20.07.2013 Views

Applications of Symplectic Geometry to Hamiltonian Mechanics

Applications of Symplectic Geometry to Hamiltonian Mechanics

Applications of Symplectic Geometry to Hamiltonian Mechanics

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Applications</strong> <strong>of</strong> <strong>Symplectic</strong><br />

<strong>Geometry</strong> <strong>to</strong> Hamil<strong>to</strong>nian<br />

<strong>Mechanics</strong><br />

Samuel Thomas Lisi<br />

A dissertation submitted in partial fulfillment<br />

<strong>of</strong> the requirements for the degree <strong>of</strong><br />

Doc<strong>to</strong>r <strong>of</strong> Philosophy<br />

Department <strong>of</strong> Mathematics<br />

New York University<br />

January 2006<br />

Helmut H. W. H<strong>of</strong>er


Dedicated <strong>to</strong> my grandparents, Thomas and Helvi Lisi.<br />

iii


Acknowledgements<br />

First and foremost, I thank my advisor Helmut H<strong>of</strong>er. His great enthusiasm for<br />

symplectic geometry couldn’t help but be contagious. His insight and intuition<br />

continue <strong>to</strong> be an inspiration <strong>to</strong> me. I also thank him for sharing many insights<br />

in<strong>to</strong> the non-mathematical aspects <strong>of</strong> being a mathematician.<br />

I thank Casim Abbas for much support and guidance. Many <strong>of</strong> the ideas in<br />

Chapter 5 trace their roots <strong>to</strong> conversations with Pr<strong>of</strong>essor Abbas. In addition,<br />

his feedback has been most helpful, and his advice on exposi<strong>to</strong>ry writing has<br />

been invaluable.<br />

I thank Joe C<strong>of</strong>fey for many helpful discussions, and for his encouragement.<br />

I am especially grateful <strong>to</strong> him for explaining dividing sets <strong>to</strong> me, and for helping<br />

me <strong>to</strong> understand Legendrian surgery.<br />

I am grateful <strong>to</strong> Richard Hind for clarifying for me the implications <strong>of</strong> Stein<br />

domains.<br />

I would like <strong>to</strong> acknowledge the help, guidance and support <strong>of</strong> Chris Wendl,<br />

Richard Siefring, Breno Madero, Joel Fish and Barney Bramham. I have learned<br />

much <strong>of</strong> what I know from our Courant <strong>Symplectic</strong> group, and for this I am<br />

very grateful.<br />

I thank all <strong>of</strong> my friends and family and everyone else who has made this<br />

possible. I thank my parents and brother for their constant encouragement and<br />

support. Without them, none <strong>of</strong> this would have been possible. Finally, and<br />

most deeply, I am grateful <strong>to</strong> Leah Tiisler for always being there and supporting<br />

me through everything.<br />

iv


Abstract<br />

In this thesis, we consider three applications <strong>of</strong> pseudoholomorphic curves <strong>to</strong><br />

problems in Hamil<strong>to</strong>nian dynamics. In a first part, we prove an existence result<br />

for homoclinic orbits on a contact-type, critical energy level <strong>of</strong> an au<strong>to</strong>nomous<br />

Hamil<strong>to</strong>nian, provided that the level is Hamil<strong>to</strong>nian displaceable. To do this, we<br />

transform the problem in<strong>to</strong> a problem <strong>of</strong> Lagrangian intersection Floer theory.<br />

This involves a construction due <strong>to</strong> Mohnke [34] and some ideas from Legendrian<br />

surgery. In particular, we prove a generalization <strong>of</strong> Séré’s result [36] on the<br />

existence <strong>of</strong> homoclinic orbits for an au<strong>to</strong>nomous Hamil<strong>to</strong>nian system.<br />

In a second part, we develop a theory <strong>of</strong> pseudoholomorphic curves in<strong>to</strong> a<br />

singular contact manifold, which represents the critical level <strong>of</strong> an au<strong>to</strong>nomous<br />

Hamil<strong>to</strong>nian. We show that a pseudoholomorphic half-plane is asymp<strong>to</strong>tic <strong>to</strong><br />

a homoclinic orbit. Furthermore, in a non-degenerate case, this convergence<br />

is <strong>of</strong> an exponential nature. This result is a first step <strong>to</strong>wards understanding<br />

the change in contact homology under Legendrian surgery. Such a surgery<br />

formula would enable the computation <strong>of</strong> contact homology for every contact<br />

three manifold.<br />

Finally, we lay the groundwork for an energy quantization result for pseudoholomorphic<br />

planes with a weaker energy bound than in the existing theory.<br />

This is a question related <strong>to</strong> the problem <strong>of</strong> understanding the compactification<br />

<strong>of</strong> the space <strong>of</strong> generalized pseudoholomorphic curves as envisioned by H<strong>of</strong>er<br />

and studied by Abbas, Cieliebak and H<strong>of</strong>er in [3]. This is part <strong>of</strong> an ongoing<br />

project with Casim Abbas and Helmut H<strong>of</strong>er.<br />

v


Contents<br />

Dedication iii<br />

Acknowledgements iv<br />

Abstract v<br />

List <strong>of</strong> Figures viii<br />

1 Introduction and Main Results 1<br />

1.1 Geometric constructions and Hamil<strong>to</strong>nian dynamics . . . . . . . 3<br />

1.1.1 Lagrangian intersections and Homoclinic orbits for Hamil<strong>to</strong>nian<br />

systems . . . . . . . . . . . . . . . . . . . . . . . 3<br />

1.1.2 Lagrangian intersections and the Weinstein conjecture . 4<br />

1.2 PDE techniques and Hamil<strong>to</strong>nian dynamics . . . . . . . . . . . 4<br />

1.2.1 Homoclinic orbits by PDE methods . . . . . . . . . . . . 4<br />

1.2.2 Finding periodic orbits by means <strong>of</strong> generalized pseudoholomorphic<br />

curves . . . . . . . . . . . . . . . . . . . . . 6<br />

2 Preliminaries 9<br />

2.1 Contact and symplectic manifolds . . . . . . . . . . . . . . . . . 10<br />

2.1.1 Contact manifolds . . . . . . . . . . . . . . . . . . . . . 10<br />

2.1.2 Legendrian submanifolds . . . . . . . . . . . . . . . . . . 12<br />

2.1.3 Weinstein domains . . . . . . . . . . . . . . . . . . . . . 12<br />

2.2 Legendrian surgery . . . . . . . . . . . . . . . . . . . . . . . . . 13<br />

2.2.1 The surgery presentation . . . . . . . . . . . . . . . . . . 14<br />

2.2.2 Weinstein’s handle attaching construction . . . . . . . . 19<br />

2.2.3 Attaching a critical handle performs −1 surgery . . . . . 21<br />

3 An existence result for homoclinic orbits 24<br />

3.1 The main theorem and overview . . . . . . . . . . . . . . . . . . 24<br />

3.2 Pro<strong>of</strong> <strong>of</strong> our result . . . . . . . . . . . . . . . . . . . . . . . . . 28<br />

3.3 <strong>Applications</strong> <strong>to</strong> the Weinstein conjecture . . . . . . . . . . . . . 34<br />

vi


4 Pseudoholomorphic curves in the singular level 38<br />

4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38<br />

4.2 A compactification <strong>of</strong> the singular level . . . . . . . . . . . . . . 39<br />

4.2.1 Construction <strong>of</strong> the compactification . . . . . . . . . . . 39<br />

4.2.2 A local model <strong>of</strong> the singularity . . . . . . . . . . . . . . 41<br />

4.2.3 A local model for the neighbourhood <strong>of</strong> a homoclinic orbit 46<br />

4.3 Type (I) and (II) ˜ J-holomorphic curves . . . . . . . . . . . . . . 49<br />

4.3.1 The singular almost complex structure . . . . . . . . . . 49<br />

4.3.2 Type (I) pseudoholomorphic curves . . . . . . . . . . . . 50<br />

4.3.3 Type (II) pseudoholomorphic curves . . . . . . . . . . . 50<br />

4.4 Asymp<strong>to</strong>tic behaviour <strong>of</strong> a Type (II) finite energy half-plane . . 52<br />

4.4.1 Convergence <strong>to</strong> a homoclinic orbit . . . . . . . . . . . . . 52<br />

4.4.2 Exponential rate <strong>of</strong> convergence . . . . . . . . . . . . . . 63<br />

4.5 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69<br />

5 Energy quantization for pseudoholomorphic curves with a relaxed<br />

area bound 70<br />

5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70<br />

5.2 “Renormalization” by the Reeb flow . . . . . . . . . . . . . . . . 73<br />

5.3 Pro<strong>of</strong> <strong>of</strong> the result . . . . . . . . . . . . . . . . . . . . . . . . . 75<br />

5.3.1 The boundary value problem . . . . . . . . . . . . . . . . 75<br />

5.3.2 Compactness <strong>of</strong> disks . . . . . . . . . . . . . . . . . . . . 76<br />

5.3.3 Fredholm theory . . . . . . . . . . . . . . . . . . . . . . 84<br />

5.4 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90<br />

Bibliography 91<br />

vii


List <strong>of</strong> Figures<br />

1.1 Relationship between homoclinic orbits, periodic orbits and Reeb<br />

chords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6<br />

2.1 Weinstein’s local handle model . . . . . . . . . . . . . . . . . . . 19<br />

2.2 The global handle attaching construction . . . . . . . . . . . . . 20<br />

3.1 Perturbing the critical level F = 0. . . . . . . . . . . . . . . . . 32<br />

3.2 The Lagrangian L . . . . . . . . . . . . . . . . . . . . . . . . . . 33<br />

4.1 Local model near the singular Legendrian . . . . . . . . . . . . . 46<br />

viii


Chapter 1<br />

Introduction and Main Results<br />

In this thesis, we apply pseudoholomorphic curve methods <strong>to</strong> the study <strong>of</strong> Hamil<strong>to</strong>nian<br />

dynamics. We focus on two main problems : the study <strong>of</strong> homoclinic<br />

orbits in a Hamil<strong>to</strong>nian system, and the Weinstein conjecture. Our pseudoholomorphic<br />

curve methods fall in<strong>to</strong> two broad categories. In a first part <strong>of</strong> this<br />

thesis, we exploit a geometric construction due <strong>to</strong> Mohnke [34]. This allows us<br />

<strong>to</strong> transform dynamical questions in<strong>to</strong> a question <strong>of</strong> whether a given immersed<br />

Lagrangian is embedded or not. In a second part, we use more analytical methods<br />

<strong>to</strong> study pseudoholomorphic curves. In each part, we address both the<br />

question <strong>of</strong> homoclinic orbits and the Weinstein conjecture.<br />

The first type <strong>of</strong> orbit we consider are periodic orbits. The Weinstein conjecture<br />

claims that for any Reeb vec<strong>to</strong>r field on any closed contact manifold,<br />

there exists a periodic orbit. In dimension 3, much progress has been made,<br />

though the conjecture is, at present, open. In higher dimensions, the body <strong>of</strong><br />

work is quite sparse. In this work, we prove the Weinstein conjecture holds<br />

for a certain class <strong>of</strong> contact manifolds realized as hypersurfaces in symplectic<br />

manifolds, for any dimension. We also prove the Weinstein conjecture holds for<br />

a contact form on a three-manifold if there exists a pseudoholomorphic plane<br />

with a weaker energy bound than previously required.<br />

The second type <strong>of</strong> orbits we consider are homoclinic orbits (homoclinic <strong>to</strong><br />

a rest point). We recall that in an au<strong>to</strong>nomous Hamil<strong>to</strong>nian system, given by<br />

a Hamil<strong>to</strong>nian function H, critical points <strong>of</strong> H are rest points for the Hamil<strong>to</strong>nian<br />

flow. An orbit which is asymp<strong>to</strong>tic <strong>to</strong> a rest point, both in forward and<br />

in backwards time, is called a homoclinic orbit. The existence <strong>of</strong> a transverse<br />

homoclinic orbit is quite interesting from a dynamical perspective, as it then<br />

implies the existence <strong>of</strong> rich dynamics nearby (by Smale’s horseshoe construction).<br />

Pseudoholomorphic curves are an essential <strong>to</strong>ol in modern <strong>Symplectic</strong> <strong>Geometry</strong>.<br />

Introduced by Gromov in 1985 [20], they allow one <strong>to</strong> see symplectic<br />

1


geometry as a generalization <strong>of</strong> Kähler geometry. A pseudoholomorphic curve<br />

is a solution <strong>to</strong> a certain partial differential equation, which can be seen as<br />

a non-linear analogue <strong>to</strong> the Cauchy–Riemann equations <strong>of</strong> complex analysis.<br />

Pseudoholomorphic curve theory combines methods from many different areas<br />

<strong>of</strong> mathematics : non-linear functional analysis, elliptic partial differential<br />

equation theory on manifolds, <strong>to</strong>pology, complex geometry, classical symplectic<br />

geometry. The study <strong>of</strong> these curves provides the essential ingredient <strong>of</strong> many<br />

results in symplectic and contact geometry, including results on Lagrangian<br />

intersections, symplectic embeddings and Hamil<strong>to</strong>nian dynamics.<br />

A first example <strong>of</strong> such an application <strong>to</strong> Hamil<strong>to</strong>nian dynamics comes from<br />

Floer’s pro<strong>of</strong> <strong>of</strong> a conjecture <strong>of</strong> Arnol ′ d. This gives a lower bound on the number<br />

<strong>of</strong> periodic orbits for a Hamil<strong>to</strong>nian system, in terms <strong>of</strong> the <strong>to</strong>pology <strong>of</strong> the<br />

underlying manifold. To obtain this result, Floer defined a homology theory,<br />

similar in flavour <strong>to</strong> Morse theory, in which pseudoholomorphic curves serve as<br />

the analogue <strong>to</strong> gradient flow lines [15].<br />

In 1993, H<strong>of</strong>er [21] introduced the appropriate notion <strong>of</strong> pseudoholomorphic<br />

curve in<strong>to</strong> a contact manifold (which is a geometric object that corresponds,<br />

roughly speaking, <strong>to</strong> a compact energy level in a classical Hamil<strong>to</strong>nian system).<br />

He established that a non-compact pseudoholomorphic curve with a certain<br />

energy bound is asymp<strong>to</strong>tic <strong>to</strong> a periodic orbit <strong>of</strong> the associated Hamil<strong>to</strong>nian<br />

system. Building on this, Eliashberg, Givental and H<strong>of</strong>er introduced symplectic<br />

field theory and its simpler cousin, contact homology, which are invariants <strong>of</strong><br />

symplectic cobordisms between contact manifolds [13]. These Floer homologylike<br />

invariants are differential graded algebras, generated by periodic orbits with<br />

a differential that counts pseudoholomorphic curves connecting them.<br />

In this thesis, we build on these results in several directions. In Chapter 2, we<br />

introduce the notation we use for the remainder <strong>of</strong> this work, and we also provide<br />

some background constructions that will be useful in the remaining sections. In<br />

Chapter 3, we present results in Hamil<strong>to</strong>nian dynamics obtained by means <strong>of</strong><br />

a geometric construction. This construction allows us <strong>to</strong> reduce the problem<br />

<strong>to</strong> one <strong>of</strong> showing a Lagrangian cannot be embedded. The non-embeddedness<br />

<strong>of</strong> this Lagrangian will then come from an argument using pseudoholomorphic<br />

disks with boundary in the Lagrangian. In Chapter 4, we introduce a class <strong>of</strong> a<br />

contact manifolds with singular contact forms that correspond <strong>to</strong> critical levels<br />

in an au<strong>to</strong>nomous Hamil<strong>to</strong>nian system. We then develop a singular pseudoholomorphic<br />

curve theory <strong>to</strong> study homoclinic orbits on the corresponding critical<br />

level in the Hamil<strong>to</strong>nian system. In Chapter 5, we study pseudoholomorphic<br />

planes in<strong>to</strong> symplectizations <strong>of</strong> contact manifolds, but with a weaker energy<br />

condition than has been considered in the literature (e.g. [21, 6]). We prove<br />

that such curves have an energy threshold. This is closely related <strong>to</strong> Abbas,<br />

Cieliebak and H<strong>of</strong>er’s programme <strong>to</strong> solve the Weinstein conjecture in dimen-<br />

2


sion 3 by means <strong>of</strong> studying a generalization <strong>of</strong> the class <strong>of</strong> holomorphic curves.<br />

The remainder <strong>of</strong> this chapter summarizes the main results and discusses the<br />

connection <strong>to</strong> existing work.<br />

1.1 Geometric constructions and Hamil<strong>to</strong>nian<br />

dynamics<br />

1.1.1 Lagrangian intersections and Homoclinic orbits for<br />

Hamil<strong>to</strong>nian systems<br />

Our main result in Chapter 3 is a new existence result for homoclinic orbits : a<br />

singular energy surface <strong>of</strong> contact type with precisely one hyperbolic singularity<br />

and finite displacement energy contains a homoclinic orbit. Séré’s important<br />

result [36] concerning such hypersurfaces in Euclidean space is then a corollary<br />

<strong>of</strong> this more general result. The pro<strong>of</strong> uses a powerful construction from Mohnke<br />

[34] and some ideas from Legendrian surgery.<br />

The main result is as follows :<br />

Theorem 3.1.2. Let (W0, ω0) be a symplectic manifold (with boundary) <strong>of</strong> dimension<br />

2n, and let F be a Morse function on W0 so that F has a unique critical<br />

point x0 <strong>of</strong> index n. Suppose there is a Liouville vec<strong>to</strong>r field Y , i.e., LY ω = ω,<br />

gradient-like for F . Suppose furthermore that the Hamil<strong>to</strong>nian vec<strong>to</strong>r field <strong>of</strong><br />

F , XF , has a hyperbolic zero at x0.<br />

We normalize F by the conditions F (x0) = 0, and dF [Y ] ≤ 1.<br />

Suppose that there exists S > 0 and a symplectic manifold (W, ω), symplectically<br />

aspherical and <strong>of</strong> bounded geometry, so that (F −1 [−S, 0), ω0) symplectically<br />

embeds in (W, ω) and furthermore, the embedding, restricted <strong>to</strong> F −1 (−S/2), induces<br />

an injection from the fundamental group <strong>of</strong> F −1 (−S/2) <strong>to</strong> the fundamental<br />

group <strong>of</strong> W .<br />

Suppose furthermore that there exists a Hamil<strong>to</strong>nian diffeomorphism Φ <strong>of</strong> W ,<br />

with compact support and H<strong>of</strong>er norm ||Φ||, displacing the image <strong>of</strong> F −1 [−S, 0).<br />

Then, our original Hamil<strong>to</strong>nian vec<strong>to</strong>r field XF has an orbit homoclinic <strong>to</strong> x0<br />

<strong>of</strong> action bounded above by ||Φ||/(1 − e −S/2 ).<br />

We obtain Séré’s result [36] as a corollary <strong>to</strong> this. Also as a corollary, we<br />

obtain that any compact, contact-type singular hypersurface with precisely one<br />

hyperbolic singularity in a subcritical Stein domain admits a homoclinic orbit.<br />

We note that our methods are very different from Séré’s. The pro<strong>of</strong> in<br />

[36] uses a variational approach. Our approach here relates the problem <strong>to</strong><br />

3


one <strong>of</strong> symplectic <strong>to</strong>pology. We exploit the fact that if we have an embedded<br />

Lagrangian submanifold, and we can displace it from itself by a Hamil<strong>to</strong>nian<br />

diffeomorphism, then there must be the bubbling <strong>of</strong>f <strong>of</strong> either a disk or a sphere<br />

in the moduli spaces <strong>of</strong> pseudoholomorphic strips, as in Lagrangian intersection<br />

Floer homology.<br />

1.1.2 Lagrangian intersections and the Weinstein conjecture<br />

In Chapter 3, we also introduce a construction that allows us <strong>to</strong> prove the<br />

Weinstein conjecture for Hamil<strong>to</strong>nian displaceable contact-type hypersurfaces <strong>of</strong><br />

symplectic manifolds. This again builds on Mohnke’s Lagrangian construction.<br />

We obtain,<br />

Theorem 3.3.1. Let (M, ξ) be a contact manifold with contact form λ.<br />

Suppose that there exists a symplectic manifold (W, ω) <strong>of</strong> bounded geometry and<br />

symplectically aspherical, and S > 0 so that we may symplectically embed ι :<br />

([−S, S] × M, d(e s λ)) ↩→ (W, ω) and furthermore, this embedding induces an<br />

injection on the fundamental groups.<br />

If there exists a Hamil<strong>to</strong>nian diffeomorphism Φ <strong>of</strong> W with compact support and<br />

finite H<strong>of</strong>er norm ||Φ|| displacing the image <strong>of</strong> [−S, S] × M, then there exists a<br />

periodic orbit <strong>of</strong> action bounded above by 4||Φ||/(e S − e −S ).<br />

In particular, this result applies for all subcritically Stein fillable contact<br />

manifolds. This also overlaps with a result <strong>of</strong> Liu and Tian [31].<br />

1.2 PDE techniques and Hamil<strong>to</strong>nian dynamics<br />

In Chapters 4 and 5, we turn <strong>to</strong> PDE methods in order <strong>to</strong> study pseudoholomorphic<br />

curves, with an eye <strong>to</strong> applications in Hamil<strong>to</strong>nian dynamics.<br />

1.2.1 Homoclinic orbits by PDE methods<br />

In Chapter 4, we study pseudoholomorphic half-planes with image in the symplectization<br />

<strong>of</strong> a singular contact manifold, <strong>of</strong> dimension 3. This singular contact<br />

manifold is a model <strong>of</strong> the critical level <strong>of</strong> an au<strong>to</strong>nomous Hamil<strong>to</strong>nian system.<br />

Such Hamil<strong>to</strong>nians arise also in a handle surgery construction, due <strong>to</strong> Weinstein<br />

(which we discuss in Chapter 2). The use <strong>of</strong> pseudoholomorphic half–planes <strong>to</strong><br />

understand homoclinic orbits (in Euclidean space) was first considered by H<strong>of</strong>er<br />

and Wysocki [22].<br />

4


In a first part, we introduce a construction <strong>to</strong> compactify the critical level<br />

(with the critical point deleted) by gluing in a Legendrian knot, L. This then<br />

provides a model for the critical level as a closed contact manifold, M. The<br />

construction also gives us a contact form, singular along the Legendrian knot.<br />

We then establish a local model near this singular Legendrian.<br />

From this local model, we are led <strong>to</strong> consider a special class <strong>of</strong> singular<br />

almost complex structures on R × M. These are singular along R × L. The<br />

local model, however, allows us <strong>to</strong> describe the almost complex structure in a<br />

neighbourhood <strong>of</strong> R × L.<br />

Finally, we introduce two types <strong>of</strong> pseudoholomorphic curve. The first type,<br />

Type (I) pseudoholomorphic curves, are pseudoholomorphic maps in<strong>to</strong> R × M<br />

in the least restrictive sense : they are pseudoholomorphic everywhere u(z) /∈ L.<br />

