Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/arxiv-cornell-university/fundamental-groups-of-formal-legendrian-and-horizontal-embedding-Vls41B8hHe
- ISSN
- 1472-2739
- eISSN
- ARCH-3343
- DOI
- 10.2140/agt.2020.20:7
- Publisher site
- See Article on Publisher Site
Abstract
FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Abstract. We compute the fundamental group of each connected component of the space of formal Legendrian embeddings in R . We use it to show that previous examples in the literature of non trivial loops of Legendrian embeddings are already non trivial at the formal level. Likewise, we compute the fundamental group of the di erent connected components of the space of formal horizontal embeddings into the standard Engel R . We check that the natural inclusion of the space of horizontal embeddings into the space of formal horizontal embeddings induces an isomorphism at {level. Contents 1. Introduction. 2 2. Spaces of embeddings of the circle into euclidean space. 3 1 3 2.1. The space Emb(S ;R ). 3 1 4 2.2. The space Emb(S ;R ). 4 3. Formal Legendrian Embeddings in R . 6 3.1. Formal Legendrian Embeddings in R . 6 3.2. The space FLeg(R ). 7 3.3. Classi cation of formal Legendrian embeddings in R . 8 3.4. Fundamental group of formal Legendrian Embeddings in R . 9 4. An application: K alm an's loop. 11 4.1. K alm an's loop. 11 4.2. Set of parametrizations of the family of loops. 13 4.3. Proof of Proposition 4.2.1. 14 5. Formal Horizontal Embeddings in R . 17 5.1. Formal Horizontal Embeddings in R . 17 5.2. The space FHor(R ). 19 Date : January 22, 2020. 2010 Mathematics Subject Classi cation. Primary: 58A30, 57R17. arXiv:1711.04320v2 [math.SG] 18 Jan 2020 2 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 5.3. The Z factor in (FHor(R )). 21 2 1 6. The Area Invariant. 25 6.1. Local models of Legendrians. 25 6.2. The Area Invariant. 30 7. h{principle at {level for horizontal embeddings. 35 7.1. The main theorem. 36 7.2. Creating exibility. 37 7.3. Cancellation by pairs of zeroes in the same connected component. 44 8. The core of the proof 48 8.1. Changing the diagram of curves of strict Legendrian immersions. 48 8.2. Globalization. 70 References 72 1. Introduction. The computation of the homotopy type of the space of Legendrian embeddings into a contact 3{fold has a long story. For a while, it was thought that the computation could be made at the formal level. We mean by that that the inclusion of the space of Legendrian embeddings into the space of formal Legendrian embeddings, ie the space of pairs: smooth embedding and formal Legendrian derivative, was a weak homotopy equivalence. This was proven to be wrong in the key article of D. Bennequin [4]; in which it was shown that the formal space associated to the standard contact R possesses some connected components that are not representable by Legendrian knots. In other words, the restriction of the induced map of the inclusion at {level was not surjective. This was the rst hint of ridigity phenomena in contact topology. Later on, there has been an industry checking how far the inclusion map is at {level from being injective or surjective, see, eg, the work of Chekanov [7], Ding and Geiges [8], Eliashberg and Fraser [10], Etnyre and Honda [13] or Osv ath, Szab o and Thurston [29]. The next step was the study of higher homotopy groups. This was developed by K alm an [23] in dimension 3 using pseudoholomorphic curves invariants and by Sablo and Sullivan [28] in dimension 2n + 1, n > 1, using generating function invariants. However, they just checked that several non trivial loops in the space of Legendrian embeddings were trivial as elements in the fundamental group of the space of smooth embeddings. We show that all K alm an's examples are non trivial in the space of formal Legendrian embeddings, see Section 4. This makes unnecessary the use of sophisticated invariants to compute these examples. In order to do that, we compute the fundamental group of the space of formal Legendrian embeddings. This is the content of Section 3. Moreover, the analogous problem for horizontal embeddings into Engel manifolds satis es an h{principle at {level, eg see Adachi [1], Geiges [16] or del Pino and Presas [27]. In Section 0 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 3 5, we compute the fundamental group of the space of formal horizontal embeddings into the standard Engel R . We check that it is ZZ where the rst component is a rotation invariant that captures the formal immersion class of the loop and the second component captures the formal embedding class of the loop. In Section 6, we provide a way of combinatorially computing this Z {invariant. Finally, in Section 7, we check that the formal fundamental group is isomorphic to the fundamental group, ie the Z {invariant and the rotation invariant completely classify the elements of the fundamental group of horizontal embeddings. Independently, Casals and del Pino have shown that there is a full h{principle for horizontal embeddings [6]. Although our proof is more speci c, it has the advantage of not using the topology of the space of smooth embeddings. Therefore, we reprove as a corollary that the space of smooth embeddings of circles into R is simply connected, see Budney [5]. The methods developed in this article may allow to compute higher rank homotopy groups of the space of embeddings of the circle in R . The strategy that we are pursuing in a forthcoming project is based on the following facts. First, we compute the higher homotopy groups of the space of horizontal embeddings by the geometrical method developed in this article, that is heavily simpli ed thanks to the work of Igusa [22]. Secondly, we use [6] in order to state that the previous computations are also computing homotopy groups of the formal horizontal embedding space. Finally, by using obstruction theory, see Hatcher [19, Chapter 4.3], we isolate the homotopy groups of the space of smooth embeddings. Acknowledgements: The authors are extremely grateful to R. Casals, V. Ginzburg and A. Del Pino for several discussions clarifying the paper. Also, they thank A. Hatcher for pointing out the useful reference [5]. Daniel Alvarez-Gavela has helped as quite a lot with enjoyable meetings whenever he comes to Madrid, he has pointed out several sharp remarks. The third author is also grateful to the organizers of the Engel Structures workshop held in April 2017 (American Institute of Mathematics, San Jose, California) for providing a nice environment in which this article was discussed. Last, but not least, we want to thank the excellent job that the referees have done: they have produced three long and deep reports that have helped to improve the readability of the paper a lot, not to speak of its soundness. The authors are supported by the Spanish Research Projects SEV{2015{0554, MTM2016{ 79400{P, and MTM2015{72876{EXP. The rst author is supported by a Master{Severo Ochoa grant and by Beca de Personal Investigador en Formaci on UCM. The second au- thor is funded by Programa Predoctoral de Formaci on de Personal Investigador No Doctor del Departamento de Educaci on del Gobierno Vasco. 2. Spaces of embeddings of the circle into euclidean space. Denote by Emb(N; M ) the space of embeddings of a manifold N into a manifold M equipped with the C {topology, r 5. 1 3 2.1. The space Emb(S ;R ). Theorem 2.1.1 (Hatcher, [18] Apendix: equivalence (15)). The space of parametrized un- knotted circles in R has the homotopy type of SO(3). 1 3 1 3 The group SO(4) acts freely on the connected component Emb (S ;S ) Emb(S ;S ) of p;q the parametrized (p; q) torus knots as 4 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 1 3 1 3 SO(4) Emb (S ;S ) ! Emb (S ;S ) p;q p;q (A; ) 7 ! A : 1 3 Thus, we have an inclusion SO(4) ,! Emb (S ;S ). The following result holds: p;q 1 3 Theorem 2.1.2 (Hatcher, [20] Theorem 1). The inclusion SO(4) ,! Emb (S ;S ) is a p;q homotopy equivalence. As a consequence of these results we obtain that 1 3 1 3 1 3 Corollary 2.1.3. Let Emb (S ;R ); Emb (S ;R ) Emb(S ;R ) be the connected compo- 0 p;q nent of the parametrized unknots or of the parametrized (p; q) torus knots, respectively. The fundamental groups of these spaces are given by 1 3 (Emb (S ;R )) = Z , 1 0 2 1 3 (Emb (S ;R )) = G o Z , 1 p;q p;q 2 where G is the knot group of the (p; q) torus knot. p;q 1 3 Proof. The case of the connected component Emb (S ;R ) follows from Theorem 2.1.1. 1 3 We need to study the connected component Emb (S ;R ) to conclude the proof. Consider p;q the following space 1 3 3 S tereo = f( ; x) : x 2= Image( )g Emb (S ;S ) S : p;q p;q We have two natural brations associated to the projection maps 1 3 3 // // Emb (S ;R ) S tereo S nK S tereo p;q p;q p;q p;q 3 1 3 S Emb (S ;S ) p;q where K is the image of the standard (p; q) torus knot in S . From the rst bration we p;q obtain 1 3 (Emb (S ;R )) = (S tereo ): 1 p;q 1 p;q 1 3 Moreover, from the second one and the fact that Emb (S ;S ) has the homotopy type of p;q SO(4) (Theorem 2.1.2), we obtain that the sequence // // // // 0 G (S tereo ) Z 0 p;q 1 p;q 2 is exact. Now, it is a simple exercise to check that this sequence is right split. 1 4 4 4 2.2. The space Emb(S ;R ). A long embedding of R into R is an embedding : R ! R 3 4 that coincides with the standard inclusion R ,! RR = R outside a compact neighborhood 4 4 of the origin. Let LEmb(R;R ) denote the space of long embeddings of R into R . 1 4 3 2 4 Lemma 2.2.1 (Budney). Emb(S ;R ) is homotopy equivalent to S S LEmb(R;R ). 1 4 Proof. It follows from [5, Proposition 2.2] that Emb(S ;R ) is homotopy equivalent to (SO(4) 4 4 ^ ^ S tereo)= SO(3), where S tereo = f(p; f ) : p 2= Image(f )g R LEmb(R;R ). Observe 4 3 that the quaternion structure in R (i; j; k) induces a homotopy equivalence S S tereo ! FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 5 (SO(4) S tereo)= SO(3); (v; (p; f )) 7! [((vjivjjvjkv); (p; f ))]. Finally, the natural bration 4 2 4 S tereo 7! LEmb(R;R ) has ber homotopy equivalent to S , since the space LEmb(R;R ) h:e: 4 2 is connected and R nR S . Moreover the bration homotopically splits since given k 4 k any family ' : S ! LEmb(R;R ), it admits a lift given by ' : S ! S tereo; z 7! ('(z)(0) + " i'(z) (0); '(z)) for some " > 0 small enough. Let us geometrically explain the homotopy equivalence stated in the last Lemma. Let 4 4 3 fe ; e ; e ; e g be the canonical basis of R and write R = Rhe i R he ; e ; e i. From 1 2 3 4 1 2 3 4 1 4 a long embedding we obtain an embedding of S into R closing it in the plane he ; pi, where 2 3 3 p 2 S R he ; e ; e i. Finally, the S factor acts by quaternionic multiplication on a xed 2 3 4 embedding. 1 4 2 4 4 It follows that (Emb(S ;R )) = (S ) (LEmb(R;R )) = Z (LEmb(R;R )). We 2 2 2 2 provide an explicit construction of the generator of the Z factor in this decomposition. For 2 3 every point p 2 S R he ; e ; e i take the standard parametrized unknot in spanfe ; pg 2 3 4 1 centered at p and tangent to the line generated by e at 0. This gives us a 2{parametric family of knots, that we denote by S , whose homotopy class is the generator (1; 0) of 1 4 (Emb(S ;R )). Explicitly, S is de ned as: 2 U 2 1 4 S : S ! Emb(S ;R ) p 7 ! (t) where, if p = ( ; ; ) 2 S , 1 2 3 (t) = sin(2t)e + (1 cos(2t))( e + e + e ): p 1 1 2 2 3 3 4 Figure 1. Construction of S (schematically). 1 4 Remark 2.2.2. In [5, Proposition 3.9(4)] Budney shows that the space Emb(S ;R ) is simply 1 4 connected and (Emb(S ;R )) = Z Z. Moreover, he provides an explicit construction of the second generator of the second homotopy group ([5, Theorem 3.13]). 1 4 In section 7 we provide an alternative proof of the fact that Emb(S ;R ) is simply connected based on the techniques developed in this paper. 6 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 3. Formal Legendrian Embeddings in R . We denote by the standard contact structure in R (x; y; z) given by = ker(dz ydx). Throughout the Section we x the Legendrian framing @ . 3.1. Formal Legendrian Embeddings in R . 1 3 0 De nition 3.1.1. An immersion : S ! R is said to be Legendrian if (t) 2 for all (t) t 2 S . If is an embedding, we say it is a Legendrian embedding. De nition 3.1.2. (a) A formal Legendrian immersion in R is a pair ( ; F ) such that: 1 3 (i) : S ! R is a smooth map. 1 3 (ii) F : S ! (TR nf0g) satis es F (t) 2 . (t) (b) A formal Legendrian embedding in R is a pair ( ; F ), satisfying: 1 3 (i) : S ! R is an embedding. 1 3 0 (ii) F : S ! (TR nf0g), is a 1{parametric family, s 2 [0; 1], such that F = and s 0 F (t) 2 . 1 (t) Use the framing h@ i to trivialize the contact distribution understood as a bundle. This 2 1 provides a bundle isomorphism ' R . From now on, we will understand the map F : S ! 1 2 2 1 2 1 1 2 2 S S \ R and the family F : S ! S with F : S ! S S \ R . We say that an s 1 immersion is strict if it is a non injective map. 3 3 3 Denote by LegImm(R ) the space of Legendrian immersions in R , by Leg(R ) the space of 3 strict 3 3 3 Legendrian embeddings in R and by LegImm (R ) = LegImm(R )nLeg(R ) the space of strict Legendrian immersions. Denote also by FLegImm(R ) the space of formal Legendrian immersions and by FLeg(R ) the space of formal Legendrian embeddings. These de nitions make sense for immersions and embeddings of the interval. We de ne LegImm([0; 1];R ), strict 3 3 3 3 Leg([0; 1];R ), LegImm ([0; 1];R ), FLegImm([0; 1];R ) and FLeg([0; 1];R ) analogously. r r 1 3 r Remark 3.1.3. The space of C {maps M aps (S ;R ) endowed with the C {topology is a Banach vector space. It has an open submanifold that is the space of immersions 1 3 3 Imm (S ;R ). Denote by the standard contact form in R ; ie ker( ) = . There is a r 1 3 1 1 3 smooth submersion ~ : M aps (S ;R ) ! (S ); 7! ( ). We have that LegImm (R ) is 1 3 a Banach submanifold of Imm (S ;R ) (see Lang [24, Corollary 5.7, page 19]). Whenever we speak about families of curves in the space of Legendrians we are implicitly asumming this Banach structure. This means that our maps are C instead of smooth but all the de nitions in the paper make sense for C {maps whenever r 5. To simplify the notation we will be writing smooth maps instead of C {maps unless it is stated otherwise. The bound r 5 comes from the fact that we are considering 3{dimensional families of Legendrians. In order to locally classify them we need to consider up to 4 derivatives of their front projection (5 of the original Legendrians). It is important to note that any 2 LegImm (R ) admits a particular type of immersed chart ' , called a Weinstein chart, which identi es via a immersed contactomorphism a Observe that is naturally (co)oriented by the contact form dz ydx and, thus, @ determines a unique oriented framing up to homotopy. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 7 1 1 tubular neighborhood of the zero section of the jet space J (S ) (see Geiges [15, Example 2.5.11]) with a tubular neighborhood of (S ). This construction provides the local structure produced by the Implicit Function Theorem applied in the last paragraph. We can just check that the exponential map in a C {neighborhood of is de ned as the following construction: r 1 r 1 1 ~ ~ take f 2 M aps (S ;R), a map f 2 M aps (S ;S ) such that f () = + f (). Then, the map is de ned as r 1 r 1 r 3 C : U M aps (S ;R)M aps (S ;R) ! V LegImm (R ) (1) 1 1 (f; g) 7 ! ' j (g f ): Note that this is a local chart (ie a di eomorphism). However, if we consider r = 1 the previous map is not surjective in the reparametrizations direction since it is well-known that 1 1 the action of the Lie algebra of Di (S ) is not surjective on Di (S ). On the other hand, if 1 3 1 we quotient out LegImm (R ) by the action of Di (S ), ie we consider non{parametrized Legendrians, the map is a local di eomorphism. This is an obvious consequence of the 1 1 characterization of the Legendrian curves in J (S ) as graphs of smooth functions. We use r < 1 in order to make sure that the parametrizations are surjective. We do not forget about reparametrizations because they are important in our algebraic topology arguments. From now on we will skip the index r = 5 in all the statements, unless it is not clear from the context. All the spaces of Legendrians are equipped with the C {topology. On the other hand, the spaces of formal Legendrians are equipped with the product topology that is the C {topology r 1 for the rst factor (the smooth immersion/embedding) and the C {topology for the second factor (the formal derivative). Remark 3.1.4. It is well{known that h{principle holds for Legendrian immersions (see, eg, 3 3 Eliashberg and Mishachev [11]). Hence, (LegImm(R )) Z, (LegImm(R )) Z and = = 0 1 3 3 (LegImm(R )) = 0; for all k 2. The connected components of LegImm(R ) are given by the rotation number. The rotation number of an Legendrian immersion is Rot( ) = deg( : 1 1 3 3 S ! S ). Let us explain the group (LegImm(R )) Z. Take a loop in LegImm(R ), the integer is just Rot ( ) = deg( 7! ( ) (0)), we call this number rotation number of the loop. These invariants make sense in the formal case and the de nitions are the obvious ones. 3 3 1 3 3.2. The space FLeg(R ). Consider the space FLeg(R ) = f( ; F )j 2 Emb(S ;R ); F 2 1 1 3 3 M aps(S ;S )g. We have a natural bration FLeg(R ) ! FLeg(R ). In order to compute 3 3 0 the homotopy groups of FLeg(R ), take 2 Leg(R ) and x ( ; ) as base point. The ber 1 2 over this point is F = F 0 = (M aps(S ;S )). We have the following exact sequence of ( ; ) homotopy groups associated to the bration: // [ (FLeg(R )) tt 3 3 // // [ (F ) (FLeg(R )) (FLeg(R )) 1 1 1 tt 3 3 // // [ // (F ) (FLeg(R )) (FLeg(R )) 0 0 0 0 8 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 3 1 3 1 3 [ [ Notice that FLeg(R ) has the homotopy type of Emb(S ;R )S Z. Hence, (FLeg(R )) 1 3 3 (Emb(S ;R )) Z, where the integer is the rotation number. Moreover, (FLeg(R )) 0 1 1 3 (Emb(S ;R )) Z and the Z factor is given by the rotation number of the loop. Finally, 3 1 3 (FLeg(R )) (Emb(S ;R )) for all k 2. k k The homotopy groups of F are easily computable. Just observe that there is a bration 1 2 2 2 2 M aps(S ;S ) ! S de ned via the evaluation map, with ber over p 2 S given by (S ). 2 1 2 As every element [f ] 2 (S ) can be lifted to an element [f ] 2 (M aps(S ;S )), de ned n n n as 1 n f (p)(t) = f (p); t 2 S ; p 2 S ; all the diagonal maps in the associated exact sequence are zero. This implies that there are 2 1 2 2 short exact sequences ( (S )) ! (M aps(S ;S )) ! (S ) for n 1. In particular, n p n n since S is simply connected, we obtain that 1 2 2 2 (F ) (M aps(S ;S )) ( (S )) (S ) Z: = = = = 0 1 1 p 2 Moreover, theses sequences are right split and, thus, split for n > 2 since the groups involved are abelian. So, we have 1 2 2 2 2 2 (F ) = (M aps(S ;S )) = (S ) ( (S )) = (S ) (S ) = Z Z 1 2 2 2 p 2 3 1 2 2 2 2 2 and (F ) (M aps(S ;S )) (S ) ( (S )) (S ) (S ) Z Z : = = = = 2 3 3 3 p 3 4 2 3 1 3 Lemma 3.2.1. (FLeg(R )) = (Emb(S ;R )) Z Z. 0 0 3 1 3 Proof. It is sucient to show that every element in (FLeg(R )) = (Emb(S ;R )) 1 1 1 1 1 3 3 (M aps(S ;S )) = (Emb(S ;R )) Z can be lifted to an element in (FLeg(R )). 1 1 1 3 0 1 1 2 Take a loop ( ; F ) in FLeg(R ). Let F = ( ) : S S (; t) ! S be the derivative 0 0 1 1 2 F (; t) = ( ) (t), we need to show that F = ( ) is homotopic to the map F : S S ! S . 0 0 1 1 1 2 Observe that the homotopy classes of maps from S S to S are classi ed by the degree and deg(F ) = 0, so we just need to show that deg(F ) = 0 to complete the proof. 1 0 Indeed, the map ( ) (t) if " = 0, (t+") (t) (2) G (; t) = if 0 < " < 1, jj (t+") (t)jj ( ) (t) if " = 1, 1 0 is well{de ned, because , 2 S , is an embedding. Thus, F = ( ) is homotopic to 0 0 F = ( ) and deg(F ) = deg( ) = 0. 0 0 3.3. Classi cation of formal Legendrian embeddings in R . We have checked that 3 1 3 (FLeg(R )) (Emb(S ;R )) Z Z. The rst Z corresponds to the rotation number 0 0 and we will show that the second one corresponds to the Thurston{Bennequin invariant. Let us re ne the de nition of formal Legendrian embedding to extend the de nition of the Thurston{Bennequin invariant to the formal case. De nition 3.3.1. A formal extended Legendrian embedding in R is a pair ( ; G ), satisfying: 1 3 (i) : S ! R is a embedding. (ii) G : S ! SO(3), is a smooth family in the parameter s 2 [0; 1], such that G = Id s 0 and G ( ) 2 . 1 (t) FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 9 3 3 We denote FELeg(R ) for the space of formal extended Legendrian embeddings in R equipped r r 1 with the C {topology in the rst factor and the C {topology in the second one. 3 3 0 Remark 3.3.2. The natural bration t : FELeg(R ) ! FLeg(R ); ( ; G ) 7! ( ; G ( )), has s s contractible bers. Thus, this map is a weak homotopy equivalence. 3 0 Given ( ; G ) 2 FELeg(R ) we have a well{de ned formal contact framing F of (G ( )) s FCont 1 0 1 given by the Legendrian condition G ( ) . Then, G (F ) de nes a framing of 1 (t) FCont the normal bundle of . On the other hand, we have a topological framing F of given Top by a Seifert surface of . De nition 3.3.3. Let ( ; G ) 2 FELeg(R ). The Thurston{Bennequin invariant is tb( ; G ) = s s 1 1 tw (G (F );F ), ie the twisting of G (F ) with respect to F . FCont Top FCont Top 1 1 3 1 3 The Thurston{Bennequin invariant is de ned over FELeg(R ) = f( ; G)j 2 Emb(S ;R ); G 2 1 0 1 1 M aps(S ; SO(3)); G( ) 2 g. Furthermore, since the unique oriented S {bundle over S (t) 3 1 3 is the trivial one, (FELeg(R )) (Emb(S ;R ))ZZ. The rst Z is just the rotation 0 0 number and the second one corresponds to the Thurston{Bennequin invariant. Now we can state the main result of this Section, which is well{known in the literature. Theorem 3.3.4. Formal Legendrian embeddings are classi ed by their parametrized knot type, rotation number and Thurston{Bennequin invariant. 