Hence according to Remark 5.4, we shall only consider sequences of bypass attachments modulo contact structures induced by isotopies.

(I)

(II)

(III)

(IV)

F IGURE 20. Four types of admissible arcs on (S 2 , ). Claim: We can isotop the sequence of bypass attachments such that only bypasses of type (I) and (IV) appear. To prove the claim, we rst show that a bypass attachment of type (III) can be eliminated. Take an admissible arc of type (III). If the bypass attachment along is trivial, then by Lemma 3.2, the bypass attachment is induced by an isotopy. Otherwise there exists an admissible arc disjoint from as depicted in Figure 21(a)4 such that if one attaches a bypass along , followed by a bypass attached along , then the later bypass attachment is trivial.

(a) F IGURE 21.

(b)

By the disjointness of admissible arcs and , we get the following isotopies of contact structures, . Observe that is a composition of type (I) and type (IV) bypass attachments. Hence a nite number of such isotopies will eliminate all bypass attachments of type (III) in a sequence. Similarly suppose that is the bypass attachment of type (II) in a sequence and is nontrivial. Choose an admissible arc disjoint from as depicted in Figure 21(b) such that if one attaches a bypass along , followed by a bypass attached along , then the later bypass attachment is trivial.
4

In literature, we say is obtained from by left rotation.

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

25

By the disjointness of and again, we get the following isotopies of contact structures: . Observe that is a composition of bypass attachments both of type (III), hence by a further isotopy will turn into a composition of bypass attachments of type (I) and (IV). A nite number of such isotopies will eliminate bypasses of type (II). The claim follows. From now on, we assume that any bypass attachment in 0 1 n1 either increases or decreases # by 2. Assume that the complexity of the bypass sequence is achieved at level S 2 {tr } for some r {0, 1, , n} and is at least 5, i.e., #S 2 {tr } = c 5. Then it is easy to see that r1 is type (I) and r is type (IV). By Proposition 6.15, we can always assume that r is disjoint from r1 modulo nitely many bypass triangle attachments. Hence we can view both r1 and r as admissible arcs on S 2 {tr1}. Let S 2 {tr1 } be the dividing circle which nontrivially intersects r1 . We do a case-by-case analysis depending on the number of points r intersecting with . Case 1: If r intersects in at most one point, then one easily check that by applying isotopy r1 r r r1 to the sequence of bypass attachments, #S 2 {tr } decreases by 2. Case 2: If r intersects in exactly two points, then once again we apply the isotopy r1 r r r1 to the sequence of bypass attachments. Now observe that r r1 is a composition of bypass attachments of type (III). In the proof of the claim above, we see that any bypass attachment of type (III) is isotopic to a composition of a bypass attachment of type (IV) followed by a bypass attachment of type (I). Such an isotopy also decreases the local maximum of # by 2. Case 3: If r also intersects in three points, we consider a disk D bounded by and r1 as depicted in Figure 22(a). If D contains no component of the dividing set in the interior, then r1 r is isotopic to a bypass triangle attachment, more precisely, there exists a trivial bypass along an admissible arc on S 2 {tr } such that r1 r is a bypass triangle attachment along r1 . Suppose D contains at least one connected component of the dividing set. Let be an admissible arc on S 2 {tr1} disjoint from r1 and r such that it intersects in two points and the dividing set contained in D in one point as depicted in Figure 22(b). We have the following isotopies of contact structures due to Lemma 5.9 and the disjointness of admissible arcs: r1 r r1 r = r1 r r1 r One can check that the last ve bypass attachments above are all of type (III). Hence we can isotop further as before to eliminate type (III) bypass attachments to decrease the complexity of the bypass sequence.

26

YANG HUANG

r1

r1

(a) F IGURE 22.

(b)

To summarize, we proved that any sequence of bypass attachments 0 1 n1 on S 2 [0, 1] is stably isotopic to another sequence of bypass attachments whose complexity is at most three, which is clearly isotopic to a power of bypass triangle attachments. Thus the proposition is proved. 8. P ROOF
OF THE MAIN THEOREM

Now we are ready to nish the proof of Theorem 0.2. Proof of Theorem 0.2. By Proposition 4.3, we can isotop and so that they agree in a neighborhood of the 2-skeleton. Without loss of generality, we can furthermore assume that there exists an embedded closed ball B3 M such that (1) B3 is convex and has a tight neighborhood in M with respect to both and . (2) = in M \ B3 . (3) The restriction of and to M \ B3 and to B3 are all overtwisted. Take a small ball B3 B3 in a Darboux chart so that both |B3 and |B3 are tight. We identify B3 \ B3 with S 2 [0, 1] and represent the contact structures |B3 \B3 and |B3 \B3 by two sequences of bypass attachments. By Proposition 7.1, both |B3 \B3 and |B3 \B3 are stably isotopic to some power of the bypass triangle attachment, in other words, there are isotopies of contact structures |B3 \B3 r n+r and |B3 \B3 s m+s for some n, m, r, s N. By assumption, the restriction of and to M \B3 are overtwisted, so there exist bypass triangle attachments along any admissible arc on B3 according to Lemma 3.1. By simultaneously attaching sufciently many bypass triangles to |B3 \B3 and |B3 \B3 , we can further assume that |B3\B3 n , |B3 \B3 m and = on M \ B3. Let d be the divisibility of the Euler class e() = e( ) H 2 (M; Z). By elementary obstruction theory and Theorem 0.1 in [10], we have d|(mn). To complete the proof of the theorem, we need to show that | M\B3 is isotopic to | M\B3 d relative to the boundary. Since d = g.c.d.{e()| H2 (M)}, it sufces to prove the following more general fact. Lemma 8.1. Let be a closed surface of genus g and be an I-invariant contact structure on [0, 1]. Then l is stably isotopic to relative to the boundary, where l = e()(). Proof. Since we only consider stable isotopies of contact structures, one can prescribe any dividing set on such that the Euler class evaluates on is equal to l. In particular, we consider the dividing set on as depicted in Figure 23, namely, there are g + 1 circles 1 g+1 dividing