In other words, their derivative is complex linear everywhere the image carries<br />

an almost complex structure. The second type, Type (II) pseudoholomorphic<br />

curves, is much more restrictive : we impose a condition <strong>to</strong> make the curve<br />

compatible with the local model, in a sense we make precise in Chapter 4.<br />

Nevertheless, it is an elliptic problem, and satisfies some useful properties. In<br />

particular, we show that in certain special circumstances, the two classes are<br />

actually the same.<br />

We prove that a Type (II) pseudoholomorphic half-plane is asymp<strong>to</strong>tic <strong>to</strong><br />

a homoclinic orbit. Furthermore, we show that with slightly stronger boundary<br />

conditions, such a half-plane converges at an exponential rate. A very nice<br />

relationship between homoclinic orbits and Legendrian chords allows us <strong>to</strong> circumvent<br />

many difficulties in this setting.<br />

We may state the main result as follows, where H = {s + it ∈ C | s ≥ 0} is<br />

the right half-plane :<br />

Theorem 4.4.4. Suppose that ũ : H → R × M is a Type (II) ˜ J holomorphic<br />

curve, has finite energy and satisfies the boundary condition at infinity :<br />

dist(u(0 + it), L) → 0 as |t| −→ ∞.<br />

Let T = u ∗ dλ. Then, for any sequence R ′ k → ∞ there is a subsequence Rk so<br />

that<br />

u(Rke it ) → x(T t/π)<br />

where x(t) is a homoclinic orbit <strong>of</strong> the (singular) Reeb vec<strong>to</strong>r field. The convergence<br />

is in C ∞ ([−π/2, π/2], M).<br />

We also obtain a result, more technical <strong>to</strong> state, that if the asymp<strong>to</strong>tic<br />

limit is a transverse homoclinic orbit, then there exist coordinates in which the<br />

rate <strong>of</strong> convergence is exponential. We show that this rate <strong>of</strong> approach may<br />

be described in terms <strong>of</strong> an eigenvalue (and corresponding eigenvec<strong>to</strong>r) <strong>of</strong> a<br />

5


Figure 1.1: This figure shows a homoclinic orbit on the critical level, <strong>to</strong>gether<br />

with the corresponding periodic orbit and Legendrian chord on nearby levels.<br />

This diagram is merely a simplistic sketch, <strong>to</strong> illustrate the type <strong>of</strong> phenomena<br />

we are interested in.<br />

self-adjoint, unbounded opera<strong>to</strong>r. These results are given precisely in Theorem<br />

4.4.12.<br />

A motivation for the study <strong>of</strong> these half-planes, in addition <strong>to</strong> the dynamical<br />

interest <strong>of</strong> finding homoclinic orbits, is that the change in contact homology<br />

under Legendrian surgery is related <strong>to</strong> the occurrence <strong>of</strong> homoclinic orbits on<br />

an associated critical energy level (see Figure 1.1). This current work is part <strong>of</strong><br />

a larger goal <strong>to</strong> encode these changes in terms <strong>of</strong> a contact homology for this<br />

critical level, which would include the homoclinic orbits among the genera<strong>to</strong>rs.<br />

Ultimately, this might provide a means <strong>of</strong> computing the change in contact<br />

homology across Legendrian surgery. This would be <strong>of</strong> particular interest in<br />

dimension three, for which there is a Legendrian surgery presentation <strong>of</strong> every<br />

contact manifold ([10]). This approach might then lead <strong>to</strong> a surgery formula for<br />

contact homology.<br />

Another interesting point is that the Floer complex <strong>of</strong> such a theory would<br />

carry extra structure. Transverse homoclinic orbits admit a concatenation operation.<br />

Corresponding <strong>to</strong> this, we should also have a gluing <strong>of</strong> half-planes. This<br />

represents an extra operation on the Floer complex, <strong>of</strong> a nature not studied<br />

before.<br />

1.2.2 Finding periodic orbits by means <strong>of</strong> generalized<br />

pseudoholomorphic curves<br />

In Chapter 5, we consider the behaviour <strong>of</strong> infinite energy pseudoholomorphic<br />

planes. We recall that in all current applications <strong>of</strong> pseudoholomorphic curves<br />

<strong>to</strong> contact geometry, we restrict our attention <strong>to</strong> finite energy curves. These are<br />

curves for which<br />

<br />

E(ũ) := sup{ ũ ∗ d(φλ) | φ : R → [0, 1] with φ ′ ≥ 0} < ∞.<br />

6


This energy was introduced in [21], and is the assumption necessary for the<br />

existing compactness theory, as in [6].<br />

In this work, we consider the case <strong>of</strong> E(ũ) = ∞, but impose instead :<br />

<br />

0 < u ∗ dλ < ∞.<br />

We refer <strong>to</strong> this weaker energy as the contact area since it measures the symplectic<br />

area <strong>of</strong> the projection <strong>of</strong> T u <strong>to</strong> the contact structure. We note that this<br />

condition means that the R–fac<strong>to</strong>r <strong>of</strong> the curve is no longer a proper map. Such<br />

infinite energy curves then behave very wildly, as compared <strong>to</strong> the finite energy<br />

curves.<br />

In this work, we prove that such curves have an energy threshold :<br />

Theorem 5.1.1. Suppose ũ : C → R × M is pseudoholomorphic with respect<br />

<strong>to</strong> an almost complex structure ˜ J, adjusted <strong>to</strong> the contact form λ, and has finite<br />

contact area :<br />

<br />

0 < u ∗ dλ = C < ∞.<br />

C<br />

We also impose the technical condition that πλ T u = 0.<br />

Then, there exists a periodic orbit <strong>of</strong> the Reeb vec<strong>to</strong>r field with action 0 < T ≤ C.<br />

The key ingredient <strong>of</strong> the pro<strong>of</strong> is a construction <strong>of</strong> renormalization. This<br />

construction makes precise the observation, due <strong>to</strong> H<strong>of</strong>er, that such infinite<br />

energy curves are merely finite energy curves that have been “smeared” out by<br />

the Reeb flow.<br />

The technical condition that πλ T u never vanish is merely introduced <strong>to</strong><br />

simplify the analysis. We note that πλ T u also satisfies a Cauchy-Riemann type<br />

equation, and so, if not identically zero, its zeros are isolated and have positive<br />

multiplicity. In practice, then, this condition may be checked by algebraic<br />

means.<br />

The more general case, in which πλ T u may have zeros, introduces a greater<br />

number <strong>of</strong> technical problems. We include, in Chapter 5, a discussion <strong>of</strong> the<br />

ingredients <strong>of</strong> the pro<strong>of</strong> in the more general case. We also remark that preliminary<br />

work indicates that C should be equal <strong>to</strong> a period <strong>of</strong> the Reeb vec<strong>to</strong>r<br />

field.<br />

This result is part <strong>of</strong> an ongoing research project with Casim Abbas and<br />

Helmut H<strong>of</strong>er. Our eventual goal is <strong>to</strong> understand precisely in what sense we<br />

may speak <strong>of</strong> these curves as being asymp<strong>to</strong>tic <strong>to</strong> a periodic orbit.<br />

A description <strong>of</strong> such planes is necessary <strong>to</strong> understand the compactification<br />

<strong>of</strong> H<strong>of</strong>er’s generalized pseudoholomorphic curve theory. These generalized pseudoholomorphic<br />

curves satisfy a PDE which resembles the non-linear Cauchy-<br />

Riemann equations, but with an extra term involving a harmonic form. A<br />

7


compactification <strong>of</strong> the space <strong>of</strong> such objects is needed for the Abbas-Cieliebak-<br />

H<strong>of</strong>er program <strong>to</strong> prove the Weinstein conjecture in dimension three (begun in<br />

[3]).<br />

8


Chapter 2<br />

Preliminaries<br />

We are interested in the situation in which we have a 2n dimensional Weinstein<br />

domain (W, ω, Y, F ). That is, we have a symplectic manifold (W, ω) and a<br />

smooth Hamil<strong>to</strong>nian function F : W → R whose level sets are contact–type.<br />

Then, a level set <strong>of</strong> F for a regular value is a contact manifold with a preferred<br />

choice <strong>of</strong> contact form, iY ω| F −1 (c). The associated Reeb vec<strong>to</strong>r field is a positive<br />

multiple <strong>of</strong> the Hamil<strong>to</strong>nian vec<strong>to</strong>r field, and so questions about qualitative<br />

behaviour <strong>of</strong> the dynamics associated <strong>to</strong> the Hamil<strong>to</strong>nian vec<strong>to</strong>r field (existence<br />

<strong>of</strong> periodic orbits, ...) can be answered by studying the dynamics <strong>of</strong> the Reeb<br />

vec<strong>to</strong>r field.<br />

The problem we are interested in is <strong>to</strong> understand what happens at a critical<br />

level for F . This is no longer a submanifold <strong>of</strong> W . Nevertheless, assuming<br />

only finitely many critical points, after removing the critical points <strong>of</strong> F , we<br />

obtain a non–compact contact manifold. In this situation, a new qualitative<br />

behaviour can emerge : a homoclinic orbit. A homoclinic orbit is an orbit <strong>of</strong><br />

the Hamil<strong>to</strong>nian vec<strong>to</strong>r field that converges, both in forward and backward time,<br />

<strong>to</strong> the same critical point <strong>of</strong> F .<br />

This problem is <strong>of</strong> interest both in the study <strong>of</strong> Hamil<strong>to</strong>nian dynamics, and<br />

also in order <strong>to</strong> understand what happens <strong>to</strong> contact homology after performing<br />

Legendrian surgery.<br />

We will begin by reviewing some results and constructions in contact geometry.<br />

This will allow us <strong>to</strong> define our terminology and fix sign conventions.<br />

A deep observation that traces back <strong>to</strong> Eliashberg [12], is that contact manifolds,<br />

unlike their symplectic cousins, allow for “cut-and-paste” constructions.<br />

Much work in contact geometry has been accomplished by these methods. This<br />

thesis lays the foundations for a project <strong>to</strong> understand how these cut-and-paste<br />

constructions interact with pseudoholomorphic curves.<br />

In this chapter, we will discuss Legendrian surgery through two different<br />

constructions. We will also prove some neighbourhood theorems that will be<br />

9


useful in our later analysis.<br />

2.1 Contact and symplectic manifolds<br />

2.1.1 Contact manifolds<br />

A contact manifold (M 2n−1 , ξ) is a smooth, odd-dimensional manifold, M, <strong>to</strong>gether<br />

with a hyperplane bundle ξ ⊂ T M, called the contact structure. This<br />

hyperplane bundle satisfies a non-integrability condition : locally, we can find a<br />

one form λ so ξ = ker λ and λ ∧ (dλ) n−1 is a volume form. We say the contact<br />

structure is co-orientable if it can be realized globally as the kernel <strong>of</strong> a one-form<br />

λ with λ ∧ (dλ) n−1 a volume form on M.<br />

Any one-form λ so that ker λ is a contact structure (equivalently, λ∧dλ > 0)<br />

is called a contact form. We note that if λ is a contact form generating the<br />

contact structure ξ, then fλ is also, for any f : M → (0, ∞). The contact<br />

structure, ξ, is a symplectic vec<strong>to</strong>r bundle over M, with symplectic structure<br />

given by dλ. We note that even though λ is not canonical, the conformal class<br />

<strong>of</strong> dλ|ξ is.<br />

A diffeomorphism Ψ : M1 → M2 between two contact manifolds, (M1, ξ1)<br />

and (M2, ξ2) is called a contact diffeomorphism, or a contac<strong>to</strong>morphism, if<br />

Ψ∗ξ1 = ξ2. If ξi = ker λi, then this is equivalent <strong>to</strong> requiring the existence<br />

<strong>of</strong> a function f : M1 → (0, ∞) so that Ψ ∗ λ2 = fλ1.<br />

On a contact manifold (M, ξ), given the choice <strong>of</strong> a contact form, we introduce<br />

the Reeb vec<strong>to</strong>r field, X. This vec<strong>to</strong>r field satisfies dλ(X, ·) = 0 and<br />

is normalized so λ(X) = 1. An essential point is that the Reeb vec<strong>to</strong>r field<br />

depends on the choice <strong>of</strong> contact form. Once we have a choice <strong>of</strong> contact form,<br />

and thus <strong>of</strong> Reeb vec<strong>to</strong>r field, we have a splitting <strong>of</strong> T M = RX ⊕ ξ. We let<br />

πλ denote the projection from T M → ξ corresponding <strong>to</strong> this splitting. Where<br />

there is no ambiguity, we will sometimes omit the subscript, and write π.<br />

In this work, we will be interested in understanding the dynamics <strong>of</strong> the<br />

Reeb vec<strong>to</strong>r field. There are two main motivations for this. The first is that<br />

Reeb dynamics are closely related <strong>to</strong> Hamil<strong>to</strong>nian dynamics (for instance, the<br />

work in [27] answers questions about geodesic flow on S 2 ). The second is that a<br />

better understanding <strong>of</strong> all possible Reeb dynamics associated <strong>to</strong> a given contact<br />

structure will shed light on contact geometry.<br />

Relation <strong>to</strong> symplectic geometry<br />

Contact manifolds arise naturally as boundaries <strong>of</strong> symplectic manifolds. Indeed,<br />

we say that a contact manifold M is a contact–type hypersurface in a<br />

10


symplectic manifold (W, ω) if there exists a vec<strong>to</strong>r field, Y , defined in a neighbourhood<br />

<strong>of</strong> M, so that LY ω = ω and Y is transverse <strong>to</strong> M. This induces a<br />

contact structure on M by taking the maximal symplectic sub-bundle <strong>of</strong> T M.<br />

We call any vec<strong>to</strong>r field Y with the property that LY ω = ω a Liouville vec<strong>to</strong>r<br />

field, a conformally symplectic vec<strong>to</strong>r field or an expanding vec<strong>to</strong>r field. We note<br />

that all three terms are used in the literature, but occasionally with different<br />

sign conventions (e.g. Weinstein, in [37], uses the opposite sign convention).<br />

If M is a contact-type hypersurface in W , and Y is a Liouville vec<strong>to</strong>r field<br />

transverse <strong>to</strong> M, we obtain a preferred choice <strong>of</strong> the contact form by λ = iY ω|M.<br />

We note that by Cartan’s formula (and since ω is closed), d iY ω = LY ω = ω,<br />

so this form is the restriction <strong>of</strong> a primitive <strong>of</strong> ω <strong>to</strong> M. We note that any two<br />

Liouville vec<strong>to</strong>r fields transverse <strong>to</strong> a given contact-type hypersurface M induce<br />

iso<strong>to</strong>pic contact forms. By Gray’s theorem (which we discuss below), the two<br />

contact structures are contac<strong>to</strong>morphic.<br />

Given such a contact–type hypersurface, M, there are many contact–type<br />

hypersurfaces M ′ nearby. Let Ψt be the flow <strong>of</strong> the vec<strong>to</strong>r field Y and f : M →<br />

R, with small enough C 0 norm that Ψf(x) is defined for all x ∈ M. Then we may<br />

define another submanifold <strong>of</strong> W by M ′ = {Ψf(x)(x) | x ∈ M}. This manifold is<br />

diffeomorphic <strong>to</strong> M, but the induced contact form, pulled back <strong>to</strong> M is given<br />

by e f(x) λ.<br />

An important result on the stability <strong>of</strong> contact structures is as follows :<br />

Theorem 2.1.1 (Gray’s Theorem). Let (M, ξ0) be a closed contact manifold.<br />

Suppose that ξt are contact structures on M for t ∈ [0, 1], varying smoothly in<br />

t. Then, there exists a one parameter family <strong>of</strong> diffeomorphisms Ψt : M → M,<br />

t ∈ [0, 1] so that Ψ0 = Id and Ψt∗ξt = ξ0.<br />

Equivalently, if ξt = ker λt, then Ψt ∗ λt = ftλ0, where ft : M → (0, ∞) is a<br />

smoothly varying family <strong>of</strong> functions with f0 ≡ 1.<br />

Every contact manifold (M, ξ) with contact form λ may be realized as a<br />

contact-type hypersurface in some (non-compact) symplectic manifold. Indeed,<br />

we take W = R × M with the symplectic form ω = d(e a λ), where a is the<br />

coordinate on the R component. M embeds as {0}×M with transverse Liouville<br />

vec<strong>to</strong>r field Y = ∂<br />

∂a .<br />

Contact manifolds, much like symplectic manifolds, admit no local invariants.<br />

Indeed, we have a number <strong>of</strong> neighbourhood theorems. The first is a<br />

Darboux chart theorem:<br />

Theorem 2.1.2. let M and M ′ be contact manifolds with contact forms λ and<br />

λ ′ respectively. For any p ∈ M and p ′ ∈ M ′ , there exist neighbourhoods U and<br />

U ′ and a diffeomorphism ψ : U → U ′ so that ψ(p) = p ′ and ψ ∗ λ ′ = λ.<br />

11


2.1.2 Legendrian submanifolds<br />

In symplectic <strong>to</strong>pology, an interesting class <strong>of</strong> submanifolds are the Lagrangian<br />

submanifolds. Likewise, in contact geometry, we have an analogous concept.<br />

Suppose (M, ξ) is a contact manifold <strong>of</strong> dimension 2n−1. We say a submanifold<br />

L ⊂ M is Legendrian if L is n − 1 dimensional and T L ⊂ ξ. Equivalently, T L<br />

is a Lagrangian sub-bundle <strong>of</strong> the symplectic vec<strong>to</strong>r bundle (ξ, dλ). (Observe<br />

that even though this statement makes use <strong>of</strong> the contact form λ, it is actually<br />

independent <strong>of</strong> λ since it only relies on the conformal class <strong>of</strong> dλ|ξ.)<br />

In this work, we will focus primarily on closed contact 3-manifolds. In this<br />

case, a (closed) Legendrian submanifold will be a link.<br />

An important property <strong>of</strong> Legendrian submanifolds is that they admit a<br />

neighbourhood theorem.<br />

Theorem 2.1.3 (Legendrian Neighbourhood Theorem). Let (M 2n−1 , λ) be a<br />

contact manifold with preferred choice <strong>of</strong> contact form. Let ℓ be a Legendrian<br />

submanifold <strong>of</strong> M. Then, for a small neighbourhood U <strong>of</strong> the Legendrian ℓ and<br />

for ɛ > 0, sufficiently small, we have a diffeomorphism Ψ : U → (−ɛ, ɛ) × T ∗ ℓ,<br />

with Ψ ∗ λ = dt + Θ, where t is the coordinate in the R variable and Θ is the<br />

canonical one form on T ∗ ℓ.<br />

We recall that the canonical one form Θ can be written as Θ = pi dqi, where<br />

qi are coordinates on the manifold and pi are coordinates on the fibre.<br />

This is [4, Thm. 2.2.4]. A slightly weaker form, which only shows that a<br />

Legendrian neighbourhood is contac<strong>to</strong>morphic <strong>to</strong> a standard model, is given by<br />

[16, Cor. 2.28].<br />

The Legendrian neighbourhood theorem is the corners<strong>to</strong>ne <strong>of</strong> many “cutand-paste”<br />

constructions on contact manifolds. We will discuss how this allows<br />

us <strong>to</strong> construct new contact manifolds by attaching handles, creating connected<br />

sums and, in the three dimensional case, performing Dehn surgery on the manifold,<br />

and creating a (canonical) contact structure on the new manifold.<br />

2.1.3 Weinstein domains<br />

In order <strong>to</strong> simplify our bookkeeping in dealing with the cut-and-paste constructions,<br />

we need the notion <strong>of</strong> a Weinstein domain.<br />

A Weinstein domain consists <strong>of</strong> a a symplectic manifold (W 2n , ω), a Liouville<br />

vec<strong>to</strong>r field Y and a Morse function F , with Y gradient–like for F . In other<br />

words, we require LY ω = ω, that the critical points <strong>of</strong> F are precisely the zeros<br />

<strong>of</strong> Y and furthermore, that the zeros <strong>of</strong> Y are non-degenerate. We also require<br />

that Y be transverse <strong>to</strong> the boundary <strong>of</strong> W . By using Y , we introduce the<br />

12


positive and negative boundary <strong>of</strong> W . We require then that the positive and<br />

negative boundaries <strong>of</strong> W be regular level sets <strong>of</strong> F .<br />

We observe that we follow the opposite sign convention from Weinstein, who<br />

first introduced the concept in [37], under the name <strong>of</strong> “Liouville pair”. We will<br />

denote a Weinstein domain by the tuple (W, ω, Y, F ).<br />

Let x0 be a critical point for F . Let W u be the unstable manifold for the<br />

flow <strong>of</strong> Y , and W s the stable manifold. Then, W u and W s are co-isotropic<br />

and isotropic, respectively. It follows from our definition that all <strong>of</strong> the critical<br />

points <strong>of</strong> F have Morse index at most n. We also observe that every regular<br />

level set <strong>of</strong> F is a contact–type hypersurface in W .<br />

If the negative boundary <strong>of</strong> W is empty, we call the Weinstein domain a<br />

Weinstein filling. A special case <strong>of</strong> a Weinstein domain is a Stein domain —<br />

indeed, in that case we may take Y = ∇F and ω = − d( d(F ◦ J)), where F is<br />

the plurisubharmonic function.<br />

A contact manifold M is said <strong>to</strong> be Weinstein fillable if there exists a Weinstein<br />

filling for which M is the positive boundary. We remark that we differ<br />

here from the terminology used in [8], where Cieliebak uses the terms Weinstein<br />

domain and relative Weinstein domain where we use, respectively, Weinstein<br />

filling and Weinstein domain.<br />

2.2 Legendrian surgery<br />

We will now present Legendrian surgery in two different ways. The first description<br />

makes it clear that Legendrian surgery is Dehn surgery along a knot<br />

<strong>to</strong>gether with a contact-theoretic construction <strong>to</strong> construct a contact structure<br />

on the surgered manifold, unique up <strong>to</strong> contac<strong>to</strong>morphism. The uniqueness<br />

(and a short-cut <strong>to</strong> existence) <strong>of</strong> the construction will require some background<br />

on convex surfaces.<br />

The second construction, due <strong>to</strong> Weinstein, will realize this same surgery by<br />

means <strong>of</strong> attaching a handle <strong>to</strong> our contact manifold. This will have the useful<br />

side-effect <strong>of</strong> constructing a Weinstein domain providing an oriented symplectic<br />

cobordism from our old contact manifold <strong>to</strong> the new contact manifold, or viceversa,<br />

depending on the sign <strong>of</strong> the surgery. We remark that this construction is<br />

more general than our application makes explicit — it allows for the construction<br />

<strong>of</strong> a connected sum <strong>of</strong> two contact manifolds, for instance.<br />

We will finally show that Legendrian ±1 surgery may be realized by means<br />

<strong>of</strong> the handle attaching construction due <strong>to</strong> Weinstein [37]. This formulation <strong>of</strong><br />

Legendrian surgery will be the most useful <strong>to</strong> us, and is <strong>of</strong> interest in its own<br />

right.<br />

13


2.2.1 The surgery presentation<br />

Given a framed knot in a three dimensional manifold and a rational number r<br />

(possibly infinite), we may construct a new manifold obtained by performing<br />

surgery on that knot. The remarkable fact is that given a Legendrian knot in a<br />

contact 3–manifold M, and if 0 = r = 1/q for some 0 = q ∈ Z, then the manifold<br />

M ′ obtained by performing 1/q surgery on M also carries a contact structure<br />