3 3 ^ \ The proof of this result follows directly using the bration F : FELeg(R ) ! FELeg(R ) 3 3 and the fact that the map t : FELeg(R ) ! FLeg(R ) is a weak homotopy equivalence. Note also that the bration F has connected ber, because its is given by (SO(3)) = 0 2 0. This completes the proof. However to get a more geometric picture, we will express the isomorphism in more concrete terms. Clearly the isomorphism preserves the rotation invariant, ie the rotation number ( ; G ) is sent to the rotation invariant of ( ; G ( )). To s s 0 3 understand the rest of the isomorphism we x a base point ( ; F = ) in FLeg(R ) with Rot( ; ) = 0, i.e we declare the base point to be a Legendrian embedding with zero rotation. s 0 1 0 Now, given an element of the ber, ie ( ; F ) with F = F = , we claim that for [( ; 0; k)] 2 3 1 0 (FLeg(R )), the isomorphism (t) is given by tb( (t) ( ; 0; k)) = tb( ; ) 2k. In other 0 0 0 words, it depends on the choice of base point. This is obvious if we check that given a double stabilization, see De nition 7.2.3 for a precise de nition, of the Legendrian knot, the value of the degree invariant in the ber increases by 1 and it is a simple computation to check that the tb decreases by 2. 3.4. Fundamental group of formal Legendrian Embeddings in R . As a consequence of Lemma 3.2.1 we have that the following sequence is exact: 3 3 // // [ (FLeg(R )) (FLeg(R )) 2 2 tt 3 3 // // // (F ) (FLeg(R )) (FLeg(R )) 0 1 1 1 z z 3 3 [ [ Take a 2{sphere ( ; F ) in FLeg(R ), the diagonal map d : (FLeg(R )) ! (F ) measures 2 1 z z z z 3 the obstruction to lifting ( ; F ) to a 2{sphere ( ; F ) in FLeg(R ), ie the obstruction to nd 1 s 10 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS z 0 2 1 2 z 0 a homotopy between the derivative map F = ( ) : S S ! S , (z; t) 7! F (z; t) = ( ) (t), 0 0 2 1 2 and the map F : S S ! S . Note that by the Legendrian condition F is nullhomotopic, 1 1 since it is not surjective. The rst obstruction to nd this homotopy is just the degree of z 0 2 2 F (z; 0) = ( ) (0) and corresponds to the rst Z factor of (F ) (S ) (S ) ZZ. = = 0 1 2 3 z 0 z 0 In particular, since ( ) (0) is homotopic to ( ) (0) this obstruction vanishes (see equation (2)). Theorem 3.4.1. The sequence 3 1 3 // // // // 0 Z Z (FLeg(R )) (Emb(S ;R )) Z 0 m 1 1 0 1 3 is exact, where m 0. In particular, if we x the connected component Emb (S ;R ) 1 3 Emb(S ;R ) of the parametrized unknot or of the parametrized (p; q) torus knot we have that m = 0 and so 3 1 3 // // // // 0 Z Z (FLeg(R )) (Emb (S ;R )) Z 0 1 1 is exact. Proof. We only need to check the particular cases mentioned above. For the connected 1 3 component Emb (S ;R ) of the parametrized unknot the result follows from Theorem 2.1.1, 3 1 3 3 [ [ since (FLeg (R )) = (Emb (S ;R )) = (SO(3)) = 0, where FLeg (R ) stands for a 2 2 0 2 0 0 formal Legendrian connected component of the smooth unknot. 1 3 On the other hand, x the connected component Emb (S ;R ) of the parametrized (p; q) p;q torus knot and consider the commutative diagram 1 3 1 3 // Emb (S ;R ) Emb (S ;S ) p;q p;q _ _ 1 3 1 3 // Imm(S ;R ) Imm(S ;S ) de ned by the natural inclusions, where Imm(N; M ) denotes the space of immersions of a manifold N into a manifold M equipped with the C {topology, r 5. 1 3 By the Smale{Hirsch Theorem for immersions (see [21] or [11]) we have that Imm(S ;R ) has 1 2 1 3 1 3 the homotopy type ofM aps(S ;S ) and Imm(S ;S ) has the homotopy type ofM aps(S ;S ) 1 2 1 3 1 3 M aps(S ;S ). Moreover, the map induced by the inclusion Emb (S ;R ) ,! Imm(S ;R ) p;q z 1 3 z 0 1 3 at {level sends the homotopy class [ ] 2 (Emb (S ;R )) to [( ) ] 2 (Imm(S ;R )) = 2 2 p;q 2 1 2 3 1 2 (M aps(S ;S )); ie coincides with the diagonal map d : (FLeg(R )) ! (F ) (M aps(S ;S )). 2 2 1 2 Consider the induced commutative diagram at {level 1 3 1 3 // (Emb (S ;R )) (Emb (S ;S )) 2 p;q 2 p;q 1 3 1 3 // (Imm(S ;R )) (Imm(S ;S )) 2 2 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 11 1 3 since (Emb (S ;S )) is trivial (see Theorem 2.1.2) it is sucient to show that the homo- 2 p;q 1 3 1 3 morphism (Imm(S ;R )) ! (Imm(S ;S )) is injective to conclude the proof; but this is 2 2 2 1 clear, by using the h{principle for immersions, since the degree of the induced map S S 3 2 1 2 to S is zero and the induced map for the derivative from S S to S is sent to itself by the inclusion. 4. An application: Kalman's loop. 4.1. K alm an's loop. K alm an has constructed a series of examples of loops of Legendrian 3 1 3 positive torus knots non{contractible in the space Leg(R ), though contractible in Emb(S ;R ) [23]. Let us prove that K alm an's examples are non trivial even as loops of formal Legendrian embeddings; that is, in the space FLeg(R ). We will prove that they are not contractible for any choice of parametrization, thus they are not contractible as loops of unparametrized oriented knots. Consider a Legendrian positive (p; q) torus knot, a loop is described in Figure Figure 2. The loop in front projection (p = 3; q = 7). Note that we have to cycle 2p = 6 times. 2. The loop takes the p strands of the knot to the cyclic rotation of them. This geometri- cally corresponds to a 2=p rotation along the core of the de ning torus. Let us consider 2p concatenations of this loop. Thus, it is generated by two full rotations along the core of the torus. 4.1.1. Simpli ed position. First, we will deform through formal loops the initial loop into a formal loop in a \simpli ed" position. 2 3 Step 1. Consider the contactomorphism f (x; y; z) = (x=r; y=r; z=r ) in the standard (R ; ker(dz ydx)). By using it, we assume that the de ning torus for the loop has arbitrarily small merid- ional radius. Therefore, the knot is C {close to the core of the torus. We are not using the standard notion of C {closeness, but a weaker one. Ie we mean that a sequence of im- 1 1 mersions ^ is C { close to an immersion ~ if for any " > 0 and for any point t 2 S : for 12 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS every k 2 Z large enough there exists a point (t) 2 S such that j ^ (t) ~( (t))j " and 0 0 ^ (t) ~ ( (t)) j j ", 0 0 jj ^ (t)jj jj ~ ( (t))jj Moreover, by further shrinking, the knot and the core , they may be assumed to be arbi- trarily close to ker dz; ie C {close to their Lagrangian projections. (a) Front projection of the (b) Knot C {close to the core. knot and the core (in red colour). Figure 3. C {approximation of the knot to the core, shown in the front projection. Step 2. Denote the initial loop of Legendrian embeddings. Understood as a formal loop, it is written as ( ; F ), where F = ( ) . Let us construct a 1{parametric family of formal s s ;u ;u loops ( ; F ), u 2 [0; 1], de ned as follows ;u (i) = , ;u 0 0 (ii) F = (1 s)( ) + s((1 u)( ) + u@ ). s y 0 2 This is a family of formal loops, since ( ) is never a negative multiple of @ . To check that, note that is C {close to . Step 3. We consider the family of rotations fr 2 SO(3)g taking the quadrant XY to v v2[0;1] ;u ;u the quadrant ZX , see Figure 4. Construct a family of formal loops ( ; F ); u 2 [1; 2], as follows ;u (i) = r , u 1 ;u ;u 0 (ii) F = (1 s)( ) + s@ . s y ;u 0 Again, this is a family of formal loops because ( ) is never a negative multiple of @ . Note that we are using that is C {close to . Step 4. Finally, we turn over the left lobe of the unknot core r by an isotopy de ned as follows. Take polar coordinates (r; ') in the plane Y Z and " > 0 small enough. We de ne the isotopy as: (x; r; ' + u (x) ) if x ", f (x; r; ') = (x; r; ') if " x where 0 u 1 and : R ! R is a non{decreasing smooth function satisfying We are assuming an orientation of the Legendrians. The argument with the opposite orientation runs in the same way by changing @ by @ . y y See de nition 5.1.3. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 13 (x) = 1 for all x 0, (x) = 0 for all x ". We apply the isotopy to r , see Figure 4. Again, the derivative is never tangent to @ 1 y and thus we can interpolate to @ . We have proven that our initial loop of Legendrian embeddings is homotopic to the loop of formal Legendrian embeddings ( ~ ; F ) de ned as follows: (i) ~ is the loop of parametrized (p; q) torus knots supported in the torus associated to the unknot contained in the plane XZ . The loop is obtained by a rotation of 4 radians of the standard (p; q){embedding in the direction of the parallel of the supporting torus. (ii) F = (1 s)( ~ ) + s@ . -Z -Z -Z X X X (a) Step 2. (b) Step 3. (c) Step 4. Figure 4. Construction of the path of loops. We represent the moves of the core . 4.2. Set of parametrizations of the family of loops. As an outcome of the previous discussion, we may assume that our formal Legendrian parametrized (p; q) torus knot can be written as ( ; F ), where 0 1 (cos(2pt) + 2) cos(2qt) @ A (t) = sin(2pt) , (cos(2pt) + 2) sin(2qt) F (t) = s@ + (1 s)( ) (t): s y One particular parametrization of the loop can be written as ( ; F ), where 0 10 1 cos(4) 0 sin(4) (cos(2pt) + 2) cos(2qt) @ A@ A (t) = 0 1 0 sin(2pt) , sin(4) 0 cos(4) (cos(2pt) + 2) sin(2qt) F (t) = s@ + (1 s)( ) (t). We will show that any possible parametrization of the loop gives raise to a non{trivial loop of parametrized formal Legendrian embeddings. Up to homotopy, the possible parametrizations of the formal Legendrian loop are given by: 14 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS ;k (t) = (t + k); ;k ;k 0 F (t) = F (t + k) = s@ + (1 s)( ) (t); + 4 where k 2 Z. This is because (Di (S )) = (SO(2)) = Z. 1 1 We will prove the following statement. ;k ;k Proposition 4.2.1. The loop of formal Legendrian embeddings ( ; F ) is non trivial for any k 2 Z. This proves that the loop is non trivial as a loop of non parametrized formal Legendrian knots. 4.3. Proof of Proposition 4.2.1. It follows from the previous discussion that the loop ;k 1 3 ;k (t) of smooth embeddings lies in SO(4) Emb (S ;S ), ie (t) = A (t), where p;q ;k p p 3 2 A 2 SO(4). More speci cally, on S ( 2), the 2 radius sphere in C , we have ;k 2ipt 1 0 ;0 (t) = : 4i 2iqt 0 e e Thus, in these coordinates the other parametrizations are given by 2ipk 2ipt e 0 e ;k ;0 (t) = (t + k) = : 2iqt 2i(2+qk) 0 e ;k 1 3 By Theorem 2.1.2 the parametrized loop is trivial in Emb (S ;S ) if and only if 2 + p;q k(p + q) is even. From now on we will assume that this is the case. Thus, there is a family ~ ~ ~ fA g such that A = A . Since (SO(4)) = 0, the disk A is unique up (r;);k (r;)2D (1;);k ;k 2 (r;);k 1 3 to homotopy xing the boundary and the same holds for the disk A in Emb (S ;S ). (r;);k p;q By Theorem 3.4.1, we have the following exact sequence: m m 1 2 3 3 // // // [ // (3) 0 (F ) (FLeg(R )) (FLeg(R )) 0 1 1 1 ;k ;k 3 Thus, in order to prove that our loop ` = [( ; F )] 2 (FLeg(R )) is non trivial we k 1 distinguish two cases: ;k 1 3 5 4.3.1. Case 1. k 6= 0. We claim that m (` ) 6= 0, ie [ ] 6= 0 2 (Emb (S ;R )) . 2 k 1 p;q 1 3 Recall from Corollary 2.1.3 that (Emb (S ;R )) = (S tereo ) and that we have an 1 p;q 1 p;q exact sequence 1 3 // // // // 0 G (S tereo ) (Emb (S ;S )) 0 p;q 1 p;q 1 p;q Thus, we must show that [(A ;1)] 2 (S tereo ) is non trivial. Ie the family of loops ;k 1 p;q 1 3 that is trivial by hypothesis in (Emb (S ;S )) does not admit a capping disk whose 1 p;q Remember that the Legendrian knots are oriented. 5 ;k ;k Observe that Rot ( ; F ) = 0: s FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 15 evaluation map avoids 1 2 S . We check it by composing with the 1{parametric family of 1 1 1 ~ ~ ~ loops A 2 SO(4), we obtain a 1{parametric family of loops (A A ; A 1), ;k (r;);k (r;);k (r;);k r 2 [0; 1]. For r = 0 we obtain the initial loop and for r = 1 we obtain the loop ( ; A 1). ;k Thus, these two loops represent the same element of (S tereo ). Moreover, [( ; A 1)] 1 p;q ;k 1 3 can be lifted to G , since it lies on the ber de ned by the element [ ] 2 (Emb (S ;S )). p;q 1 p;q So, we are reduced to check whether [A 1] 2 G is trivial. The knot group of the (p; q) p;q ;k p q pk torus knot is G = ha; b : a = b i. Thus, [A 1] = b 6= 0, since b is a non torsion element p;q ;k of G . p;q ;k Figure 5. Visualization of the loop for a (5; 2) torus knot (k 6= 0). 4.3.2. Case 2. k = 0. Since m (` ) 2 (FLeg(R )) is zero, there exists A 2 (F ) such that 2 0 1 1 m (A) = ` . We are going to geometrically check that A 6= 0 and therefore, by the injectivity 1 0 of m provided by the sequence (3), the non triviality of ` follows. 1 0 ;0 ;0 3 Write ( ; F ) = ( ; F ). Note that in (R (x; y; z); ker(dz ydx)) the parametrized loop is written as: 0 10 1 cos(4) 0 sin(4) (cos(2pt) + 2) cos(2qt) @ A@ A (t) = B (t) = 0 1 0 sin(2pt) , sin(4) 0 cos(4) (cos(2pt) + 2) sin(2qt) F (t) = s@ + (1 s)( ) (t). ;0 Figure 6. Visualization of the loop for a (5; 2) torus knot (k = 0). 16 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS (r;) ~ ~ ~ Moreover, is bounded by ~ (t) = B (t), where B 2 SO(3), such that B = B (r;) (r;) (1;) ~ ~ and is homotopic to A inside SO(4). Thus, we have a disk D(r; ) = f(B ; @ )g in (r;) r; 3 3 FLeg(R ) that bounds m (` ). We try to lift it to a disk in FLeg(R ) that bounds ` . There 2 0 0 is no homotopical obstruction to lifting it for the punctured disk D(r; )j . Therefore, the r>0 homotopy obstruction is represented by an element of (F ). So we obtain a loop ( ; F ) 1 0 over the ber of ( ; @ ) where = B . It follows by construction that A = [( ; F )]. Let 0 y 0 0 0 us perform the computation. 0 1 B ( ) if 0 s ; 2s; (4) F = 0 1 F = (2s 1)@ + (2 2s)B ( ) if s 1: 2s 1 0 2 ~ ~ ~ ~ Observe that F = and F = @ . Thus, we can understand F as a map F : S ! 0 0 1 s 1 2 1 2 2 2 M aps(S ;S ). It follows that A = [F ] 2 (F ) (M aps(S ;S )) (S ) ( (S )), = = 1 2 2 2 p that is the fundamental group of the ber. We already computed this group in Theorem 3.4.1. Moreover, we also showed that the morphism to (FLeg(R )) induced by the inclusion is injective. To conclude the proof we must check that [F ] 6= 0. In order to see this, we will verify that 2 2 ~ ~ deg(F (0) : S ! S ) is nonzero. This degree is the rst coordinate of [F ] 2 ZZ = (F ) = 2 2 (S ) ( (S )). 2 2 p We give an explicit description of fB g in SO(3). Identify SO(3) with RP in the usual r; way. Ie understand RP as the 3{ball of radius with its boundary points identi ed via the antipodal map. Then a point p = (p ; p ; p ) in the described 3{ball corresponds in SO(3) to 1 2 3 2 2 2 the rotation of angle p + p + p in R (x; y; z) around the axis described by its position 1 2 3 vector p. In these coordinates, see the left drawing in Figure 7, the loop B is given by (0; 4; 0) if 0 ; 1 3 B = (0; 2 + 4; 0) if ; 4 4 (0; 4 + 4; 0) if 1: De ne the disk fB g as the intersection of the plane fz = 0g with the {radius ball. It r; 2 3 3 1 2 produces an RP = RP fz = 0g RP . We have B = fB : 2 S g RP . We obtain RP n B is an embedded 2-disk. See Figure 7. We have that (0) = 2p@ + 6q@ . Substituting in equation (4) for t = 0, we get y z B (2p@ + 6q@ ) if 0 s ; 2s; y z F (0) = F = (2s 1)@ + (2 2s)B (2p@ + 6q@ ) if s 1: y y z 2s 1 Moreover, the maps, u 2 [0; 1], B ((1 u)(2p@ + 6q@ ) + u@ ) if 0 s ; 2s; y z z G(s; ; u) = F = (2s 1)@ + (2 2s)B ((1 u)(2p@ + 6q@ ) + u@ ) if s 1 y y z z 2s 1 are always non zero. Thus, the map F (0) = G(s; ; 0) is homotopic to B (@ ) if 0 s ; 2s; z G(s; ) = G(s; ; 1) = (2s 1)@ + (2 2s)B (@ ) if s 1: y z In order to compute the degree of G(s; ), we write G(s; ) = (g (s; ); g (s; ); g (s; )) 2 x y z 2 1 S (@ ; @ ; @ ) and we check that #G ( @ ) = 1: x y z y FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 17 If s 1 then G(s; ) is a linear combination with positive coecients between 1 2 @ and B (@ ) 2 S (@ ; @ ) S (@ ; @ ; @ ) thus g (s; ) 0. y z x z x y z y If 0 s then G(s; ) is just a rotation around an axis in the XY {plane acting over @ . The rotation of angle around the X {axis is the unique rotation that sends @ to @ . Thus, G ( @ ) = f(s ; )g where (s ; ) is the only point that satis es z y y 0 0 0 0 that B is the mentioned rotation. 2s ; 0 0 The map G is a local di eomorphism in a neighborhood of the point (s ; ). Thus, @ is 0 0 y a regular value for G and j deg(F (0))j = j deg(G(s; ))j = 1 6= 0. Figure 7. Explicit construction of the capping disk. 5. Formal Horizontal Embeddings in R . In this Section, we denote by D the standard Engel structure in R (x; y; z; w) given by D = ker(dy zdx)\ ker(dz wdx). Throughout the Section we x the framing of D given by the kernel of the Engel structure W = h@ i. 5.1. Formal Horizontal Embeddings in R . 1 4 0 De nition 5.1.1. An immersion : S ! R is said to be horizontal if (t) 2 D for all (t) t 2 S . When is an embedding, we say it is a horizontal embedding. Remark 5.1.2. The projection 4 3 : R (x; y; z; w) ! R (x; z; w) is called the Geiges projection. Notice that the Geiges projection maps horizontal embeddings to Legendrian immersions in (R (x; z; w); ker(dz wdx)). Moreover, since the front projection 3 3 for (R (x; z; w); ker(dz wdx)) is the Lagrangian projection for (R (x; y; z); ker(dy zdx)) the Geiges projection of a horizontal embedding 2 Hor(R ) is a Legendrian immersion which satis es the following area condition zdx = 0: The composition of the Geiges projection and the front projection is called the front Geiges projection. The front Geiges projection of is denoted by . The front projection of a FG Legendrian immersion is denoted by and the Lagrangian projection is denoted by . F L 1 2 De nition 5.1.3. Let 2 Imm(S ;R ). By a lobe of we mean any segment of the curve 2 1 that encloses a topological disk in R n (S ). 18 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS When we do not specify it explicitly, by a lobe of a horizontal embedding we mean a lobe of . FG De nition 5.1.4. (a) A formal horizontal immersion in R is a pair ( ; F ) such that: 1 4 (i) : S ! R is a smooth map. 1 4 (ii) F : S ! (TR nf0g) satis es F (t) 2 D . (t) (b) A formal horizontal embedding in R is a pair ( ; F ) satisfying: 1 4 (i) : S ! R is an embedding. 1 4 0 (ii) F : S ! (TR nf0g), is a 1{parametric family, s 2 [0; 1], such that F = and s 0 F (t) 2 D . 1 (t) 2 1 3 Use h@ i to identify D R . From now on, we will understand the family F : S ! S with w s 1 1 3 2 F : S ! S S \ R . 4 4 Denote by HorImm(R ) the space of horizontal immersions and by Hor(R ) the space of 4 r horizontal embeddings in R . All these spaces are endowed with the C {topology. De- 4 4 4 note by FHorImm(R ) the space of formal horizontal immersions in R and by FHor(R ) the space of formal horizontal embeddings in R . All these spaces are endowed with the r r 1 (C ; C ){topology. These de nitions make sense for immersions and embeddings of the in- 4 4 4 4 terval. We de ne HorImm([0; 1];R ), Hor([0; 1];R ), FHorImm([0; 1];R ) and FHor([0; 1];R ) analogously. Remark 5.1.5. In the same vein as in the Legendrian case (see Remark 3.1.3) we can equip 4 4 4 4 the spaces HorImm(R ); Hor(R ); FHorImm(R ) and FHor(R ) with a structure of Banach manifolds. r 4 Like in the Legendrian case it is important to note that any 2 HorImm (R ) that is transverse to the kernel W of the Engel structure admits a particular type of immersed chart ' which identi es via an immersed Engelmorphism a tubular neighborhood of the zero 2 1 1 section of the jet space J (S ) (see [27, Lemma 1]) with a tubular neighborhood of (S ). This construction provides local charts in the C {topology just by de ning the map r 1 r 1 4 C : U M aps (S ;R)M aps (S ;R) ! V HorImm (R ) (5) 1 2 (f; g) 7 ! ' j (g f ): This proves that the open subset of horizontal immersed curves that are transverse to the kernel of the Engel structure is a Banach manifold. 4 4 Remark 5.1.6. In [27] it is proved that the inclusion HorImm(R ) ,! FHorImm(R ) is 4 4 a weak homotopy equivalence. Thus, (HorImm(R )) Z, (HorImm(R )) Z and = = 0 1 (HorImm(R )) = 0 for all k 2. As in the Legendrian case the connected components of 4 4 HorImm(R ) are classi ed by a rotation number. The rotation number of 2 HorImm(R ) is 0 1 1 4 Rot( ) = deg( : S ! S ). In the same way the homotopy type of a loop in HorImm(R ) is determined by the integer Rot ( ) = deg( 7! ( ) (0)) which is called rotation number of the loop. Moreover, it is known that horizontal embeddings in R are also classi ed by their rotation number. This result was proven by Adachi [1] and Geiges [16]. Note that these invariants are de ned in the formal setting in the obvious way. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 19 4 4 4 5.2. The space FHor(R ). Consider the natural bration FHor(R ) ! FHor(R ) = f( ; F ) : 1 4 1 1 4 0 4 2 Emb(S ;R ); F 2 M aps(S ;S )g. Take 2 Hor(R ) and x ( ; ) 2 FHor(R ) as base point to compute the homotopy of FHor(R ). The ber over this point is F = 1 3 F 0 = 0 (M aps(S ;S )). To compute some homotopy groups of F use the bration ( ; ) 1 3 3 M aps(S ;S ) ! S de ned by the evaluation map. Observe that, since every element 3 1 3 [f ] 2 (S ) can be lifted to an element [f ] 2 (M aps(S ;S )), de ned as n n n 1 n f (p)(t) = f (p); t 2 S ; p 2 S ; all the diagonal maps in the associated exact sequence are zero and the associated short exact sequences are right split and, thus, split for n > 1. Since S is 2{connected we conclude that 1 3 1 3 3 3 (F ) (M aps(S ;S )) = 0, (F ) (M aps(S ;S )) ( (S )) (S ) Z and = = = = = 0 1 1 2 2 p 3 1 3 3 3 3 3 (F ) (M aps(S ;S )) (S ) ( (S )) (S ) (S ) Z Z . = = = = 2 3 3 3 p 3 4 2 4 1 4 1 Furthermore, the space FHor(R ) has the homotopy type of Emb(S ;R ) S Z. Thus, 4 4 1 4 4 1 4 [ [ [ (FHor(R )) Z, (FHor(R )) (Emb(S ;R ))Z and (FHor(R )) (Emb(S ;R )) = = = = 0 1 1 2 2 Z[S ] (LEmb(R;R )) (see Lemma 2.2.1). U 2 4 4 The exact sequence associated to the bration FHor(R ) ! FHor(R ) takes the following form: // Z (LEmb(R;R )) ss 4 1 4 // // Z (FHor(R )) (Emb(S ;R )) Z 1 1 ss // // // 0 (FHor(R )) Z 0 In particular, this proves that formal horizontal embeddings are classi ed by their rotation number. We can state the next result concerning the fundamental group of each connected component of FHor(R ) : Lemma 5.2.1. The sequence 4 1 4 // // // // 0 Z (FHor(R )) (Emb(S ;R )) Z 0 2 1 1 is exact. Proof. It is sucent to show that the image of the diagonal map d : (FHor(R )) ! (F ) is 2 1 z z 4 2Z. Observe that d measures the obstruction to lifting an element [( ; F )] 2 (FHor(R )) 4 z 0 2 1 3 z 0 to (FHor(R )). In other words, let F = ( ) : S S ! S , (z; t) 7! F (z; t) = ( ) (t), 2 0 0 be the derivative map. The homomorphism d measures the obstruction to nd a homotopy z 0 2 1 3 z between the derivative map F = ( ) and the map F : S S ! S , (z; t) 7! F (t). Since 0 1 z z 6 the map F is null{homotopic, we conclude that d[( ; F )] = deg(F ). 