A PROOF OF THE CLASSIFICATION THEOREM OF OVERTWISTED CONTACT STRUCTURES

27

into two punctured disks, in each of which there are p and q isolated circles respectively. We call the left most circles in the sets of p and q isolated circles 0 and 1 respectively. We also choose admissible arcs {1 , 2 , , p1 } and {1 , 2, , q1 }, and orient i , 1 i g + 1, in a way as depicted in Figure 23.
1 + 1 0 + 1 2 1 + 2 +

...

q1

g+1

... ...
p1 +

F IGURE 23. An easy calculation shows that l = 2(p q). Choose small disks D2 , D2 in such that ,0 ,1 = 0 and D2 = 1 . Observe that the bypass triangle attachment along any i ,1 and j consists of three trivial bypass attachments, hence is isotopic to contact structures induced by a pure braid of the dividing set. More precisely, let i , i = 1, 2, , g + 1, be an oriented loop in the negative region which is parallel to i . We have the following isotopies of contact structures 2 1 2 p1 (0 ,D2 ,1 g+1 ) (0 ,D2 ,1 ) (0 ,D2 ,g+1 ) , where we think of ,0 ,0 ,0 1 g+1 as an oriented loop homologous to the union of the i s. Similarly one can study the bypass triangle attachments along the j s, but with an opposite orientation. Let i+ be an oriented loop in the positive region which is parallel to i for 1 i g + 1. We have the following (stable) + + + + isotopies of contact structures 2 2 (1 ,D2 ,1 g+1 ) (1 ,D2 ,1 ) (1 ,D2 ,g+1 ) . 1 q1 ,1 ,1 ,1 To summarize the computations above, we get the following (stable) isotopies of contact structures: D2 ,0 l (2 1 2 p1 ) (2 2 ) 1 q1
+ + ((0 ,D2 ,1 ) (0 ,D2 ,g+1 ) ) ((1 ,D2 ,1 ) (1 ,D2 ,g+1 ) ) ,0 ,0 ,1 ,1 + + ((0 ,D2 ,1 ) (1 ,D2 ,1 ) ) ((0 ,D2 ,g+1 ) (1 ,D2 ,g+1 ) ) ,0 ,1 ,0 ,1

where the last step follows from the fact that isotopies that parallel transport D2 and D2 are ,0 ,1 disjoint. Now it sufces to prove that (0 ,D2 ,i ) (1 ,D2 ,i+ ) is stably isotopic to an I-invariant contact ,1 ,0 structure for 1 i g + 1. To see this, take an annular neighborhood Ai of i containing D2 ,0 and D2 and an admissible arc i which intersects 0 , 1 , and i as depicted in Figure 24. We can ,1 assume that the isotopies (0 , D2 , i ) and (1 , D2 , i+) are supported in Ai . For simplicity of ,0 ,1 notation, we denote the composition (0 ,D2 ,i ) (1 ,D2 ,i+ ) by i . ,1 ,0 By pushing down the bypass attachment i through i , we have the following isotopies of contact structures: i i = i i i i i (i ) i i i i = i i

28

YANG HUANG

i + i 0 1 +

F IGURE 24. An annulus neighborhood Ai of i containing 0 and 1 . where i is the push-down of i which is isotopic to i , and the (i ) is easily seen to be isotopic to an I-invariant contact structure. The argument works for all i {1, 2, , g+1}, hence we establish the stable isotopy as desired. Acknowledgements. The author is very grateful to Ko Honda for inspiring conversations throughout this work. The author also thank MSRI for providing an excellent environment for mathematical research during the academic year 2009-2010. R EFERENCES
[1] D. Bennequin, Entrelacements et equations de Pfaff, Ast risque, 107-108 (1983), 87-161. e [2] Y. Eliashberg, Classication of overtwisted contact structures on 3-manifolds, Invent. Math. 98 (1989), 623-637. [3] Y. Eliashberg, Contact 3-manifolds, twenty years since J. Martinets work, Ann. Inst. Fourier 42 (1992), 165-192. [4] Y. Eliashberg and M. Gromov, Convex symplectic manifolds, Proceeding of Symposium Pure Math., vol.52, Amer. Math. Soc., Providence, RI, (1991), 165-192 [5] J. Etnyre and K. Honda, On connected sums and Legendrian knots, Adv. Math. 179 (2003), 59-74. [6] H. Geiges, An Introduction to Contact Topology, Cambridge University Press, (2008) [7] E. Giroux, Convexit en topologie de contact, Comm. Math. Helv. 66 (1991), 637-677. e [8] E. Giroux, Sur les transformations de contact au-dessus des surfaces, Essays on geometry and related topics, Vol. 1,2, Monogr. Enseign. Math., 38, Enseignement Math., Geneva, (2001), 329350. [9] K. Honda, On the classication of tight contact structures I, Geom. Topol. 4 (2000), 309368 (electronic). [10] Y. Huang, Bypass attachments and homotopy classes of 2-plane elds in contact topology, preprint 2011. arXiv:1105.2348v1 [math.GT] [11] I. Torisu, On the additivity of the Thurston-Bennequin invariant of Legendrian knots, Pacic J. Math. 210 (2003) 359-365 U NIVERSITY OF S OUTHERN C ALIFORNIA , L OS A NGELES , CA 90089 E-mail address: huangyan@usc.edu