(which is independent <strong>of</strong> the choices made in the construction). Furthermore,<br />

away from a neighbourhood <strong>of</strong> the Legendrian knot, the two manifolds are<br />

contac<strong>to</strong>morphic.<br />

We will first discuss the construction involved in 1/q surgery, and refer <strong>to</strong><br />

the key results from the literature that establish some <strong>of</strong> the key properties <strong>of</strong><br />

this Legendrian surgery. We will sketch a pro<strong>of</strong> <strong>of</strong> the result that 1/q surgery<br />

can be realized by a sequence <strong>of</strong> +1 or −1 surgeries. In our exposition, we will<br />

follow the notation <strong>of</strong> Ding, Geiges and Stipsicz, and will cite some <strong>of</strong> their key<br />

results on Legendrian surgery presentations <strong>of</strong> contact 3–manifolds.<br />

We refer the interested reader <strong>to</strong> the papers by Ding, Geiges and then also<br />

Stipsicz : [9, 10, 11]. In these papers, Ding and Geiges introduce p/q surgery<br />

(rather than just 1/q as we discuss here). They also show in [10] that every<br />

contact three manifold may be obtained by 1/q surgery on a link in the standard<br />

tight S 3 . Then, in [11], an algorithm is given <strong>to</strong> turn these r = 1/q surgeries<br />

<strong>to</strong> a sequence <strong>of</strong> +1 and −1 surgeries. The authors <strong>of</strong> these papers, and others,<br />

have used these constructions <strong>to</strong> find various contact structures that are tight,<br />

but with interesting properties with regards <strong>to</strong> symplectic fillability. Many <strong>of</strong><br />

the ideas and constructions are due <strong>to</strong> Eliashberg and then <strong>to</strong> Gompf [12, 19].<br />

In order <strong>to</strong> perform 1/q surgery, we must start with a framed knot. We note<br />

that in a contact 3-manifold (M, ξ = ker λ), a Legendrian knot L comes with<br />

a canonical framing from the contact structure. Indeed, a framing is given by<br />

the choice <strong>of</strong> a vec<strong>to</strong>r field along K, positively transverse <strong>to</strong> ξ. Any two choices<br />

will then be homo<strong>to</strong>pic.<br />

To simplify the exposition <strong>of</strong> the construction (at least conceptually), we<br />

will introduce some results <strong>of</strong> Giroux and Honda on convex surfaces. Furthermore,<br />

these results will be key in establishing that that the resultant (M ′ , ξ ′ ) is<br />

independent <strong>of</strong> the choices made. Honda has written a good survey article that<br />

provides a self-contained collection <strong>of</strong> definitions and results [29].<br />

We will first introduce the concept <strong>of</strong> a characteristic foliation. Suppose<br />

Σ ↩→ M is an embedded surface in a contact 3–manifold (M, ξ). Then, we<br />

introduce a singular foliation F on Σ by taking, at each point p ∈ Σ where ξ is<br />

not tangent <strong>to</strong> Σ, the line in TpΣ that lies in ξp. The singularities are the points<br />

where ξp = TpΣ. In other words, F(p) = ξp ∩ TpΣ.<br />

The characteristic foliation is important because it describes the neighbour-<br />

14


hood <strong>of</strong> Σ up <strong>to</strong> contac<strong>to</strong>morphism [17]:<br />

Proposition 2.2.1. Let ξ and ξ ′ be two contact structures that induce the same<br />

characteristic foliation on Σ. Then ξ and ξ ′ are iso<strong>to</strong>pic through contact structures,<br />

where the iso<strong>to</strong>py fixes Σ.<br />

A (closed) convex surface in a contact 3-manifold (M, ξ) is an embedded<br />

(closed) surface Σ ↩→ M for which there exists a transversal contact vec<strong>to</strong>r field<br />

Z. We say that a vec<strong>to</strong>r field Z is contact if the flow <strong>of</strong> Z preserves the contact<br />

structure ξ. Equivalently, if ξ = ker λ, we have LZλ = fλ for some function<br />

f : M → R. We will only consider the case <strong>of</strong> closed surfaces. We note that<br />

among embedded closed surfaces, the property <strong>of</strong> being convex is C ∞ -generic.<br />

Given a convex surface Σ, we introduce the dividing set,<br />

ΓΣ := {x ∈ Σ | Z(x) ∈ ξx}.<br />

This will be a collection <strong>of</strong> properly embedded smooth curves in Σ. The iso<strong>to</strong>py<br />

class <strong>of</strong> ΓΣ does not depend on choice <strong>of</strong> Z.<br />

We say that a singular foliation F is adapted <strong>to</strong> a dividing set ΓΣ if there<br />

exists a contact structure ζ defined in a neighbourhood <strong>of</strong> Σ so that ΓΣ is<br />

the dividing set for ζ and F is the characteristic foliation for ζ. (We remark<br />

that Giroux has characterized which singular foliations come as characteristic<br />

foliations [17]. Furthermore, we may intrinsically characterize which pairs <strong>of</strong><br />

characteristic foliation and dividing set are adapted [10]).<br />

The main interest <strong>of</strong> convex surfaces and their dividing sets comes from a<br />

result <strong>of</strong> Giroux’s [17] :<br />

Theorem 2.2.2 (Giroux’s Flexibility Theorem). Assume Σ is convex with characteristic<br />

foliation F0, transverse contact vec<strong>to</strong>r field Z and dividing set ΓΣ. Let<br />

F1 be another singular foliation adapted <strong>to</strong> ΓΣ. Then, there exists an iso<strong>to</strong>py φt<br />

<strong>of</strong> Σ in M, t ∈ [0, 1] so that :<br />

1. The iso<strong>to</strong>py fixes the dividing set : φt|ΓΣ = Id for all t.<br />

2. The iso<strong>to</strong>py is transverse <strong>to</strong> the transverse contact vec<strong>to</strong>r field : φt(Σ) ⋔ Z<br />

for all t ∈ [0, 1].<br />

3. The final surface φ1(Σ) has characteristic foliation F.<br />

The essential point <strong>of</strong> this theorem is that the dividing set contains the<br />

complete information about the contact structure in a neighbourhood <strong>of</strong> the<br />

convex surface.<br />

For our application, we need the following result, the key ideas <strong>of</strong> which are<br />

due <strong>to</strong> Kanda [30] and <strong>to</strong> Makar-Limanov [32], but whose formulation in these<br />

terms is due <strong>to</strong> Honda [28]. This is Proposition 4.1 in [29] (and a pro<strong>of</strong> is also<br />

provided there) :<br />

15


Theorem 2.2.3. Let n ∈ Z. Suppose we have a solid <strong>to</strong>rus S 1 × D 2 with<br />

boundary T 2 = ∂(S 1 × D 2 ), a candidate dividing set Γ T 2 and a characteristic<br />

foliation on T 2 that is adapted <strong>to</strong> Γ T 2. Suppose furthermore that we have :<br />

(a) the number <strong>of</strong> components <strong>of</strong> the dividing set #Γ T 2 = 2<br />

(b) with respect <strong>to</strong> a framing <strong>of</strong> the <strong>to</strong>rus in which the meridian has zero slope<br />

and the longitude (x, y) = (0, 0) has infinite slope, slope(Γ T 2) = 1/n.<br />

Then there exists a tight contact structure on S 1 × D 2 , unique up <strong>to</strong> iso<strong>to</strong>py<br />

fixing the boundary, realizing T 2 as a convex surface, and so that its dividing<br />

set is given by our candidate Γ T 2.<br />

We note that by Giroux’s flexibility theorem, the requirement <strong>of</strong> the existence<br />

<strong>of</strong> an adapted characteristic foliation is a compatibility condition on the<br />

dividing set : what it is does not matter since it is iso<strong>to</strong>pic <strong>to</strong> any other one.<br />

We also observe that this result is only stated for n > 0 in [29]. Nevertheless,<br />

the result is true for all n ∈ Z. Indeed, suppose we have a contact structure ξ<br />

on S 1 × D 2 so that the slope <strong>of</strong> Γ T 2 is 1/n. Then, we push ξ forward by means<br />

<strong>of</strong> a diffeomorphism <strong>of</strong> the solid <strong>to</strong>rus corresponding <strong>to</strong> m Dehn twists along the<br />

meridian <strong>to</strong> obtain a contact structure ξ ′ on S 1 × D 2 . We compute then that<br />

the dividing set for this contact structure has slope 1/(n + m).<br />

Remark 2.2.4. A standard model for each <strong>of</strong> these is given by the following<br />

contact forms on S 1 × D 2 :<br />

and by<br />

αn = cos nφ dx + sin nφ dy for n = 0<br />

α0 = αstd = dx + y dφ.<br />

We recall that, by the Legendrian neighbourhood theorem, a Legendrian<br />

knot has a tubular neighbourhood admitting a standard local model. We will<br />

take the model :<br />

Nδ = {(x, y, φ) ∈ R × R × S 1 | x 2 + y 2 ≤ δ 2 }<br />

α0 = dx + y dφ<br />

L0 := {(0, 0, φ) | φ ∈ S 1 }.<br />

Here, and in the following, we will take S 1 = R/Z. Then, by the Legendrian<br />

neighbourhood theorem, there exists a δ0 > 0 and a diffeomorphism on<strong>to</strong> its<br />

image Ψ : N2δ0 −→ M so that Ψ ∗ λ = α0 and Ψ(L0) = L. We will let ξ0 = ker α0<br />

be the contact structure on Nδ.<br />

16


We will now work in our local model. We consider the boundary <strong>of</strong> Nδ,<br />

which we denote Tδ := ∂Nδ. This is now a convex <strong>to</strong>rus. Indeed, we observe<br />

that the vec<strong>to</strong>r field Z := x ∂ ∂ + y is a contact vec<strong>to</strong>r field transverse <strong>to</strong> Tδ.<br />

∂x ∂y<br />

A meridian m <strong>of</strong> this <strong>to</strong>rus is given by φ = const. We will take as longitude l<br />

one <strong>of</strong> the two curves that form the dividing set, l = {(0, 1, φ)}. A 1/q surgery<br />

takes the meridian m and maps it <strong>to</strong> m+ql. The exact behaviour <strong>of</strong> l under the<br />

mapping does not matter, since any two choices differ by a Dehn twist, which<br />

then extends <strong>to</strong> a diffeomorphism <strong>of</strong> the solid <strong>to</strong>rus.<br />

In order <strong>to</strong> construct a contact structure on the surgered manifold, we will<br />

need <strong>to</strong> study a tubular neighbourhood <strong>of</strong> Tδ. We let A := N2δ0 \ int Nδ0, a<br />

tubular neighbourhood <strong>of</strong> the convex surface Tδ. We now construct a map<br />

g : A → A so that Tδ0 is mapped <strong>to</strong> itself, as is T2δ0. We want the map g <strong>to</strong><br />

send m <strong>to</strong> m + ql. This will be our gluing map. We now consider the contact<br />

form on A given by α1 = g∗α0. We let ζ := ker α1 be the corresponding contact<br />

structure. We must now extend ζ <strong>to</strong> the solid <strong>to</strong>rus N2δ0.<br />

The goal now is <strong>to</strong> extend ζ <strong>to</strong> a smooth contact structure on the solid <strong>to</strong>rus<br />

N2δ0. To show that this is possible, we will use the result on convex surfaces<br />

we cited earlier. By our construction, we have a characteristic foliation and<br />

dividing set on the boundary <strong>of</strong> N2δ0. It now follows by Theorem 2.2.3 that<br />

there exists a unique contact structure on N2δ0. We note that the existence<br />

<strong>of</strong> such a filling may be shown by more direct, computational methods — the<br />

use <strong>of</strong> Theorem 2.2.3 will become much more useful in establishing uniqueness.<br />

(The computations, in a special case, will be carried out below.)<br />

Our surgered manifold will then be given by<br />

M ′ = M \ Ψ(Nδ0) ∪ N2δ0/ ∼<br />

where we identify the points w ∈ N2δ0 \ Nδ0 with the points Ψ(g(w)) ∈ M.<br />

Since Ψ : N2δ0 → M and g : N2δ0 \ Nδ0 → N2δ0 \ Nδ0 are contac<strong>to</strong>morphisms,<br />

the contact structure on M \ Ψ(Nδ0) ⊔ N2δ0 given by ξ ⊔ ζ descends <strong>to</strong> a smooth<br />

contact structure ξ ′ on M ′ .<br />

An important property <strong>of</strong> this construction is that up <strong>to</strong> contac<strong>to</strong>morphism,<br />

the resultant (M ′ , ξ ′ ) only depends on (M, ξ) and on L. In other words, the<br />

result is independent <strong>of</strong> the choices <strong>of</strong> Ψ, <strong>of</strong> g and <strong>of</strong> continuation <strong>of</strong> ζ. This is<br />

a result <strong>of</strong> Ding and Geiges [9, Proposition 7]. The pro<strong>of</strong> <strong>of</strong> this result hinges<br />

on the observation that up <strong>to</strong> contac<strong>to</strong>morphism, we may assume that the two<br />

surgeries are performed in the same Legendrian model neighbourhood. The<br />

result then follows as an application <strong>of</strong> Theorem 2.2.3.<br />

Ding and Geiges also establish that 1/q and −1/q Legendrian surgery are<br />

inverses <strong>of</strong> each other. Specifically, if we perform 1/q Legendrian surgery on M<br />

at L <strong>to</strong> obtain M ′ , L ′ , then, performing −1/q surgery on M ′ at L ′ gives us M<br />

and L again, up <strong>to</strong> contac<strong>to</strong>morphism [9, Proposition 8].<br />

17


To prove this result, by the uniqueness <strong>of</strong> the surgery, we need only consider<br />

the behaviour <strong>of</strong> the surgery in the local model. We then show that the boundary<br />

<strong>of</strong> local model surgered twice has the same dividing set as the the boundary <strong>of</strong><br />

the local model with no surgery. The result now follows by Theorem 2.2.3.<br />

The same idea can be pushed further <strong>to</strong> obtain the following result [9, Proposition<br />

9]:<br />

Proposition 2.2.5. If (M ′ , ξ ′ ) is obtained from (M, ξ) by 1/q surgery for q = 0,<br />

we may also obtain (M ′ , ξ ′ ) by a sequence <strong>of</strong> |q| applications <strong>of</strong> q/|q| surgeries.<br />

This proposition is essential in that it allows us <strong>to</strong> focus all <strong>of</strong> our attention<br />

on ±1 Legendrian surgery. This surgery, however, can be realized by a handleattaching<br />

construction due <strong>to</strong> Weinstein [37].<br />

±1 surgery, in coordinates<br />

We may carry out ±1 surgery directly, in coordinates, without appeal <strong>to</strong> the<br />

theory <strong>of</strong> dividing sets, if we do not concern ourselves with uniqueness.<br />

Indeed, we consider a Legendrian neighbourhood D 2 × S 1 with contact form<br />

given by :<br />

cos φ dx + sin φ dy.<br />

We delete the Legendrian {(0, 0)} × S 1 and introduce exponential polar coordinates<br />

by x = e s cos θ and y = e s sin θ. We then obtain a deleted neighbourhood<br />

diffeomorphic <strong>to</strong> R − × S 1 × S 1 with contact form<br />

α− = e s (cos(φ − θ) ds + sin(φ − θ) dθ) .<br />

We will now compactify R − × S 1 × S 1 with this contact structure ξ = ker α−<br />

<strong>to</strong> S 1 × D 2 , by gluing in a different Legendrian.<br />

Indeed, we introduce ψ = φ − θ and η = θ. Let now<br />

We then have<br />

u = 2e s/2 cos ψ v = 3e s/2 sin ψ.<br />

α+ := du + v dη = e s/2 α−.<br />

The new contact form α+ extends smoothly over the u = v = 0 circle.<br />

We now observe that this construction maps a transverse translate <strong>of</strong> the<br />

Legendrian {θ = φ} (parallel <strong>to</strong> the dividing set) <strong>to</strong> {ψ = 0, η = φ} and the<br />

meridian {φ = 0} <strong>to</strong> {η = −ψ}. On R − × S 1 × S 1 , we have orientations given<br />

by ds ∧ dφ ∧ dθ = ds ∧ dψ ∧ dη. Thus, we have performed −1 surgery.<br />

We have now performed −1 surgery explicitly from α− <strong>to</strong> α+. To perform<br />

+1 surgery, it suffices <strong>to</strong> reverse these steps.<br />

18


N +<br />

δ<br />

L +<br />

N −<br />

δ<br />

Figure 2.1: The local handle attaching model. The arrows represent the Liouville<br />

vec<strong>to</strong>r field, Y . The shaded region corresponds <strong>to</strong> Wδ (see Figure 2.2).<br />

L −<br />

L −<br />

N −<br />

δ<br />

L +<br />

2.2.2 Weinstein’s handle attaching construction<br />

Weinstein’s general construction takes a 2n−1 dimensional contact manifold M<br />

and attaches a k ≤ n handle <strong>to</strong> it, obtaining a manifold M ′ . The construction<br />

realizes the new manifold as a contact manifold, and furthermore provides a<br />

symplectic cobordism between the old manifold and the new one. More specifically,<br />

the construction realizes a Weinstein domain (W, ω, Y, F ) so that M is the<br />

negative boundary <strong>of</strong> W and M ′ is the positive boundary. Furthermore, in the<br />

case <strong>of</strong> an n handle, which we call a critical handle, there are Legendrian spheres<br />

L and L ′ in M and M ′ so that M \ nbd(L) is contac<strong>to</strong>morphic <strong>to</strong> M ′ \ nbd(L ′ ).<br />

Let (M, ξ) be a 2n − 1 dimensional contact manifold and let L be a Legendrian<br />

sphere in M. Let λ be a contact form with ξ = ker λ. By the Legendrian<br />

neighbourhood theorem, there is a neighbourhood <strong>of</strong> L which is contac<strong>to</strong>morphic<br />

<strong>to</strong> a standard model. Our construction will take place entirely within this<br />

neighbourhood. This is illustrated in Figure 2.1.<br />

As in the discussion <strong>of</strong> surgery, we let Nδ be the model neighbourhood <strong>of</strong> a<br />

Legendrian knot. We then have a contac<strong>to</strong>morphism Ψ : N2δ → U, where U is<br />

19<br />

N +<br />

δ


M −<br />

Figure 2.2: The global attaching construction. The shaded region is given by<br />

the local model in Figure 2.1. The unshaded part is given by I × (M \ Nδ).<br />

a neighbourhood <strong>of</strong> the Legendrian knot. We let U ′ be the image <strong>of</strong> Nδ.<br />

The idea <strong>of</strong> the construction is very straightforward, though managing all <strong>of</strong><br />

the details becomes cumbersome. We first split our contact manifolds M in<strong>to</strong><br />

two pieces : M = (M \U ′ )∪N2δ/ ∼, where ∼ is the identification provided by Ψ.<br />

We will now construct our symplectic manifold (W, ω) in two pieces. The first<br />

piece will be a trivial symplectic cobordism, W1 := [a, b]×(M \U ′ ), ω = d(etλ). The Liouville vec<strong>to</strong>r field here will be ∂ . This is organized in Figure 2.2.<br />

∂t<br />

The second piece, (W2, ω2), will be obtained by means <strong>of</strong> a local construction<br />

in R2n , which we will discuss next. The straightforward, but technical, part <strong>of</strong><br />

the construction is then <strong>to</strong> arrange a smooth identification <strong>of</strong> points near the<br />

boundaries <strong>of</strong> W1 and <strong>of</strong> W2. For the full details, we refer the reader <strong>to</strong> [37].<br />

We now present the local model in R2n . We consider R2n with the standard<br />

symplectic form ω0 = dxi ∧ dyi. Let H : R2n → R by (x, y) ↦→ |x| 2 − 1/2|y| 2 .<br />

Then, Y := ∇H = 2xi ∂ ∂ − yi is a Liouville vec<strong>to</strong>r field. We observe<br />

∂xi ∂yi<br />

that H has an index n critical point at 0 and no other critical points. Thus,<br />

H = c = 0 is a contact–type submanifold <strong>of</strong> R2n . We have then that the induced<br />

contact form is given by:<br />

λ = 2xidyi + yidxi<br />

restricted <strong>to</strong> the tangent space <strong>to</strong> H = c. We also observe that the intersection<br />

<strong>of</strong> any regular level set H = c = 0 with the stable or unstable manifold <strong>of</strong> 0<br />

for the flow <strong>of</strong> Y is a Legendrian sphere in the contact manifold H = c. Let us<br />

denote this Legendrian sphere by L c .<br />

20<br />

M +


We let Ψt be the flow <strong>of</strong> the vec<strong>to</strong>r field Y . We observe that if c, d > 0<br />

or c, d < 0, then Ψ induces a contac<strong>to</strong>morphism from the level H = c <strong>to</strong> the<br />

level H = d. Furthermore, if c > 0 and d < 0, we have that Ψ induces a<br />

contac<strong>to</strong>morphism from H −1 (d) \ L d <strong>to</strong> H −1 (c) \ L c .<br />

Let L − = {(x, y) | x = 0, |y| 2 = 2} be the intersection <strong>of</strong> the stable manifold<br />

<strong>of</strong> 0 for the vec<strong>to</strong>r field Y with the level set H = −1. This is a Legendrian sphere<br />

in H = −1. We will now consider a tubular neighbourhood <strong>of</strong> this Legendrian<br />

N −<br />

δ := {(x, y) | |x| 2 − 1/2|y| 2 = −1 and |x| ≤ δ}. Let L + = {(x, 0) | |x| 2 = 1}<br />

be the intersection <strong>of</strong> the unstable manifold <strong>of</strong> Y with H = 1.<br />

Then, there exists a neighbourhood N +<br />

δ <strong>of</strong> L+ so that Ψ induces a diffeomor-<br />

phism from N −<br />

δ \ L− → N +<br />

δ \ L+ . We note that N +<br />

δ = {(x, y) | |x|2 − 1/2|y| 2 =<br />

1 and |y| ≤ δ ′ }, where δ ′ = δ ′ (δ) is a mono<strong>to</strong>ne function <strong>of</strong> δ, with δ ′ (0) = 0.<br />

We now let Wδ be the region in R2n between N − +<br />

δ and N δ and bounded by the<br />

flow lines <strong>of</strong> Ψ through the boundary <strong>of</strong> N −<br />

δ . This now gives us our local model<br />

(Wδ, ω0, Y, H) that we glue in <strong>to</strong> the symplectization <strong>of</strong> M \ U.<br />

We refer the reader <strong>to</strong> [37] and <strong>to</strong> [8] for details in the gluing. Weinstein<br />

and Cieliebak, in these articles, also show that the result <strong>of</strong> the construction is<br />

independent <strong>of</strong> the choices made (up <strong>to</strong> symplec<strong>to</strong>morphism and contac<strong>to</strong>morphism).<br />

2.2.3 Attaching a critical handle performs −1 surgery<br />

We will now show how this handle attaching construction realizes −1 Legendrian<br />

surgery. To show this, we need only consider the local model. We also note that<br />

we may choose a slightly different model neighbourhood <strong>to</strong> replace the N −<br />

δ from<br />

above. Indeed, we will take instead :<br />

N −<br />

δ := {(x, y) ∈ R 2n | |x| ≤ δ and |y| = 1}.<br />

For δ small enough, this is transverse <strong>to</strong> the vec<strong>to</strong>r field<br />

Y = ∂ ∂<br />

2xi − yi .<br />

∂xi ∂yi<br />

We then have that N −<br />

δ is a contact submanifold <strong>of</strong> R 2n , with the contact form<br />