1 0 6 4 3 We are arguing as in Lemma 3.2.1, but now the antipodal map in S(R ) = S has degree 1. 20 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 4 1 4 4 Recall that (FHor(R )) (Emb(S ;R )) Z[S ] (LEmb(R;R )), see Lemma 2.2.1. = = 2 2 U 2 We have that d[S ] = 2. Indeed, d[S ] is the degree of the map U U 2 1 3 F = : S S ! S jj jj (( ; ; ); t) 7 ! cos(2t)e + sin(2t)( e + e + e ); 1 2 3 1 1 2 2 3 3 4 4 3 where fe ; e ; e ; e g is the canonical basis of R S . It follows that 1 2 3 4 1 3 F (e ) = fq = ((0; 0; 1); ); q = ((0; 0; 1); )g: 4 1 2 4 4 We want to compute the orientation class of the image of the tangent space into the tangent 2 1 3 1 3 4 space. We understand S S as a submanifold of R S and S as a submanifold of R . We extend the tangent spaces of those two submanifolds by adding the outward normal vectors 2 1 2 1 3 3 to both spheres TS TS T (S r@r) TS and TS T (S r@r). We declare that 3 1 4 the orientations in the submanifolds are induced from the orientations in R S and R . In order to compute the degree of the di erential at the two points, we realize that F linearly 3 1 4 extends to F : R S ! R . The degree of F is computed by that of F . We obtain 0 1 0 0 0 2 sin(2t) B C sin(2t) 0 0 2 cos(2t) 1 4 B C det(dF ) = det = 2(sin(2t)) @ A 0 sin(2t) 0 2 cos(2t) 0 0 sin(2t) 2 cos(2t) is non zero and positive in (F ) (e ). Thus, e is a regular value for F and d[S ] = deg F = 4 4 U z 1 4 To conclude the proof we must check that any 2{sphere 2 Emb(S ;R ) which comes from the space of long embeddings satis es that d[ ] 2 2Z. In order to see this observe that | 1 4 if (t) = (x(t); y(t); z(t); w(t)) 2 Emb(S ;R ) comes from a long embedding then we may assume the following: 1 3 4 (i) (( ; )) Int(( 1; 1) ), 4 4 1 3 0 1 0 3 (ii) ( ) = e , ( ) = e and ( ) = ( ) = e , 1 1 1 4 4 4 4 1 1 3 (iii) (S n( ; )) is contained in the plane spanned by e and e . 1 2 4 4 (iv) 1 7 parametrizes the semicircle of radius R and center R e joining ( ) = jS n( ; ) 8 8 8 R e + R e and ( ) = R e + R e which passes trough (0) = 2R e , 1 2 1 2 2 where R >> 1 is a constant independent of . Let be any 2{sphere which comes from the space of long embeddings. The associated map z z (t+h) (t) 0 z 0 h 0 = ( ) is homotopic to (z; t) = , for any h 2 (0; 1). Thus, is homotopic z z jj (t+h) (t)jj z z (t+h(t)) (t) to (z; t) = , where h : [0; 1] ! (0; 1) is any continuous map which satis es z ~ z jj (t+h(t)) (t)jj ~ ~ ~ h(0) = h(1). Consider the family of functions h , 0 < " < , de ned as 1 7 > " if t 2 [0; "][ [ "; 1] 4 8 7 1 7 t if t 2 [ ; "] h (t) = " 8 4 8 1 1 t +" t +" 5 1 1 4 4 (1 )" + if t 2 [ "; ] " " 8 4 4 ~ ~ h z 0 h " " It follows that e 2= Image( ). Hence, d[ ] = deg( ) = deg( ) = 0 2 2Z. 3 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 21 5.3. The Z factor in (FHor(R )). Let us explain the Z factor in the exact sequence 2 1 2 described in Lemma 5.2.1. Observe that Z is just the kernel of the homomorphism 4 4 f : (FHor(R )) ! (FHor(R )); 1 1 4 4 [ [ induced by the bration FHor(R ) ! FHor(R ) and that ker(f ) = (F )=d( (FHor)) (see 1 2 Lemma 5.2.1). Consider the unique isomorphism : ker(f ) ! Z : De nition 5.3.1. The {invariant of a loop ( ; F ) 2 ker(f ) is ( ; F ) 2 Z . s s 4 4 Take a loop ( ; F ) in FHor(R ) and assume that ( ; F ) is trivial in FHor(R ), ie is s 1 z z trivial as a loop of smooth embeddings and Rot ( ; F ) = 0. Then, there is a disk ( ; F ) s 1 2i 2i 4 e e in FHor(R ), z 2 D, such that ( ; F ) = ( ; F ). Now we try to lift it to a disk 1 1 z z 4 ( ; F ) in FHor(R ). This can be done for all the values of the parameter except for z = 0, 1 0 0 0 where we have S di erent limits. So, we have a loop ( ; F ) in the ber of ( ; F ). We s 1 de ne F as 1 1 3 F : [0; 1] S ! M aps(S ;S ) ~ ~ (s; ) 7 ! F (s; ) = F : 0 0 0 2 1 3 ~ ~ ~ Since F = ( ) and F = F , we can understand F as a map from S to M aps(S ;S ). 0 1 1 1 3 3 Ie [F ] 2 (F ) (M aps(S ;S )) (S ). The {invariant measures the degree mod 2 = = 1 2 3 2 1 3 of the associated map from S S to S , ie ( ; F ) = deg (F ). This is because the el- s 2 ements in (F ) = (S ) are classi ed by the degree but the construction of F depends 1 3 on the choice of disk, inside FHor(R ), bounding ( ; F ). The di erence between two [ ~ disks is measured by the diagonal map d : (FHor(R )) ! (F ); ie F is unique, up to 2 2 homotopy, inside (F )=d( (FHor(R ))). It follows from Lemma 5.2.1 that the elements 1 2 (F )=d( (FHor(R ))) = Z are classi ed by the degree mod 2 of the associated map. 1 2 2 Figure 8. The base point (a horizontal knot) of the are twist loop in the front Geiges projection. Let us describe explicitly ker(f ). The rst element of ker(f ) is just the trivial loop and the second one is given by the homotopy class of a loop of horizontal embeddings that we name area twist loop and which is described in Figure 9 (we give a precise de nition below). Observe that this loop is trivial as a loop of horizontal immersions. Let be a horizontal immersion with exactly one generic self{intersection. We are going to construct a loop of horizontal embeddings that is supported on a very small neighborhood of the self{intersection. Informally speaking (it could be completely formalized by adapting to the horizontal case the Subsection 6.1, in particular we would need the equivalent of Lemma 6.1.5), on an neighborhood of inside the space of horizontal immersions, the subset of strict 22 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Figure 9. The area twist loop in the front Geiges projection with the stan- dard capping disk through horizontal immersions. Note that the unique strict horizontal immersion in the disk is the center of it. 2 4 horizontal immersions has codimension 2. Thus, there exists a disk D ! HorImm(R ) that intersects the strict immersion just at the center . There is a natural orientation on it and we take the loop @D with the induced orientation. Let us give a more hands{on approach. Assume that the self{intersection times are t ; t and x t = 0. Now, select "; > 0 such 0 1 1 that << " < 1. By [27, Lemma 1] we may assume that (t) = (t t ; 0) for t 2 Op(ft g) and FG 0 0 (t) = (t; t ) for t 2 ( 2"; 2"). FG 2 u;v 4 For each (u; v) 2 [ ; ] de ne 2 HorImm(R ) in terms of the front Geiges projection u;v u;v u;v (t) = (x (t); z (t)) as follows: FG 0;0 = , u;v (t) = (t) for t 2 S n( 2"; 2"), FG FG u;v 0;0 x = x , u;v z (t), t 2 ( 2"; 2"), is de ned by the following conditions: u;v 2 " " (i) z (t) = v (t u) , t 2 [ ; ], 2 2 u;v 2 (ii) z (t) = t , t > j"j, R R " " u;v 0;0 (iii) z (t)dt = z (t)dt, " " R R u 0 u;v 0;0 (iv) z (t)dt < z (t)dt, for 0 < u , " " R R u 0 u;v 0;0 (iv) z (t)dt > z (t)dt, for u < 0. " " u;v 4 Observe that conditions (iii), (iv) and (v) imply 2 Hor(R ) for (u; v) 6= (0; 0). De nition 5.3.2. An area twist loop is any loop of horizontal embeddings which is homotopic cos(2); sin(2) to := . Observe that the notion of area twist loop is well{de ned since the space of choices in the u;v construction of is contractible. The following lemma states that the area twist loop is actually the second element of ker(f ). FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 23 0 0 Lemma 5.3.3. Let be an area twist loop. Then, ( ; ( ) ) 2 ker(f ) and ( ; ( ) ) = 1. Proof. We prove the result for the area twist described in Figure 9. The general case is analogous. ;u ;u Consider the 1{parametric family of formal horizontal embeddings ( ; F ), u 2 [0; 1], de ned by ;u (i) = , ;u 0 0 (ii) F = (1 s)( ) + s((1 u)( ) + u@ ). s w ;u 0 0 Note that, F = ( ) , u 2 [0; 1], and ( ) is never a positive multiple of @ . Hence, we ;1 can assume that the area twist loop in FHor(R ) is ( ; F ). To compute the Z {invariant s 2 r; 1 of the area twist loop we are going to construct a disk f( ~ ; @ ) : r 2 [0; 1]; 2 S g in 4 1; FHor(R ) such that ~ = . For each 2 R consider the 3{plane R with equation x = . The plane intersects each embedding in zero, one or two points. Consider the segment G = f : # \ = 1 or 2g and write \ = fA ; B g for all 2 G. Denote by A B the unique segment (in R ) de ned by the pair of points A , B . The union of all these segments allows us to construct an embedded 2{disk D = [ A B , whose boundary is . We call the foliation, 2G in the so constructed 2{disk, provided by the segments the \ruling". 1 0 Assume that the left cusp of happens at time t 2 S . Observe that (t) = (t) for t 2 Op(ft g). Thus, by construction, the intersection D \Op(f (t )g) = D is independent 0 0 ;t of . Fix q 2 Int(D) and carefully take polar coordinates D = fp : 2 [0; 1]; t 2 S g such that 1;t p = (t), 0;t p = q . ;t for any and xed, the loop fp g 1 intersects the \ruling" provided by A B in t2S ;t zero, one or two points. In other words, the curve fp g 1 becomes tangent to the t2S ruling just two times. Moreover, x " > 0 small enough and let Q : D(") ! Int(D) be an embedding such that ;t 2it 1 Q (0) = q . Assume that Q (e ) = p , 0 ", 2 S . Finally, consider a smooth 0 0 0 r; increasing function ~ : [0; 1] ! ["; 1] such that ~(0) = " and ~(1) = 1. Then, the disk f ~ g ~(r);t r; is de ned as ~ (t) = p . Observe that the tangent space at each point of D is given by (6) h@ + ; @ + @ + @ i; x z y w where 2 spanf@ ; @ ; @ g and @ + @ + @ = B A as vectors in R . Obviously the y z w z y w r; embedding ~ is tangent to D . 1 1 3 Consider the map F : [0; 2] S S (s; ; t) ! S de ned by s; 0 ( ~ ) (t) if s 2 [0; 1] F (s; ; t) = ;1 F (t) if s 2 [1; 2]: s 1 24 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 0; Figure 10. Disk D with ~ = ~ in its rst lobe in the front Geiges projection. 0 1 ~ ~ ~ Since F (0; ; ) = ( ~) ( ) and F (1; ; ) = @ , for 2 S , the map F can be regarded as a 2 1 3 map from S S ! S . The {invariant of the area twist is given by the degree mod 2 of this map. In order to compute the degree of this map we compute the preimages of @ . Since @ 2= D and F (s; ; t) 2 D, for s 1, the equality F (s; ; t) = @ is only possible if s 2 [0; 1). z z s; 0 Ie we need to study the equation ( ~ ) (t) = @ , s 2 [0; 1), this equality implies that s; 0 r; (a) ( ~ ) (t) 2 ker(dx), ie the front Geiges projection of ( ~ ) has either a cusp or a vertical r; tangency at time t, in other words, ~ becomes tangent to the \ruling" (see Equation (6)); s; 0 (b) ( ~ ) (t) 2 ker(dy), ie y(B ) y(A ) = 0, in other words, the area enclosed by the front Geiges projection of between A and B is zero; and s; 0 (c) ( ~ ) (t) 2 ker(dw), ie w(B ) w(A ) = 0, in other words, the slopes of the tangent lines of the front Geiges projection of at A and B are the same. Figure 11. Example where the rst condition is ful lled for a point p but not the other two. After these observations, it is clear that the {invariant of the area twist loop is 1, see Figure (a) The tangent is @ . (b) The tangent is @ . z z Figure 12. Cases where the three conditions are ful lled. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 25 6. The Area Invariant. In this Section we de ne a Z -invariant, called Area Invariant, which coincides with the { invariant de ned in the previous Section in the horizontal case. Ie for any loop of horizontal embeddings the Area Invariant satis es Area( ) = ( ; ( ) ) 2 Z . This invariant is de ned studying the space Hor(R ) as a subspace of the space of horizontal immersions HorImm(R ), which is a exible space [27]. The key point is that the strict horizontal immersions are encoded by means of the zeroes of a \suitable" function, called the area function. We keep the same notation as in Section 5. 6.1. Local models of Legendrians. Our goal is to understand generic local models of fami- lies of Legendrian immersions. Before studying the di erent models of Legendrian immersions depending on the codimension, we rst de ne a proper subset of Legendrian immersions which has in nite codimension in the space LegImm(R ) and, thus, does not intersect a generic - nite dimensional family. A point in a Legendrian is not injective if the preimage of the image has cardinality greater than one. The set of non injective images is the image of the set of non injective points. A Legendrian is special if the cardinality of the non injective images is non{ nite. We denote the space of special Legendrian immersions by SpLegImm(R ). If a Legendrian is special, there exists a Weinstein chart (x; y; z) such that the Legendrian has at least two branches on the chart, one of them being the curve (t; 0; 0) and the other one is (t; f (t); f (t)), where f (t) is a function with in nite zeroes such that t = 0 is a zero and is limit of a sequence of zeroes. This implies that f has an in nite order zero at t = 0. This is an in nite codimension condition in the space of LegImm(R ). Therefore, the set SpLegImm(R ) is a subset of an in nite codimension strati ed submanifold of the space of Legendrian immersions. We can dismiss this set when considering nite dimensional families, since any nite dimensional family can be generically perturbed to be disjoint from an in nite codimension subset. First, consider the following space 3 strict 3 3 3 1 1 SLegImm(R ) =f( ; v; t ; t ) 2 LegImm (R )nSpLegImm(R ) R S S : 0 1 (t ) = (t ) = vg; 0 1 3 3 3 1 1 equipped with the natural inclusion SLegImm(R ) LegImm(R )R S S . Denote by 3 3 1 1 3 : LegImm(R )R S S ! LegImm(R ) the projection onto the rst factor. Observe strict 3 3 3 that (SLegImm(R )) = LegImm (R )nSpLegImm(R ). 3 strict 3 The projection 3 : SLegImm(R ) ! LegImm (R ) is a local di eomorphism. jSLegImm(R ) We just need to check that the ber of is transverse to SLegImm(R ); ie given = 3 3 1 ( ; v; t ; t ) 2 SLegImm(R ), then T F \ T SLegImm(R ) = f0g, where F = ( ). 0 1 This is obvious since we are considering Legendrians with isolated non injective points. Thus, we have the following diagram: 3 3 1 1 // SLegImm(R ) LegImm(R ) R S S loc. di . strict 3 3 // LegImm (R ) LegImm(R ) strict 3 3 3 The space LegImm (R )nSpLegImm(R ) is not a strati ed submanifold while SLegImm(R ) it is. Thus, the rst set can be regarded as the image by a local di eomorphism of a strati- ed submanifold. We can now compute local models of R{versal deformations (see Arnold, 26 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Gusein-Zade and Varchenko [3, page 147]) upstairs and translate them downstairs just by using the local di eomorphism. De nition 6.1.1. [3, page 49] Consider a germ of a smooth real function f : R ! R 1;:::;1 (j) (k+1) satisfying f (0) = 0. We say that f 2 if f (0) = 0, for j k. If f (0) 6= 0, then k k 1;:::;1;0 1;:::;1 we say that f 2 . Fix (t ) = (t ) a non injective image for a curve 2 LegImm(R ). By Weinstein's tubular 0 1 neighborhood theorem, we can assume that there is a chart in which the branch of through t becomes the zero section in the Weinstein model. So, we can locally regard the other branch as a smooth real map with an order 2 zero at t = 0. Let us now introduce some concepts that will be useful along the upcoming discussion. Remark 6.1.2. [3, pages 147{148] Given f : (R ; 0) ! R a smooth map germ, we say that a deformation with parameter space K = R is a germ at the origin of a map-germ F : n k R R ; 0 ! R such that F (x; 0) f (x). We say that a deformation G is R equivalent to F if n k m G(x; k) F (h(x; k); k) with h : R R ; 0 ! (R ; 0) a smooth germ with h(x; 0) x: We say that a deformation F of f is R versal if any other deformation G can be expressed as G(x; ) F (h(x; ); '()) with h(x; 0) x and '(0) = 0: In order to understand the local models that we will produce, we consider the space of Legendrians up to reparametrizations. Note that this is equivalent to say that the local model representing the non injective point of the Legendrian that is given by a smooth function has to be considered up to R{versal deformations. 3 I We can then say that a point ( ; v; t ; t ) 2 SLegImm(R ) belongs to if and only if when 0 1 we express as the germ of a smooth real function around the origin near the tangency (t ; t ), this germ of function belongs to . Note that this is well de ned. 0 1 Let us introduce some notation. De ne the con guration space of two points in S as 2 1 1 1 k 3 Conf (S ) = f( ; ) 2 S S : 6= g. If e : D ! LegImm(R ) is a k{disk of Legendrian 0 1 0 1 immersions we want to check that Thom's Transversality Theorem can be applied to e in order to obtain a disk which is transverse to . By a map transversal to a strati ed set we mean a map that is transversal to each stratum. We say that a property P for maps is generic if the set of maps that satisfy the property is a Baire set in the total space of maps. k 3 We are going to check that a property of the map e : D ! LegImm(R ) is generic. Thus, by a standard compactness argument, we can further assume that the disk is very small. Ie take any nite covering for which we check that the property is satis ed in a Baire subset, then we are able to check the property in the initial disk since the nite intersection of Baire subsets is Baire. k 1 By the previous discussion, there is an associated map e : D S ! R that is constructed 1 1 by identifying a tubular neighborhood of e(0) with J (S ;R), where the image of e(0) corre- sponds to the zero section. Thus, e (z; ) is a function whose associated 1{jet represents the k 2 Legendrian immersion e(z) in a Weinstein chart. De ne e ^ : D Conf (S ) ! R; (z; t ; t ) 7! 0 1 e (z; t ) e (z; t ), that is considered as a k{parametric family of real functions. Now, x 0 1 r 2 Z the amount of regularity of the Banach manifold of maps in which we are working, ie FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 27 we are working in the space of C Legendrian immersions, and consider the holonomic lift of r k 2 1 r k 2 1 r k 2 1 e ^ into J (D Conf (S );R) and denote it by j e ^ : D Conf (S ) ! J (D Conf (S );R). r k 2 1 1;:::;1 We have a strati ed submanifold inside J (D Conf (S );R) de ned by F 2 , k < r, (j) (j) de ned by the intersection of hyperplanes F = F for j 2 f0; 1; : : : ; kg. In particular, the 0 1 1;:::;1 k 1 holonomic lift of e ^ intersects if there exist z 2 D and (t ; t ) 2 Conf (S ) such that 0 1 (j) (j) e(z) (t ) = e(z) (t ) for j 2 f0; 1 : : : ; kg. 0 1 Now we further reduce our domain. For this denote by U an open neighborhood of the 1 1 r I diagonal of S S . Thus, j e ^ does not intersect . So, by compactness, we may jD (U n) assume that there is a nite covering by squares of Conf (S )n(U n) j j j j j j j j fS = [t "; t + "] [t "; t + "] : [t "; t + "]\ [t "; t + "] = ;; j 2 f1; : : : ; Ngg: 0 0 1 1 0 0 0 1 Now, we try to perturb e ^ along the domains D S in such a way that e ^ becomes jD S transverse to . But in that domain perturbations of e ^ are in one to one correspondence with perturbations of e . Therefore, we can apply the standard Thom's Transversality Theorem (see [11, Theorem 2.3.2]) to conclude that the space of perturbations transverse to the strati ed submanifold is a Baire set. Thus, we have proven the following statement k 3 3 Lemma 6.1.3. Assume 0 k r 2. Let e : D ! LegImm(R )nSpLegImm(R ) be a disk. r k 3 3 Then, there is a C {perturbation e ~ : D ! LegImm(R )nSpLegImm(R ) of e such that strict (i) e ~ is transverse to LegImm (R ) and 3 I (ii) the lift of e ~ to SLegImm(R ) is transverse to each , jIj r. From the previous discussion we conclude that the classi cation of generic local models of self{intersections of Legendrians tantamounts to classifying local germs of smooth real maps. In fact, we obtain a systematic method for classifying local models of generic k{dimensional disks of Legendrians by taking into account Arnold's classi cation of local germs of smooth functions (see [3]). In order to be precise we need to work on SLegImm(R ). Codimension 0. A generic germ of a smooth function vanishing at the origin has gener- ically non zero rst derivative at the origin. This just re ects the fact that Legendrian immersions are generically embeddings. 1;0 Codimension 1. A generic 1{parametric disk of smooth maps at the origin intersects 1;1 at isolated points and in the empty set. By an application of Hadamard's lemma (see Milnor [26, Lemma 2.1]) we can assume that the intersection with this stratum takes place at the germ at the origin of the map f : R ! R; x 7! x . We have that an R{versal deformation 2 2 of x at 0 is given by x + a, where a 2 Op(f0g) (see [3, page 151] for further details). This corresponds to the local model represented in Figure 13. Figure 13. Local model in codimension 1 representing a simple crossing. 28 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Codimension 2. A generic 2{parametric family D of smooth maps at the origin intersects 1 1;1;0 1;1;0 at curves and at isolated points. Fix a point in \D, we have that the crossing can be identi ed with the germ at the origin of the function f : R ! R; t 7! t . We know 3 3 2 that an R{versal deformation of t at 0 is given by (t; a; b) = t +at +b, where (a; b) 2 D (") for " > 0 small enough (see [3, page 151] for details). Realize that the intersection of the disk with the stratum is given by the curve (seen as a curve of germs of functions at the origin) determined by the equations (t; a; b) = 0 () t + at + b = 0 (t; a; b) = 0 () 3t + a = 0 dt with solutions t = and the parameters satisfying the relationship 3 1=2 3=2 2 2 a (3 3 ) + b = 0 which describes a cuspidal curve. This local model is represented in Figure 14b. (a) Two simultane- (b) Cuspidal curve. ous simple tangen- cies. Figure 14. Local models in codimension 2. 