We also take :<br />

λ − = 2xi dyi + yi dxi.<br />

N +<br />

δ := {(x, y) ∈ R 2n | |x| = 1 and |y| ≤ δ}.<br />

We have that L − := {(0, y) | |y| = 1} and L + := {(x, 0) | |x| = 1}.<br />

21


We observe that N −<br />

δ \ L− is diffeomorphic <strong>to</strong> N + √ \ L<br />

δ + . We also have that<br />

N −<br />

δ and N + √ are contac<strong>to</strong>morphic <strong>to</strong> the model neighbourhoods we used in the<br />

δ<br />

previous section (hence our re-use <strong>of</strong> the same names).<br />

From these coordinates, we have a framing <strong>of</strong> the solid <strong>to</strong>rus N −<br />

δ<br />

that Z := x1 ∂<br />

∂x1<br />

. We have<br />

+ x2 ∂<br />

∂x2 is a contact vec<strong>to</strong>r field transverse <strong>to</strong> ∂N − . With<br />

respect <strong>to</strong> this framing, we have then that the dividing set has slope 1.<br />

On N + , we have a different framing. In these coordinates, the dividing set<br />

has slope −1. This then amounts <strong>to</strong> a −1 surgery along the knot L − . We can<br />

also see this since, after a performing a Dehn twist on each local model, we<br />

obtain the same local model as in our previous example, in which we worked<br />

out −1 surgery explicitly in coordinates.<br />

Computations<br />

It will be convenient for our later work <strong>to</strong> study the behaviour <strong>of</strong> the surgery in<br />

explicit coordinates. We have a diffeomorphism Ψ from R − ×S 1 ×S 1 → N − \L −<br />

by (s, θ, φ) ↦→ (e s cos(θ), e s sin(θ), cos(φ), sin(φ)). This pulls back the contact<br />

form λ − = 2xi dyi + yi dxi <strong>to</strong> the form<br />

α − = e s (cos(θ − φ) ds + 2 sin(θ − φ) dφ − sin(θ − φ) dθ).<br />

Remark 2.2.6. We observe this isn’t quite one <strong>of</strong> the standard forms on a Legendrian<br />

neighbourhood. However, it is merely a Dehn twist away from being<br />

<strong>of</strong> the form dx + y dφ. Indeed, this follows if we take x = e s cos(θ − φ) and<br />

y = e s sin(θ − φ).<br />

Now, the flow <strong>of</strong> Y induces a diffeomorphism from the image <strong>of</strong> Ψ <strong>to</strong> N + \L + .<br />

Thus, we obtain a diffeomorphism from R − × S 1 × S 1 <strong>to</strong>N + \ L + by<br />

(s, θ, φ) ↦→ (cos(θ), sin(θ), e s/2 cos(φ), e s/2 sin(φ)).<br />

This pulls λ + back <strong>to</strong> the contact form<br />

α + = e s/2 (cos(θ − φ) ds + 2 sin(θ − φ) dφ − sin(θ − φ) dθ).<br />

We also observe that the flow <strong>of</strong> Y induces a diffeomorphism from the image<br />

<strong>of</strong> Ψ <strong>to</strong> the critical level Msing := {(x, y) | |x| 2 − 1<br />

2 |y|2 = 0}. The induced contact<br />

form is<br />

αsing = ce 2s/3 (cos(θ − φ) ds + 2 sin(θ − φ) dφ − sin(θ − φ) dθ).<br />

22


At this point, we observe a fact that will be key in the following section. If<br />

we take the map<br />

x = 3<br />

2 e2s/3 cos(θ − φ)<br />

y = 1<br />

2 e2s/3 sin(θ − φ)<br />

ψ = θ + φ<br />

Then, we obtain that αsing is the pull back <strong>of</strong> dx+y dψ. As our map is a double<br />

covering, this shows that this singular level has a neighbourhood that double<br />

covers the standard Legendrian neighbourhood.<br />

This will turn out <strong>to</strong> be important for our work in Chapter 4. We will<br />

exploit this double cover <strong>to</strong> be able <strong>to</strong> study pseudoholomorphic curves in<strong>to</strong> the<br />

symplectization <strong>of</strong> the singular level in this Weinstein domain.<br />

23


Chapter 3<br />

An existence result for<br />

homoclinic orbits<br />

3.1 The main theorem and overview<br />

In this section, we prove the existence <strong>of</strong> a homoclinic orbit on a compact,<br />

contact–type critical level <strong>of</strong> a Hamil<strong>to</strong>nian, assuming that a neighbourhood <strong>of</strong><br />

this level is displaceable by a Hamil<strong>to</strong>nian with finite oscillation. Furthermore,<br />

we provide an upper bound for the length <strong>of</strong> this homoclinic orbit. This is then<br />

a generalization <strong>of</strong> a result <strong>of</strong> Séré’s, which gives this result in R2n (where the<br />

condition on displaceability is a consequence <strong>of</strong> compactness).<br />

First, we recall some important definitions.<br />

We recall that a symplectic manifold (W, ω) is said <strong>to</strong> be symplectically<br />

aspherical if ω|π2(M) = 0. This condition is useful in that it guarantees that all<br />

pseudoholomorphic spheres are constant.<br />

We also recall that a symplectic diffeomorphism Φ : W → W is said <strong>to</strong><br />

be a Hamil<strong>to</strong>nian diffeomorphism if it is the time 1 map <strong>of</strong> a time dependent<br />

Hamil<strong>to</strong>nian. To such a Φ, we may associate the H<strong>of</strong>er norm, denoted ||Φ||.<br />

This is given by taking the infimum <strong>of</strong> the oscillation over all Hamil<strong>to</strong>nians that<br />

generate Φ :<br />

||Φ|| = inf<br />

Ht generates Φ<br />

1<br />

0<br />

(sup Ht − inf Ht) dt<br />

W W<br />

Finally, we recall the definition <strong>of</strong> bounded geometry, essentially due <strong>to</strong><br />

Gromov [20] :<br />

Definition 3.1.1 (Bounded geometry). We say that the non-compact symplectic<br />

manifold (W, ω) is geometrically bounded if there exists an almost complex<br />

structure J such that ω(·, J·) is a Riemannian metric on W whose sectional<br />

24


curvature is bounded above and whose injectivity radius is bounded below.<br />

We also refer <strong>to</strong> such almost complex structures as having bounded geometry.<br />

The result we will prove is :<br />

Theorem 3.1.2. Let (W0, ω0, Y, F ) be a Weinstein domain <strong>of</strong> dimension 2n,<br />

so that F has a unique critical point x0 <strong>of</strong> index n and so that the corresponding<br />

Hamil<strong>to</strong>nian vec<strong>to</strong>r field XF has a hyperbolic zero at x0. Let us take F <strong>to</strong> be<br />

normalized by the conditions F (x0) = 0, and dF [Y ] ≤ 1<br />

Suppose that there exists S > 0 and a symplectic manifold (W, ω) <strong>of</strong> bounded<br />

geometry and symplectically aspherical so that (F −1 [−S, 0), ω0) symplectically<br />

embeds in (W, ω) and furthermore, the embedding, restricted <strong>to</strong> F −1 (−S/2), induces<br />

an injection from the fundamental group <strong>of</strong> F −1 (−S/2) <strong>to</strong> the fundamental<br />

group <strong>of</strong> W .<br />

Suppose furthermore that there exists a Hamil<strong>to</strong>nian diffeomorphism Φ <strong>of</strong> W ,<br />

with compact support and finite H<strong>of</strong>er norm ||Φ||, displacing the image <strong>of</strong><br />

F −1 [−S, 0).<br />

Then, our original Hamil<strong>to</strong>nian vec<strong>to</strong>r field XF has an orbit homoclinic <strong>to</strong> x0,<br />

<strong>of</strong> action bounded above by ||Φ||/(1 − e −S/2 ).<br />

Remark 3.1.3. The condition that dF [Y ] ≤ 1 is so that we may use a change in<br />

F <strong>to</strong> estimate the time it takes <strong>to</strong> flow from one point <strong>to</strong> another with the flow<br />

<strong>of</strong> Y . Indeed, if F (p) = a and F (q) = b > a are on the same flow line for Y ,<br />

then it will take at least time b − a <strong>to</strong> go from p <strong>to</strong> q. This condition normalizes<br />

the scaling <strong>of</strong> F .<br />

In a certain number <strong>of</strong> cases, we may immediately conclude that the hypotheses<br />

<strong>of</strong> Theorem 3.1.2 are verified. The first such case is in R 2n . In this<br />

case, every set with finite diameter in R 2n is Hamil<strong>to</strong>nian displaceable. This<br />

recovers a result due <strong>to</strong> Séré, [36, Theorem 1.1]. We recall that a hypersurface<br />

in a symplectic manifold is <strong>of</strong> restricted contact-type if there exists a global<br />

Liouville vec<strong>to</strong>r field transverse <strong>to</strong> the hypersurface.<br />

Corollary 3.1.4 (Séré). Suppose H : R 2n → R is a smooth Hamil<strong>to</strong>nian function,<br />

so that :<br />

• The level set H −1 (0) is compact<br />

• H has no critical points on H −1 \ {0}<br />

• dH(0) = 0, d 2 H(0) is non-degenerate and the linearization <strong>of</strong> XH at zero<br />

is hyperbolic.<br />

• the level set H −1 (0) \ {0} is <strong>of</strong> restricted contact–type.<br />

25


Then, there is a trajec<strong>to</strong>ry <strong>of</strong> XH homoclinic <strong>to</strong> 0.<br />

Pro<strong>of</strong>. The fact that M is <strong>of</strong> restricted contact-type gives us the existence <strong>of</strong> a<br />

global primitive <strong>of</strong> ω, which restricts <strong>to</strong> the contact form on M. This removes the<br />

need for the embedding <strong>of</strong> M in<strong>to</strong> R 2n <strong>to</strong> induce an injection on the fundamental<br />

groups.<br />

The next corollary uses a result <strong>of</strong> Biran and Cieliebak [5, Lemma 3.2].<br />

Suppose (W, ω, Y, F ) is a subcritical Weinstein filling. Denote by M = ∂W , its<br />

positive boundary (we recall a Weinstein filling has no negative boundary). We<br />

may now construct ( ˜ W , ˜ω, ˜ Y , ˜ F ), a (non-compact) subcritical Weinstein manifold<br />

for which the flow <strong>of</strong> Y exists for all time by gluing on the symplectization<br />

R×M <strong>to</strong> the boundary <strong>of</strong> W and extending Y, F and ω over this. This construction<br />

is called the completion <strong>of</strong> (W, ω, Y, F ). The result <strong>of</strong> Biran and Cieliebak<br />

may now be stated as follows : Suppose K is a compact set in ˜ W . Then, K is<br />

Hamil<strong>to</strong>nian displaceable in ˜ W .<br />

From this, we obtain :<br />

Corollary 3.1.5. Suppose (W, ω, Y, F ) is a Weinstein filling with a unique critical<br />

point <strong>of</strong> index n, which we label x0. We normalize F by the condition that<br />

F (x0) = 0. Then, the level F = 0 carries an orbit homoclinic <strong>to</strong> x0. (In other<br />

words, we require that F ≤ −δ is a subcritical Weinstein filling for all δ > 0.)<br />

Pro<strong>of</strong>. We then take W0 = F −1 (−ɛ, ɛ) for ɛ > 0 small. The result now follows<br />

from the fact that F −1 (−ɛ) is the positive boundary <strong>of</strong> a subcritical filling and<br />

is thus Hamil<strong>to</strong>nian displaceable in the completion <strong>of</strong> F −1 (−∞, −ɛ].<br />

Finally, the pro<strong>of</strong> <strong>of</strong> Theorem 3.1.2 leads us <strong>to</strong> a slight generalization :<br />

Corollary 3.1.6. Suppose (W0, ω0, Y, F ) and (W, ω) as in Theorem 3.1.2, but<br />

so that F has N index n critical points on the level F = 0, again satisfying the<br />

conditions <strong>of</strong> Theorem 3.1.2.<br />

Then the F = 0 level admits a homoclinic chain. Furthermore, the action <strong>of</strong><br />

the homoclinic chain is bounded by N times the bound in Theorem 3.1.2.<br />

We will present the pro<strong>of</strong> <strong>of</strong> this corollary after we prove the main result,<br />

since the result is actually a corollary <strong>of</strong> the pro<strong>of</strong> <strong>of</strong> Theorem 3.1.2.<br />

We recall our definition <strong>of</strong> a Weinstein domain from the previous chapter.<br />

A Weinstein domain (W, ω, Y, F ) consists <strong>of</strong> the following data :<br />

• A symplectic manifold (W, ω)<br />

26


• A vec<strong>to</strong>r–field Y on W with the property that LY ω = ω (i.e. Y is a<br />

Liouville vec<strong>to</strong>r field)<br />

• A Morse function F with the property that Y is gradient–like for F .<br />

We also take F = const on the boundary <strong>of</strong> W , and that the boundaries are<br />

regular levels <strong>of</strong> F . We call such a domain critical if at least one <strong>of</strong> the critical<br />

points has index n = 1 dim W , and subcritical if all <strong>of</strong> the critical points have<br />

2<br />

index k < n = 1 dim W .<br />

2<br />

We are particularly interested in the critical case. We suppose that there<br />

is only one such critical point, which we will call x0. We will normalize by<br />

assuming that F (x0) = 0. Furthermore, we assume that there are no other<br />

critical points with F ≥ −1.<br />

We consider the critical level M := F −1 (0) \ {x0}. This is a submanifold<br />

<strong>of</strong> W <strong>of</strong> contact–type, so it has a preferred contact form λ = iY ω. We wish<br />

<strong>to</strong> study the Reeb orbits which start and end on the singular point — we will<br />

call such orbits, homoclinics. Indeed, these correspond <strong>to</strong> homoclinic orbits <strong>of</strong><br />

any Hamil<strong>to</strong>nian system admitting M ∪ {x0} as a (critical) energy level (with<br />

unique critical point at x0) — for instance, F . We note that since Reeb chords<br />

are parametrized by action ( x∗λ), and since homoclinics will have finite action,<br />

these Reeb orbits only exist for finite time.<br />

The pro<strong>of</strong> <strong>of</strong> the result is very similar <strong>to</strong> the pro<strong>of</strong> <strong>of</strong> a result by Mohnke [34],<br />

in which he proves the existence <strong>of</strong> Legendrian chords, also sometimes called<br />

Reeb chords, and provides a similar bound on their length. In his case, we<br />

are given a Legendrian knot (or other, more complicated Legendrian, in higher<br />

dimensions) ℓ, and ask the question as <strong>to</strong> whether there exists a trajec<strong>to</strong>ry <strong>of</strong><br />

the Reeb vec<strong>to</strong>r field that intersects the Legendrian twice.<br />

As we discussed previously, a homoclinic orbit is, in some sense, a Legendrian<br />

chord for a singular Legendrian. As we will see below, in a suitable sense, the<br />

“unit” stable and unstable manifolds <strong>of</strong> an index n critical point are Legendrian<br />

submanifolds. A homoclinic orbit is then a Legendrian chord from the unit<br />

unstable manifold <strong>to</strong> the unit stable manifold. We will make this idea more<br />

precise in our pro<strong>of</strong> <strong>of</strong> the result.<br />

Mohnke’s result may be stated as follows:<br />

Theorem 3.1.7 (Mohnke). Let (M, ξ) be a contact manifold with contact form<br />

λ. Let L be a Legendrian submanifold <strong>of</strong> M.<br />

Suppose that there exists a symplectic manifold (W, ω), symplectically aspherical,<br />

and <strong>of</strong> bounded geometry (in the sense <strong>of</strong> Gromov) and S > 0 so that we<br />

may symplectically embed ([−S, 0] × M, d(e s λ)) ↩→ (W, ω) and furthermore, this<br />

embedding induces an injection on the fundamental groups.<br />

If there exists a Hamil<strong>to</strong>nian diffeomorphism Φ <strong>of</strong> W with compact support and<br />

27


finite H<strong>of</strong>er norm ||Φ|| displacing the image <strong>of</strong> [−S, 0] × M, then there exists a<br />

Legendrian chord <strong>of</strong> action bounded above by ||Φ||/(1 − e −S )<br />

The idea <strong>of</strong> the pro<strong>of</strong> is <strong>to</strong> construct an immersed Lagrangian submanifold<br />

<strong>of</strong> [−S, 0] × M, depending on a parameter T , so the Lagrangian is embedded if<br />

and only if no Reeb chord has action less than T . Furthermore, by exploiting<br />

the fact that this Lagrangian lies in a neighbourhood which is a piece <strong>of</strong> the<br />

symplectization <strong>of</strong> M, we can show that the area <strong>of</strong> any disk with boundary<br />

in our Lagrangian has symplectic area equal <strong>to</strong> an integer times a constant<br />

depending explicitly on T .<br />

If the Lagrangian is embedded, a result <strong>of</strong> Chekanov (Theorem 3.2.1, below),<br />

<strong>to</strong>gether with our assumption on the existence <strong>of</strong> our Hamil<strong>to</strong>nian diffeomorphism,<br />

gives the existence <strong>of</strong> a non–constant holomorphic disk (in W ) with<br />

boundary in the Lagrangian, <strong>of</strong> symplectic area no greater than ||Φ||. By the<br />

assumption on the induced injection on fundamental groups, this means there<br />

is a disk in [S, 0] × M with boundary on the Lagrangian. By the symplectically<br />

aspherical condition on W , this must have the same symplectic area as<br />

the holomorphic one. We then obtain an inequality giving an upper bound on<br />

T .<br />

Our result on homoclinic orbits will use a similar argument, except this<br />

time, the failure <strong>of</strong> our Lagrangian <strong>to</strong> be embedded will force the existence <strong>of</strong> a<br />

homoclinic orbit.<br />

3.2 Pro<strong>of</strong> <strong>of</strong> our result<br />

We will now present the pro<strong>of</strong> <strong>of</strong> theorem 3.1.2. The argument will come in<br />

three steps : the first step is <strong>to</strong> show that there is a suitable sense in which the<br />

unit unstable manifold <strong>of</strong> our critical point is a Legendrian. The second step<br />

will be <strong>to</strong> assume that we have no homoclinic <strong>of</strong> action less than or equal <strong>to</strong> T ,<br />

and use this assumption <strong>to</strong> construct an embedded Lagrangian. Finally, we use<br />

Legendrian surgery <strong>to</strong> arrange the problem so that we may apply the following<br />

theorem <strong>of</strong> Chekanov, proved in [7] :<br />

Theorem 3.2.1. Let (W, ω) be a symplectic manifold, either closed or <strong>of</strong><br />

bounded geometry (in the sense <strong>of</strong> Gromov). We denote by J the set <strong>of</strong> all<br />

almost complex structures J on W , tamed by ω and <strong>of</strong> bounded geometry. Let<br />

L be a closed Lagrangian submanifold <strong>of</strong> W . For each J ∈ J , let σ S 2(J) be the<br />

minimal symplectic area <strong>of</strong> a non-constant J-holomorphic sphere and σD(J) be<br />

the minimal symplectic area <strong>of</strong> a non-constant J-holomorphic disk with bound-<br />

ary in L. Let now<br />

σ = sup min{σS2(J), σD(J)}.<br />

J∈J<br />

28


Suppose φ : W → W is a Hamil<strong>to</strong>nian diffeomorphism whose H<strong>of</strong>er norm<br />

satisfies<br />

||φ|| < σ.<br />

If φ(L) intersects L transversely then<br />

#L ∩ φ(L) ≥ dim H∗(L, Z2).<br />

Pro<strong>of</strong> <strong>of</strong> Theorem 3.1.2. Our first step is the construction <strong>of</strong> the Lagrangian<br />

submanifold. We have assumed that XF has a hyperbolic rest point at x0.<br />

Hence, we have the existence <strong>of</strong> a global stable and unstable manifold. Let us<br />

denote the stable and unstable manifolds as W s and W u , respectively. Furthermore,<br />

we have that each <strong>of</strong> these two manifolds has dimension n. We consider<br />

now the linearized flow. In the linear case, the stable and unstable subspaces<br />

are isotropic subspaces. Hence, W u and W s are Lagrangian.<br />

In order <strong>to</strong> carry out the construction, we need <strong>to</strong> show that there is a a<br />

suitable notion <strong>of</strong> a “unit unstable manifold” that can be arranged <strong>to</strong> be a<br />

Legendrian sphere. In other words, we need the following lemma :<br />

Lemma 3.2.2. Suppose (W 2n , ω, Y, F ) is a Weinstein domain with critical point<br />

x0 on the level F = 0. Suppose XF (0) is a hyperbolic zero. Let W s and W u<br />

denote the stable and unstable manifolds for the flow <strong>of</strong> XF . Then, for any<br />

neighbourhood <strong>of</strong> x0, there exists a Legendrian embedding <strong>of</strong> S n−1 in<strong>to</strong> W u [<br />

W s ], everywhere transverse <strong>to</strong> XH, with image in the chosen neighbourhood.<br />

Furthermore, for any constant c sufficiently small, we may arrange the Legendrian<br />

so that for any path γ connecting x0 <strong>to</strong> any point on the Legendrian<br />

sphere, the path has action γ ∗ λ = c.<br />

Pro<strong>of</strong> <strong>of</strong> Lemma 3.2.2. By the Stable/Unstable Manifold Theorem, we have an<br />

immersion p : Rn → W u , with p(0) = x0. Since W u is Lagrangian, we have<br />

p∗ω = 0. We let λ := iY ω. Thus, d(p∗λ) = p∗ω = 0. Since we are working<br />

in Rn , we have that p∗λ = dG for some function G : Rn → R. The vec<strong>to</strong>r–<br />

field XF is tangent <strong>to</strong> W u , and so we may pull XF |W u back <strong>to</strong> Rn using p−1 .<br />

We will write ZF = p −1 ∗XF |W<br />

u. We then obtain that G is non-constant since<br />

dG[ZF ] = λ(XF ) = ω(Y, XF ) = dF [Y ] ≥ 0. We recall that dF [Y ] > 0 away<br />

from x0. Thus, G has no critical points away from 0, and level sets <strong>of</strong> G are<br />

transverse <strong>to</strong> ZF . We normalize G by the condition that G(0) = 0. Thus, we<br />

have G(x) ≥ 0 with G = 0 only at 0.<br />

We now claim that p(G −1 (const)) is a Legendrian sphere in the unstable<br />

manifold, transverse <strong>to</strong> XF .<br />

29


By the Stable/Unstable manifold theorem, there is a positive quadratic form<br />

N(x) on R n so that in a neighbourhood <strong>of</strong> 0, the norm is increasing along orbits<br />

<strong>of</strong> Z. We consider the sphere S := {N(x) = δ0} for δ0 sufficiently small. Let c<br />

be the minimum <strong>of</strong> G on this sphere. Then c > 0 and G −1 (c) is contained in<br />

{x |, N(x) ≤ δ0} and thus is compact. Furthermore, G −1 (c) is diffeomorphic <strong>to</strong><br />

S, where the diffeomorphism is given by following the flow lines <strong>of</strong> Z. We then<br />

see that G −1 (c) is a sphere for c > 0 sufficiently small. (Then, it follows that<br />

all level sets <strong>of</strong> G are spheres.)<br />

We now have that ℓ := p(G −1 (c)) is a Legendrian sphere in the unstable<br />

manifold, transverse <strong>to</strong> XF . The action <strong>of</strong> any path γ between x ∈ ℓ and x0 is<br />

then given by G(p −1 (x)) = c.<br />

Let us denote the unit unstable Legendrian given by Lemma 3.2.2 by ℓ. Let<br />

us choose a T1 > 0 for the action necessary <strong>to</strong> get <strong>to</strong> ℓ from x0.<br />