3 1 1;1 Codimension 3. A generic 3{parametric family D intersects at surfaces, at 1;1;1;0 1;1;1;0 3 curves and at isolated points. A point in \D can be identi ed with the germ 4 4 at the origin of the map f : R ! R; x 7! x . An R{versal deformations of x at 0 is given 4 2 3 by x + ax + bx + c, where (a; b; c) 2 D (") (see [3, page 151] for further details). This gives raise to the local model represented in Figure 15 (see [3, pages 47{48]). strict 3 3 In order to understand deformations around points of LegImm (R )nSpLegImm(R ) we 3 3 1 1 3 project the models from LegImm(R ) R S S to LegImm(R ). Since the projection is a local di eomorphism, we get the same local models, but di erent branches upstairs may come together downstairs. So, we have the same models but with the addition of cartesian strict product models that show up at points 2 LegImm (R ) that lie at the intersection of 2 or more branches. For instance, in codimension 2 we get a new model represented in Figure 14a. We have shown that any continuous disk in the space of Legendrian immersions can be perturbed to be in generic position. However, in this article we usually have a disk of horizontal immersions and we want to check that a perturbation of this disk renders a generic perturbation of its Geiges projection inside the space of Legendrian immersions. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 29 Figure 15. Local model in codimension 3. k 4 Lemma 6.1.4. Assume that 0 k r 2. Let e : D ! HorImm(R ) be a disk such k 3 4 r that e = e : D ! LegImm(R )nSpLegImm(R ). Then, there is a C {perturbation G G k 4 e ~ : D ! HorImm(R ) of e such that k 3 3 (i) e ~ : D ! LegImm(R )nSpLegImm(R ), strict 3 (ii) e ~ is transverse to LegImm (R ) and 3 I (iii) the lift of e ~ to SLegImm(R ) is transverse to each , jIj r. Proof. It follows directly from Lemma 6.1.3 and Lemma 7.2.1. This last lemma will be proven later on and it allows to lift multiparametric families of Legendrian immersions. For later use we want to check that 1;0 3 Lemma 6.1.5. LegImm(R ) is a locally closed codimension 1 immersed Banach submanifold. 1;0 3 Note that the lift of to SLegImm(R ) is, in fact, an embedded submanifold. However, it does not help our purposes. 3 1 3 Proof. We already noted that LegImm(R ) is an immersed Banach submanifold of Imm(S ;R ) 1;0 (see Remark 3.1.3). In order to check that is a codimension 1 locally closed immersed submanifold we just need to de ne an atlas of LegImm(R ) with charts satisfying that the 1;0 pullback by the chart map of the intersection of with the chart is a closed codimension 1 Banach subspace. 1;0 For a point 2 , the exponential map is a local di eomorphism because its di erential at is the identity (see [24, Theorem 5.2, page 15]). By using a Weinstein chart around we just de ne the exponential by the formula 1. We can simply check that given a vector r 1 r 1 1;0 (f; g) 2 M aps (S ;R) M aps (S ;R), the condition for it to be tangent to reads as 30 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS f ( ) = 0, where is the self{intersection time. This is obviously a closed codimension 1 0 0 subspace, this concludes the proof. 6.2. The Area Invariant. Fix a loop in Hor(R ) with Rot ( ) = 0. This means that is trivial as a loop of horizontal immersions. Thus, there exists a disk D = f ~ g in 2i 4 e HorImm(R ) such that ~ = . We want to study the obstruction for that disk to belong to Hor(R ). By Lemma 6.1.4 we can assume that D is in generic position, ie D contains curves G G (1{dimensional connected manifolds), smooth away from the cusps, of strict Legendrian immersions which may intersect in Int(D ). Moreover, we may assume that in @D \ G G strict 3 1;0 LegImm (R ) we have only elements in , hence the number of strict Legendrian immersions in @D is even. See Figure 16. -- Figure 16. Generic disk D . The red curves represent strict Legendrian immersions. z z | Let ~ 2 D be a strict Legendrian immersion. Write ~ (t) = (x(t); z(t); w(t)) and assume G G z z 1;0 1 that p = ~ (t ) = ~ (t ) is a transverse self{intersection point, ie of type ; t ; t 2 S . 0 1 0 1 G G We de ne z 0 " ( ~ ; p) = z(t)x (t)dt; where the factor 2 f1g corresponds to the following sign convention: we always integrate along a segment that starts at the lower branch of the curve in the front projection and follow the orientation of the curve. Notice that this convention is well{de ned since we 1;0 z are working with self{intersections of type . If ~ has only one tangency point, then z z we write " ( ~ ) = " ( ~ ; p). We call " the area function. The sign of " over a curve A A A A G G of strict Legendrian immersions is reversed when it passes through a cusp (a point of type 1;1;0 ). Indeed, note that when crossing these points the lower branch becomes the upper branch and vice versa, thus producing a change of sign in the area function. In particular, the absolute value of the area function is well{de ned over a cusp. Note that, by genericity, we may assume that " is non zero over the cusp points. Hence, the area function is continuous over the curves of strict Legendrian immersions except at the cusp points. Moreover, if ~ z z is a strict Legendrian immersion, then ~ is a horizontal embedding if and only if " ( ~ ;) is non zero over all tangency points of ~ . Thus, to study the obstruction for D to belong 7 3 3 The tangent space of the two branches in R at the intersection point p 2 R de nes a framing of . The choice just ensures that the orientation of the framing is always positive. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 31 to Hor(R ) we need to study the zero set of the area function. Observe that, since @D is 2i 4 e the Geiges projection of 2 Hor(R ), we obtain that " ( ~ ) is non zero. We de ne the z z + sign of ~ as the sign of " ( ~ ). Denote by P (D ) the set of positive points in @D . The A G G G G following lemma is just a simple exercise: + + (a) Capping disk in (b) Curve of strict LegImm(R ). Legendrian immer- sions in the front Geiges projection. Figure 17. Geiges projection of the standard capping disk through horizon- tal immersions of the area twist loop. Lemma 6.2.1. Let C D be a curve of strict Legendrian immersions. Assume that one of the three following conditions holds: (i) The boundary points of C have the same sign and the number of cusps in C is odd. (ii) The boundary points of C have di erent signs and the number of cusps in C is even. (iii) The boundary of C is empty and the number of cusps in C is odd. Then, D contains at least one strict horizontal immersion. + + (a) Curves with the same sign in (b) Curves with di erent signs the boundary and an odd number in the boundary and an even of cusps. number of cusps. Figure 18. Some examples of curves of strict Legendrian immersions that contains the Geiges projection of a strict horizontal immersion. De nition 6.2.2. Let C usp(D ) be the set of cusps of D . We de ne the area of D as G G Area(D) = (#P (D ) + #C usp(D )) mod 2: G G Lemma 6.2.3. Assume that Area(D) = 1, then D contains at least one strict horizontal immersion. 32 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Proof. We proceed by induction over the number of strict Legendrian immersions in the boundary of D . Observe that, since Area(D) = 1 there has to be a curve of strict Legendrian immersions in D . 0 points in the boundary. Since Area(D) = 1 then #C usp(D ) must be odd. Hence, there exists a closed curve with an odd number of cusps. By Lemma 6.2.1(iii), we conclude this case. 2 points in the boundary. We have two cases: (a) Assume that the two points in the boundary have the same sign. Further assume there exists a curve of strict Legendrian immersions with the same sign in the boundary and an odd number of cusps. Use Lemma 6.2.1(i) to conclude. Otherwise, there is a closed curve of strict Legendrian immersions with an odd number of cusps, by Lemma 6.2.1(iii) we conclude this case. (b) Assume that there exist two points in the boundary which have di erent signs. If the curve of strict Legendrian immersions that connects them has an even number of cusps, we conclude by Lemma 6.2.1(ii). In other case, there must be a closed curve with an odd number of cusps. By Lemma 6.2.1(iii) we conclude this case. 2n + 2 points in the boundary. We have two cases: (a) Assume that the points in the boundary connected by a curve of strict Legendrian im- mersions have the same sign. Further assume there exists a curve with the same sign in the boundary and an odd number of cusps. We conclude by Lemma 6.2.1(i). Oth- erwise, there exists a closed curve with an odd number of cusps by Lemma 6.2.1(iii) and we conclude this case. (b) Assume that we have a curve C with opposite signs in the boundary. Then, if the number of cusps is even, we apply Lemma 6.2.1(ii) to conclude. On the other hand, if the number of cusps is odd, consider as a new nite set of curves the old one minus C and apply the induction hypothesis. We want to show that the assignment 7! Area(D) de nes a homotopy invariant. Thus, let and two homotopic loops of horizontal embeddings with rotation number zero. For 0 1 2i z z 4 e that, choose D = f ~ g and D = f ~ g two disks in HorImm(R ) such that ~ = and 0 1 0 1 0 0 2i ~ = . We need to study the relationship between the con guration of curves of strict 1 1 Legendrian immersions in (D ) and (D ) . Denote by the homotopy between and . 0 G 1 G t 0 1 Since (HorImm(R )) = 0 we can assume that we have a solid cylinder Cyl = [0; 1] D in HorImm(R ) such that ftg @D coincides with , D = f0g D and D = f1g D. Let 0 1 Cyl = [0; 1] D be the Geiges projection of the cylinder. Denote by L = [0; 1] @D the G G lateral boundary of the cylinder and L = (L) its Geiges projection. In particular, every point in L = [0; 1] @D is a Legendrian embedding. By genericity, we can assume that 3 3 every Legendrian in Cyl LegImm(R ) is the projection of an element in SLegImm(R ) via the canonical projection 3 3 1 1 3 : LegImm(R ) R S S ! LegImm(R ): 1;1;1;0 In addition, we can assume that S := ( ) \ Cyl is a set of isolated points, S := 0 G 1 1;1;0 1;0 ( )\Cyl a set of immersed curves and S := ( )\Cyl a set of immersed surfaces. G 2 G In addition, by genericity, we have that di erent branches of S do not intersect, di erent 1 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 33 Figure 19. Solid cylinder in HorImm(R ) connecting D and D 0 1 branches of S and S intersect in isolated points S = S \ S , and di erent branches of S 1 2 12 1 2 2 intersect in a set of embedded curves S . We can also assume that (i) S \ L = ;, (ii) S := S \ L is a nite set of points, 1L 1 (iii) S := S \ L is a set of embedded curves. 2L 2 Assume that the \height function" for the cylinder h : Cyl = [0; 1]D ! [0; 1] is a Morse G G function for S ; S and S . Thus, with these assumptions, the 1{parametric family of slices 1 2L 22 1;0 ftg D ; t 2 [0; 1], induces a movie of curves de ned by the intersections of ( ) with strict 3 each slice. Let us discuss the topology of the con guration of LegImm (R ) at each slice (see Figure 20). The changes of the topology of the curves happen at speci c times due to the following phenomena: Points in S . If such a point lies on a slice ftg D the movie around that time 12 G corresponds to the crossing of a curve with a cusp with another immersed curve (see Figure 20, First elementary change). Points in S This corresponds to the crossing of two immersed curves (see Figure 20, Second elementary change). Points in S . This corresponds to the crossing of a cusp with the boundary (see 1L Figure 20, Third elementary change, cases 1 and 2). Appearance (disappearance) of points in S \ (ftg D ) for some t 2 [0; 1]. This 2L G generically corresponds to the appearance (or disappearance) of an immersed curve during the 1{parametric family. The time t 2 [0; 1] represents the time where this curve appears in the movie of slices (Fourth elementary change, cases 1 and 2). Critical points of h for S . A critical point in S for h represents either the appear- 2 2 ance/disappearance of a closed embedded curve: the case of a minimum/maximum; or an index 1 surgery in the set of curves: the case of an index 1 critical point (see Figure 20, Fifth and Sixth elementary changes). Critical points of h for S . This corresponds to the appearance/disapearance of an embedded curve (respectively minimun/maximum). What is special is that the curve has a double cusp (See Figure 20, Seventh elementary change and Eighth elementary change). Points in S . This corresponds to the codimension 3 swallow{tail singularity (see Figure 20, Ninth elementary change). Corollary 6.2.4. Let and be two homotopic loops of horizontal embeddings with ro- 0 1 z z 4 tation number zero. Let D = f ~ g and D = f ~ g be two disks in HorImm(R ) such that 0 1 0 1 34 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS + - Fi First elementary cha rst elementary change. nge. Second elementary change. Second elementary change. Third el Third elementar ementary change. Cas y change. Case 1. e 1. + - + + Third el Third elementar ementary change. Cas y change. Case 2 e 2 F Fourth elementary cha ourth elementary change. Case 1 nge. Case 1. . F Fourth elementary change. C ourth elementary change. Case 2 ase 2.. . . Sev Seventh elementary change. enth elementary change. Fi Fifth el fth elementar ementary change. y change. Si Sixth elementary change. xth elementary change. + + + + - - - - - - - - Eighth el Eighth elementar ementary change. y change. Ni Ninth elementary change. nth elementary change. Figure 20. List of elementary changes. The boundary of the cylinder is represented in black. Blue circles represent local neighborhoods of the slices. The signs in the boundary points are de ned by the area function. Of course the same pictures with signs reversed could happen. 2i 2i e e ~ = and ~ = . Then, 0 0 1 1 Area(D ) = Area(D ): 0 1 Proof. By the previous discussion, any generic homotopy between two disks is a composition of the nine elementary changes: they come with suitable signs in order to be geometrically realizable, see Figure 20. Since these nine elementary changes preserve the area the result follows. We claim: 4 z Theorem 6.2.5. Let be a loop in Hor(R ) such that Rot ( ) = 0. Let D = f ~ g be some 2i 4 e disk in HorImm(R ) such that ~ = . The assignment ! Area( ) := Area(D); de nes a homotopy Z {invariant, called the Area Invariant. Alternatively, we de ne the Area Invariant as the number of strict horizontal immersions in D mod 2. Both de nitions coincide. Proof. We are left with checking that the Area invariant is computed by the number of strict horizontal immersions mod 2. We proceed by induction in the number of zeroes of the Area function over the strict immersed curves. If there are no zeroes, then Lemma 6.2.1 implies that the Area invariant is zero. Assume that it is true for k = 0; : : : ; n. Then, we want to study a con guration with n + 1 zeroes. If all of them lie in the same curve, we have that Lemma 6.2.1 implies the result. If not, the curves can be divided into two subsets such that FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 35 they have n k > 0 and n k > 0 zeroes each with n + 1 = k + k . By induction, we 1 2 1 2 conclude the proof. Remark 6.2.6. Let us check that the Area Invariant coincides with the {invariant de ned in the previous Section in the horizontal case. Ie for any loop of horizontal embeddings the Area Invariant satis es Area( ) = ( ; ( ) ) 2 Z : The {invariant describes the Z factor given by the kernel of the homomorphism 4 4 f : (FHor(R )) ! (FHor(R )): 1 1 4 4 4 1 Consider the natural inclusion i : Hor(R ) ,! FHor(R ) and let (Hor(R )) := (i) (ker(f )). 1 1 In order to prove that the Area Invariant computes the {invariant in the formal case, we must check that the next diagram commutes: (i) // (Hor(R )) ker(f ) 1 f Area %% 4 z Let be a loop of horizontal embeddings which lies in (Hor(R )) . Let D = f ~ g be a 2i 4 e disk in HorImm(R ) such that ~ = . Assume that D has exactly n strict horizontal 1 n immersions ; : : : ; . Let D be the Geiges projection of D, note that by genericity, we 1 n 1;0 can assume that ; : : : ; belong to . Because of De nition 5.3.2 any small disk whose G G diagrams of strict Legendrians is as in Figure 17a has as boundary circle an area twist. Then collapsing the disk into small neighborhoods of the curves , as we show in Figure 21, is homotopic to a concatenation of n area twist loops. Observe that the Area Invariant and the {invariant of an area twist loop is one. Thus, Area( ) = ( ; ( ) ) = n mod 2 (see Lemma 5.3.3). This concludes the argument. Figure 21. Homotopy between and the concatenation of n area twist loops for the case n = 4 represented in D . We have drawn the zeroes of the area function in blue. The concatenation of 4 area twist loops is represented by the green curve in the last step. 7. h{principle at {level for horizontal embeddings. We keep the same notation as in Sections 5 and 6. 36 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 7.1. The main theorem. 4 4 Theorem 7.1.1. The inclusion Hor(R ) ,! FHor(R ) induces an isomorphism of fundamen- tal groups. We are assuming a base point in the previous inclusion. We claim that the isomorphism works for any choice of base point, ie for all the connected components: they are given by the rotation number of the horizontal embedding (see Remark 5.1.6). To prove this result we proceed as follows. Consider the homomorphism Rot : (Hor(R )) ! Z [ ] 7 ! Rot ( ) and de ne (Hor(R )) as the kernel of Rot . The homomorphism Rot is surjective. 1 0 L L Indeed, it is enough to check that for any horizontal embedding 2 Hor(R ) and any integer k 2 Z there exists a loop of horizontal embeddings , based at , with Rot ( ) = k. De ne as the Legendrian immersion obtained by a rotation of angle 2k of in the Lagrangian projection. The lift of to Hor(R ) satis es the desired property. It follows from Theorem 6.2.5 that the Area Invariant de nes an homomorphism Area : (Hor(R )) ! Z . Moreover, we will prove the following 1 0 2 Theorem 7.1.2. The Area Invariant de nes an isomorphism (7) Area : (Hor(R )) ! Z : 1 0 2 In particular, (Hor(R )) = Z Z . 1 2 Moreover, the techniques developed in this Section will allow us to prove (it will be proven at the end of Subsection 7.2) 4 1 4 Lemma 7.1.3. The homomorphism (Hor(R )) ! (Emb(S ;R )), induced by the inclu- 1 0 1 4 1 4 sion Hor(R ) ,! Emb(S ;R ), is surjective. Thus, since the two elements in (Hor(R )) (the trivial loop and the area twist loop) are 1 0 1 4 trivial as loops of smooth embeddings and the space Emb(S ;R ) is path{connected, we conclude the following 1 4 Corollary 7.1.4. The space Emb(S ;R ) is simply connected. This is a \Legendrian" proof of a well known result (see [5] Proposition 3:9(4)). We expect to be able to compute higher rank homotopy groups of this space generalizing the discussion below since we know the space of Legendrian immersions behaves nicely. Ie the crossings types correspond to tangencies of smooth functions with the horizontal line. It is known that these tangencies can be assumed to be either Morse or birth{death, see [22]. For higher dimensional families this is a rather non trivial result. In a forthcoming project, we will use this fact to collect non trivial information about the high rank homotopy groups of the space 1 4 Emb(S ;R ). It is clear that Theorem 7.1.2 and Corollary 7.1.4, together with the exact sequence stated in Lemma 5.2.1, implies Theorem 7.1.1. The purpose of this Section is to reduce the proof of Theorem 7.1.2 to a technical cancellation Theorem, see Proposition 8.1.22, that is of independent interest and is proven in Section 8. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 37 This Section presents several results needed in orden to reduce and to develop the proof of the cancellation Theorem. An overview of what follows is In Subsection 7.2, we introduce several results: Lemma 7.2.1 states that the area function over a connected component (see below for a precise de nition of connected component) can be easily deformed without a ecting the other connected compo- nents; then we focus on deforming the immersed Legendrians produced by Geiges projection. The main result states that the only obstruction to having an h{principle for embeddings of Legendrians lies on the fact that the double stabilization of a Leg- endrian (main tool to approximate a smooth curve) changes the embedding class of the Legendrian. The key remark (Lemma 7.2.4) shows that this is not the case in the horizontal case. We give a multiparametric proof of the fact that (double) sta- bilizations allow to approximate any smooth curve: Lemma 7.2.5. What is left to complete a full h{principle is to deal with the positions of the cusps points. This is done just in 1{parametric families that is enough for the purposes of this article. The proof contained in [6] controls those cups in higher dimensional families. In Subsection 7.3, we start studying the injectivity of the morphism (7). This will be the task of the remaining part of the article. For that we need to show that the parity of the zeroes (Area invariant) of a capping disk on HorImm(R ) for a loop on Hor(R ) fully recovers the homotopy class of the loop. Thus, we need to cancel the zeroes by pairs. In this Subsection, we assume that the two zeroes to be cancelled do lie in the same connected component. In that case, a careful use of Lemma 7.2.1 is all we need to conclude. In Subsection 8.1, we explain how to change the diagram of the capping disk associ- ated to an element in (Hor(R )) in order to place two zeroes in the same connected 1 0 component. We prove in these pages a previous key result (Proposition 8.1.22) ex- plaining how to deform a path connecting the two zeroes through curves that, except for a nite number of points, are Legendrian embeddings in Geiges projection. The idea is to build a map of the square into Hor(R ). The bottom side of the square is the initial path. The other three edges correspond to a \much better" path that connects the two zeroes by a curve of strict immersions; ie we have found a curve making them live in the same connected component. Finally, we explain how to exploit the success in order to use the previous square to change the diagram of curves in a suitable way (creating an index one surgery in the diagram). In Subsection 8.2, we take a suitable path joining the two zeroes, then the index 1 surgery previously built allows to connect the connected components of the two di erent zeroes. 7.2. Creating exibility. s 4 Lemma 7.2.1. Let K be a compact parameter space and 2 HorImm([0; 1];R ), s 2 K . Let A : K ! R be a continuous map. Let p 2 [0; 1], s 2 K , be any continuous family of regular points for the front Geiges projection of ; and " > 0 suciently small. Then, there s;u 4 exists a 1{parametric family 2 HorImm([0; 1];R ), u 2 [0; 1], satisfying: s;0 s (i) = . 38 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS s;u Figure 22. Construction of via its front Geiges projection. s s;u 0 8 (ii) The Geiges projection of and are C {close and (0; 0; 0; 0) in [0; p "], s;u s (0; uA(s); 0; 0) in [p + "; 1]. s;u Moreover, if is an embedding then is also an embedding. G G s;u s;u s s s s Proof. De ne = in [0; p "] and = in [p + "; 1]. It is enough to de ne s;u s s + the front Geiges projection of in [p "; p + "]. Let N 2 Z , consider the equispaced s;u s s s s s s s partition fa = p " < a < : : : < a = p + "g of [p "; p + "]. De ne in 0 1 N +1 FG uA(s) s s s s s [p "; p +"] just adding a Reidemeister I move of area to at each point (a ), FG FG i i = 1; : : : ; N ; see Figure 22. The C {closeness follows from choosing N large enough. Remark 7.2.2. Lemma 7.2.1 also applies to horizontal loops, provided that the zero total area condition is preserved. We do it as follows: let 2 HorImm(R ) be a horizontal immersion, p; n 2 S di erent regular points of the front Geiges projection of and A 2 R. We write # (A)# ( A) 2 HorImm(R ) p n to denote the horizontal immersions obtained from when, using Lemma 7.2.1, we add area A near the point (p) and A near the point (n). Moreover, if 2 LegImm(R ) satis es 4 1 the zero total area condition, ie = for 2 HorImm(R ), and p; n 2 S are regular points of the front projection of we write # (A)# ( A) 2 LegImm(R ) p n to denote the Geiges projection of # (A)# ( A) 2 HorImm(R ). p n De nition 7.2.3. Let 2 LegImm(R ). A double stabilization (DS) of is a 1{parametric u 3 family 2 LegImm(R ), u 2 [0; 1], such that (i) = , (ii) is stabilized once negatively and once positively. The homotopy ; u 2 [0; 1], is explicitly described in Figure 23. R1 R1 R2 Figure 23. Sequence of moves in the front projection of creating a double stabilization. 8 4 We are working in the standard Engel structure in R (x; y; z; w) given byD = ker(dy zdx)\ker(dz wdx). FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 39 We can de ne the DS of a horizontal immersion via the Geiges projection, just being careful enough not to change the total area condition. To do this pick t ; t 2 S two regular points 0 1 of the front Geiges projection of . Use t to add a DS and use Lemma 7.2.1 to add the necessary area over (t ) to ensure that the area condition is ful lled. We can do it even better as the next key result shows. Lemma 7.2.4. The double stabilization is a well{de ned operation for horizontal embeddings. Proof. The total area enclosed by the only tangency point created in a DS is nonzero. We do not claim that this construction is unique. In fact, it is possible to produce two paths not formally homotopic, connecting the end of the DS with the initial loop: just draw a di erent picture movie with a di erent area balance and then it is clear that the concatenation of the rst path with the inverse of the second one is a loop with non-trivial area invariant. However, this is enough for our purposes. The next result shows that DSs can be used in order to approximate families of smooth curves by Legendrian curves. k 3 Theorem 7.2.5. Given K = [0; 1] [0; 1], C = [0; 1]f0g K and a family : [0; 1] ! R , k 2 K , such that (i) are embeddings, (ii) (t) are Legendrian for t 2 Opf0; 1g, (iii) are Legendrian for every k 2 C . k;u 3 Then, there exists a family of curves : [0; 1] ! R , k 2 K , u 2 [0; 1], such that k;0 k (i) = , k;u 0 k (ii) are immersions and C {close to , k;1 (iii) are Legendrian embeddings, k;u k (iv) (t) = (t) for t 2 Opf0; 1g, (v) there exist a nite sequence 0 < u < u < : : : < u < 1 and > 0 small enough 1 2 N such that k;u (a) are Legendrian embeddings for every k 2 C , u 6= u ; : : : ; u , 1 N k;u k;u (b) , u 2 [u ; u + ], is a DS of for every k 2 C and i = 1; : : : ; N . i i This Theorem together with Lemma 7.2.1 easily implies the following k 4 Corollary 7.2.6. Given K = [0; 1][0; 1], C = [0; 1]f0g K and a family : [0; 1] ! R , k 2 K , such that (i) are embeddings, (ii) (t) are horizontal for t 2 Opf0; 1g, (iii) are horizontal for every k 2 C . k;u 4 Then, there exists a family of curves : [0; 1] ! R , k 2 K , u 2 [0; 1], such that k;0 k (i) = , k;u 0 k (ii) are embeddings and C {close to , k;1 (iii) are horizontal embeddings, k;u k (iv) (t) = (t) for t 2 Opf0; 1g. 40 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS k;u Proof. Apply the previous Theorem to the family to obtain a family , u 2 [0; 1]; G G of immersed curves in R . To conclude the proof we just need to approximate the area coordinate (the w{coordinate). This can be easily done by using Lemma 7.2.1. We can relax the hypothesis (i) for later use. We will prove a particular result just with the interval as the parameter space. s 1 3 Lemma 7.2.7. Let 2 Imm(S ;R ), s 2 [ 1; 1], be a 1{parametric family of immersions such that s 1 3 (i) 2 Emb(S ;R ) for s 6= 0, (ii) has a generic self{intersection. s;u 1 3 Then, there exists a deformation 2 Imm(S ;R ), u 2 [0; 1], satisfying s;0 s (a) = , s;u 1 3 (b) 2 Emb(S ;R ) for s 6= 0, 0;u (c) , u 2 [0; 1], has a generic self{intersection, s;1 3 (d) 2 LegImm(R ) for s 2 Op(f0g), s;u s (e) = for s 2= Op(f0g). 1 0 Proof. Let t ; t 2 S be the self{intersection times of . After applying an isotopy we may 0 1 assume that there exists 0 < " such that 0 0;u There exists a time{dependent vector eld v , supported near (t ), such that , 0 0 u 2 (0; "), is obtained from by pushing the branch through (t ) in the direction of v . is Legendrian. jOp(ft ;t g) 0 1 Apply Theorem 7.2.5 to the two smooth embedded arcs to construct a path S nOp(ft ;t g) 0 1 strict 0;u 1 3 0;1 3 ~ 2 Imm(S ;R ), u 2 [0; 1], where ~ 2 LegImm (R ) has a generic self{intersection. u 0;u 0 0 De ne a family of vector elds v ~ , u 2 [0; 1], supported near (t ) such that v ~ = v and 1 s;u 1 3 v ~ is a contact vector eld. De ne ~ 2 Emb(S ;R ), (s; u) 2 (0; ") [0; 1], by pushing the 0;u u s;1 3 branch through ~ (t ) in the direction of v . It follows that ~ 2 Leg(R ) for s 6= 0. Finally, consider a smooth bump function : [ "; "] ! [0; 1] such that 1 and jOp(f0g) 0. The desired family is jOp(f ";"g) if s 2 [ 1; 1]n[ "; "] s;u s; (s) ~ if s 2 [ "; "]. s 4 Corollary 7.2.8. Let : [0; 1] ! R be a 1{parametric family of embeddings such that the i s curves , i 2 f0; 1g; are horizontal and (t), s 2 [0; 1]; are horizontal for t 2 Opf0; 1g. s;u 4 2 Then, there exists a homotopy : [0; 1] ! R , (s; u) 2 [0; 1] , satisfying s;0 s (i) = , s;u 0 s (ii) are embeddings and C {close to , s;1 (iii) are horizontal embeddings, s;u s (iv) (t) = (t) for t 2 Opf0; 1g, i;u (v) , i 2 f0; 1g, are horizontal embeddings. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 41 Proof. We may assume that there exist 0 < s < < s < 1 such that 1 N are embeddings for s 2 [0; 1]nfs ; : : : ; s g, 1 N , i 2 f1; : : : ; Ng, has a generic self{intersection. Moreover, after applying Lemma 7.2.7 we may also assume that is Legendrian for s 2 Op(fs ; : : : ; s g). Now the result follows by an application of Corollary 7.2.6 to the paths 1 N , s 2 [0; 1]nOpfs ; : : : ; s g. 1 N Remark 7.2.9. The result can be extended to any compact parameter space K and any closed subset C K . A direct consequence is a parametric version of the Fuchs{Tabachnikov's result (see [14]) that asserts that any two Legendrian embeddings which are smoothly isotopic are, after a nite number of stabilizations, Legendrian isotopic. If we assume that we have a k 4 k 4 k{disk D FHor(R ) such that @D Hor(R ), it follows that we can homotope this ball 4 4 inside FHor(R ) into a ball in Hor(R ). For more details, see [6]. Nevertheless, our argument does not assume that we have a disk of formal horizontal embeddings but just a disk of k 4 horizontal immersions; ie D HorImm(R ). Thus, we need to homotope this disk, through the space of horizontal immersions, into a disk of horizontal embeddings. In other words, we are not using as initial datum the homotopy type of the space of smooth embeddings of the circle into R : part of the data provided by having a disk of formal horizontal embeddings. In a sense, we are studying the relative homotopy type of the space of horizontal embedding inside the space of horizontal immersions. The following result works only for 1-parametric families over the segment. This is the main obstacle in our proof to generalize to a complete h-principle. In [6] a generalization to deal with the cusp points in higher dimensional families is provided and this proves a complete h-principle. Our result just claims that for a 1{parametric family of Legendrian immersions the cusps can be assumed to be constant (in time and position) for the whole family. s 3 Lemma 7.2.10. Let , s 2 [0; 1], be a path in Leg([0; 1];R ) such that Opf0; 1g are regular s;u 3 points for the front projection. Then, there exists a 2{parametric family 2 Leg([0; 1];R ), s; u 2 [0; 1]; satisfying s;0 s s;u s (i) = and (t) = (t) for t 2 Opf0; 1g. s;u s 0 (ii) and are C {close. s;1 (iii) is generated by a sequence of Legendrian Reidemeister moves of (only) type II and III . s 3 s Proof. The family is understood as : [0; 1] [0; 1] ! R , (t; s) 7! (t). By Thom's Transversality Theorem (see [11, Theorem 2.3.2]) we assume that the sets of cusp points in [0; 1] [0; 1] are embedded curves. Moreover, we assume that the height function h(t; s) = s, restricted to these curves, has a nite number of maximum and minimum points: all of them non degenerate. These points correspond to Reidemeister I moves. Take the point with the lowest height among all the minima. Since there is not any other minimum with a lower height, we can nd a curve C joining this point with fs = 0g that does not intersect any other curve of cusp points and which does not have any critical point. Now, we can remove this minimum adding a (C {small) family of Reidemeister I moves over C . Keep going to remove all the minima. To remove the maxima we do likewise. Thus we obtain 42 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Corollary 7.2.11. Up to reparametrization, there exist c ; : : : ; c 2 [0; 1] such that the cusp 1 m s;1 points of are at xed times c ; : : : ; c ; for all s 2 [0; 1]. 1 m Proof of Theorem 7.2.5. Assume rst that , k 2 C , does not have any cusp point. Since the h{principle for smooth immersions works relative to the domain and the parameters and 0 k is C {dense (see [11]), we can assume that the front projection of each is immersed. We declare the slope of the formal derivative of the front projection of to be the value of the coordinate z (@ is the projection direction). If is the projection of a Legendrian this z F de nition of formal derivative coincides with the derivative of the front projection. Therefore, this equips the whole family with a formal Legendrian structure. 1 N Take an equidistant partition u = 0 < u = < : : : < u = < u = 1 of [0; 1]. 0 1 N N +1 N +1 N +1 Consider the collection of intervals I = [u ; u ], i = 0; : : : ; N and let K (N ) 2 Z . De ne i i i+1 k;u inductively over fI g as follows k;u k (i) In I and I de ne = . k;u k;u (ii) In I , i = 1; : : : ; N 1, de ne to be K (N ) DSs of approximating the k;u 2i+1 formal derivate of at time t = as described in Figures 24 and 25 (for 2(N +1) K (N ) = 2). Let us provide the details of the construction. The depicted blue segment k 1 represents (I ) that can be assumed to be (C {close to) a straight segment for N large enough. The green segments represent the formal derivative, again for N large enough is C {close to a constant. We further assume that the formal derivative has angle zero with respect to the horizontal axis, but we are reduced to that case composing with a linear transformation in R (x; y) to make the formal derivative zero, ie the formal derivative is de ned as (1; z(t)), we just change coordinates by using the unique linear transformation 2 2 A : R ! R z(t) 0 0 1 1 satisfying A = and A = . This canonically lifts z(t) z(t) 1 1 z(t) 0 to a contact transformation. This allows us to assume that the formal derivative is zero. This is the reason why we are depicting only the case of formal derivative horizontal. Coming back to the gure, the number below each image is the angle between the curve (straight segment) and the formal derivative. The black curves are the 0 k K (N ) DSs that C {approximate with prescribed derivative given by the green curve (in our pictures slope zero). The geometric idea is to perform rst the negative stabilizations and later on the positive ones. The rule is for the case in which the curve to be approximated points downwards we make the negative stabilizations small and the positive ones big. In particular, for K (N ) suciently large, we can make the derivative of the approximating curve arbitrarily close to the formal one (horizontal). (iii) The rule for the case in which the angle between the curve to be approximated and the formal derivative (horizontal in our pictures) tends to radians is the same. First negative stabilizations and afterwards positive ones. The key point is that the front projected curve cannot remain an embedding when reaching : increasing the angle. The reason is that the y coordinate of the ending point of the curve for angle slightly over is below the y coordinate of the starting point. Since the end of the curve is decreasing (positive stabilizations), there is a moment in which the positive We follow the standard convention that a negative stabilization is an upward zig{zag and viceversa for the positive ones. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 43 stabilization has to cross (through K (N ) 1 Reidemeister type II moves) over the curve that connects the negative with the positive stabilizations. Note that no front tangencies are created. This is done in order to be able to balance the DS and place them in the middle at angle and then, push them to the beginning when going over radians. This makes possible to proceed simetrically when coming from angles bigger than . In other words, the approximation of a curve with angle is the symmetry with respect to the y axis of the one with angle 2 . The construction is clearly continuous in the angle, therefore it can be done in a canonical way. This readily implies that it works parametrically. Moreover, the approximation coincides with the formal derivative at the beginning and the end of the interval I , thus the constructed approximations are smooth. k;1 k;1 We claim that the are embeddings. The segments are clearly embedded: just by jI visual inspection and since they just possess R II front moves. However, intersections may k;1 k;1 show up between the end of and the beginning of . By suciently increasing N , we jI jI i i+1 assume that the angle of the two consecutive segments is almost the same. In other words, we are in the same position in the clock gure (Figure 24). There are no tangencies between the two consecutive segments if the angle is away from radians since the front curve of the union is clearly embedded. In the neighborhood of radians, we need to be much more careful and depict the precise movie of the two consecutive segments. The problem is that there are intersections in front projection between the end of the rst approximating curve k;1 k;1 and the beginning of the next one. In Figure 26, is depicted in red and in black. jI jI i i+1 The key point is to make sure that the big{sized stabilization boundary, i.e the beginning of the approximating curve for angle smaller than (respectively the end for angle bigger than ) has the shape of a scimitar with a fat blade. The fatness of the blade allows to t inside its bell the whole end of the previous approximating curve (consisting of very small stabilizations) and the convexity of the bottom part of the scimitar blade allows to cross with R II moves to the exterior part of the scimitar. Once there, the small stabilizations are running away from the danger area. It is left to check the crossing of the rst stabilization of the second curve with the last stabilization of the rst one. They do not create tangencies thanks to the convexity of the scimitar shape. In the general case, apply Corollary 7.2.11, to assume that the cusp points of each , k 2 C , are at times c ; : : : ; c . It is very simple to deform each , k 2 K , at small 1 m neighborhoods of times c ; : : : ; c , making the family Legendrian and having a cusp point 1 m over each c ; : : : ; c . Finally, add c = 0 and c = 1 and apply the rst case over the 1 m 0 m+1 family , i = 0; : : : ; m, to conclude the proof. j[c ;c ] i i+1 We are able to prove the following: Proof of Lemma 7.1.3. Let be a horizontal embedding such that is a Legendrian em- s 1 4 0 1 bedding and let 2 Emb(S ;R ), s 2 [0; 1], be any loop based at , ie = = . Without loss of generality we may assume that the Geiges projection of is embedded except for a nite sequence of times 0 < s < < s < 1 and the strict immersions has a generic 1 N self{intersection. By Corollary 7.2.8 there exists a loop ^ of horizontal embeddings with 0 1 s ^ = ^ = homotopic to . Note that the total zero area condition for a closed horizontal curve can be easily achieved with Lemma 7.2.1 since the map A : [0; 1] ! R; s 7! ( ^ ) zdx; is continuous. Moreover the approximation procedure xes the cusps of and, thus, the rotation number of this loop is zero so the result follows. 44 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Figure 24. Interpolation. Note that the orientation of the formal derivative (green) is determined by the Legendrian segments in C . 7.3. Cancellation by pairs of zeroes in the same connected component. For a generic 2{disk of Legendrians D, the intersection of the disk with the space of strict immersions is a set of curves with crossings. They come from a set of curves in SLegImm(R ) by projection. We FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 45 Figure 25. Interpolation between 0:9 and 1:1 in Figure 24. strict 3 declare a connected component of the set D\LegImm (R ) to be a connected component of the set seen in SLegImm(R ). In other words, the two branches of a crossing (possibly) are in di erent connected components. 4 z Let ' : D ! HorImm(R ); z 7! , be a disk of horizontal immersions satisfying: '(@D) Hor(R ), ' : D ! LegImm(R ) is in generic position, 46 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Figure 26. Checking that the gluing process does not create tangency points. 1 strict 3 ( ') (LegImm (R )) is a nite union of connected components C ; : : : ; C , G 1 n C contains, at least, two zeroes of the area function. Note that Lemma 6.1.4 implies that there is no loss of generality in thesew assumptions. Let C C be a connected arc containing exactly two zeroes of the area function. Parametrize 1 3 s 4 4 the curve ( ')(C ) as f : s 2 [0; 1]g such that and are the two zeroes of the area G G G 1 3 function. Moreover, we can assume that f : s 2 [0; ][ [ ; 1]g, is smooth (does not contain G 4 4 any cusp) in LegImm(R ). FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 47 Consider " > 0 small enough and the closed "{neighborhood K of C in D (see Figure 27), satisfying that K does not contain any other zero of the area function for a di erent connected component and such that K \ @D = ;. We have a well{de ned continuous map Figure 27. Compact neighborhood K of C in D. 1 1 s s s s : C ! S S ; s 7! (t ; t ), where the components of the ordered pair (t ; t ) are the 0 1 0 1 s s s s | tangency times of the front projection of (t) = (x (t); z (t); w (t)) , s 2 [0; 1], and the 0 0 order of the pair is determined by the following rule: (t ; t ) are chosen with the usual 0 1 s s rule to de ne the area function " and the pair of points (t ; t ) are extended in the unique 0 1 1 3 4 4 continuous way. In order to apply Lemma 7.2.1 to cancel the zeroes and it is necessary G G 1 1 k k to nd an extension : K ! S S ; k 7! (t ; t ), of . Just take any retraction r : K ! C " " 0 1 and de ne := r = r. Write K = K . We state the main result of this section. Proposition 7.3.1. There exists a 1{parametric family ' : D ! HorImm(R ), u 2 [0; 1], satisfying: (i) ' = ', (ii) ' (z) = '(z), z 2 DnK , strict strict 1 3 1 3 (iii) ( ' ) (LegImm (R )) = ( ') (LegImm (R )), G u G (iv) " ( (' (z))) 6= 0, z 2 C K . A G Proof. Assume that the area function over is positive in s 2 [0; ). De ne G 4 Z s s s s 0 " ~ (s) = " ~ ( ~ ) = z (t)(x ) (t)dt: A A Observe that " ~ is just the continuous version of the area function over C and satis es j" j = j" ~ j. The function " ~ : C ! R is depicted in Figure 28. A A A De ne ' = ' over DnK . The problem is clearly to de ne ' over K . Proceed as follows: u u k k 1 k In order to add area choose two points p ; n 2 S in the domain of , k 2 K . We require these points to have the following properties k k k p and n are regular for the front projection of , k k k if has more tangency points, di erent from (t ; t ), the area function over these G 0 1 k k points remains unchanged after adding area over p and n . 