At this point, it is more convenient <strong>to</strong> work with the Reeb vec<strong>to</strong>r field than<br />

with the Hamil<strong>to</strong>nian vec<strong>to</strong>r field. In this case, the Reeb vec<strong>to</strong>r field is given<br />

by :<br />

X =<br />

1<br />

ω(Y, XH) XH.<br />

Observe that X blows up at x0, and so it is defined on a non–compact space,<br />

F −1 (0) \ {x0}.<br />

Let T > 0. Assume that the orbits <strong>of</strong> the Reeb vec<strong>to</strong>r field with initial data<br />

on ℓ exist for time T . Observe that this is implied by the stronger assumption<br />

that there are no homoclinic orbits (i.e. no orbit starting at ℓ intersecting the<br />

stable manifold). Let φt be the flow associated <strong>to</strong> the Reeb vec<strong>to</strong>r field, X.<br />

Thus, the image <strong>of</strong> [0, T ] × ℓ by (t, x) ↦→ φt(x) is compact in F −1 (0) \ {x0}.<br />

We may then take a small neighbourhood U ⊂ W0 <strong>of</strong> x0 whose closure is disjoint<br />

from the image <strong>of</strong> [0, T ] × ℓ.<br />

We will now construct a new contact manifold, N, realized as a hypersurface<br />

in W0, so that there is a neighbourhood U ′ so that N \ U ′ = F −1 (0) \ U. The<br />

essential idea is that we will see F −1 (0) as the critical level in a Legendrian<br />

surgery, where the surgery has support entirely in the neighbourhood U. Then,<br />

since the Reeb orbit through any initial data on ℓ up <strong>to</strong> time t avoids U, we may<br />

substitute N for F −1 (0) without any loss <strong>of</strong> relevant information. (See Figure<br />

3.1.)<br />

To construct N, we will consider a new Weinstein domain (W0, ω0, Y, ˜ F )<br />

obtained by perturbing F inside <strong>of</strong> U. We wish <strong>to</strong> take ˜ F : W0 → R satisfying<br />

the following properties :<br />

• ˜ F ≥ F on all <strong>of</strong> W0.<br />

• ˜ F = F on W0 \ U.<br />

30


• Y is gradient-like for ˜ F .<br />

• ˜ F has no critical points for ˜ F = 0.<br />

Then, if we consider the level set N := { ˜ F = 0}, this will lie entirely in<br />

F −1 [−δ, 0] \ U, where δ > 0 is small (and given by our construction). Furthermore,<br />

N will be a smooth contact manifold, with contact form given by<br />

iY ω0. The corresponding Reeb vec<strong>to</strong>r field X ′ will agree with X on N \ U. Let<br />

φ ′ t be the flow <strong>of</strong> X ′ . Thus, we have that the map<br />

[0, T ] × ℓ → N ⊂ W0<br />

(t, x) ↦→ φt(x)<br />

is an embedding whose image does not intersect the closure <strong>of</strong> U.<br />

An illustration <strong>of</strong> the perturbation ˜ F is given in figure 3.1. An example <strong>of</strong><br />

such a perturbation can be constructed as follows. Since x0 is a Morse critical<br />

point for F <strong>of</strong> index n, we may find a neighbourhood <strong>of</strong> x0 in W0 and coordinates<br />

ξ1, . . . , ξn and ζ1, . . . , ζn on this neighbourhood so that F = n<br />

i=1 ξ2 i − ζ 2 i . Let<br />

β : [0, 1] → [0, 1] so that β(r) = 1 for 0 ≤ r ≤ 1/4 and β(r) = 0 for 3/4 ≤ r ≤ 1,<br />

and −4 ≤ ∂<br />

∂r β(r) ≤ 0. Let δ0 > 0 be small enough so that these coordinates are<br />

defined on a ball <strong>of</strong> radius δ0 and furthermore, so that their image is entirely in<br />

the neighbourhood U <strong>of</strong> x0. We set<br />

P :=<br />

<br />

1<br />

β<br />

δ2 (ξ<br />

0<br />

2 1 + · · · + ξ 2 n + ζ 2 1 + · · · + ζ 2 <br />

n)<br />

˜F := F + ɛP,<br />

where ɛ > 0 will be chosen later in order <strong>to</strong> satisfy the properties we need.<br />

Then, ˜ F ≥ F , and ˜ F = F outside <strong>of</strong> U. Furthermore, we check that if ɛ < δ 2 0/4,<br />

we do not introduce any new critical points for ˜ F , and thus 0 is not a critical<br />

value <strong>of</strong> ˜ F . Finally, we need <strong>to</strong> check that Y is gradient-like for ˜ F . We observe<br />

that d ˜ F [Y ] = df[Y ] + ɛ dP [Y ]. We have that dP has support in a compact<br />

region given by<br />

1<br />

4 δ2 0 ≤ ξ 2 1 + · · · + ξ 2 n + ζ 2 1 + · · · + ζ 2 n ≤ 3<br />

4 δ2 0.<br />

In this region, df[Y ] is bounded away from zero. We may thus take ɛ sufficiently<br />

small so that d ˜ F [Y ] is bounded away from zero on this region. It then follows<br />

that Y is gradient like for ˜ F .<br />

We let N = ˜ F −1 (0). This is now a smooth, compact contact manifold. We<br />

may symplectically embed [−S/2, 0] × N in (W, ω) by fac<strong>to</strong>ring through the<br />

31


ℓ u<br />

x0<br />

ℓ u<br />

F = 0<br />

˜F = 0<br />

F = 1<br />

F = −1<br />

Figure 3.1: The level sets F = 0 and ˜ F = 0. The two level sets agree away<br />

from x0. The zero level <strong>of</strong> ˜ F and <strong>of</strong> F are represented in dotted and solid lines,<br />

respectively.<br />

Weinstein domain (W0, ω0). We note that since the image <strong>of</strong> F −1 [−S, 0] was<br />

displaced by Φ, we have that the image <strong>of</strong> [−S/2, 0] × N is also.<br />

We now have, by construction, a Legendrian ℓ in N with no Legendrian chord<br />

<strong>of</strong> action less than or equal <strong>to</strong> T . Indeed, as ℓ lies in the unstable manifold <strong>of</strong><br />

x0, any chord must run through the surgered region. By assumption, this does<br />

not happen.<br />

For the sake <strong>of</strong> completeness, we will now provide the construction <strong>of</strong> the<br />

Lagrangian, due <strong>to</strong> Mohnke. We could, however, conclude immediately by citing<br />

Theorem 3.1.7, and deriving a contradiction.<br />

Our step now is <strong>to</strong> construct the Lagrangian submanifold (see Figure 3.2).<br />

We first embed<br />

¯q : [0, T ] × [−S/2, 0] × ℓ ↩→ [−S/2, 0] × N<br />

(t, s, x) ↦→ (s, φt(x))<br />

where φt is the flow <strong>of</strong> the Reeb vec<strong>to</strong>r field. This is clearly an immersion. It<br />

is an embedding by our construction, which guarantees no chords and hence no<br />

self-intersection points. We also have that [−S/2, 0] × N ↩→ W0. Let q be the<br />

composition <strong>of</strong> these two embeddings, so q : [0, T ] × [−S/2, 0] × ℓ ↩→ W0.<br />

We now have q ∗ (e s λ) = e s dt and q ∗ ω0 = e s ds ∧ dt. Fix δ > 0. We find a<br />

circle C that approximates the boundary <strong>of</strong> [0, T ] × [−S/2, 0]. We may choose<br />

32


s<br />

this circle so that<br />

T (1 − e −S/2 <br />

) − δ ≤<br />

L<br />

Figure 3.2: The Lagrangian L.<br />

W u<br />

e<br />

C<br />

s dt ≤ T (1 − e −S/2 ).<br />

Let us call the action <strong>of</strong> this circle, C = <br />

C es dt.<br />

We now introduce our Lagrangian L = C × ℓ. We observe that any disk<br />

v : D → W0 with boundary in L has symplectic area <br />

D v∗ω = <br />

∂D v∗ (esλ). This is then given by kC, where k ∈ Z is the winding number <strong>of</strong> v|∂D around<br />

the non-trivial circle C.<br />

We now use the deep result <strong>of</strong> Chekanov, Theorem 3.2.1. Let J be an almost<br />

complex structure on W , tamed by ω and <strong>of</strong> bounded geometry. Then, since<br />

dim H∗(L, Z2) > 0 and Φ displaces L from itself, we have the existence <strong>of</strong> a nonconstant<br />

pseudoholomorphic disk u with boundary in L so that its symplectic<br />

area u∗ω ≤ ||Φ||, the H<strong>of</strong>er norm <strong>of</strong> Φ.<br />

By the fact that the embedding N induces an injection on fundamental<br />

groups, there is a disk u ′ lying entirely in the image <strong>of</strong> [−S/2, 0] × N with<br />

the same boundary as u. Furthermore, as W is symplectically aspherical, the<br />

symplectic areas u ′∗ω = u∗ω agree. By our construction <strong>of</strong> L, however, we<br />

have that u ′∗ω = kC for some integer k. The pseudoholomorphic curve u has<br />

positive symplectic area so k ≥ 1. Thus, we have the inequality<br />

<br />

C ≤ u ′∗ ω ≤ ||Φ||.<br />

Since δ > 0 was arbitrary, we have (1 − e −S/2 )T ≤ ||Φ||.<br />

Thus, we must have a homoclinic orbit <strong>of</strong> action no greater than T1 +<br />

||Φ||/(1 − e −S/2 ). As T1 was arbitrarily small, the result now follows.<br />

We will now prove the corollary 3.1.6. This is merely using the previous<br />

argument <strong>to</strong>gether with a pigeonhole principle <strong>to</strong> get a homoclinic chain.<br />

33


Pro<strong>of</strong> <strong>of</strong> Corollary 3.1.6. Let us label the critical points xi, for i = 1, . . . , N.<br />

For each one, we construct a unstable manifold Legendrian, as in Lemma 3.2.2.<br />

We will denote this by ℓi. Let B denote the bound from the theorem : B =<br />

||Φ||/(1 − e −S/2 ).<br />

We begin with some xi. The argument used in the pro<strong>of</strong> <strong>of</strong> Theorem 3.1.2<br />

allows us <strong>to</strong> conclude that there is either a homoclinic orbit <strong>of</strong> action no greater<br />

than B or that there is an orbit starting on ℓi, asymp<strong>to</strong>tic <strong>to</strong> some ℓj, <strong>of</strong> action<br />

no greater than B.<br />

We thus conclude that from each critical point, there is either a homoclinic<br />

orbit <strong>of</strong> action bounded by B or a heteroclinic orbit <strong>of</strong> action bounded by B. By<br />

the pigeonhole principle, it follows that there is at least one homoclinic chain,<br />

<strong>of</strong> length no greater than N. We then obtain that the action <strong>of</strong> the chain is<br />

bounded above by NB as claimed.<br />

3.3 <strong>Applications</strong> <strong>to</strong> the Weinstein conjecture<br />

Surprisingly, Mohnke’s result has some important consequences for the existence<br />

<strong>of</strong> periodic orbits. In particular, this provides an alternate pro<strong>of</strong> for many special<br />

cases <strong>of</strong> a result due <strong>to</strong> Liu and Tian in [31].<br />

Theorem 3.3.1. Let (M, ξ) be a contact manifold with contact form λ.<br />

Suppose that there exists a symplectic manifold (W, ω), symplectically aspherical,<br />

<strong>of</strong> bounded geometry (in the sense <strong>of</strong> Gromov) and S > 0 so that we may<br />

symplectically embed ι : ([−S, S] × M, d(e s λ)) ↩→ (W, ω) and furthermore, this<br />

embedding induces an injection on the fundamental groups.<br />

If there exists a Hamil<strong>to</strong>nian diffeomorphism Φ <strong>of</strong> W with compact support and<br />

finite H<strong>of</strong>er norm ||Φ|| displacing the image <strong>of</strong> [−S, S] × M, then there exists a<br />

periodic orbit <strong>of</strong> action bounded above by 4||Φ||/(e S − e −S ).<br />

This implies, in particular, a result that overlaps with Liu and Tian’s result<br />

about the “stabilized” Weinstein conjecture.<br />

Theorem 3.3.2 (Liu–Tian [31, Theorem 1.3]). Let (W, ω) be a symplectic manifold<br />

(either closed, or <strong>of</strong> bounded geometry). Then, there exists a non-negative<br />

integer l0 so that for any l ≥ l0, any compact, contact-type separating hypersurface<br />

M <strong>of</strong> (W ⊕ C l , ω ⊕ ω0) carries a closed Reeb orbit.<br />

This result follows as a corollary <strong>of</strong> Theorem 3.3.1 if the symplectic form on<br />

(W, ω) is exact. (This, in particular, requires that W be non-compact.) In this<br />

case, we may take l0 = 1.<br />

34


This result also follows as a corollary <strong>of</strong> Theorem 3.3.1 if W is symplectically<br />

aspherical and the embedding <strong>of</strong> M in<strong>to</strong> W ⊕ C l induces an injection on the<br />

fundamental groups. Again, in this case, we may take l0 = 1.<br />

We also obtain that there exists a symplectic structure on W ⊕ C l so that<br />

any contact-type hypersurface carries a closed Reeb orbit. Indeed, this follows<br />

by a result <strong>of</strong> Eliashberg :<br />

Theorem 3.3.3 (Theorem 1.3.1 in [12]). Suppose (X 2n , J) is an almost complex<br />

manifold, with n > 2. If X admits an exhausting Morse function F : X → R,<br />

with each critical point <strong>of</strong> index no greater than n, then there exists an integrable<br />

complex structure J ′ , homo<strong>to</strong>pic <strong>to</strong> J through almost complex structures, so that<br />

(X, J ′ , F ) is Stein.<br />

We recall that a Morse function F : X → R is said <strong>to</strong> be exhausting if F<br />

is proper and bounded from below. We say that (X, J ′ , F ) is a Stein domain<br />

if J ′ is an integrable complex structure, F : X → R is an exhausting Morse<br />

function and F is plurisubharmonic with respect <strong>to</strong> J ′ . From this, we obtain a<br />

symplectic form by ωF := − d( dF ◦ J ′ ).<br />

We put a Morse function H on W . Let N ≤ 2n be the maximal index <strong>of</strong> its<br />

critical points. Let k = 0 if N ≤ n, and k = N − n otherwise.<br />

We take l > k. Then there exists a Morse function ˆ H : W ⊕ C l → R whose<br />

critical points all have index strictly less than n+l. (We may take ˆ H = H+|z| 2 .)<br />

By Eliashberg’s theorem, it then follows that there exists a J ′ on W ′ for<br />

which F is plurisubharmonic. Let Ω = − d d C F be the corresponding symplectic<br />

form on W ′ . This gives us a subcritical Weinstein domain. This establishes the<br />

claim.<br />

Pro<strong>of</strong> <strong>of</strong> Theorem 3.3.1. The key point in the pro<strong>of</strong> <strong>of</strong> this result is <strong>to</strong> transform<br />

the periodic orbit problem <strong>to</strong> a Reeb chord problem. We will then use Theorem<br />

3.1.7 <strong>to</strong> conclude. We note that we have changed the piece <strong>of</strong> symplectization<br />

we use from [−S, 0] <strong>to</strong> [−S, S]. This is merely a notational convenience.<br />

In the following, we continue the notation from the previous sections, and<br />

assume that M and W satisfy the hypotheses <strong>of</strong> Theorem 3.3.1. Let Y = ι∗ ∂<br />

∂s .<br />

This is a (local) vec<strong>to</strong>r field on W , defined on the image <strong>of</strong> [−S, 0] × M. We<br />

note that this is a Liouville vec<strong>to</strong>r field.<br />

Given our symplectic manifold, W , we can construct a larger one W, which<br />

admits W as a Lagrangian submanifold. Let W = W × W with the symplectic<br />

form Ω = ω ⊕ −ω. Then, embedding W as the diagonal gives us a Lagrangian<br />

embedding. We note in general that the graph <strong>of</strong> any symplec<strong>to</strong>morphism<br />

W → W is a Lagrangian embedding <strong>of</strong> W in<strong>to</strong> W.<br />

We now have that ι ⊕ ι induces a symplectic embedding <strong>of</strong><br />

ι ⊕ ι : [−S, 0] × M × [−S, 0] × M ↩→ W.<br />

35


The symplectic form on the domain, [−S, 0] × M × [−S, 0] × M, is given by<br />

We now observe that Z := ∂<br />

∂s1<br />

Ω0 = d(e s1 λ) ⊕ − d(e s2 λ).<br />

+ ∂<br />

∂s2<br />

is a Liouville vec<strong>to</strong>r field for this<br />

symplectic form. This is transverse <strong>to</strong> the level sets <strong>of</strong> (s1, x, s2, y) ↦→ s1 + s2.<br />

Thus, we may find an embedding <strong>of</strong> M := [−S/2, S/2] × M × M in<strong>to</strong> R × M ×<br />

R × M as a contact-type hypersurface by :<br />

ϱ : M → R × M × R × M<br />

(s, x, y) ↦→ (s, x, −s, y)<br />

Thus, M becomes a contact manifold with contact form Λ := ϱ ∗ iZΩ0. We have<br />

and thus<br />

iZΩ0 = e s1 λ ⊕ −e s2 λ<br />

Λ = e s λ ⊕ −e s λ.<br />

We observe that M is not compact. This will not affect our argument, as we<br />

only need a “germ” <strong>of</strong> a submanifold near (0, x, 0, y).<br />

Let R be the Reeb vec<strong>to</strong>r field on M corresponding <strong>to</strong> Λ. We recall that<br />

X is the Reeb vec<strong>to</strong>r field on M corresponding <strong>to</strong> the original form λ. The<br />

associated Reeb vec<strong>to</strong>r field is then given by<br />

Rs,x,y = 1<br />

2 e−s Xx − 1<br />

2 es Xy.<br />

In particular, we observe that the submanifolds <strong>of</strong> s = const are invariant under<br />

the flow <strong>of</strong> R. In particular then, while M is not compact, only the compact<br />

subset {s0} × M × M is <strong>of</strong> relevance in studying the Reeb flow. We also note<br />

that for s = 0, R = X ⊕ −X.<br />

We now have a symplectic embedding <strong>of</strong> [−S/2, S/2] × M in<strong>to</strong> W by composing<br />

the various embeddings we have introduced so far. We first have<br />

[−S/2, S/2] × M → [−S, S] × M × [−S, S] × M<br />

(τ, s, x, y) ↦→ (τ + s, x, τ − s, y).<br />

Now, we also have that [−S, S] × M × [−S, S] × M symplectically embeds in W<br />

by using ι ⊕ ι. We have that the image <strong>of</strong> [−S/2, S/2] × M is contained in the<br />

image <strong>of</strong> [−S, S] × M × [−S, S] × M in W. This also allows us <strong>to</strong> conclude that<br />

the image <strong>of</strong> this embedding is Hamil<strong>to</strong>nian displaceable. Indeed, if Φ is our<br />

Hamil<strong>to</strong>nian diffeomorphism on W , then Φ⊕Φ is a Hamil<strong>to</strong>nian diffeomorphism<br />

36


on W. This new Hamil<strong>to</strong>nian diffeomorphism may be generated by Ht + Ht,<br />

where Ht generates Φ. Thus its H<strong>of</strong>er norm, ||Φ ⊕ Φ|| ≤ 2||Φ||.<br />

We also have a Legendrian embedding <strong>of</strong> M in<strong>to</strong> M by x ↦→ (0, x, x). We<br />

are interested in Legendrian chords for this Legendrian. Indeed, suppose γ :<br />

[0, T ] → M is such a Legendrian chord, <strong>of</strong> action T . Then, writing γ(t) =<br />

(0, x(t), y(t)),we have ˙x(t) = X and ˙y(t) = −X. Furthermore, since the orbit<br />

starts and ends on M, γ(0) = (0, x0, x0) and γ(T ) = (0, y0, y0). From this, we<br />

obtain that x(t) is an orbit <strong>of</strong> X that starts at x0 and ends at y0 and that<br />

y(T − t) is an orbit starting at y0 and ending at x0. It follows then that the<br />

orbit through x0 is a periodic orbit <strong>of</strong> action 2T .<br />

While the contact–type hypersurface we construct is not compact, the Lagrangian<br />

we construct as in the pro<strong>of</strong> <strong>of</strong> Theorem 3.1.2 lies in a compact region<br />

(independent <strong>of</strong> T ). We also have that the Lagrangian will always lie in a Hamil<strong>to</strong>nian<br />

displaceable region. We now apply the result <strong>of</strong> Mohnke, Theorem 3.1.7,<br />

in order <strong>to</strong> conclude the existence <strong>of</strong> periodic orbits <strong>of</strong> action no greater than<br />

4||Φ||/(e S − e −S )<br />

37


Chapter 4<br />

Pseudoholomorphic curves in<br />

the singular level<br />

4.1 Overview<br />

In this chapter, we will study properties <strong>of</strong> pseudoholomorphic curves in<strong>to</strong> the<br />

symplectization <strong>of</strong> a critical level in a four dimensional Weinstein domain.<br />

We consider (W 4 , ω, Y, F ), where (W, ω) is a symplectic manifold (with<br />

boundary), Y is a Liouville vec<strong>to</strong>r field and F is a Morse function F : W → R.<br />

We have that Y is gradient-like for F . We also assume that F has a unique<br />

critical point x0 <strong>of</strong> index 2. We normalize F by F (x0) = 0.<br />

In this chapter, we will first study the critical level in the Weinstein domain,<br />

F −1 (0). After deleting the critical point x0, the critical level becomes a<br />

non-compact contact manifold. The contact form coming from the Weinstein<br />

domain, ıY ω, becomes singular at this critical point, as does the associated<br />

Reeb vec<strong>to</strong>r field. Instead <strong>of</strong> working in this setting, we introduce a compactification<br />

<strong>of</strong> the critical level by gluing in a Legendrian knot. The contact structure<br />

will extend smoothly across the compactification, but the contact form will not<br />

extend as a contact form.<br />

With certain technical assumptions on the Morse function F in the Weinstein<br />

domain, we will have a neighbourhood theorem for the singularity. We<br />

obtain that a deleted neighbourhood <strong>of</strong> the singularity double-covers a deleted<br />

neighbourhood <strong>of</strong> a standard Legendrian knot. We will then define a suitable<br />

class <strong>of</strong> (singular) almost complex structures, compatible with this double cover.<br />

From this class <strong>of</strong> almost complex structures, we are led <strong>to</strong> study two different<br />

Cauchy-Riemann type first order elliptic PDE problems. We will call these<br />

type (I) pseudoholomorphic curves and type (II) pseudoholomorphic curves. We<br />

will show that under certain conditions, type (I) curves are actually instances<br />

38


<strong>of</strong> type (II) curves. Furthermore, we will show that type (II) pseudoholomorphic<br />

curves satisfy many nice properties relevant <strong>to</strong> the study <strong>of</strong> Hamil<strong>to</strong>nian<br />

dynamics. Specifically, a type (II) finite energy half-plane is asymp<strong>to</strong>tic <strong>to</strong> a<br />

homoclinic orbit (Theorem 4.4.4). We will also prove that if the asymp<strong>to</strong>tic<br />

limit is a transverse homoclinic orbit, then the convergence is at an exponential<br />

rate (Theorem 4.4.12).<br />

This result, both in its statement and in its method <strong>of</strong> pro<strong>of</strong>, is quite similar<br />

<strong>to</strong> analogous facts about punctured pseudoholomorphic curves in<strong>to</strong> symplectizations<br />

<strong>of</strong> smooth contact manifolds (as in [21, 2, 6]).<br />

In a final section, we discuss the directions <strong>of</strong> future work indicated by the<br />

results <strong>of</strong> this chapter. Of particular interest is the development <strong>of</strong> a contact<br />

homology–like theory generated by homoclinic orbits. Then, the differential<br />

in this theory would be given by counting pseudoholomorphic curves <strong>of</strong> type<br />

(II). This introduces a very interesting structure on the Floer complex : just<br />

as transverse homoclinic orbits may be glued, so <strong>to</strong>o pseudoholomorphic halfplanes<br />

may be glued. This will introduce a new operation on the Floer complex,<br />

unlike any previously studied. Additionally, a good understanding <strong>of</strong> the relationship<br />

between type (I) and type (II) curves is needed. This would involve a<br />

generalization <strong>of</strong> Gromov’s removal <strong>of</strong> singularities theorem. Finally, a complete<br />

description <strong>of</strong> the compactification <strong>of</strong> these spaces is needed (analogous <strong>to</strong> the<br />

description in [6]).<br />

Ultimately, we wish <strong>to</strong> understand the relationship between homoclinic orbits<br />

on the singular level and periodic orbits on nearby regular levels, and thus<br />

the relationship between the curves we study here and the ones in<strong>to</strong> the symplectizations<br />

<strong>of</strong> smooth contact manifolds.<br />

4.2 A compactification <strong>of</strong> the singular level<br />

4.2.1 Construction <strong>of</strong> the compactification<br />

We recall that we are working in a Weinstein domain, (W 4 , ω, Y, F ), with positive<br />

and negative boundary ∂W = M + ⊔ M − . We assume that F has a unique<br />

critical point x0, with F (x0) = 0, and <strong>of</strong> Morse index 2. We are interested in<br />

orbits <strong>of</strong> the Hamil<strong>to</strong>nian vec<strong>to</strong>r field XF homoclinic <strong>to</strong> x0. In order <strong>to</strong> study<br />

these, we will study orbits <strong>of</strong> the Reeb vec<strong>to</strong>r field X = 1<br />

dF [Y ] XF . The orbits <strong>of</strong><br />

the two vec<strong>to</strong>r fields will be the same up <strong>to</strong> a time re-parametrization.<br />

We recall, that by definition, a homoclinic orbit γ(t) : R → F −1 (0) ⊂ W is a<br />

trajec<strong>to</strong>ry for XF that satisfies limt→±∞ γ(t) = x0. However, we note that this<br />

same orbit, when re-parametrized so that γ(τ) is an orbit for X, has existence<br />

only on a finite τ interval. Indeed, X is singular at x0 (it has a singularity <strong>of</strong><br />

39


order 1/r). Thus, for us, a homoclinic orbit is an orbit that exists for a finite<br />

time interval. This is <strong>to</strong> be expected, since the Reeb vec<strong>to</strong>r field parametrizes<br />

its orbits by action, x ∗ λ, and the action <strong>of</strong> a homoclinic orbit is finite.<br />

In our Weinstein domain, (W 4 , ω, Y, F ), we recall that F has a unique critical<br />

point, x0 ∈ W , with F (x0) = 0. We suppose that this critical point has Morse<br />

index 2 and that x0 is a hyperbolic zero <strong>of</strong> Y . We further normalize F by<br />

∂ − W = F −1 (−1) =: M − and ∂ + W = F −1 (1) =: M + . These boundaries, M ± ,<br />

are smooth contact manifolds with contact forms<br />

λ ± 0 := iY ω|M ±.<br />

We are interested in studying the qualitative behaviour <strong>of</strong> the Hamil<strong>to</strong>nian<br />

vec<strong>to</strong>r field XF on the critical level, F −1 (0). We observe that<br />

M sing := F −1 (0) \ {x0}<br />

is a smooth, non-compact, contact manifold with contact form<br />

λ sing = iY ω| M sing.<br />

We recall a key observation from Chapter 2. The flow <strong>of</strong> Y defines a contac<strong>to</strong>morphism<br />

between level sets <strong>of</strong> F . Indeed, if c, c ′ are both positive or both<br />

negative, then by following the flow lines <strong>of</strong> Y , one obtains a contac<strong>to</strong>morphism<br />

from F −1 (c) <strong>to</strong> F −1 (c ′ ). Let W u and W s be the unstable and stable manifolds<br />

<strong>of</strong> x0 with respect <strong>to</strong> the flow <strong>of</strong> Y . Let ψt be the flow <strong>of</strong> Y . Then, as discussed<br />

in Chapter 2, W u and W s are Lagrangian submanifolds <strong>of</strong> W . They intersect<br />

regular levels F −1 (c), c = 0, in Legendrian submanifolds, which we denote by<br />

Lc . We observe now that for any c and c ′ , F −1 (c) \ Lc and F −1 (c ′ ) \ Lc′ are<br />

diffeomorphic by means <strong>of</strong> following the flow lines <strong>of</strong> Y . If we denote this diffeomorphism<br />

by ψ, we have ψ∗λ = et(p) λ, where t(p) is the time it takes <strong>to</strong> flow from<br />

p ∈ F −1 (c) <strong>to</strong> the level F −1 (c ′ ). In particular then, ψ is a contac<strong>to</strong>morphism.<br />

We use this construction <strong>to</strong> obtain a contac<strong>to</strong>morphism between the smooth<br />

part <strong>of</strong> the singular level M sing = F −1 (0) \ {x0} and M ± \ L ±1 . We let t :<br />

(M ± \ L ± ) → R associate <strong>to</strong> a point p, the time it takes <strong>to</strong> flow from p <strong>to</strong> the<br />

level F = 0. The diffeomorphism induced by the flow <strong>of</strong> Y pulls the contact<br />

form on the singular level λsing back <strong>to</strong> the contact forms<br />

We let<br />

λ ± = e t(p) λ ± 0 on M ± \ L ± .<br />

f + := e t(p) |M + and f − = e t(p) |M −.<br />

This construction allows us identify M sing with an open subset <strong>of</strong> a smooth,<br />

closed contact manifold. Indeed, the contac<strong>to</strong>morphism induced by the flow <strong>of</strong><br />

40


Y identifies M sing with M + \ L +1 or with M − \ L −1 . We may then compactify<br />

M sing by gluing in the missing Legendrian knot, L ±1 . We then have that M ±<br />

are two possible compactifications <strong>of</strong> M sing . The price we pay is that we no<br />

longer have a contact form defined on all <strong>of</strong> M ± . Instead, we obtain a contact<br />

form that is singular along L ±1 .<br />

We observe that the diffeomorphism induced by the flow <strong>of</strong> Y induces a singular<br />

contact form on M ± , but the singularity is <strong>of</strong> a different nature depending<br />

on whether we consider M + or M − . Indeed, we have that f + extends smoothly<br />

<strong>to</strong> be 0 along the singular Legendrian L + . However, f − becomes unbounded<br />

as we approach the singular Legendrian. Instead, 1/f − extends smoothly <strong>to</strong> be<br />

zero along the singular Legendrian L − . In the following analysis, we will need<br />

a notion <strong>of</strong> bounds on gradients. It will be much more convenient <strong>to</strong> work with<br />

the compactification <strong>to</strong> (M + , ξ) with singular contact form λ + = f + λ + 0 . We will<br />

point this out explicitly when we discuss gradient bounds (the work leading <strong>to</strong><br />

Proposition 4.4.6).<br />

In the following, we will then consider a smooth contact manifold (M, ξ)<br />

with smooth contact form λ0. We have a Legendrian knot L ∈ M and a smooth<br />

function G : M → [0, ∞) so that G = 0 on L, and G > 0 on M \L. We consider<br />

then the singular contact form λ = Gλ0. This will be the model we use for all<br />

<strong>of</strong> the subsequent work.<br />

4.2.2 A local model <strong>of</strong> the singularity<br />

For our analysis, we will need <strong>to</strong> have a good model <strong>of</strong> the behaviour <strong>of</strong> λ = Gλ0<br />

near the Legendrian. In the process, we will place a symmetry condition on the<br />

functions G we allow — this in turn, will represent a condition on the class <strong>of</strong><br />

Weinstein domains (W, ω, Y, F ) we consider. For the purpose <strong>of</strong> understanding<br />

Legendrian surgery, this is not a difficulty, since the standard construction from<br />

Weinstein (as discussed in Section 2.2.2) satisfies our symmetry condition. In<br />

general, we are also interested in studying homoclinic orbits on the critical<br />

level in a given (W, ω, Y, F ). We will show that in the case <strong>of</strong> a strictly nondegenerate<br />

pair (Y, F ), we may make an arbitrarily small perturbation <strong>of</strong> F<br />

(small in the C 1 <strong>to</strong>pology), so that the resulting Fɛ satisfies our symmetry<br />

condition, and so F = Fɛ outside a small neighbourhood <strong>of</strong> the critical point.<br />

Thus, if we can conclude that a homoclinic orbit exists by virtue <strong>of</strong> the existence<br />

<strong>of</strong> a pseudoholomorphic curve, then we may use a limiting argument <strong>to</strong> establish<br />

existence on the critical level for F .<br />

We recall that the pair (Y, F ) is non-degenerate if F is a Morse function, Y is<br />

gradient-like with respect <strong>to</strong> F and furthermore, the zeros <strong>of</strong> Y are hyperbolic<br />

and the zeros <strong>of</strong> XF are hyperbolic. We say the pair (Y, F ) is strictly nondegenerate<br />

if, in a neighbourhood <strong>of</strong> each critical point p <strong>of</strong> F , there exists a<br />

41


Darboux chart on (W, ω) and a constant so that Y [F ] ≥ c(|x| 2 + |y| 2 ).<br />

We will now show that if G arises as an f + for a strictly non-degenerate<br />

Liouville pair, then it is very close <strong>to</strong> being <strong>of</strong> the form described by the standard<br />

model in Section 2.2.2.<br />

We recall that, as we computed in Section 2.2.2, if we introduce exponential<br />

polar coordinates <strong>to</strong> map R − × S 1 × S 1 → Nbd(L), we may write<br />

α − = e s (cos θ ds − sin θ dθ + sin θ dφ)<br />

α + = e s/2 (cos θ ds − sin θ dθ + sin θ dφ)<br />

αsing = e 2s/3 (cos θ ds − sin θ dθ + sin θ dφ)<br />

In particular, in these coordinates, G(s, φ, θ) = G(s) = e s/6 .<br />

We will denote by Nδ ⊂ R 2 × S 1 the standard Legendrian neighbourhood,<br />

and αstd = dx+y dφ the standard contact form. Let (s, θ, φ) be the exponential<br />

polar coordinates on Nδ \ Lstd by x = e s cos θ, y = e s sin θ. (Thus, αstd =<br />

e s (cos θ ds + sin θ dφ − sin θ dθ).)<br />

Let us now consider the more general situation <strong>of</strong> an arbitrary pair (Y, F ),<br />

where the critical points are strictly non-degenerate. Then, by definition, we<br />

may take a Darboux chart in which dF [Y ] ≥ c(|x| 2 + |y| 2 ). Thus, if we write<br />

Y = Y0 + Y1, where Y0 is the linear part <strong>of</strong> Y , and Y1 is higher order, we obtain<br />

that Y0 is also Liouville (and thus, Y1 is a Hamil<strong>to</strong>nian vec<strong>to</strong>r field). Let us<br />

write F = F0 + F1, where F0 is quadratic and F1 is higher order. We then have<br />

that Y0 is locally gradient-like for both F and F0. We may thus assume that in<br />

our local coordinates, Y = Y0.<br />

Lemma 4.2.1. Suppose Y0 is a linear Liouville vec<strong>to</strong>r field on (R 2n , ω0). If<br />

Y0(0) = 0 and has n expanding directions and n contracting directions, then<br />

there exist positive constants b1, b2, . . . , bn so that :<br />

n ∂ ∂<br />

Y (x, y) = (1 + bi)xi − biyi<br />

∂xi ∂yi<br />

i=1<br />

<br />

n<br />

= ∇ (1 + bi) 1<br />

<br />

1<br />

.<br />

i=1<br />

2 x2i − bi<br />

2 y2 i<br />

(Here, ∇ denotes the gradient with respect <strong>to</strong> the standard metric on R 2n .)<br />

Pro<strong>of</strong>. By assumption Y0(0) = 0 and the zero is hyperbolic, with n contracting<br />

directions and n expanding directions. Its stable and unstable manifolds are<br />

then Lagrangian submanifolds <strong>of</strong> R 2n , intersecting transversally at 0.<br />

42


We may now find a symplec<strong>to</strong>morphism on R2n , fixing 0, so that the stable<br />

manifold is mapped <strong>to</strong> Rn and the unstable manifold <strong>to</strong> J0Rn .<br />

Writing<br />

n n<br />

Y0 = a j<br />

i xi<br />

∂<br />

− b<br />

∂xj<br />

j<br />

i yi<br />

∂<br />

∂yj<br />

j=1<br />

i=1<br />

and using the fact that d(iY0ω0) = LY0ω0 = ω0, we compute : a i j = 0 and b i j for<br />

i = j, a i i − b i i = 1. In order <strong>to</strong> satisfy the requirement that Y0 have the correct<br />

index, we also have that b i i > 0. The result now follows.<br />

Write z = (x, y) ∈ R 2n . Let B be the matrix so that Y (z) = Bz. Let<br />

H(z) = |x| 2 − 1/2|y| 2 . We write F (z) = F0(z) + F1(z), where F0(z) =< Az, z ><br />

for some self-adjoint matrix A, and F1(z) = O(|z| 3 ). By assumption, AB is<br />

positive. Let | · | be the norm < ABz, z > 1/2 . This norm is equivalent <strong>to</strong> the<br />

standard norm on R 2n .<br />

Let α = ıY ω|{H=0}. Let λ = ıY ω|{F =0}. Let λ ′ = ıY ω|{F0=0}. By following<br />

the orbits <strong>of</strong> Y , we have that {H = 0} \ {0} is diffeomorphic <strong>to</strong> {F0 = 0} \ {0}<br />

and <strong>to</strong> {F = 0} \ {0}. We wish <strong>to</strong> estimate the amount <strong>of</strong> time it takes <strong>to</strong> flow<br />

from {H = 0} \ {0} <strong>to</strong> {F = 0} \ {0}. To do this, we will first estimate the time<br />

it takes <strong>to</strong> flow from {H = 0} \ {0} <strong>to</strong> {F0 = 0} \ {0}. Then, we will estimate<br />

the time it takes <strong>to</strong> flow from {F0 = 0} \ {0} <strong>to</strong> {F = 0} \ {0}.<br />

We first notice that {H = 0} \ {0} and {F0 = 0} \ {0} are cones. Suppose<br />

p is a point so that H(p) = 0 and S(p) is the time so that F0(ψS(p)(p)) = 0.<br />

Then, since the flow <strong>of</strong> Y is linear, we have for any r = 0, F0(ψS(p)(rp)) = 0.<br />

Thus, S(p) = S(p/||p||).<br />

Now, we consider the time it takes <strong>to</strong> flow from {F0 = 0} \ {0} <strong>to</strong> {F =<br />

0} \ {0}. We denote this time by T (x) for points x with F0(x) = 0. Let x0 so<br />

that F0(x0) = 0. Let x1 = ψT (x0)(x0) be the point on {F = 0}, and x(t) the<br />

orbit between them. Let y(t) = µx(t) for some µ > 0. We then obtain:<br />

F (y(t)) − F (y(0)) = F (y(t)) − F1(y(0))<br />

=<br />

t<br />

= µ 2<br />

dF [Y ]|µx(t) dt<br />

dF0[Y ]|x(t) + 1<br />

µ dF1(µx(t))[Y<br />

<br />

(x(t))] dt .<br />

0<br />

t<br />

0<br />

We now observe that by construction, | dF1(z)| ≤ C|z| 2 , so we obtain :<br />

µ 2<br />

t<br />

dF [Y ]|x(t) − µ<br />

0<br />

3 C ′<br />

t<br />

0<br />

|x(t)| 3 ≤ F (y(t)) − F1(y(0))<br />

43<br />

≤ µ 2<br />

t<br />

dF [Y ]|x(t) + µ<br />

0<br />

3 C ′<br />

t<br />

0<br />

|x(t)| 3


With tµ = T (µx0), we obtain :<br />

µ 2<br />

tµ<br />

0<br />

|x| 2 − µ 3 C ′ |x(t)| 3<br />

<br />

≤ −F1(µx0) ≤ µ 2<br />

tµ<br />

0<br />

|x| 2 + µ 3 C ′ |x(t)| 3<br />

<br />

.<br />

We restrict our attention <strong>to</strong> a sufficiently small neighbourhood <strong>of</strong> 0 so that<br />

C|x| 3 ≤ 1/2|x| 2 . We also consider only those µ ∈ (0, 1). We recall that<br />

|F1(µx0)| ≤ Cµ 3 |x0| 3 .<br />

Then, we obtain :<br />

so we have<br />

thus,<br />

µ 2 tµ<br />

| |x|<br />

0<br />

2 | dt ≤ C ′′ µ 3 |x0| 3<br />

|tµ| inf |x(t)|<br />

0≤t≤T (x0)<br />

2 ≤ C ′′ µ|x0| 2<br />

|tµ| ≤ C(x0)µ.<br />

Thus, if we consider x = e s v, where ||v|| = 1, then we have<br />

T (x) ≤ C(v)e s .<br />

Thus, it takes time τ(p) = S(p) + T (ψS(p)(p)) <strong>to</strong> flow from a point p in<br />

{H = 0} <strong>to</strong> the corresponding point q ∈ {F = 0}. This satisfies the decay<br />

estimate (for s < 0):<br />

|τ(e s p) − S(p)| ≤ Ce s ,<br />

This gives us the following result:<br />

Lemma 4.2.2. Suppose that (Y, F ) are a strictly non-degenerate Liouville pair,<br />

defined in a neighbourhood <strong>of</strong> 0 ∈ R 4 . Suppose furthermore, that Y is linear.<br />

Then, with the local construction and coordinates as in Section 2.2.2, we have<br />

the following estimate :<br />

f + = e s/6+E(s,φ,θ)+B(φ,θ)<br />

where |E(s, φ, θ)| ≤ Ce s for −s sufficiently large.<br />

44


(We recall that the coordinates (s, θ, φ) come from taking exponential polar<br />

coordinates in a deleted tubular neighbourhood <strong>of</strong> the model Legendrian in the<br />

Legendrian neighbourhood theorem.)<br />

From this, it also becomes clear that we can make a C 1 small perturbation <strong>of</strong><br />

F that is supported in an arbitrarily small neighbourhood <strong>of</strong> L, so that G = f +<br />

becomes only a function <strong>of</strong> s. To do this, we can make use <strong>of</strong> an averaging<br />

construction.<br />

We are now able <strong>to</strong> prove our most important result on the local structure<br />

<strong>of</strong> the singularity.<br />

Proposition 4.2.3. Let β := cos θ ds+sin θ dψ. This is a contact form on R − ×<br />

S 1 × S 1 . Suppose G : R − × S 1 × S 1 → (0, ∞) with G(s, θ, ψ) = e 2s/3 H(s, θ, ψ),<br />

with H(s, θ, ψ) = H(s, θ, ψ + π) and c ≤ H ≤ 1/c for some constant c > 0.<br />

Suppose also that lims→−∞ H(s, θ, ψ) exists for all θ, and ψ, and is independent<br />

<strong>of</strong> ψ and θ.<br />

Then, there exists a double covering local diffeomorphism<br />

so that there exists a function<br />

Ψ : R − × S 1 × S 1 → (D 2 \ {(0, 0)}) × S 1<br />

F : D 2 × S 1 → (0, ∞) with<br />

Ψ ∗ F ( dx + y dη) = Gβ.<br />

The smoothness <strong>of</strong> F depends on how fast H decays <strong>to</strong> its limit as s → −∞.<br />

Pro<strong>of</strong>. We observe that the following map :<br />

Ψ : (s, θ, ψ) ↦→ (e 2s/3 cos θ, e 2s/3 sin θ, 3θ + 2ψ)<br />

satisfies Ψ ∗ ( dx + y dη) = e 2s/3 β.<br />

We notice that this is a double covering map. In particular, (s, θ, ψ) and<br />

(s, θ, ψ + π) are mapped <strong>to</strong> the same point. By assumption, G(s, θ, ψ) =<br />

G(s, θ, ψ+π) and so it pushes forward <strong>to</strong> a function F on (D 2 \{(0, 0)})×S 1 . Our<br />

conditions on H now guarantee that this function F extends across {(0, 0)}×S 1 .<br />

H must have an exponential decay rate for F <strong>to</strong> become differentiable at 0 (the<br />

higher the decay rate, the more derivatives F has).<br />

This gives us that in our situation, if G is sufficiently symmetric, then a<br />

neighbourhood <strong>of</strong> the singular Legendrian double covers a neighbourhood <strong>of</strong><br />

45


ℓ s<br />

ℓ u<br />

ℓu (a) Slice <strong>of</strong> the Legendrian neighbourhood in<br />

the singular case, given by φ =const.<br />

ℓ s<br />

Ψ(ℓ s )<br />

Ψ(ℓ u )<br />

(b) Image <strong>of</strong> the slice under Ψ, in the standard<br />

Legendrian neighbourhood<br />

Figure 4.1: Observe that in the singular case (left), ℓ u and ℓ s are embedded<br />

Legendrians, but they each intersect the slice twice. They each double cover<br />

their image in the standard neighbourhood (right).<br />

a Legendrian with a smooth contact form. In particular, we will take G <strong>to</strong><br />

be independent <strong>of</strong> φ and θ, as in the model case <strong>of</strong> a Weinstein handle, and<br />

this construction becomes somewhat overkill. We illustrate the double cover in<br />

Figure 4.1.<br />

This double cover allows us <strong>to</strong> define a new notion <strong>of</strong> Reeb orbit.<br />

Definition 4.2.4 (Generalized Reeb orbit). We will say that a continuous map<br />

γ : [a, b] → M is an orbit <strong>of</strong> the Reeb vec<strong>to</strong>r field if:<br />

• for t so that γ(t) /∈ L, we have that γ is differentiable with γ ′ (t) = X(γ(t))<br />

and<br />

• for t so that γ(t) ∈ L, we have a neighbourhood (t − δ, t + δ) on which<br />

Ψ(γ(t)) is defined and Ψ(γ(t)) is an orbit <strong>of</strong> the standard Reeb vec<strong>to</strong>r field<br />

∂<br />

∂z .<br />

In particular, this allows for certain homoclinic chains <strong>to</strong> be viewed as orbits<br />

<strong>of</strong> the Reeb vec<strong>to</strong>r field. This definition gives that γ(t) ∈ L for isolated t.<br />

Finally, we note that this definition destroys uniqueness <strong>of</strong> orbits through a<br />

given point. Instead, we have local uniqueness for points p /∈ L, but when we<br />

hit L, the orbit has two “choices” for how <strong>to</strong> leave L.<br />

4.2.3 A local model for the neighbourhood <strong>of</strong> a homoclinic<br />

orbit<br />

46


From the local model for the singular Legendrian, we may obtain a model for<br />

a neighbourhood <strong>of</strong> a homoclinic orbit (in the generalized sense <strong>of</strong> definition<br />