48 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 1/4 3/4 Figure 28. Possible graph of " ~ before (blue) and after (red) adding area using Lemma 7.2.1. k k k k For a given , choose two regular points p and n very close to t , one to the left and another G 0 one to the right of t . These two points satisfy the properties above. By compactness, there i i i 1 1 exists a nite open covering fU g of K and an election (p ; n ) 2 S S compatible with each k i i i , k 2 U . Take a partition of unity f g subordinated to fU g and a continuous function : K ! [0; 1] such that 0 in @K and 1 s 1 3 1 in ( ') (f : s 2 [ ; ]g)\ K . 4 4 Finally, x some positive number A > maxfj" ~ ( ~ )j : s 2 [0; 1]g. De ne ' (k), k 2 K , as A u the lift of k i i # (u (k)(k)A)# ( u (k)(k)A); i i i i G p n to HorImm(R ). The family ' satis es the required properties. 8. The core of the proof We explain now how for a given capping disk of horizontal immersions we are able to cancel zeroes of the area function by deforming the disk relative to the boundary. In particular, we prove that any pair of zeroes can be cancelled out without creating new zeroes: there will be no assumptions about the zeroes. This completes the proof of the main Theorem. 8.1. Changing the diagram of curves of strict Legendrian immersions. The goal is to create a 2{square (2{disk) in the space of LegImm(R ) such that the bottom of the square s 0 1 is a path in the disk D whose endpoints are two strict Legendrian immersions and . Moreover, there exists a nite set of values of the parameter 0 < s < < s < 1 such that s s is also a strict immersed Legendrian. The rest of the points , s 2 (0; 1)nfs ; : : : ; s g are Legendrian embeddings. The other three sides of the square will produce a path of strict s s Legendrian immersions that has the property that the self{intersections points ^ (t ) = ^ (t ) 0 1 form a continuous family. In other words, we are constructing a curve of Legendrians that makes the two initial Legendrians live in the same connected component of strict immersions. This is the key point and is the content of Proposition 8.1.15. Afterwards, we play with the result: given an embedded curve in the disk whose ends live in two di erent connected components in the disk, we nd a square transverse to the disk D that creates on the other three sides a curve of strict immersions joining the two strict Legendrian immersions. Then, FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 49 we take a small open neighborhood of it removing the interior in order to produce a new disk (see Figure 39) in which the curves of strict Legendrian immersions have changed their topology (see Figure 38) . Proposition 8.1.13 explains how to build the disk in the smooth category. Finally, we just use Theorem 7.2.5 to conclude in the Legendrian setting. strict 3 strict 3 Denote by LegImm (R ) LegImm (R ) the open stratum of simple transverse self{intersections. 8.1.1. Legendrian ngers. 1 3 1 De nition 8.1.1. Fix 2 Emb(S ;R ) and b 2 S . A nger bone is an embedded path : 3 1 [0; 1] ! R such that (0) = (b) and the intersection is transverse. Moreover, (0; 1]\ (S ) is either one point or the empty set. In the former case the intersection must be transverse. 1 3 1 De nition 8.1.2. Fix 2 Emb(S ;R ) and b 2 S . A nger germ is a pair ( ; v) satisfying (a) : [0; 1] ! R is a nger bone such that (0) = (b), (b) v 2 TR satis es (i) v(t) and (t) are linearly independent, (ii) v(0) = (b), Without loss of generality assume that is geodesic on a very small neighborhood of b. For " > 0 small enough, x an embedding Q : [ "; "] [0; 1] ! R de ned as Q(x; z) = exp (x v(z)). Choose a bump function : [ "; "] ! [0; 1] satisfying (z) is an even function, is increasing in [ "; 0] (j) + ( ") = 0, ( ") = 0 for all j 2 Z and (j) + (0) = 1 and (0) = 0 for all j 2 Z . 1 3 De nition 8.1.3. The nger deformation #( ; v) 2 Imm(S ;R ) associated to the nger germ ( ; v) is given by the curve (t) t 62 (b "; b + "); #( ; v)(t) = Q(t b; (t b)) t 2 [b "; b + "]: This is a well{de ned operation that works for compact families of curves and nger germs, provided that " > 0 is chosen small enough and uniform for the whole family. 1 3 Note that and #( ; v) are homotopic in Imm(S ;R ). Denote by ( ; v ) = ( ; v ) u u j[0;u] j[0;u] the obvious homotopy of germs of ngers. We say that #( ; v ) is obtained from adding u u the nger germ ( ; v ). u u De nition 8.1.4. Fix 2 Leg(R ) and ( ; v) a nger germ such that (a) is a transverse embedding, ie (t) 2= for each t 2 [0; 1], (t) (b) v is Legendrian, ie v(t) 2 for each t 2 [0; 1]. (t) We say that ( ; v) is a pre{Legendrian nger germ. The picture looks like a bypass picture in contact topology, though geometrically it has nothing to do with a bypass. Formally, we play the same game: our initial disk plays the formal role of the convex surface, the strict immersions play the formal role of the dividing sets. The bypass changes the strict immersions/dividing set combinatorial structure. 50 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS Figure 29. Finger addition process #( ; v ); u 2 [0; 1]. u u Figure 30. Legendrian Finger construction process. The key tool is to create a huge tower of double stabilizations. Let us explain how to add the Legendrian nger associated to a pre{Legendrian nger germ. We are working on the standard (R (x; y; z); = ker(dz ydx)) with the trivialization of given by the Legendrian framing h@ i. Take a contactomorphism : (C ; ) ! (D (0; ") y " [0; 1](x; y; z); = ker(dz ydx)). The domain C , that we will call a cylindrical shape, is a suciently small tubular neighborhood of and xing " > 0 small enough, we have that (t) = (0; 0; t), v = @ ; (t) = (t b; 0; 0), for jt bj ". The existence of this contactomorphism is a direct consequence of the Tubular Neighborhood Theorem for Contact Submanifolds (see [15, Theorem 2.5.15]). From now on we work in these coordinates. The front projection in (D (0; ") [0; 1](x; y; z); = ker(dz ydx)) is given by the projection in the XZ {plane. The front projection of the original Legendrian in a neighborhood of the base of the nger is the horizontal axis in front coordinates. jOp(fbg) The front of the transverse nger bone is the vertical axis, see Figure 30. To approximate the nger bone by a Legendrian nger just add N = N (") 2 Z DSs to the Legendrian over (b) as in Figure 30 in such a way that the derivative of the new Legendrian #( ; v) at time b is v(1). This operation is well de ned up to Legendrian immersions homotopy and works for compact families. Note that this follows just taking the radius " of the tubular neighborhood and the number N of DSs in the nger addition process constant in the whole family. This means k k k k that given a family ( ; ( ; v ); C ), k 2 K , we obtain a continuous family of Legendrian " FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 51 k k k nger deformations #( ; v ). In particular, if there is some k 2 K such that the pre{ k k 0 0 Legendrian nger germ ( ; v ) is trivial, ie k k k k 0 0 0 0 0 ( ; v ) ( (b); ( ) (b)); k k k k 0 0 0 0 and N 6= 0, then the Legendrians and #( ; v ) are di erent. In other words, to interpolate between a trivial pre{Legendrian nger germ and a non{trivial one we need to add DSs to the trivial nger. Another important observation is that the DSs operation is well{de ned for horizontal embeddings (see Lemma 7.2.4), this is relevant since we are going to work with the Geiges projection of families of horizontal curves. Remark 8.1.5. Another possible de nition of Legendrian nger is by using Reidemeister I moves instead of DSs. Note that, in both cases, the transverse nger bone is approximated by half of the Legendrian nger which is a very stabilized Legendrian curve. Moreover, the de nition using Reidemeister I moves does not work well for our purposes because the two halves of the nger are linked and in the later argument is very important to not have linking between the two halves. A stabilization of a transverse curve is an operation that changes the curve in such a way that in front projection a small arc is replaced by a 2 rotation as in Figure 31. The stabilization does not depend on the chosen small segment (see Etnyre [12]). So we can speak of a k stabilization of a transverse curve that we will denote by . Figure 31. Stabilization of a transverse curve. The next Lemma will be useful in the next subsection and ensures that we can interpolate between a small nger with nger bone a small Reeb chord and another one with nger bone a very stabilized Reeb chord. strict 3 Lemma 8.1.6. Let 2 LegImm (R ) be a Legendrian immersion with a self{intersection 1 3 at times t ; t 2 S and ~ 2 Leg(R ) a Legendrian embedding such that 0 1 (i) There is a small Reeb chord : [0; 1] ! R between ~(b) and ~(e) for some times b; e 2 S ; (ii) = ~#( ; v) where v = @ + y@ . x z + u strict 3 Then, for any k 2 Z there exists a 1{parametric 2 LegImm (R ), u 2 [0; 1], with a 1 0 1 self{intersection at times t ; t 2 S , such that = and = ~#( ; v). Moreover, the 0 1 k family is constructed from just by adding a sequence of DSs to the initial nger, ie the number N of DSs in the de nition of Legendrian nger can be chosen in such a way that ~#( ; v) = ~#( ; v). Proof. Since is C {close to we may assume that the nger associated to ( ; v) is k k contained in the tubular neighborhood C of . Moreover, (C ; ) is contactomorphic to " " 3 3 (R ; = ker(dz ydx)) by Eliashberg's classi cation of contact structures in R [9]. In this sense the right side of the ngers associated to ( ; v) and ( ; v), denoted by R and R k k 11 0 We may assume that ~ (b) = v . (b) 52 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS respectively, can be understood as positively stabilized long Legendrian unknots. In the same way the left side, denoted as L and L , as negatively stabilized long Legendrian unknots. Thus, there exist two non{negative numbers M; M 2 Z such that k 0 M k S (R) is Legendrian isotopic to S (R ), + k M k S (R) is Legendrian isotopic to S (R ); where S (S ) denotes a positive (negative) stabilization of a Legendrian curve. This follows from the classi cation of Legendrian unknots given in [10]. The Lemma follows from the fact that the number of positive and negative stabilizations are the same. Remark 8.1.7. Another possible proof of the previous Lemma is to use Theorem 7.2.5 to each half of the nger and, then, slide the DSs over the nger nails. A Legendrianization of a transverse knot/curve (with xed framings at the ends) is an induced Legendrian knot/curve that is created by drawing the characteristic foliation of the boundary of a small solid torus xed around the transverse knot for a radius such that the slope is rational and pick one integral curve. The transverse push o of a Legendrianization is always transverse isotopic to the original one (see [12]). The next lemma ensures the existence of 2{parametric families of pre{Legendrian nger germs, it generalizes the 1{parametric result explained in Lemma 3.3.3 in [15]. It is the key lemma that we will use to produce controlled deformations on our families of curves. s;u 3 s;0 Lemma 8.1.8. Fix 2 Emb([0; 1];R ), (s; u) 2 [0; 1][0; 1], assume that is transverse s;u and is also transverse, then there exist a constant k > 0 and a family of transverse jOp(f0;1g) s;0 s;u s;u 3 0 s;u s;0 ~ ~ ~ embeddings : [0; 1] ! R , C {close to , such that = and = k jOp(f0;1g) s;u jOp(f0;1g) s;0 s;0 s;u Proof. Perform a Legendrianization of that we call , that extends to a family s;u that is smoothly isotopic to . Further asume that we Legendrianize the small segments s;u s;u . Abusing notation, we still call a perturbation of the original family in jt2Op(0)[Op(1) order to make it Legendrian on a very small neighborhood of a time t . s;u s;u Now, apply Theorem 7.2.5 to the two segments and . This produces a family jt2[0;t ] jt2[t ;1] 0 0 s;u s;0 ^ that is Legendrian. The process introduces a high number of DSs on the family . We denote by k the number of introduced DSs,. s;0 s;0 Construct a new family ~ that it is just built by performing k DS at time t on . It is well known that DSs can be moved around a Legendrian (they are Legendrian homotopic [12]). s;0 s;0 This, in particular implies, that ~ and ^ are homotopic through Legendrian embeddings s;0 s;0 s;0 s;0 s;0 , r 2 [0; 1], ie = ~ and = ^ . So, we have proven that 0 1 s;0 ; u 2 [0; ] s;u 2u s;2u 1 ^ ; u 2 [ ; 1] s;u s;0 s;0 is a family of Legendrians approximating such that ~ = . s;u Perform a transverse push o to obtain a family of transverse knots. Since the transverse push o of k DSs of a Legendrian knot corresponds to k stabilizations and since the transverse push o of a Legendrianization coincides with the original transverse knot, we have shown s;0 s;0 that = . k FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 53 The unknotting number of an embedding is nite (see Adams [2]). We can think of the unknotting number as the number of ngers that we need to add to the knot in order to make it the unknot. Moreover the necessary number of ngers in order to unlink two embeddings is also nite. We summarize this information as 1 3 Proposition 8.1.9. Fix ; 2 Emb(S ;R ) disjoint embeddings. There exists a parametrized 1 3 unknot ~ 2 Emb(S ;R ), obtained from just by adding a nite number of nger germs, which is unlinked from . We are interested in applying this result to the two resolutions of a strict immersion with 1 3 exactly one (generic) self{intersection. More precisely, let 2 Imm(S ;R ) be an immersion 1 1 with exactly one self{intersection at times (t ; t ) 2 S S . Assume without loss of generality 0 1 that there exists " > 0 small enough such that ((t "; t + ")[ (t "; t + ")) is contained 0 0 1 1 0 0 2 in the ane plane (t ) + spanf (t ); (t )g. Take coordinates (x; y) 2 R in the plane 0 0 1 0 0 (t ) + spanf (t ); (t )g such that (t + r) = (r; 0) and (t + r) = (0; r) for r 2 ( "; "). 0 0 1 0 1 Finally, x any smooth non increasing function : [ "; "] ! [0; 1] such that ( ") = 1 and (") = 0. 1 3 De nition 8.1.10. The resolution of over [t ; t ] is the embedding 2 Emb(S ;R ) 0 1 Res(t ;t ) 0 1 de ned as: (t) if t 2 (t + "; t "), 0 1 (t) = (t t )(0; t t ) + (1 (t t ))(t t ; 0) if t 2 [t ; t + "); Res(t ;t ) 0 0 0 0 0 0 0 1 (t t )(0; t t ) + (1 (t t ))(t t ; 0) if t 2 (t "; t ]: 1 1 1 1 1 1 where S = [t ; t ]= . 0 1 Remark 8.1.11. The resolution is well{de ned up to isotopy. Moreover, observe Res(t ;t ) 0 1 1 1 that (S )\ (S ) = ;. Res(t ;t ) Res(t ;t ) 0 1 1 0 In these terms, Proposition 8.1.9 applied over just says that, up to adding a nite number of ngers to , we may assume that the two resolutions and are unlinked Res(t ;t ) Res(t ;t ) 0 1 1 0 and one of them is an unknot. In fact, we can add ngers in order two obtain two unlinked unknots. In the same fashion we can de ne the resolution of a strict Legendrian immersion 2 strict 3 LegImm (R ). Assume that (t ) = (t ). Take adapted coordinates such that the 0 1 front (t + r) = (r; 0) and (t + r) = (r; r ) for r 2 ( "; "). Fix a non increasing F 0 F 1 smooth function : [ "; 0] ! [0; 1] such that ( ") = 1 and (0) = 0. Finally, take a decreasing smooth function : [ "; 0] ! [0; ], where > 0 is very small, satisfying that j) + ( ") = and (0) = 0 for j 2 Z [f0g. De nition 8.1.12. The resolution of over [t ; t ] is the Legendrian embedding 2 0 1 Res(t ;t ) 0 1 Leg(R ) de ned as follows: (i) If (t + r) = (r; r ) for r 2 ( "; ") then F 1 (t) if t 2 [t ; t "), F 0 1 (t) = Res(t ;t );F 0 1 (r; (r)r + (1 (r))(r)) if t = t + r 2 (t "; t ], 1 1 1 54 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS (ii) If (t + r) = ( r; r ) for r 2 ( "; ") then F 1 (t) if t 2 [t ; t "), F 0 1 (t) = Res(t ;t );F 0 1 ( r; (r)r + (1 (r))r ) if t = t + r 2 (t "; t ], 1 1 1 where S = [t ; t ]= . 0 1 (a) Resolution (i). (b) Resolution (ii). Figure 32. Resolutions in De nition 8.1.12. is represented in red. Res(t ;t ) 0 1 We have the following result for 1{parametric families of smooth curves. s 1 3 Proposition 8.1.13. Fix 2 Imm(S ;R ), s 2 [0; 1], such that (i) , s 2 (0; 1) n fs ; : : : ; s ; 1g, for 0, where 0 < s < < s < 1, is an 1 1 embedding, i 1 1 (ii) , i 2 f0; 1g, has exactly one (generic) self{intersection at times (t ; t ) 2 S S and 0 1 s 0 1 , j 2 fs ; : : : ; s g, has exactly one (generic) self{intersection at times ( ; ) 2 s s j j 1 1 0 1 S S . The times t ; t ; ; , j 2 f1; : : : ; g, are pairwise di erent. 0 1 s s j j s s s;u Then, there exist k 1{parametric families f ; : : : ; g of nger germs and a family 2 1 k 1 3 s;0 s s;1 s s s Imm(S ;R ), (s; u) 2 [0; 1] [0; 1], such that = and = # # # , ie the 1{ 1 k s;1 s parametric family is obtained from by adding in consecutive times the nite sequence of ngers, satisfying s;1 1 1 (i) , s 2 [0; 1], has one self{intersection at times (t ; t ) 2 S S . This self- 0 1 intersection is unique for s 2 [0; 1]nfs ; : : : ; s g. i;u 1 1 s ;u (ii) , i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S S and , j 2 0 1 0 1 1 1 f1; : : : ; g, has one self{intersection at times ( ; ) 2 S S . s s j j (iii) There exists a nite sequence 0 < u < : : : < u < 1 such that 1 N s;u 1 3 1 3 2 Imm(S ;R )nEmb(S ;R ) for each j 2 f1; : : : ; Ng and s 2 (0; 1), s;u 1 3 (iv) 2 Emb(S ;R ) for u 6= 1; u ; : : : ; u and s 2 (0; 1)nfs ; : : : ; s g. 1 N 1 Moreover, if there is c 2 (t ; t ) such that is constant then the construction can be 0 1 jOp(fcg) s;u done relative to Op(f (c)g); ie is constant. jOp(fcg) s s;u Note that we start with a segment and we create a square in such a way that one of the edges is the original segment and the other three edges constitute a curve of self{ intersections. Therefore, this is the key result in the smooth category and we will adapt it to the Legendrian case in Proposition 8.1.15. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 55 Proof of Proposition 8.1.13. . Case 1: fs ; : : : ; s g = ;. The proof is divided in three steps. 0 0 1 Step I. Unknot : For and , we declare t to be the point in the lower branch Res(t ;t ) 0 1 in front projection. Apply Proposition 8.1.9 and create a sequence of nger germs 0 0 0 0 ( ; v ); : : : ; ( ; v ) 1 1 K k 0 0 0 0 1 1 to unknot and unlink from . Let (b ; e ) 2 S S be the base point and 1 1 Res(t ;t ) Res(t ;t ) 0 1 1 0 0 0 0 0 0 0 0 0 the end point of the nger ( ; v ); ie the nger germ ( ; v ) joins (b ) with (e ). We 1 1 1 1 1 1 extend this nger germ to a whole family of nger germs for the family . In order to do this, 1 1 s s s s 1 1 choose any continuous map F : [0; 1] ! S S , s 7! (b ; e ); such that b 6= e and b ; e 6= t , 1 1 1 1 1 1 s s s s s s i 2 f0; 1g, this provides a continuous family ( ; v ) of nger germs joining (b ) with (e ). 1 1 1 1 s;u 1 s; s s 3K This produces a family , with s 2 [0; 1] and u 2 [0; ], where = # . 3K s;u Proceed in the same way with the rest of the nger germs to construct a family , s 2 [0; 1] and u 2 [0; ], satisfying s;0 s (i) = , s; s s s (ii) = # # # , 1 K i;u 1 1 (iii) , i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S S , 0 1 (iv) There exists a nite sequence 0 < u < < u < such that s;u 1 3 1 3 2 Imm(S ;R )nEmb(S ;R ) for each j 2 f1; : : : ; Kg and s 2 (0; 1), s;u 1 3 (v) 2 Emb(S ;R ) for u 6= u ; : : : ; u and s 2 (0; 1), 1 K 1 1 0; 0; 3 3 (vi) is a parametrized unknot and is unlinked from . Res(t ;t ) Res(t ;t ) 0 1 1 0 1; Step II. Unknot : Do the same as in Step I but be careful in order to add families of Res(t ;t ) 0 1 0; ngers which restricted to s = 0 end up in and to not relink this resolution from the Res(t ;t ) 1 0 0; 3 1 1 1 1 other resolution . More precisely, let ( ; v ); : : : ; ( ; v ) be the sequence of K+1 K+1 N N Res(t ;t ) 0 1 1 1 1; 1; 3 3 nger germs obtained from Proposition 8.1.9 to unknot and unlink from . Res(t ;t ) Res(t ;t ) 0 1 1 0 1 1 1 1 As in the previous step, begin with (b ; e ) 2 S S the base and the end points of the K+1 K+1 1 1 1 1 nger ( ; v ) and extend this nger to the whole family via a map F : [0; 1] ! S S , K+1 K+1 s s 1 1 s 7! (b ; e ). We need to make sure that the nger germ ( ; v ) is extended to K+1 K+1 K+1 K+1 s s a 1{parametric family ( ; v ) such that K+1 K+1 1 1 1 0 0; 0 0 0; 0 0; 1 0 0 3 3 3 (0) = (b ); (1) = (e ) 2 (S n(t ; t )); ie b ; e 2 0 1 K+1 K+1 K+1 K+1 K+1 K+1 S n(t ; t ), and 0 1 0; 0 3 is unlinked from . K+1 Res(t ;t ) 0 1 12 0 1 Being precise we should write t (0) and t (1) since the intersection times of and could be di erent. 0 0 We avoid this to simplify the notation. 56 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 0; In other words, denote by D an embedded 2{disk whose boundary is given by . Res(t ;t ) 0 1 Then, the condition [0; 1]\ D = ; is sucient to keep the unlinking property. We say K+1 that a 1{parametric family of nger germs that satis es this condition is valid. Hence, we must show that there always exists a valid 1{parametric family of nger germs which extends 1 1 ( ; v ). K+1 K+1 s s 0 Assume that the family ( ; v ) does not satisfy this condition, ie [0; 1] \ D = K+1 K+1 K+1 fp ; : : : ; p g. Let > 0 be small enough and let D ) D be a small deformation of D such 1 k 1 1 ; ; ~ ~ ~ ~ 3 3 that D\ (S ) = [t ; t ] ( @D. Take any 2 Di (D; @D), isotopic to the identity, such 0 1 that (p ) 2 DnD, i 2 f1; : : : ; kg, and extend it to a di eomorphism of R with compact support close enough to D isotopic to the identity. Let " > 0 be small enough and let , s 2 [0; 1], be any isotopy between = and = Id, such that = Id for s ". 