4.2.4).<br />

Definition 4.2.5 (Simple homoclinic orbit). We say that a homoclinic orbit γ<br />

is simple if γ : [0, T ] → M has γ(0) ∈ L and γ(T ) ∈ L but γ(t) /∈ L for all<br />

0 < t < T .<br />

We remark that our definition <strong>of</strong> Reeb orbit (in our generalized sense) allows<br />

us <strong>to</strong> have Reeb orbits that intersect L multiple times. The intersection must<br />

occur at discrete times, however. For any homoclinic orbit, we may decompose<br />

it in<strong>to</strong> its simple components.<br />

Now, in order <strong>to</strong> study the neighbourhood <strong>of</strong> the homoclinic orbit, we will<br />

need the following lemma we proved earlier :<br />

Lemma 3.2.2. Suppose (W 2n , ω, Y, F ) is a Weinstein domain with critical<br />

point x0 on the level F = 0. Suppose XF (0) is a hyperbolic 0. Let W s and<br />

W u denote the stable and unstable manifolds for the flow <strong>of</strong> XF . Then, for any<br />

neighbourhood <strong>of</strong> x0, there exists a Legendrian embedding <strong>of</strong> S n−1 in<strong>to</strong> W u [<br />

W s ], everywhere transverse <strong>to</strong> XH, with image in the chosen neighbourhood.<br />

Furthermore, for any constant c sufficiently small, we may arrange the Legendrian<br />

so that for any path γ connecting x0 <strong>to</strong> any point on the Legendrian<br />

sphere, the path has action γ ∗ λ = c.<br />

Proposition 4.2.6. Let γ : [0, T ] → M be a simple homoclinic orbit for the<br />

(singular) Reeb vec<strong>to</strong>r field, with γ(0) ∈ L and γ(T ) ∈ L.<br />

Suppose γ(0) = γ(T ). Then, there exists a neighbourhood U <strong>of</strong> γ([0, T ]) in<br />

M and neighbourhoods V0 ∋ γ(0) and V1 ∋ γ(T ) and a local diffeomorphism<br />

Φ : U → U ′ ⊂ R 3 so that:<br />

1. Φ(γ(t)) = (0, 0, t).<br />

2. The restriction Φ|U\(V0∪V1) is a diffeomorphism.<br />

3. Φ ∗ αstd = λ.<br />

4. Φ(L ∩ V0) = {(x, 0, 0)} ∩ U ′ for |x| small, and<br />

5. Φ(L ∩ V1) = {(f1(τ), f2(τ), f3(τ)} ∩ U ′ ,<br />

for τ ∈ (−δ, δ), and (f1(0), f2(0), f3(0)) = (0, 0, T )<br />

6. Φ∗ ˜ J is an almost complex structure on R × U ′ adjusted <strong>to</strong> the standard<br />

contact form α0.<br />

47


In the case that γ(0) = γ(T ), there exists a lift<br />

h : U → Nbd(γ([0, T ])) ⊂ M<br />

so that h ∗ λ = λ ′ , where λ ′ is a singular contact form with all the properties <strong>of</strong><br />

λ, defined on U. Then there exists a local diffeomorphism Φ : U → U ′ , as above.<br />

Pro<strong>of</strong>. We will prove the existence <strong>of</strong> Φ by construction, starting with the existence<br />

<strong>of</strong> Ψ. In order <strong>to</strong> simplify the notation, we will only consider the case<br />

γ(0) = γ(T ). The idea carries through identically in the more general case.<br />

To do this construction, we will first consider the unit stable/unstable Legendrian<br />

manifolds given by Lemma 3.2.2. The homoclinic orbit is a Legendrian<br />

chord between these two Legendrian knots. We will use a tubular neighbourhood<br />

theorem in order <strong>to</strong> construct Φ on a neighbourhood <strong>of</strong> the orbit between<br />

these two Legendrians. Then, we will use Ψ, the double cover that exists in a<br />

neighbourhood <strong>of</strong> the singular Legendrian L, <strong>to</strong> continue Φ in<strong>to</strong> a neighbourhood<br />

<strong>of</strong> L.<br />

The construction will use the fact that if we have a diffeomorphism F : U ⊂<br />

R 3 → V ⊂ R 3 with F ∗ αstd = αstd, then F (φt(p)) = φt(F (p)) for p ∈ U and<br />

t small enough so that φt(p) ∈ U. Thus, we may extend F <strong>to</strong> {φt(p) | p ∈ U}<br />

by F (φt(p)) := φtF (p). If U and V are chosen so that no Reeb orbit exits U<br />

or V and then re-enters at a later time, then this is well-defined. In particular,<br />

this is well defined if U and V are convex. Furthermore, with this restriction<br />

on U and V , this defines the extension <strong>of</strong> F uniquely. Indeed, any two such<br />

diffeomorphisms that agree on U and V must then agree on {φt(p) | p ∈ U}.<br />

Let U be the neighbourhood <strong>of</strong> L on which Ψ is defined, and let Nδ \ Lstd<br />

be its image. Recall that αstd = dx + y dη and Lstd = {(0, 0, η)}.<br />

Take ℓu and ℓs, unit unstable and stable Legendrian manifolds, as in Lemma<br />

3.2.2, sufficiently close <strong>to</strong> L that they intersect U. We see that Ψ(ℓu) and Ψ(ℓs)<br />

are given by {(−c, 0, η)} and {(c, 0, η)} respectively, where c > 0 is small.<br />

We have then that p := γ(c) ∈ ℓu and q := γ(T − c) ∈ ℓs. We have that<br />

restricted <strong>to</strong> a neighbourhood <strong>of</strong> p, Ψ is a diffeomorphism on<strong>to</strong> its image. We<br />

then have Ψ −1 : B((0, 0, c), ɛ) ⊂ R 3 → U1, a neighbourhood <strong>of</strong> p. We have that<br />

(Ψ −1 ) ∗ λ = αstd. We also take ɛ sufficiently small that φT −c(U1) ⊂ U \ L.<br />

We now introduce the map<br />

F1 : Dɛ × [c, T − c] −→ M<br />

(x, y, z) ↦→ φz−c(Ψ −1 (x, y, c)).<br />

This is a diffeomorphism on<strong>to</strong> its image. Furthermore, we have that F1 ∗ λ = αstd.<br />

We now have that F −1<br />

1 , in a neighbourhood <strong>of</strong> γ(τ ′ ), maps ℓ s <strong>to</strong> a Legendrian<br />

arc in Dɛ × [τ, τ ′ ]. Let us denote this arc by ˇ ℓ. Then, there exists<br />

48


a diffeomorphism G : Nbd( ˇ ℓ) → Nbd((0, 0, 0)) so that G( ˇ ℓ) = Nbd((0, 0, 0)) ∩<br />

{(0, 0, η) | |η| small.}. We now extend G <strong>to</strong> a neighbourhood <strong>of</strong> {(0, 0)}×[−T, T ]<br />

by using the Reeb flow. In particular, we have that G −1 ◦Ψ is defined on a neigh-<br />

bourhood <strong>of</strong> γ([T − c, T ]).<br />

Observe that G −1 ◦ Ψ : Nbd(ℓ s ) → R 3 and F −1<br />

1<br />

sufficiently small neighbourhood <strong>of</strong> ℓ s .<br />

: Nbd(ℓ s ) → R 3 agree on a<br />

We now take our map Φ = Ψ near p, and Φ = F −1<br />

1 in the region in which it is<br />

defined. Finally, we take Φ = G −1 ◦Ψ in a neighbourhood <strong>of</strong> γ([T −c, T ]). These<br />

maps agree where their domains <strong>of</strong> definition overlap, so Φ is well defined. By<br />

construction, this is a local diffeomorphism from a neighbourhood <strong>of</strong> γ([0, T ])<br />

on<strong>to</strong> its image. Furthermore, it is a diffeomorphism on<strong>to</strong> its image once a<br />

neighbourhood <strong>of</strong> L is deleted from its domain <strong>of</strong> definition.<br />

4.3 Type (I) and (II) ˜ J-holomorphic curves<br />

In this section, we introduce a special class <strong>of</strong> almost complex structures, compatible<br />

with the double cover Ψ introduced in Prop 4.2.3. We then define the<br />

two classes <strong>of</strong> pseudoholomorphic curve we will be interested in. Finally, we<br />

point out some cases in which the two classes agree.<br />

4.3.1 The singular almost complex structure<br />

By Proposition 4.2.3, there exist a δmax > 0, and a local diffeomorphism Ψ<br />

defined in a neighbourhood U <strong>of</strong> L, so that Ψ : (U \ L) → (Nδmax \ Lstd) is a<br />

double cover. Furthermore, we have Ψ ∗ αstd = λ. We will let Uδ = Ψ −1 (Nδ \<br />

Lstd) ∪ L and Mδ = M \ Uδ. We observe then that Mδ is a compact contact<br />

manifold, with convex boundary.<br />

In the following, we will abuse notation and let Ψ denote either be the map<br />

Ψ : U −→ Nδmax<br />

or the map Ψ : R × U −→ R × Nδmax,<br />

by extending Ψ by the identity on the R component.<br />

Our local model <strong>of</strong> the singularity allows us <strong>to</strong> define what we mean by an<br />

almost complex structure adjusted <strong>to</strong> the singular contact form λ.<br />

Definition 4.3.1. We say that ˜ J is an almost complex structure adjusted <strong>to</strong><br />

the singular contact form λ = f ± λ0 if ˜ J is adjusted <strong>to</strong> λ away from L, and<br />

furthermore, T Ψ ◦ ˜ J ◦ ( T Ψ) −1 is a complex structure adjusted <strong>to</strong> αstd and<br />

extends smoothly across Lstd.<br />

49


We recall that a smooth almost complex structure ˜ J is adjusted <strong>to</strong> a<br />

(smooth) contact form α if ˜ J is R invariant, ˜ J ∂<br />

∂s = X and ˜ Jξ = ξ with ˜ J|ξ<br />

compatible with the symplectic form dλ. This terminology is as in [6], though<br />

this class <strong>of</strong> almost complex structures was previously introduced by H<strong>of</strong>er in<br />

[21].<br />

In the following, we will refer <strong>to</strong> an almost complex structure on M \ L,<br />

adjusted <strong>to</strong> the singular contact form λ as a singular almost complex structure<br />

if it satisfies this Definition 4.3.1.<br />

Remark 4.3.2. The class <strong>of</strong> almost complex structures defined by 4.3.1 is not<br />

empty. The key observation is that the space <strong>of</strong> almost complex structures<br />

compatible with a symplectic form is contractible. Then, given any two almost<br />

complex structures adjusted <strong>to</strong> a given contact form, we may find a path <strong>of</strong><br />

almost complex structures between them. In particular, this allows us <strong>to</strong> do<br />

an “averaging” construction <strong>to</strong> glue <strong>to</strong>gether local constructions. Thus, we<br />

construct an almost complex structure on U. We then take an almost complex<br />

structure on Mδ. We may then adjust the two almost complex structures <strong>to</strong><br />

agree on Uδmax \ Uδmax/2 and then obtain an almost complex structure on M<br />

satisfying definition 4.3.1.<br />

To obtain an almost complex structure on U, we start by choosing an almost<br />

complex structure on Nδmax, adjusted <strong>to</strong> αstd. By means <strong>of</strong> Ψ, we may pull this<br />

complex structure back <strong>to</strong> a complex structure on U \ L.<br />

4.3.2 Type (I) pseudoholomorphic curves<br />

The first type <strong>of</strong> pseudoholomorphic curve we introduce is the most natural.<br />

Suppose ˜ J is an almost complex structure as in Definition 4.3.1. Then we say<br />

that a continuous map ũ : (Σ, j) → R × M is a Type (I) pseudoholomorphic<br />

curve if T ũ(z) ◦ j = ˜ J(u(z)) T ũ(z) for all z ∈ Σ so that u(z) /∈ L.<br />

In particular, we note that if we have a sequence <strong>of</strong> pseudoholomorphic<br />

curves that are pseudoholomorphic for a sequence <strong>of</strong> smooth almost complex<br />

structures converging <strong>to</strong> ˜ J in C ∞ loc<br />

on compact subsets <strong>of</strong> M \ L, and they<br />

satisfy a uniform gradient bound, then a subsequence will converge <strong>to</strong> a Type<br />

(I) pseudoholomorphic curve.<br />

We will call the set {z | u(z) ∈ L}, the set <strong>of</strong> singularities <strong>of</strong> ũ, and denote<br />

it by Γ(u, L).<br />

Unfortunately, these curves are difficult <strong>to</strong> deal with. This will lead us <strong>to</strong><br />

the second class <strong>of</strong> pseudoholomorphic curve we consider.<br />

4.3.3 Type (II) pseudoholomorphic curves<br />

50


In light <strong>of</strong> the double cover we obtained in Proposition 4.2.3, another natural<br />

elliptic PDE <strong>to</strong> consider is that the curve satisfy the non-linear Cauchy-Riemann<br />

equations when pushed down by the double cover.<br />

Definition 4.3.3. We say that a curve ũ = (a, u) : (Σ, j) → R × M is Type<br />

(II) pseudoholomorphic with respect <strong>to</strong> the singular almost complex structure<br />

˜J, satisfying the definition 4.3.1 if<br />

• ũ is continuous,<br />

• ũ is pseudoholomorphic on {z | u(z) ∈ L}, and<br />

• Ψ(ũ) is pseudoholomorphic as a map in<strong>to</strong> R × Nδmax, where defined.<br />

As above, we will call the set {z | u(z) ∈ L}, the set <strong>of</strong> singularities <strong>of</strong> ũ, and<br />

denote it by Γ(u, L).<br />

We see that a Type (I) pseudoholomorphic curve is au<strong>to</strong>matically a Type (II)<br />

pseudoholomorphic curve. The converse is not known in general. Nevertheless,<br />

we may conclude that in certain circumstances, the two types coincide.<br />

Proposition 4.3.4. Let ũ : (Σ, j) → R × M be a Type (I) pseudoholomorphic<br />

curve. Let Γ ′ be the set <strong>of</strong> accumulation points <strong>of</strong> Γ(u, L). If Γ ′ accumulates in<br />

a discrete set, then ũ is a Type (II) pseudoholomorphic curve.<br />

This result is actually a corollary <strong>of</strong> Gromov’s theorem on the removal <strong>of</strong><br />

singularities for a pseudoholomorphic curve :<br />

Theorem 4.3.5. Suppose (W, ω) is a symplectic manifold (either compact or<br />

<strong>of</strong> bounded geometry) and J is an almost complex structure compatible with ω.<br />

Then, if we have a pseudoholomorphic map u from the punctured, closed disk<br />

<strong>of</strong> finite symplectic area :<br />

u : ¯ <br />

D \ {0} → W with u ∗ ω < ∞<br />

then, u has a smooth (and hence pseudoholomorphic) extension across 0.<br />

(For a pro<strong>of</strong>, we refer the reader <strong>to</strong> [33].)<br />

We will use two simple corollaries <strong>of</strong> this result. The first applies this theorem<br />

<strong>to</strong> the symplectization <strong>of</strong> a contact manifold. The second discusses what<br />

happens if we have more than a single singularity.<br />

Corollary 4.3.6 (Contact Version). Suppose M is a contact manifold with<br />

contact form λ and ˜ J is an R–invariant almost complex structure on the symplectization,<br />

adjusted <strong>to</strong> the contact form. Any pseudoholomorphic map from a<br />

punctured disk in<strong>to</strong> the symplectization, ũ = (a, u) : ¯ D \ {0} → R × M with<br />

u ∗ dλ < ∞ and a bounded admits a smooth extension across 0.<br />

51


This follows immediately from estimating the symplectic area <strong>of</strong> the form<br />

d(e t λ) in terms <strong>of</strong> the bound on a and the area u ∗ dλ. We may then apply<br />

Gromov’s theorem.<br />

Corollary 4.3.7. Suppose (W, ω) is a symplectic manifold (either compact or<br />

<strong>of</strong> bounded geometry) and J is an almost complex structure compatible with ω.<br />

Let Γ be a closed subset <strong>of</strong> the open disk, whose set <strong>of</strong> accumulation points<br />

is precisely {0}.<br />

Then, if we have a pseudoholomorphic map u from the closed disk, with<br />

singularities at Γ, but with finite symplectic area :<br />

u : ¯ <br />

D \ Γ → W with u ∗ ω < ∞<br />

then, u has a smooth (and hence pseudoholomorphic) extension across Γ.<br />

The pro<strong>of</strong> is by iterated application <strong>of</strong> Gromov’s removal <strong>of</strong> singularities.<br />

In order <strong>to</strong> prove the result 4.3.4, we consider Ψ ◦ ũ : U → R × Nδ, where<br />

U ⊂ Σ is the set on which Ψ ◦ u is defined. Since Γ ′ has a discrete set <strong>of</strong><br />

accumulation points, we may apply Gromov’s removal <strong>of</strong> singularities iteratively.<br />

We then have that Ψ◦ũ admits a pseudoholomorphic extension across Γ. Thus,<br />

we have that Ψ ◦ ũ is pseudoholomorphic where defined, and hence ũ is a Type<br />

(II) pseudoholomorphic curve.<br />

4.4 Asymp<strong>to</strong>tic behaviour <strong>of</strong> a Type (II) finite<br />

energy half-plane<br />

In this section, we prove some results about the asymp<strong>to</strong>tic behaviour <strong>of</strong> a Type<br />

(II) pseudoholomorphic curve. Of greatest interest is that a Type (II) pseudoholomorphic<br />

half-plane is asymp<strong>to</strong>tic <strong>to</strong> a homoclinic orbit. The convergence<br />

will be at an exponential rate if the asymp<strong>to</strong>tic limit is transverse.<br />

4.4.1 Convergence <strong>to</strong> a homoclinic orbit<br />

In this section, we prove that Type (II) ˜ J-holomorphic half-plane, in the sense<br />

<strong>of</strong> definition 4.3.3, with finite energy, is asymp<strong>to</strong>tic <strong>to</strong> a homoclinic orbit.<br />

We write H = {z ∈ C | ℜ(z) ≥ 0}.<br />

Theorem 4.4.1. Let M, λ and ˜ J as in Definition 4.3.1. Suppose that ũ : H →<br />

R × M is Type (II) ˜ J holomorphic, in the sense <strong>of</strong> Definition 4.3.3, with finite<br />

52


energy E(ũ) = T < ∞ and boundary condition u(∂H) ∈ L. Let T = u ∗ dλ.<br />

Then, for any sequence R ′ k → ∞ has a subsequence Rk so that<br />

u(Rke it ) → x(T t/π)<br />

where x(t) is a homoclinic orbit <strong>of</strong> the (singular) Reeb vec<strong>to</strong>r field. The convergence<br />

is in C ∞ ([−π/2, π/2], M).<br />

We are especially interested in the limit if this is a transverse homoclinic<br />

orbit. We will use the term non-degenerate <strong>to</strong> emphasize the similarity with<br />

non-degenerate periodic orbits, as in Floer theory or Contact Homology, or<br />

non-degenerate Reeb chords (as in [2]).<br />

Definition 4.4.2 (Non-degenerate homoclinic orbit). Let ℓu and ℓs be the unit<br />

unstable and stable Legendrians given by Lemma 3.2.2, <strong>of</strong> action ɛ > 0 (smaller<br />

than the shortest homoclinic). Let γ be a simple homoclinic orbit. Suppose<br />

γ is parametrized so that γ(0) ∈ ℓu and γ(τ) ∈ ℓs. Then, the orbit γ is nondegenerate<br />

if T γ(τ) T γ(0)ℓu ⊕ T γ(τ)ℓs = ξγ(τ).<br />

Equivalently, the map R×ℓu → M : (t, p) ↦→ φt(p) intersects ℓs transversally<br />

at (τ, γ(0)).<br />

A homoclinic orbit γ is non-degenerate if each <strong>of</strong> its simple components are.<br />

We remark that this is equivalent <strong>to</strong> requiring that Φ(γ) be a non-degenerate<br />

Legendrian chord.<br />

In the case that the asymp<strong>to</strong>tic limit is a non-degenerate homoclinic orbit,<br />

we have :<br />

Theorem 4.4.3. Let ũ : H → R × M as in the hypotheses <strong>of</strong> 4.4.1. If the<br />

asymp<strong>to</strong>tic limit, x : [0, T ] → M is a non-degenerate, simple homoclinic orbit,<br />

then<br />

lim<br />

R→∞ u(Reit ) = x(T t)<br />

where x(t) is a homoclinic orbit <strong>of</strong> the (singular) Reeb vec<strong>to</strong>r field. The convergence<br />

is in C ∞ ([−π/2, π/2], M).<br />

We also establish that a pseudoholomorphic curve with a weaker boundary<br />

condition also converges <strong>to</strong> a homoclinic orbit :<br />

Theorem 4.4.4. Let M, λ and ˜ J as in Definition 4.3.1. Suppose that ũ : H →<br />

R × M is Type (II) ˜ J holomorphic, in the sense <strong>of</strong> Definition 4.3.3, with finite<br />

energy E(ũ) < ∞ and boundary condition at infinity<br />

d(u(0 + it), L) → 0 as |t| −→ ∞.<br />

53


Let T = u ∗ dλ. Then, for any sequence R ′ k → ∞ there is a subsequence Rk so<br />

that<br />

u(Rke it ) → x(T t)<br />

where x(t) is a homoclinic orbit <strong>of</strong> the (singular) Reeb vec<strong>to</strong>r field. The convergence<br />

is in C ∞ ([−π/2, π/2], M).<br />

In order <strong>to</strong> prove Theorem 4.4.1, we must first establish some properties <strong>of</strong><br />

Type (II) pseudoholomorphic curves for a singular almost complex structure,<br />

in the sense <strong>of</strong> Definition 4.3.3. Due <strong>to</strong> the existence <strong>of</strong> the double cover, as<br />

in Proposition 4.2.3, we will be able <strong>to</strong> show that ˜ J-holomorphic curves, for<br />

our singular almost complex structure, share many <strong>of</strong> the same properties as<br />

pseudoholomorphic curves for a regular almost complex structure.<br />

Our results are all <strong>of</strong> an analytical nature, and <strong>to</strong> develop them, we must<br />

introduce various Sobolev spaces. Of greatest importance <strong>to</strong> us are the W 1,p<br />

spaces, for p > 2. It will be convenient <strong>to</strong> realize our function spaces as linear<br />

spaces, so we will need a way <strong>to</strong> embed our problem in<strong>to</strong> a high dimensional<br />

R N . This is where we use the specific compactification <strong>of</strong> the singular level<br />

<strong>to</strong> M + with contact form f + λ + 0 . We equip M + with the metric coming from<br />

λ + 0 (specifically g = λ + 0 ⊗ λ + 0 + dλ(·, J·) where J is a suitable almost complex<br />

structure on the contact structure). We now embed this isometrically in R N for<br />

N sufficiently large. By the explicit form <strong>of</strong> the local diffeomorphism Ψ given<br />

by Proposition 4.2.3, we see that || T Ψ|| ≤ C where it is defined. This will allow<br />

us <strong>to</strong> obtain gradient bounds “downstairs” if we have them “upstairs”.<br />

Our first result will be <strong>to</strong> show that gradient bounds imply C ∞ bounds.<br />