0 1 s s s s Figure 33. Family of nger bones before ( ) and after ( = ) K+1 K+1 K+1 applying the described isotopy. s s s s s s The family of nger germs ( ; v ~ ) = ( ) ( ; v ) = ( ; d v ) s s s K+1 K+1 K+1 K+1 K+1 K+1 0 1 1 satis es the condition \ D = ; and extends ( ; v ) so we nd a valid family of K+1 K+1 K+1 1 1 nger germs which extends ( ; v ), see Figure 33. K+1 K+1 s s In conclusion, we may assume that the family of nger germs ( ; v ) which extends K+1 K+1 1 1 s;u 1 1 ( ; v ) is valid. Use this family to construct , s 2 [0; 1], u 2 [0; + ], such K+1 K+1 3 3(N K) 1 1 s; + s; 3 3(N K) that = # . K+1 Proceed in the same way with the rest of nger germs (N K in total), ie extending them s;u to valid families of nger germs, to obtain a 2{parametric family with s 2 [0; 1] and u 2 [0; ], such that s;0 s (i) = , s; s s s (ii) = # # # , 1 N i;u 1 1 (iii) , i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S S , 0 1 (iv) There exists a nite sequence 0 < u < < u < such that 1 N s;u 1 3 1 3 2 Imm(S ;R )nEmb(S ;R ); for each j 2 f1; : : : ; Ng and s 2 (0; 1), s;u 1 3 (v) 2 Emb(S ;R ) for u 6= u ; : : : ; u and s 2 (0; 1), i; (vi) The resolution , i 2 f0; 1g, is a parametrized unknot and is unlinked from Res(t ;t ) 0 1 i; the other resolution . Res(t ;t ) 1 0 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 57 s; Step III. Create a 1{parametric family of self{intersections using : s;u 2 s;u We will further extend the family , u 2 [0; ] to a new family , u 2 [0; 1]. i; There exists a disk which bounds . Interpret this disk as a band B : [0; 1] [0; 1] ! Res(t ;t ) 0 1 R , i 2 f0; 1g, satisfying: 2 2 i; i; 3 3 (i) B (1; h) = (t ) = (t ), i 0 1 i; t +t 0 1 (ii) B (0; h) = ( ), i; t +t t t 0 1 1 0 (iii) B (t; 1) = ( + t ), 2 2 i; t +t t t 0 1 1 0 (iv) B (t; 0) = ( t ), 2 2 i; (v) B ((0; 1) (0; 1))\ (S ) = ;. Then, there exists a 1{parametric family of bands B : [0; 1] [0; 1] ! R interpolating between B and B satisfying the properties (ii),(iii),(iv),(v) and 0 1 2 2 2 s; s; s; 1 3 3 3 (i)' B (1; 0) = (t ); B (1; 1) = (t ) and B (f1g (0; 1))\ (S ) = ;. s 0 s 1 s This is just built by continuously collapsing the two initial bands into small bands B : 3 i i [0; 1] [0; 1] ! R de ned as B (t; h) = B ( t; h). From there, it is simple to interpolate 0 1 ~ ~ between B and B for > 0 small enough. s s @B In particular, ( ( ); v ( )) = (B (1; ); (1; )) de nes a continuous family of N +1 N +1 @t s; s;u nger germs for . Use this family to create the required self{intersections. Ie de ne , s 2 [0; 1], u 2 [0; 1], such that s;0 s (i) = s;1 s; s s s s (ii) = # = # # # , N +1 1 N +1 i;u 1 1 (iii) , i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S S , 0 1 s;1 1 1 (iv) has exactly self{intersection at times (t ; t ) 2 S S , 0 1 (v) There exists a nite sequence 0 < u < : : : < u < 1 such that 1 N s;u 1 3 1 3 2 Imm(S ;R )nEmb(S ;R ) for each j 2 f1; : : : ; Ng and s 2 (0; 1), s;u 1 3 (vi) 2 Emb(S ;R ) for u 6= 1; u ; : : : ; u and s 2 (0; 1). 1 N This completes the argument to prove the rst part of the Proposition. For the relative 1 s construction observe that since S nfcg is connected all the ngers germs , j 2 f1; : : : ; N + s s s s s 1 1g, can be chosen in such a way that (0); (1) 6= (c), ie b ; e 2 S nfcg. j j j j Case 2: fs ; : : : ; s g =6 ;. Observe that in the proof of the rst case all the deformations over the family of immer- s;u 1 3 sions : S ! R take place in a neighborhood of at most two points in the domain z (s; u); z (s; u) 2 S . There is a nite number of steps at times 0 < u < u 1, N 0, 0 1 1 N s;u 1 3 1 3 such that 2 Imm(S ;R )nEmb(S ;R ), s 2 [0; 1]. We may assume in all the steps that 0 1 the points z (s ; u ); z (s ; u ); ; , (j; k) 2 f1; : : : ; gf1; : : : ; Ng, are pairwise di erent. 0 j k 1 j k s s j j Observe that a generic choice of the times z (s; u); z (s; u) satis es this assumption. Thus, 0 1 this case follows immediately from the previous one. 58 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 8.1.2. Construction of the square. Theorem 7.2.5 and Lemma 8.1.8 allow us to state the last Proposition in the Legendrian setting, which provides the answer that we are looking for. To simplify the statement we introduce the following de nition: De nition 8.1.14. Let , s 2 [0; 1], be a path of Legendrian immersions such that (i) , s 2 [0; 1]nf0; s ; : : : ; s ; 1g, for 0, where 0 < s < < s < 1, is a Legendrian 1 1 embedding, i strict 3 1 (ii) 2 LegImm (R ), i 2 f0; 1g; has a self{intersection at times (t ; t ) 2 S 0 1 strict 1 s 3 S ; and 2 LegImm (R ), j 2 f1; : : : ; g; has a self{intersection at times 0 1 1 1 ( ; ) 2 S S . s s j j s;u s;0 s A homotopy , (s; u) 2 [0; 1] [0; 1], of paths of Legendrian immersions between = s;1 and is admissible if there exists a nite sequence 0 < u < : : : < u 1 such that 1 N s;u strict 3 (i) 2 LegImm (R ) for j 2 f1; : : : ; Ng and s 2 (0; 1), s;u 3 (ii) 2 Leg(R ) for u 6= u ; : : : ; u , and s 2 (0; 1)nfs ; : : : ; s g, 1 N 1 strict i;u 3 1 1 (iii) 2 LegImm (R ) has a self{intersection at times (t ; t ) 2 S S for i 2 f0; 1g 0 1 and u 2 [0; 1], s ;u strict 3 0 1 (iv) 2 LegImm (R ), j 2 f1; : : : ; g; has a self{intersection at times ( ; ) 2 0 s s j j 1 1 s ;u j k S S and u 2 [0; 1]. Moreover, the intersection times of , for k 2 f1; : : : ; Ng, 0 1 do not coincide, ie the real numbers t ; t ; ; are pairwise di erent. 0 1 s s j j strict i;u 3 1;0 (v) Moreover, the segments 2 LegImm (R ) has self{intersections of type except at an even nite number of values u 2 (0; 1) in which the self{intersections 1;1;0 will be of type . 1;0 Let us detail some co{orientation issues in . There is a standard orientation of the normal 1;0 1;1;0 1;0 1;1;0 bundle of n . It is de ned at 2 n as follows. Denote (t ) = (t ) the self{ 0 1 intersection point. Let us order the two branches. We declare that the branch through (t ) 0 0 goes rst and the branch through (t ) second if and only ifh (t ); (t )i is a positive basis of 1 0 1 1;0 1;1;0 3 . We want to de ne the normal vector to n at that is a eld in T LegImm(R ), (t ) ie a vector eld on TR . We de ne it as a vector eld supported around (t ) such that it extends to a contact vector eld in a small ball B centered at (t ) 2 R . This contact vector eld has as associated Hamiltonian function a cut-o function : B ! [0; 1] satisfying that (0) = 1, d (0) = 0 and j = 0. In other words, we are pushing through the Reeb Op(@B) vector eld the second branch. 1;1;0 This de nition does not extend continuously to . However, it provides a co{orientation 1;0 1;1;0 s for n . We call this co{orientation the standard one. Let the path of Leg- i 1;0 i 1 endrian immersions such that 2 , i 2 f0; 1g. The vectors v = ( 1) j 2 i s=i ds 3 1;0 1;1;0 T iLegImm(R ), i 2 f0; 1g, determine two orientations of the normal bundle of n 0 1 at and . We need the co{orientation de ned by these two vectors to coincide with the standard one. In these terms, the main result of the section is the following: Proposition 8.1.15. Let , s 2 [0; 1], be a path of Legendrian immersions such that (i) , s 2 (0; 1)nfs ; : : : ; s g, is a Legendrian embedding, 1 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 59 strict i 3 1 (ii) 2 LegImm (R ), i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S 0 1 1 s strict 3 S and 2 LegImm (R ), i 2 fs ; : : : ; s g, has a self{intersection at times 0 1 1 1 ( ; ) 2 S S , s s j j i 1 t i (iii) moreover, v = ( 1) j de nes the standard co{orientation at , i 2 f0; 1g. i t=i dt s;u 3 Then, there exists an admissible homotopy 2 LegImm(R ), (s; u) 2 [0; 1] [0; 1], with s;0 s s;1 strict 3 = such that 2 LegImm (R ) has one self{intersection at times (t ; t ) 2 0 1 1 1 S S . We need a sequence of simple lemmas. We assume them to be true and we will prove them later. The following lemma is fairly general in homotopy theory saying that if you have a retraction K ! L then, by precomposing with the retraction, any map on L extends to K . For our purpose K = [0; 1] [0; 1] and L K , such that K deformation retracts to L, and we have a map L ! LegImm(R ). This allows us to extend local deformations in a path of Legendrian immersions to produce a complete square. s;u 3 Lemma 8.1.16. Let 0 < < s be a positive number. Fix 2 LegImm(R ), (s; u) 2 L = [0; ] [0; 1][ [0; 1] [0; ]. Assume that strict i;u 3 (i) 2 LegImm (R ), i 2 f0; 1g, s;u 3 (ii) 2 Leg(R ) for s 2 (0; 1)nfs ; : : : ; s g. s;u 3 Then, there exists a family ~ 2 LegImm(R ), (s; u) 2 [0; 1] [0; 1], satisfying s;u s;u (i) ~ = if (s; u) 2 L, s;u 3 (ii) ~ 2 Leg(R ) if s 2 (0; 1)nfs ; : : : ; s g, 1;u 1; s ;u s ; j j (iii) ~ = and ~ = , j 2 f1; : : : ; g, if u 2 [; 1]. The following Lemma allows us to assume, up to reparametrization, that there is a xed cusp in any family of Legendrian immersions parametrized by the segment (we need to add parametric Reidemeister type 1 Legendrian moves, see the proof for the details). Lemma 8.1.17. Let be a path of Legendrian immersions such that (i) , s 2 (0; 1)nfs ; : : : ; s g, is a Legendrian embedding, i strict 3 1 (ii) 2 LegImm (R ), i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S 0 1 1 s strict 3 S ; and 2 LegImm (R ), s 2 fs ; : : : ; s g, has a self{intersection at times 0 1 1 1 ( ; ) 2 S S . s s j j s;u s;0 There exists an admissible homotopy , (s; u) 2 [0; 1] [0; 1], with = satisfying s;u 3 (i) 2 Leg(R ) for s 2 (0; 1)nfs ; : : : ; s g, 1 s;1 (ii) there exists a sequence of times c ; : : : ; c 2 S such that all the cusps of are at 1 2k times c . The following lemma plays the role of Steps I and II in the proof of Proposition 8.1.13 and solves our problem modulus the creation of the interpolating bands. Lemma 8.1.18. Let be a path of Legendrian immersions such that 60 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS (i) , s 2 (0; 1)nfs ; : : : ; s g, is a Legendrian embedding, i strict 3 1 (ii) 2 LegImm (R ), i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S 0 1 strict 1 s 3 S ; and 2 LegImm (R ), s 2 fs ; : : : ; s g, has a self{intersection at times 0 1 1 1 ( ; ) 2 S S . s s j j s;u s;0 s There exists an admissible homotopy , (s; u) 2 [0; 1] [0; 1], with = such that i;u (i) The self{intersection times (t (i; u); t (i; u)) of , i 2 f0; 1g, satisfy that t (i; 1) < 0 1 1 0 1 ; < t (i; 0), for j 2 f1; : : : ; g. s s j j s;1 3 (ii) 2 Leg(R ) for s 2 (0; 1)nfs ; : : : ; s g, i;1 i;1 (iii) , i 2 f0; 1g, is a Legendrian (stabilized) unknot and is unlinked from . Res(t ;t ) Res(t ;t ) 0 1 1 0 We will need to extend a strict Legendrian immersion into a path in which the self{intersection points cross a positive (negative) stabilization. This will be key to change the formal invari- ants associated to the resolutions and : since we will add/substract a Res(t ;t ) Res(t ;t ) 0 1 1 0 stabilization to each of the resolutions. The price that you pay is that the new path has two 1;1;0 cusp points ( ). We are being precise here: moving the self{intersection point through two cusps in the front projection creates two cusps in the induced curve in the moduli of strict Legendrian immersions. Lemma 8.1.19. Assume that there is a 2{disk ' : D ! LegImm(R ) such that it posseses just one connected curve C , parametrized as , s 2 [0; 1], of strict immersions that is a properly embedded segment on the disk. Denote by t and t the self{intersection times. 0 1 strict s 3 Moreover, assume that 2 C LegImm (R ) possesses a positive DS at time t + 0 0 for > 0 small enough. Then: 0 3 (1) There is a local C deformation of the disk ' : D ! LegImm(R ), t 2 [0; 1], relative to the boundary, such that for t = 1, we have that the curve C is deformed into 1;1;0 a curve C with 4 cusps: points of type (a DS in the moduli). Moreover, d 0 ('; ' ) = O(). s 1=2 (2) Parametrize the curve C by ~ , s 2 [0; 1]. Assume that ~ lies between the second and the third cusp in the moduli, and denote its self-intersection times t and t , then 0 1 1=2 1 1=2 1 ( ~ ) inherits a negative stabilization and ( ~ ) inherits a positive Res(t ;t ) Res(t ;t ) 0 1 1 0 1=2 stabilization with respect to the resolutions associated to . 1;0 The standard normal vector for a strict Legendrian immersion of type is preserved when 1;1;0 we cross over a cusp (a point of type in ) in the moduli space (see Figure 37), so we can easily assume that the orientations are preserved in the previous construction. Ie the standard normal vector is the same as the vector eld normal to the curve in the 1{parametric family constructed by the rule of avoiding tangencies with the curve of strict immersions. The next lemma provides a way to match the formal invariants of the resolutions of two generic strict Legendrian immersions. This is a necessary condition to join two generic strict Legendrian immersions by a path of generic strict Legendrian immersions. In other words we want the ends of the path to be stabilized Legendrian unknots, with the same formal invariants. Therefore, by the classi cation theorem of Legendrian unknots [10], they are the same Legendrian knot. Thanks to that fact, it makes sense to mimic the smooth proof and look for a 1{parametric family of ngers interpolating between both of them. Lemma 8.1.20. Let be a path of Legendrian immersions such that FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 61 (a) Change in the topology of the disk (b) Birth of in the moduli space via ' . four cusps in the moduli space. Figure 34. The 1{parametric family of disks ' : D ! LegImm(R ), t 2 [0; 1], described in Lemma 8.1.19. (i) , s 2 (0; 1)nfs ; : : : ; s g, is a Legendrian embedding, strict i 3 1 (ii) 2 LegImm (R ), i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S 0 1 strict 1 s 3 S ; and 2 LegImm (R ), s 2 fs ; : : : ; s g, has a self{intersection at times 0 1 1 1 ( ; ) 2 S S . s s j j s;u s;0 s There exists an admissible homotopy , (s; u) 2 [0; 1] [0; 1], with = satisfying s;1 3 (i) 2 Leg(R ), for s 2 (0; 1)nfs ; : : : ; s g, 0;1 1;1 0;1 1;1 (ii) tb( ) = tb( ) and Rot( ) = Rot( ) . Res(t ;t ) Res(t ;t ) Res(t ;t ) Res(t ;t ) 0 1 0 1 0 1 0 1 The next Lemma provides the last step in the proof, we show how to mimic the band con- struction in the Legendrian setting. It allows us to connect two strict Legendrian immersions by a path of strict Legendrian immersions. Thus, to prove Proposition 8.1.15 we just need to deform our initial path, via an admissible homotopy, to a path which satis es the hypothesis of this lemma. Lemma 8.1.21. Let , s 2 [0; 1], be a path of Legendrian immersions such that (i) , s 2 (0; 1)nfs ; : : : ; s g, is a Legendrian embedding, strict i 3 1 (ii) 2 LegImm (R ), i 2 f0; 1g, has one self{intersection at times (t ; t ) 2 S 0 1 1 s strict 3 S ; and 2 LegImm (R ), s 2 fs ; : : : ; s g has a self{intersection at times 0 1 1 1 0 1 ( ; ) 2 S S . Moreover, t < ; < t , for j 2 f1; : : : ; g. 1 0 s s s s j j j j 1 s s (iii) There exists a time c 2 (t ; t ) S such that has a cusp at time c and 1 0 1 1 jOp(fc g) is constant, i i (iv) , i 2 f0; 1g, is a Legendrian (stabilized) unknot and is unlinked from , Res(t ;t ) Res(t ;t ) 0 1 1 0 0 1 (v) and are formally isotopic. Res(t ;t ) Res(t ;t ) 0 1 0 1 s;u s;0 s Then, there exists an admissible homotopy , (s; u) 2 [0; 1] [0; 1], with = satisfying s;u (i) is constant, jOp(fc g) strict s;u 3 (ii) 2 LegImm (R ), (s; u) 2 f0; 1g [0; 1][ [0; 1]f1g, has one self{intersection 1 1 at times (t ; t ) 2 S S . 0 1 Proof of Proposition 8.1.15. The proof is divided in four steps. 62 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS s;u Step I. Fix the cusp times: Apply Lemma 8.1.17 to produce a family , (s; u) 2 [0; 1][0; ], such that s;0 s (i) = , s;u 3 (ii) 2 Leg(R ) for s 2 (0; 1)nfs ; : : : ; s g, strict i;u 3 1 1 (iii) 2 LegImm (R ) has a self{intersection at times (t ; t ) 2 S S for i 2 0 1 strict s;u 3 f0; 1g; and 2 LegImm (R ), s 2 fs ; : : : ; s g, has a self{intersection at times 0 1 1 1 ( ; ) 2 S S , s s j j 1 s; (iv) there exists a sequence of times c ; : : : ; c 2 S such that, the cusps of are at 1 2k times c . s; Assume that c 2 (t ; t ) and that is constant. This property will be preserved all 1 0 1 jOp(fc g) over the construction. s;u Step II. Unknotting process: Apply Lemma 8.1.18 to nd an extended family , (s; u) 2 [0; 1] [0; ], satisfying s;u (i) is an admissible homotopy, i;u s;u (ii) , i 2 f0; 1g, has one self{intersection at times (t (i; u); t (i; u)) and , s 2 0 1 0 1 0 1 fs ; : : : ; s g, has a self{intersection at times ( ; ). Moreover, t (i; 1) < ; < 1 1 s s s s j j j j t (i; 0), for (i; j) 2 f0; 1gf1; : : : ; g. s;u (iii) is constant for u , jOp(fc g) 4 i; (iv) The resolution , i 2 f0; 1g, is an (stabilized Legendrian) unknot unlinked Res(t ;t ) 0 1 from the other one. 2 2 0; 1; 4 4 Step III. Match the formal invariants of the resolutions and : Use Lemma Res(t ;t ) Res(t ;t ) 0 1 0 1 s;u 8.1.20 to extend our family to , (s; u) 2 [0; 1] [0; ], such that s;u (i) is an admissible homotopy, i;u s;u (ii) , i 2 f0; 1g, has one self{intersection at times (t (i; u); t (i; u)) and , s 2 0 1 0 1 0 1 fs ; : : : ; s g, has a self{intersection at times ( ; ). Moreover, t (i; 1) < ; < 1 1 s s s s j j j j t (i; 0), for (i; j) 2 f0; 1gf1; : : : ; g. s;u (iii) is constant for u , jOp(fc g) 4 i; (iv) The resolution , i 2 f0; 1g, is an unknot unlinked from the other one, Res(t ;t ) 0 1 3 3 0; 1; 4 4 (v) and are formally isotopic Legendrian embeddings (stabilized un- Res(t ;t ) Res(t ;t ) 0 1 0 1 knots). s; Step IV. Create a 1{parametric family of self{intersections: Finally, observe that satis es s;u the hypothesis of Lemma 8.1.21. So, apply this Lemma to extend our family to , (s; u) 2 s;u [0; 1] [0; 1]. The family satis es the required properties. 8.1.3. Proofs of the Lemmas 8.1.16, 8.1.17, 8.1.18, 8.1.19, 8.1.20 and 8.1.21. Case 1: fs ; : : : ; s g = ;. We assume that the set of points fs ; : : : ; s g is empty. We explain afterwards the reason why the same proofs work for the general case. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 63 Proof of Lemma 8.1.16. Take a retraction r : [0; 1] [0; 1] ! L such that: (i) r(1; u) = (1; ) for all u 2 [; 1]. (ii) r ((0; 1) [0; 1]) (0; 1) [0; 1]: s;u r(s;u) Then de ne ~ = . Proof of Lemma 8.1.17. Lemma 7.2.10 and Corollary 7.2.11 also work for . Proof of Lemma 8.1.18. The rst two steps in the proof of Proposition 8.1.13 readily imply, via Lemma 8.1.8, the result. Proof of Lemma 8.1.19. Draw in the Lagrangian projection a picture with two intersecting branches: one of the branches with slope 1 and orientation from right to left and the other one horizontal with a stabilization on it. The stabilization lies on the right of the intersection point. We continuously move the intersection point from left to right, making sure that the area of the intersection point is zero. So we get an actual intersection in the Legendrian above. Visual inspection shows that if the track intersection point has area zero all the other intersections in the Lagrangian projection have non zero area, therefore they are not actual intersections in the Legendrian. There are 2 times t < < < t in which the branch 0 0 1 1 1;1;0 becomes tangent. Those times correspond to cusps in the moduli space: curves of type . See Figure 35 for a description of half of the deformation (crossing of the rst stabilization). Figure 34b for a description of the 2-dimensional family as a subset of a 3-dimensional ball of Legendrian immersions. Figure 35. Description of the movie in Lagrangian projection. The last statement follows if we are able to extend the standard orientation over the cusps. We want to create a vector eld v 2 T LegImm(R ) transverse to the curve that de nes the orientation in the moduli space given by the Figure 37. Flowing along this vector, we obtain the required band. First, observe that away from the 2 cusp points this just corresponds to the standard orientation. Denote by j the part of the curve that crosses the rst s s2[0;1=2] two cusps in the moduli and x and to be those cusps. For s 2 [0; "], with " > 0 1=6 2=6 64 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS small enough, we interpolate between the standard normal vector for s = 0 and for s = " the vector de ned by a Hamiltonian function constant equal to 1 in the image of a segment 0 0 joining two points t ; t 2 S and supported on an arbitrarily small neighborhood of it. We 0 1 0 0 require t < t < t < t < t . This last condition allows to homotope those two vectors 0 1 0 0 1 through transverse vectors. The vector v is clearly equivalent the same as the one obtained by adding a positive bump (area) to the lagrangian projection at t and a negative bump at 0 0 t . Thus, the deformation has support in [t ; t ]. The algorithm to extend the deformation 1 0 1 to the cusp points of the moduli space is depicted in Figure 36 and it corresponds to change 0 0 0 the bump function from t to t + 2, i 2 f0; 1g, creating a family t , s 2 [0; 1=2], such that i i i;s 0 0 0 0 t is strictly increasing with sand such that t = t and t = t Likewise, t = t 0 1 0 i;s 0;1=6 0;2=6 1;1=6 and t = t . See Figure 36. This clearly produces a continuous family of the deformations 1;2=6 that removes the self{intersections. See Figure 37 to visualize the deformation as a tangent vector in the moduli space. Figure 36. Algorithm in Lagrangian projection to extend the deformation to the cusp points of the moduli space. Figure 37. Deformation in the moduli space. s;u Proof of Lemma 8.1.20. De ne , (s; u) 2 [0; 1] [0; 1], as follows: FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 65 0 1 s;u s (i) If the parity of tb( ) and tb( ) coincide de ne = for u 2 [0; ]. Res(t ;t ) Res(t ;t ) 3 0 1 0 1 In other case add one DS in such that for s = 0; 1 it is inserted on the segment 1 s ;v 0 (t ; t ) for u 2 [0; ] . Then, apply Lemma 8.1.19 to create a band ^ s 2 [0; 1], 0 1 1 1 0;v v; s;u 6(u );s 6 6 v 2 [ "; "], such that ^ = for v 2 [0; "]. Thus we de ne = ^ , 1 1 s;u s 2 [0; "], u 2 [ ; ]. Finally, apply Lemma 8.1.16 to extend for all s 2 [0; 1], 6 3 1 1 0; 1; 1 1 3 3 u 2 [ ; ]. Thus, we have made sure that tb( ) tb( ) = 2k, k 2 Z. 6 3 Res(t ;t ) Res(t ;t ) 0 1 0 1 Recall from [12, Section 2.7] that the Thuston{Bennequin invariant decreases by 1 after a stabilization and that the rotation number increases (decreases) by 1 whenever the stabilization is positive (negative). (ii) If k > 0, for each i 2 f1; : : : ; kg, take a path B : [0; 1] ! S ; s 7! B (s), such that i i 1 s;u 1 1 i B (0) 2 (t ; t ) and B (1) 2 S n[t ; t ]. De ne recursively in [0; 1] [ ; + ] i 0 1 i 0 1 3 3 3k 1 i 1 s; + 3 3k creating a DS on at time B (s). Observe that a DS decreases the Thurston{Bennequin invariant by 2. Thus, we obtain 2 2 0; 1; 3 3 (8) tb( ) = tb( ): Res(t ;t ) Res(t ;t ) 0 1 0 1 Do the symmetric sequence of moves in case k < 0 to obtain again (8). (iii) For any 2 Leg(R ) we have tb( ) + Rot( ) 2 2Z + 1 (see [15, Proposition 3.5.23]). 2 2 0; 1; 3 3 Thus, since tb( ) = tb( ) we conclude that Res(t ;t ) Res(t ;t ) 0 1 0 1 2 2 0; 1; 3 3 (9) Rot( ) Rot( ) = 2m: Res(t ;t ) Res(t ;t ) 0 1 0 1 Assume that m 0, the other case is totally analogous. Add a DS in (t ; t ) and 0 1 s;u consecutively apply Lemma 8.1.19 and Lemma 8.1.16.This adds a square , s 2 2 2 1 [0; 1], u 2 [ ; + ]. Now add a DS in (t ; t ) and consecutively apply Lemma 1 0 3 3 6m 2 1 s; + 3 3m 8.1.19 and Lemma 8.1.16. This adds a second square. In the new top , the di erence (9) has decreased by 2. Repeat m 1 more times to make m = 0 in that equation. 0;1 1;1 It follows that and have the same formal invariants. Res(t ;t ) Res(t ;t ) 0 1 0 1 Proof of Lemma 8.1.21. The proof is divided in several steps. Step 0: Preparation. By hypothesis, the standard co{orientations are matched, so we may assume that the self{ i 0 i 0 intersection times t and t satisfy that f( ) (t ); ( ) (t )g, i 2 f0; 1g; is a positive basis of 0 1 0 1 i i at the point of self{intersection (t ) = (t ), and that there exists " > 0 such that 0 1 " 0 0 is obtained from by pushing (Op(ft g)) in the Reeb direction for time ", 1 " 1 1 is obtained from by pushing (Op(ft g)) in the Reeb direction for time ". i i i i i We may also think that = ^ #( ; v ) where ^ is a Legendrian embedding and is 0 0 i i a very small Reeb chord between ^ (t ) and ^ (t ). Moreover, the hypothesis about the 0 1 co{orientations allows us to further assume that s s s 0 0 (A.1) = ^ #( ((1 )t); v ((1 )t)), for s 2 [0; "]; 0 " " s s 1 s 1 1 (A.1)' = ^ #( ((1 )t); v ((1 )t)), for s 2 [0; "]; " " 66 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS s 0 (A.2) ^ is constant for s 2 ["; 2"], 1 s 1 (A.2)' ^ is constant for s 2 ["; 2"]. s t +t 0 1 Assume that the xed cusp point of the curves is placed at time c = in the whole family. Choose a capping disk for D for , i 2 f0; 1g. Choose small neighborhoods Res(t ;t ) 0 1 U of each disk. By [9], U is contactomorphic to the standard contact R and there is a i 1 0 1 0 1 contactomorphism : U ! U such that ( (t )) = (t ) and ( (c )) = (c ). By 0 1 0 0 1 1 1 0 [10] we can further assume that = . In other words, the construction Res(t ;t ) Res(t ;t ) 0 1 0 1 that we are going to do is symmetric (there is a symmetry s ! 1 s for s 2 [0; 2"]). Ie all the constructions that we are going to do near s = 0 immediately translate to s = 1 by means of the contactomorphism . We do not detail the construction in a neighborhood of s = 1, just declaring it to be the symmetric construction to the one that we provide for s = 0 2 3 Use the disk D to construct a band B : [0; 1] ! R with the following properties: 0 0 B (0; v) = (v), B (1; v) = ^ (c ), 0 1 B (; 0) = ^ ((1 )t + c ), 0 0 1 B (; 1) = ^ ((1 )t + c ), 0 1 1 2 0 1 B ((0; 1) )\ ^ (S ) = ;, for each 2 (0; 1] the map B : [0; 1] ! R ; y 7! B (; v); is an embedding. Let > 0 be small enough in such a way that is constant (independent of s). [c ;c +] 1 1 Consider the linear interpolation : [0; 1] ! [t ; c ]; s 7! (1 s)t + s(c ). The family 0 1 0 1 (s) s 0 0 = B de nes a family of smooth nger bones joining ^ ( (s)) with ^ (t + t (s)). 1 0 0 0 0 0 Note that = . Use Lemma 8.1.8 to homotope this family through nger bones to 0 0 s 0 0 a family of transverse nger bones . Observe that = by the construction of the k k k interpolation. There is no obstruction to nd a family of vector elds v in ( ) such that v = v and ( ; v ) is a pre{Legendrian nger germ for ^. 0 s s Finally, by Lemma 8.1.6 there exists a non{negative integer K 2 Z in such a way that we 0 0 0 0 can interpolate through strict Legendrian immersions between ^ #( ; v ) and ^ #( ; v ) 0 0 0 k just by adding a tower of K DSs to the initial nger (by using Lemma 8.1.19). This number K is going to be xed for the rest of the proof. Step 1: Adding DSs to the family. s;u s;0 s De ne , (s; u) 2 [0; 1] [0; ]; by adding a tower of K DSs (see Figure 30) to = based at the point (T (s)) where T (s) satisfy T (s) t + for s 2 [0; "] and, symmetrically, for s 2 [1 "; "]; T (s) linearly interpolates between t + and c for s 2 ["; 2"] and, symmetrically, 0 1 for s 2 [1 2"; 1 "] ; T (s) c for s 2 [2"; 1 2"]. Step II: Crossing the rst half of the tower of DSs. s;u 1 2 Use Lemma 8.1.19 to de ne , (s; u) 2 [0; 1] [ ; ]; just by crossing half of the tower of 3 3 i; DSs through the self{intersection of , i 2 f0; 1g; at time t . Moreover, do this in such a s; 1 2 s;u way that = for (s; u) 2 ["; 1 "] [ ; ]. 3 3 Note that the path of Legendrians is constant over ["; 2"] and over [1 2"; 1 "]. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 67 Step III: Creating the self{intersections. s;u 2 To de ne , (s; u) 2 [0; 1] [ ; 1]; proceed as follows: strict s;u s;1 3 De ne , s 2 [0; "], in such a way that 2 LegImm (R ), s 2 [0; "] interpo- 0; 0 0 lates between and ^ #( ; v ). This is a direct application of Lemma 8.1.6 and s;u the fact that we have a tower of K DSs to produce this interpolation. De ne , s 2 [1 "; 1], symmetrically. s s s;u Use the family of pre{Legendrian nger germs ( ; v ) to de ne , s 2 ["; 2"]. s;u Do the symmetric construction to de ne for s 2 [1 2"; 1 "]. Note that the 2";1 1 2";1 self{intersection of and, symmetrically, of happens at times c and c + . s;u 1 1 De ne , s 2 [2"; 1 2"], just by using the pre{Legendrian nger germ ( ; v ) to create a self{intersection at times c and c + . Note that this is trivial because 1 1 the family is constant near the cusp point. This concludes the construction of the required family. Case 2: fs ; : : : ; s g =6 ;. Reasoning as in the smooth case, we conclude. Let us provide the details. In all the previous proofs we have made deformations in the moduli space LegImm(R ). These deformations s;u 1 3 over a point : S ! R of the moduli space change the type of Legendrian by creating tangencies at a nite number of values of the parameter 0 < u < u 1, N 0, ie 1 N strict s;u 3 2 LegImm (R ), s 2 [0; 1] . The tangencies are created identifying at most two points in the domain z (s; u ); z (s; u ) 2 S . We assume in all the steps that the points 0 j 1 j 0 1 z (s ; u ); z (s ; u ); ; , (j; k) 2 f1; : : : ; g f1; : : : ; Ng, are pairwise di erent. This 0 j 1 j k k s s j j 0 1 implies that there is no interference between the deformation and the pair of points ; . s s j j Observe that a generic choice of the times z (s; u); z (s; u) satis es this assumption. So, we 0 1 need to change nothing if we add the set fs ; : : : ; s g in the proofs, except for the fact that we need to make the choices of times (mainly for the nger creation process) generic in the previous sense. 8.1.4. Index 1 surgeries. 2 4 Proposition 8.1.22. Let D(x; y) R be the disk of radius 2 and let ' : D(x; y) ! Hor(R ) be a map into the horizontal embeddings space such that (i) the con guration of ' is given by + 2 vertical segments, ie 1 strict 3 ( ') (LegImm (R )) = (fx = 0g[fx = s g[[fx = 1g)\ D(x; y); G 1 for 0 and 0 < s < < s < 1; (ii) the area function has the same sign over the rst and the last segments. Then, there exists a 1{parametric family of disks ' : D ! Hor(R ), t 2 [0; 1], satisfying that (i) ' = ', (ii) ' = ' , tj@D 0j@D (iii) The con guration of ' is given by the right diagram of Figure 38. G 1 68 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS The changes in the con guration of curves of strict immersions in ' moving from G t ' to ' correspond to an ambient index one surgery whose attaching points are G 0 G 1 the middle points (0; 0) and (1; 0) of the two vertical segments fx = 0g and fx = 1g. The attached disk is given by [0; 1] [ "; "], with " > 0 small enough, and the surgery replaces the pair of 1{disks f0; 1g [ "; "] by the pair of 1{disks [0; 1]f"g. The new 1{disks, that are curves of strict immersions, have an even number of cusp points and they do not add new zeroes to their the area functions. The other segments fx = s g; : : : ;fx = s g remain untouched in the diagram. Moreover, a nite number of closed circular components of type 1 2 2 2 S = f(x; y) 2 D : (x ) + y = r g, with 0 < r < < r < c. These new closed n 1 n components do not have cusp points and have positive area function everywhere. ++ + Figure 38. Description of the homotopy ' . Observe that the cusp points appear in pairs, therefore no zeroes of the area function appear. Proof. Denote the two curves of strict Legendrian immersions in D = '(D) by C G G 0 i s and C . Let i 2 f0; 1g and x 2 C . Choose a path , s 2 [ 1; 2], transversal to the 1 i s i strict immersions curves at C such that C \ f : s 2 [ 1; 2]g = f g. A neighborhood i i s s s s K = f : s 2 [ 1; 2]; r 2 [ 1; 1]g D of the path may be assumed to satisfy for r r s 2 [ 1; 2]. s;u Apply Proposition 8.1.15 to nd a disk E = f : s; u 2 [0; 1]g such that s;0 s = , s 2 [0; 1]. strict s;u 3 C = f 2 LegImm (R ) : (s; u) 2 f0; 1g [0; 1][ [0; 1]f1gg is a continuous s;u s;u curve of strict Legendrian immersions, with self-intersection times t ; t 2 S , 0 1 s;u s;u t 6= t . 0 1 s ;u S = f : u 2 [0; 1], j 2 f1; : : : ; g, is a curve of strict Legendrian immersions, There is a nite number of curves C of strict Legendrian immersions de ned by s;u C = f : (s; u) 2 [0; 1]fu gg; n n s;u s;u s;u n n 1 where 0 < u < < u < 1. The self-intersection times of are ; 2 S , 1 n 0 1 s;u s;u n n with 6= . These curves do not have cusp points. 0 1 s;u s;u n n 6= t , for each i; j 2 f0; 1g and (s; u ) 2 C \C . n n i j Apply Lemma 7.2.1 to lift E into a disk of horizontal immersions. Now, apply again Lemma s;u s;u 7.2.1 over C in t " and t + ", for " > 0 small enough in order to increase the area 0 0 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 69 function at will. Do likewise over the self{intersection points in C . Note that there is no interaction between the two deformations because the self{intersection times are di erent. By construction the outward normal vector eld X to C = @E coincides with the standard s;u orientation. Push C in the direction of X to obtain an extended disk E = f : s 2 3 4 s;0 s [ 1; 2]; u 2 [0; 2]g E in LegImm(R ), which lifts to Hor(R ), such that = for each s 2 [ 1; 2]. We de ne the map 4 s;u s;u : [ 1; 2] [0; 2] [ 1; 1] ! Hor(R ) : (s; u; r) 7! = : See Figure 39, to visualize the piecewise linear disk that we are about to de ne in [ 1; 2] 3 2 [0; 2] [ 1; 1] R . First, we de ne the base (a closed piecewise linear curve in R ): = f(s; r) 2 [ 1; 2] [ 1; 1] : jrj g(s)g; for 1 + s; if s 2 [ 1; 0]; g(s) = 1; if s 2 [0; 1]; 2 s; if s 2 [1; 2]: And now the piecewise linear graph in R as 8 8 9 2 + s jrj; if s 2 [ 1; 0]; < < = = (s; u; r) 2 [ 1; 2] [1; 2] [ 1; 1] : (s; r) 2 ; u = 2 jrj; if s 2 [0; 1]; : : : ; 3 s jrj; if s 2 [1; 2]: We de ne a new (piecewise linear) disk = (@( ) [0; 1])[ [ 1; 2] [0; 2] [ 1; 1]. Figure 39. Construction of D. Observe that the boundary of coincides with @ , therefore we can de ne a new disk ~ ^ D = (Dn )[ gluing along the boundaries. We can de ne a new morphism ' : D ! Hor(R ) ' (p); p 2 (Dn ); 0 B p 7! (p); p 2 : It is left to check that ' and ' are homotopic. This is performed by building any homotopy 0 1 between the two disks and relative to the boundary @ inside [ 1; 2] [0; 2] B B [ 1; 1]. 70 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS 8.2. Globalization. We use the theory developed in the last two Subsections to prove the main theorem of Section 7. We rst adapt a previous construction. Lemma 8.2.1. [Corollary of Lemma 8.1.19] Let ' : D(x; y) ! HorImm(R ) be a disk of horizontal immersions such that: (i) '(@D) Hor(R ), strict 1 3 (ii) ( ') (LegImm (R )) = C is a connected curve with a unique zero of the area function 2 C . t 4 Then, there exists a 1{parametric family of disks ' : D ! HorImm(R ), t 2 [0; 1], satisfying (i) ' = ', t 0 (ii) ' = ' , j@D j@D (iii) there is a C deformation C , t 2 [0; 1], of C = C and a family of connected curves t 0 S such that t 1 strict 3 { ( ' ) (LegImm (R )) = C [ S , G t t { 2 C is the unique zero of the area function in ' and the curves S encircle t t { C = C for t 2 [0; 1=2), t 0 { C is a DS of C in the moduli (see Lemma 8.1.19) and lies between the rst 1 1=2 and the second cusp. { S = ; for t 2 [0; 1=2), 2 2 1 2 { S = fx + y = ( t) g for t 2 [1=2; 3=4] and { S = S for t 2 [3=4; 1]. t 3=4 See Figure 40. Figure 40. Schematic description of Lemma 8.2.1. The blue point in the middle of the circle represents and the red middle circle in the cylinder the self{intersection (with non zero area) created in the DS. Proof. Since the problem is local we may assume that the disk is small and the self{intersection times for the Legendrian projections in the curve C are constant and equal to t and t . 0 1 FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 71 Choose t 2 S a time value di erent from t and t . Use polar coordinates D = D(r; ) 2 0 1 and de ne parametrically for each horizontal immersion '(r; ) a DS that is denoted B = D [0; 1=2] ! HorImm(R ), where B(r; ; h), h 2 [0; 1=2], is the DS of '(r; ). Now we apply Lemma 8.1.19 in order to further deform and create 4 cusp points in the moduli: just sliding one branch (t ) over the other (t ). Apply Lemma 7.2.1 to place the 0 1 zero in the required position. This second deformation produces a family of disks B(r; ; h) with h 2 [1=2; 1]. De ne L = fB(r; ; h) : r = ; h 2 [0; 1]g Image(B); T = D r fh = 1g Image(B); that builds a piecewise linear disk D = L[ T whose boundary coincides with the boundary 1 1 0 1 0 of D(r ) D. Replace D(r ) by D , ie de ne D = (DnD(r ))[D . The associated 2 2 2 diagram of strict immersions conforms the statement of the Lemma. See Figure 40. Proof of Theorem 7.1.2. Let be a loop of horizontal embeddings which lies in (Hor(R )) , 1 0 ie Rot ( ) = 0. Assume that Area( ) = 0, we must check that is trivial. Let D be a disk in HorImm(R ) whose boundary is given by f g. Apply successively Proposition 7.3.1 to assume that each curve of strict Legendrian immersions in D has at most one zero of the area function. Since Area( ) = 0 the number of zeroes is even. If D has not zeroes then we are done. Hence, assume that D has a positive number of zeroes and denote them ; : : : ; . 0 2n 1 Let C D be the curve of strict Legendrian immersions that contains , k 2 f0; : : : ; 2n k G k 1g. It is sucient to show that we can do a surgery along any two curves C and C to 0 1 produce a new curve of strict Legendrian immersions that contains the zeroes and 0 1 without creating any new zero. Application of Proposition 7.3.1 will provide the required result. To formalize the previous discussion, note that C is naturally oriented by the gradient 1;0 vector g of the Area function at . Together with the standard co-orientation of at i i , denoted by w , it provides an oriented basis at T D namely fg ; w g. We can assume i i i i that the orientations induced at the two points and are the same. If it is not the case, 0 1 apply Lemma 8.2.1 in order to change the orientation of the curve C . This works because the Area function changes signs when crossing a cusp in the curve C and thus the gradient of the area function over the zero gets multiplied by 1. Select a pair of points arbitrarily close to , whose area function is positive. Fix an i i embedded curve : [0; 1] ! D such that (i) = and it is transverse to C . Moreover, i i d d t 1;0 t we assume that de nes the standard coorientation at 2 . Respectively dt jt=0 dt jt=1 de nes the opposite orientation to the standard one (see Figure 41 to learn how to deal with the wrong coorientation case). A small tubular neighborhood D of the path satis es the hypothesis of Proposition 8.1.22. This implies that we can get a deformed disk with the same number of zeroes such that the deformation is performed compactly supported on the disk D following the Diagram explained in Figure 38. This new diagram makes the two zeroes be part of the same connected component of curves of strict immersions, see Figure 42. This is ensured by the 72 EDUARDO FERNANDEZ, JAVIER MARTINEZ-AGUINAGA, AND FRANCISCO PRESAS ++ Figure 41. Embedded path connecting and conforming to the coori- 0 1 entations rule. compatibility of the orientations that we have enforced in the basis fg ; w g of T D. Apply i i Proposition 7.3.1 to cancel the two zeroes. + + + + g g 0 1 - - - - Figure 42. Schematic description of the neighborhood of (t). The cusps between the concentric circles are not drawn to make the picture easier to understand. References [1] J. Adachi. Classi cation of horizontal loops in standard Engel space. Int. Math. Res. Not. 2007; Vol. 2007: article ID rnm008, 29 pages, doi:10.1093/imrn/rnm008. [2] C. C. Adams. The knot book. An elementary introduction to the mathematical theory of knots. Amer- ican Mathematical Society, Providence, RI, 2004. xiv+307 pp. [3] V. I. Arnold, S. M. Gusein-Zade, A. N. Varchenko. Singularities of Di erentiable Maps. Volume I. Birkh auser, Boston, Basel, Sttutgart. 1985. [4] D. Bennequin. Entrelacements et equations de Pfa . Third Schnepfenried geometry conference, Vol. 1 (Schnepfenried, 1982), 87{161, Ast erisque, 107-108, Soc. Math. France, Paris, 1983. [5] R. Budney. A family of embedding spaces. Geometry & Topology 13 (2008): 41{83. [6] R. Casals, A. del Pino. Classi cation of Engel knots. arXiv:1710.11034. [7] Y. Chekanov. Di erential algebra of Legendrian links. Invent. Math. 150 (2002), no. 3, 441{483. FUNDAMENTAL GROUPS OF FORMAL LEGENDRIAN AND HORIZONTAL EMBEDDING SPACES 73 1 2 [8] F. Ding, H. Geiges. The di eotopy group of S S via contact topology. Compos. Math. 146 (2010), no. 4, 1096?1112. [9] Y. Eliashberg. Classi cation of contact structures on R . Internat. Math. Res. Notices 1993, no. 3, 87{91. [10] Y. Eliashberg, M. Fraser. Topologically trivial Legendrian knots. J. Symplectic Geom. 7 (2009), no. 2, 77{127. [11] Y. Eliashberg, N. Mishachev, Introduction to the h{Principle. Graduated Studies in Mathematics, 48. AMS, Providence, RI, 2002. [12] J. Etnyre. Legendrian and transversal knots. Handbook of knot theory, 105{185, Elsevier B. V., Ams- terdam, 2005. [13] J. Etnyre and K. Honda. Knots and contact geometry I: Torus knots and the gure eight knot. J. Symplectic Geom. 1 (2001), 63-120. [14] D. Fuchs, S. Tabachnikov. Invariants of Legendrian and transverse knots in the standard contact space. Topology 36 (1997), no. 5, 1025{1053. [15] H. Geiges. An Introduction to Contact Topology. Cambr. Studies in Adv. Math. 109. Cambr. Univ. Press 2008. [16] H. Geiges. Horizontal loops in Engel space. Math. Ann. Oct 2008, Vol. 342, no. 2, 291{296. [17] R. S. Hamilton. The inverse function theorem of Nash and Moser. Bull. Amer. Math. Soc. (N.S.) 7 (1982), no. 1, 65222. [18] A. Hatcher. A proof of the Smale conjecture, Di (S ) ' O(4). Ann. of Math. (2) 117 (1983), no. 3, 553{607. [19] A. Hatcher, Algebraic topology. Cambridge University Press, Cambridge, 2002. xii+544 pp. ISBN: 0-521-79160-X; 0-521-79540-0 [20] A. Hatcher. Spaces of knots. arXiv:math/9909095. [21] M.W. Hirsch. Immersions of manifolds. Trans. Amer. Math. Soc. 93 1959 242{276. [22] K. Igusa. Higher singularities of smooth functions are unnecessary. Ann. of Math. (2) 119 (1984), no. 1, 1?58. [23] T. K alm an. Contact homology and one parameter families of Legendrian knots. Geometry & Topology 9 (2005): 2013{2078. [24] S. Lang, Fundamentals of di erential geometry. Graduate Texts in Mathematics, 191. Springer-Verlag, New York, 1999. xviii+535 pp. ISBN: 0-387-98593-X 53-01 (58-01) [25] L. Ng. Computable Legendrian invariants. Topology 42 (2003), no. 1, 55{82. [26] J. Milnor. Morse theory. Based on lecture notes by M. Spivak and R. Wells. Annals of Mathematics Studies, No. 51 Princeton University Press, Princeton, N.J. 1963 vi+153 pp. 57.50 (53.72) [27] A. del Pino, F. Presas. Flexibility for tangent and transverse immersions in Engel manifolds. arXiv:1609.09306. [28] J. M. Sablo , M. G. Sullivan. Families of Legendrian submanifolds via generating families. Quantum Topol. 7 (2016), no. 4, 639{668. [29] P. Ozsv ath, Z. Szab o, D. Thurston. Legendrian knots, transverse knots and combinatorial Floer homol- ogy. Geometry & Topology 12 (2008), no. 2, 941{980. Universidad Complutense de Madrid, Departamento de Algebra, Geometra y Topologa, Facul- tad de Matematicas, and Instituto de Ciencias Matematicas CSIC-UAM-UC3M-UCM, C. Nicolas Cabrera, 13-15, 28049 Madrid, Spain. E-mail address : eduarf01@ucm.es;eduardo.fernandez@icmat.es Universidad Complutense de Madrid, Departamento de Algebra, Geometr a y Topolog a, Facul- tad de Matematicas, and Instituto de Ciencias Matematicas CSIC-UAM-UC3M-UCM, C. Nicolas Cabrera, 13-15, 28049 Madrid, Spain. E-mail address : javier.martinez.aguinaga@icmat.es Instituto de Ciencias Matematicas CSIC-UAM-UC3M-UCM, C. Nicolas Cabrera, 13-15, 28049 Madrid, Spain. E-mail address : fpresas@icmat.es
Journal
Mathematics – arXiv (Cornell University)
Published: Nov 12, 2017