We first recall the result for smooth almost complex structures on compact<br />

manifolds. A good reference for the pro<strong>of</strong> is Appendix B <strong>of</strong> [33]. It is also<br />

proved in Chapter 6 <strong>of</strong> [4].<br />

Theorem 4.4.5 (Theorem 4.1.1 in [33]). Suppose (W, J) is an almost complex<br />

manifold. We denote by D, the unit disk in C. Assume un : D → W is<br />

a sequence <strong>of</strong> J-holomorphic curves with a uniform C 1 bound. Then, for all<br />

0 < r < 1, for all p > 2 and for all k ∈ N, there exists a constant C = C(r, p, k)<br />

so that<br />

||ũn|| W k,p (Dr) ≤ C.<br />

In particular, a subsequence converges in C ∞ loc .<br />

From this, we will deduce a comparable result for our singular setting :<br />

Proposition 4.4.6. Let ũn = (an, un) : D → R × M be a sequence <strong>of</strong> ˜ Jholomorphic<br />

curves for a singular almost complex structure ˜ J, in the sense <strong>of</strong><br />

Definition 4.3.1.<br />

54


Suppose the sequence has a uniform gradient bound : |∇ũn| ≤ C and a bound<br />

on an(0) (say, |an(0)| ≤ C ′ for all n). In other words, we require a C 1 bound<br />

on the sequence.<br />

Then, there exists an 0 < rmax ≤ 1, so that for all 0 < r < rmax, for all p > 2<br />

and for all k ∈ N, there exists a C = C(r, p, k, l) so that<br />

||ũn|| W k,p (Dr) ≤ C.<br />

In particular, a subsequence converges in C ∞ loc .<br />

Pro<strong>of</strong>. Our result will follow from Theorem 4.4.5, and from the fact that Ψ<br />

is Lipschitz (as observed above). This uses, in an essential way, the fact that<br />

we are working with the compactification <strong>to</strong> M + with singular contact form<br />

λ = Gλ0, with G = 0 along L.<br />

Indeed, we introduce rmax = min(1, δmax/4C), where C is the bound on the<br />

gradients <strong>of</strong> ũn. Then, for any ũ in our sequence, by the bound on ∇ũ, and the<br />

triangle inequality, we have one <strong>of</strong> the two following cases :<br />

1. The image <strong>of</strong> u is <strong>to</strong>tally contained in the domain <strong>of</strong> the double cover :<br />

u(D) ⊂ Uδmax<br />

2. The image <strong>of</strong> u is <strong>to</strong>tally contained in the Mδmax/2.<br />

In either case, the bound follows immediately from the result 4.4.5.<br />

We will prove Theorems 4.4.1, 4.4.4 by building up the following propositions.<br />

In all <strong>of</strong> them, we take M, λ and ˜ J as in Definition 4.3.1.<br />

Proposition 4.4.7. Suppose that ũ : H → R × M is Type (II) ˜ J holomorphic,<br />

in the sense <strong>of</strong> Definition 4.3.3, with finite energy E(ũ) < ∞ and boundary<br />

condition u(∂H) ∈ L. If, furthermore,<br />

<br />

u ∗ dλ = 0.<br />

Then, ũ must be constant.<br />

C<br />

In order <strong>to</strong> prove Theorem 4.4.1, it will be convenient <strong>to</strong> re-parametrize<br />

ũ : H → R × M by<br />

˜v : R × [−π, π] −→ R × M<br />

z ↦→ ũ(e z ). (4.1)<br />

This strip ˜v then satisfies the boundary condition v(s ± π) ∈ L. Then, we<br />

will also prove a slightly different result, having <strong>to</strong> do with the behaviour <strong>of</strong> a<br />

pseudoholomorphic strip with no contact area :<br />

55


Proposition 4.4.8. Suppose that ũ : R × [−π/2, π/2] → R × M is Type (II) ˜ J<br />

holomorphic, in the sense <strong>of</strong> Definition 4.3.3, with finite energy E(ũ) < ∞ and<br />

boundary condition u(s ± πi) ∈ L. If, furthermore,<br />

<br />

u ∗ dλ = 0<br />

and |∇ũ| ≤ C < ∞, then,<br />

C<br />

ũ = (cs + d, γ(ct))<br />

where γ : [0, T ] → M is a homoclinic orbit and c and d are real constants.<br />

We will use then Proposition 4.4.7 <strong>to</strong> obtain<br />

Proposition 4.4.9. If ˜v is a finite energy strip as in (4.1), then ˜v has bounded<br />

gradient.<br />

Finally, this will allow us <strong>to</strong> establish Theorem 4.4.1.<br />

Then, in order <strong>to</strong> prove Theorem 4.4.4, we will need the following result :<br />

Proposition 4.4.10. Suppose ũ : H → R × M is Type (II) pseudoholomorphic,<br />

as in Definition 4.3.3, with the boundary condition at infinity<br />

u(0 + it) → L as t → ±∞<br />

and finite energy. Suppose ˜v(z) = ũ(e z ) is the corresponding strip, as in (4.1).<br />

Then, ˜v(z) has bounded gradient.<br />

This result, in turn, will allow us <strong>to</strong> conclude Theorem 4.4.4.<br />

The key fact underpinning all <strong>of</strong> these results is that Proposition 4.4.6 tells<br />

us that if we have uniform gradient bounds on a sequence <strong>of</strong> curves, we have a<br />

limit (after taking a subsequence, perhaps). We need then <strong>to</strong> understand what<br />

happens if we don’t have gradient bounds.<br />

An important technical <strong>to</strong>ol in all <strong>of</strong> these bubbling <strong>of</strong>f pro<strong>of</strong>s is the following<br />

lemma, due <strong>to</strong> H<strong>of</strong>er (in [21]), whose pro<strong>of</strong> is elementary, but whose use greatly<br />

simplifies a certain number <strong>of</strong> otherwise difficult arguments. A pro<strong>of</strong> may be<br />

found in [4, Chapter 6], or in [33, Lemma 4.6.4].<br />

H<strong>of</strong>er’s Lemma. Let (X, d) be a metric space, f : X → R a non-negative,<br />

continuous function. Fix x ∈ X and δ > 0.<br />

Suppose the closed ball B2δ(x) is complete. Then, there exists a ξ ∈ X and a<br />

positive number ɛ ≤ δ so that :<br />

d(x, ξ) < 2δ, sup f ≤ 2f(ξ),<br />

Bɛ(ξ)<br />

ɛf(ξ) ≥ δf(ξ).<br />

56


Pro<strong>of</strong>s <strong>of</strong> our results<br />

We are now able <strong>to</strong> build up the pro<strong>of</strong>s <strong>of</strong> our propositions, culminating in the<br />

pro<strong>of</strong> <strong>of</strong> Theorem 4.4.4. The pro<strong>of</strong>s themselves are very similar <strong>to</strong> the standard<br />

theory <strong>of</strong> pseudoholomorphic curves in symplectizations, though some extra care<br />

is needed near R × L, our singular region.<br />

Pro<strong>of</strong> <strong>of</strong> Proposition 4.4.7. Since ũ is pseudoholomorphic (in the smooth sense)<br />

away from its singularities Γ(u, L), we have u ∗ dλ is a non-negative multiple <strong>of</strong><br />

the volume form away from these singularities. Thus, for u <strong>to</strong> have no contact<br />

area, there must exist an orbit <strong>of</strong> the Reeb vec<strong>to</strong>r field, γ, so that<br />

ũ = (a(z), γ(f(z)).<br />

In order for the boundary condition on u <strong>to</strong> be satisfied, we must have γ is a<br />

homoclinic orbit. By Proposition 4.2.6, we have then the existence <strong>of</strong> a local<br />

diffeomorphism Φ from a neighbourhood <strong>of</strong> γ <strong>to</strong> a standard neighbourhood<br />

V ⊂ R3 with the standard contact form. By the symmetry condition on ˜ J<br />

given by Definition 4.3.1, we have that the diffeomorphism Φ pushes the almost<br />

complex structure ˜ J forward <strong>to</strong> an almost complex structure ˜ M on R×V , which<br />

extends smoothly <strong>to</strong> all <strong>of</strong> R×V . We then have that ˜w(z) = (a(z), Φ(u(z))) is a<br />

pseudoholomorphic half-plane in<strong>to</strong> R×V ⊂ R×R 3 with the standard structure.<br />

Furthermore, w(z) satisfies the boundary condition that w(it) ∈ ℓ0 ∪ ℓ1, for all<br />

t ∈ R. Thus, f(it) = const for all t. In order for ˜w(z) <strong>to</strong> be pseudoholomorphic,<br />

we must have that a + if is holomorphic. We have the boundary condition :<br />

(a+if)(iR) ∈ R, so we may apply the Reflection Principle <strong>to</strong> obtain an extension<br />

<strong>to</strong> a holomorphic plane. In the following, we will use the same notation <strong>to</strong> denote<br />

this extended holomorphic function, a + if.<br />

We wish <strong>to</strong> show that a must be constant. We thus consider the case it<br />

is non-constant. We now consider two sub-cases. First we assume |∇a| is<br />

bounded. Then, by Liouville’s theorem, a ′ + if ′ is a constant function on C.<br />

Thus, (a + if) is an affine function on C. We write (a + if)(z) = ic1z + c2. We<br />

have ic1it + c2 ∈ R for all t ∈ R, thus, c1, c2 ∈ R. From this, we conclude that<br />

a(s + it) = −c1t + c2. We now check the finite energy condition :<br />

<br />

ũ<br />

H<br />

∗ <br />

d(φλ) = ˜w<br />

H<br />

∗ d(φαstd)<br />

<br />

= φ ′ <br />

(a(z)) ( ∂<br />

∂s a)2 + ( ∂<br />

∂t a)2<br />

<br />

ds ∧ dt<br />

= 1<br />

2<br />

H<br />

<br />

φ<br />

C<br />

′ (−c1t + c2)c 2 1 ds ∧ dt<br />

57


For this <strong>to</strong> be finite, we need that c1 = 0, which contradicts the assumption<br />

that a was non-constant.<br />

We now consider the case that |∇a| is unbounded. Let Rn > 0 be a mono<strong>to</strong>ne<br />

sequence <strong>of</strong> values with Rn → ∞ and Rn > |∇(a + if)(0)|. Choose then a<br />

sequence ɛn so that ɛn → 0 and ɛnRn → ∞. Then, there exist a sequence<br />

<strong>of</strong> points zn ∈ C (without loss <strong>of</strong> generality, we may assume zn ∈ H) so that<br />

|∇(a + if)(zn)| = Rn.<br />

By H<strong>of</strong>er’s Lemma, we may move the zn slightly and adjust the Rn and ɛn<br />

so that we we have that there exist zn, ɛn and Rn with the properties :<br />

|∇(a + if)(zn)| = Rn,<br />

Rn → ∞ and ɛn → 0 with ɛnRn → ∞<br />

|∇(a + if)(z)| ≤ 2Rn for |z − zn| ≤ ɛn.<br />

We consider now the sequence <strong>of</strong> holomorphic functions<br />

Fn := (a + if)(zn + z<br />

Rn<br />

) − (a + if)(zn).<br />

We observe that |∇Fn(0)| = 1 and |∇Fn(z)| ≤ 2 for all |z| ≤ ɛnRn. For each<br />

R > 0, the sequence <strong>of</strong> functions Fn is bounded in C1 (BR) (once n is large<br />

enough that Fn is defined on BR). We obtain then higher order bounds by<br />

Cauchy’s integral formula. Thus, for each N, a subsequence <strong>of</strong> Fn converges in<br />

C∞ on BN <strong>to</strong> a holomorphic function. By a diagonal subsequence argument, we<br />

obtain that the sequence Fn converges in C ∞ loc<br />

<strong>to</strong> a pseudoholomorphic plane F<br />

with uniform gradient bound and |∇F (0)| = 1. This limiting curve has no more<br />

energy than the original (a + if), so the same argument as in the previous case<br />

gives that F must be constant. This contradicts however that |∇F (0)| = 1.<br />

Pro<strong>of</strong> <strong>of</strong> Proposition 4.4.8. As in the pro<strong>of</strong> <strong>of</strong> Proposition 4.4.7, we may write<br />

our pseudoholomorphic curve as :<br />

ũ(z) = (a(z), γ(f(z))),<br />

where γ is a an orbit <strong>of</strong> the singular Reeb vec<strong>to</strong>r field and (a + if) is a holomorphic<br />

function. The boundary condition gives that γ intersects L. We use<br />

Proposition 4.2.6, <strong>to</strong> obtain a pseudoholomorphic curve ˜w = (a(z), Φ(u(z))).<br />

We have then<br />

˜w : R × [−π/2, π/2] −→ R × V<br />

w(s ± iπ/2) ∈ ℓ0 ∪ ℓ1.<br />

58


We then have that f(s ± iπ/2) is constant on each boundary component. By<br />

assumption, f must have bounded gradient. Furthermore, we have that f is<br />

harmonic, so we have that f(s + it) = c1t + c2, for a pair <strong>of</strong> real constants c1, c2.<br />

It follows then that a(s + it) = c1s + c ′ 2, where c ′ 2 is another real constant. By<br />

changing the initial condition <strong>of</strong> γ, we obtain then that<br />

ũ(z) = (c1s + c2, γ(c1t)).<br />

Pro<strong>of</strong> <strong>of</strong> Proposition 4.4.9. We will prove this result by contradiction. Then,<br />

there exists a sequence <strong>of</strong> points zn = (sn, tn) so that |∇ũ(zn)| → ∞. Since we<br />

assumed ũ was obtained by re-parametrizing a half-plane, we have sn → +∞.<br />

Furthermore, by H<strong>of</strong>er’s lemma, we may assume that we have<br />

|∇ũ(zn)| = Rn, Rn → ∞ and ɛn → 0 with ɛnRn → ∞<br />

|∇un(z)| ≤ 2Rn for z ∈ B(zn, ɛn) ∩ H.<br />

There are now two cases : in case (1), we have interior bubbling, and in case<br />

(2), we have boundary bubbling.<br />

(1) We have that B(zn, ɛn) ⊂ R × (−π/2, π/2), for n large. We introduce a<br />

re-scaled sequence <strong>of</strong> maps by :<br />

ũn(z) := (a(zn + z<br />

Rn<br />

) − a(zn), u(zn + z<br />

By construction, these maps have uniform gradient bounds on B(0, ɛnRn)<br />

and have |∇un(0)| = 1. We also verify that ũn|B(0,ɛnRn) has energy<br />

bounded above by E(ũ). We then obtain that ũn converge in C ∞ loc <strong>to</strong><br />

a Type (II) pseudoholomorphic plane ˜w : C → R × M, in the sense <strong>of</strong><br />

Definition 4.3.3. Furthermore, |∇ ˜w(0)| = 1, so ˜w is non-constant.<br />

We have again one <strong>of</strong> two possibilities : after taking a subsequence, we<br />

may take either B(zn, ɛn) ∩ Γ(u, L) = ∅ for all n, or Γ( ˜w, L) = ∅. Let us<br />

consider these sub-cases separately.<br />

(i) If B(zn, ɛn) ∩ Γ(u, L) = ∅ for all n, the re-scaled curves have image<br />

in the almost complex manifold R × (M \ L). Thus, the theory <strong>of</strong><br />

pseudoholomorphic curves in<strong>to</strong> symplectizations, with smooth almost<br />

complex structures, applies. In this case, a finite energy plane with<br />

no dλ contact area must be constant (by [4, Prop. 6.4.2], whose<br />

statement and pro<strong>of</strong> is very similar <strong>to</strong> our Proposition 4.4.7). This<br />

contradiction eliminates this case.<br />

59<br />

Rn<br />

))


(ii) We have that <br />

C w∗ dλ = 0. Hence w(z) = (b(z), γ(f(z))), where<br />

a + if is holomorphic and γ is an orbit <strong>of</strong> the Reeb vec<strong>to</strong>r field.<br />

By virtue <strong>of</strong> the fact that Γ(w, L) = ∅, we must have that γ<br />

is a homoclinic orbit. We may then consider the curve ˜v(z) =<br />

(b(z), Φ(γ(f(z)))). This is a pseudoholomorphic plane in<strong>to</strong> the symplectization<br />

R × V , holomorphic with respect <strong>to</strong> a smooth almost<br />

complex structure. Then, again by [4, Prop. 6.4.2], it follows that<br />

˜v must be constant. Hence, ˜w is constant. This contradiction eliminates<br />

this case, and thus the possibility <strong>of</strong> interior bubbling. We are<br />

now only left with the case <strong>of</strong> boundary bubbling.<br />

(2) After taking a subsequence, we may assume that zn = sn + itn have tn →<br />

±π/2. Let us assume tn → π/2. (The other case is similar.) Furthermore<br />

(by taking a subsequence), we may assume that |π/2 − tn|Rn → c < ∞.<br />

Otherwise, we could make ɛn < |π/2 − tn| and we would be in the case <strong>of</strong><br />

interior bubbling.<br />

Let ζn = sn + iπ/2. Let δn = ɛn − |π/2 − tn|. Then, we have δnRn → ∞<br />

and B(ζn, δn) ⊂ B(zn, ɛn). We now introduce a sequence <strong>of</strong> curves, ũn as<br />

follows :<br />

ũn : H ∩ B(0, ɛnRn) −→ R × M<br />

z ↦→ (a(ζn + −iz<br />

) − a(ζn), u(ζn + −iz<br />

)).<br />

Rn<br />

By construction, these curves satisfy a uniform gradient bound. Thus,<br />

we obtain that ũn converge in C∞ loc <strong>to</strong> a Type (II) pseudoholomorphic<br />

curve ˜w in the sense <strong>of</strong> Definition 4.3.3. We have that ˜w : H → R × M<br />

and ˜w maps ∂H in<strong>to</strong> R × L, with w∗ dλ = 0. Furthermore, we have<br />

|∇ũn(0+iRn(tn −π/2))| = 1. Since |tn −π/2|Rn → c < ∞, by considering<br />

a subsequence, we have that Rn(tn − π/2) → z∞, with |∇ ˜w(z∞)| = 1.<br />

Thus ˜w is non-constant. This contradicts Proposition 4.4.7. This last<br />

contradiction completes the pro<strong>of</strong>.<br />

Pro<strong>of</strong> <strong>of</strong> Theorem 4.4.1. Let ˜v = (b, v) be the pseudoholomorphic strip obtained<br />

by the change <strong>of</strong> variables (4.1). Let Sk = log Rk. We will show that after con-<br />

T t<br />

sidering a subsequence, v(Sk +t) is asymp<strong>to</strong>tic <strong>to</strong> γ( ), where γ is a homoclinic<br />

2π<br />

orbit <strong>of</strong> the Reeb vec<strong>to</strong>r field.<br />

By Proposition 4.4.9, we have that there exists a constant so that |∇v(z)| ≤<br />

C < ∞ for all z ∈ R × [−π/2, π/2].<br />

60<br />

Rn


We consider now the sequence <strong>of</strong> curves :<br />

˜vn : R × [−π/2, π/2] −→ R × M<br />

z ↦→ (b(Sk + z) − b(Sk), v(Sk + z)).<br />

We have then that |∇˜vn| ≤ C < ∞. Thus, there exists a Type (II) pseudoholomorphic<br />

curve in the sense <strong>of</strong> Definition 4.3.3 ˜w, so that ˜vn → ˜w in C∞ loc . We<br />

observe that<br />

<br />

<br />

w<br />

[−R,R]×[−π/2,π/2]<br />

∗ dλ = lim<br />

n→∞<br />

= lim<br />

n→∞<br />

= 0<br />

<br />

vn<br />

[−R,R]×[−π/2,π/2]<br />

∗ dλ<br />

[Sn−R,Sn+R]×[−π/2,π/2]<br />

v ∗ dλ<br />

and thus we obtain a finite energy strip with no contact area and bounded<br />

gradient. It then follows from Proposition 4.4.8 that ˜w = (cs + d, γ(ct)), where<br />

γ is a homoclinic orbit and c and d are constants.<br />

From this, we now compute that for any φ : R → [0, 1] with φ ′ ≥ 0 and<br />

φ(−∞) = 0 and φ(+∞) = 1, we have :<br />

<br />

T =<br />

˜w<br />

R×[−π/2,π/2]<br />

∗ d(φλ)<br />

<br />

=<br />

φ ′ (cs + d)c 2 ds ∧ dt<br />

= cπ<br />

R×[−π/2,π/2]<br />

Hence, we have that w(Sk + it) → γ(T t/π) as Sk → ∞, and the convergence is<br />

in C ∞ ([−π/2, π/2], M).<br />

Pro<strong>of</strong> <strong>of</strong> Proposition 4.4.10. We will prove this result by contradiction. Then<br />

there are a sequence <strong>of</strong> points zn = (sn, tn) so that |∇˜v(zn)| = Rn → ∞. We<br />

observe that the argument in the pro<strong>of</strong> <strong>of</strong> Proposition 4.4.9 that establishes the<br />

contradiction from interior bubbling applies in this case as well. Thus, after<br />

taking a subsequence, we may assume that zn = sn + itn have tn → ±π/2. Let<br />

us assume tn → π/2. (The other case is similar.) Furthermore, we may assume<br />

that |π/2 − tn|Rn → c < ∞. Let ζn = sn + iπ/2. Let δn = ɛn − |π/2 − tn|. Then,<br />

we have δnRn → ∞ and B(ζn, δn) ⊂ B(zn, ɛn). We now introduce a sequence <strong>of</strong><br />

curves, ũn as follows :<br />

ũn : H ∩ B(0, ɛnRn) → R × M<br />

z ↦→ (a(ζn + −iz<br />

) − a(ζn), u(ζn + −iz<br />

)).<br />

61<br />

Rn<br />

Rn


By construction, these curves satisfy a uniform gradient bound. Thus, we<br />

obtain that ũn converge in C∞ loc <strong>to</strong> a Type (II) pseudoholomorphic curve ˜w in<br />

the sense <strong>of</strong> Definition 4.3.3. By re-parametrizing the domain, we have that<br />

˜w : H → R × M and ˜w maps ∂H in<strong>to</strong> R × L, with w∗ dλ = 0.<br />

We have<br />

w(0 + it) = lim un(0 + it) = lim u(ζn + t<br />

= lim u(sn + t<br />

Rn<br />

+ i π<br />

2 )<br />

By the boundary condition on u, we have that there is a subsequence along<br />

which u(sn + t ) converges <strong>to</strong> a point on L. Thus, w(0 + it) ∈ L for each t ∈ R.<br />

Rn<br />

We then have that ˜w satisfies the hypotheses <strong>of</strong> Proposition 4.4.7. However,<br />

˜w is non-constant, by construction.<br />

Pro<strong>of</strong> <strong>of</strong> Theorem 4.4.4. Let ˜v = (b, v) be the pseudoholomorphic strip obtained<br />

by the change <strong>of</strong> variables (4.1). Let Sk = log Rk. We will show that after con-<br />

T t<br />

sidering a subsequence, v(Sk +t) is asymp<strong>to</strong>tic <strong>to</strong> γ( ), where γ is a homoclinic<br />

2π<br />

orbit <strong>of</strong> the Reeb vec<strong>to</strong>r field.<br />

By Proposition 4.4.10, we have that there exists a constant so that |∇v(z)| ≤<br />

C < ∞ for all z ∈ R × [−π/2, π/2].<br />

We consider now the sequence <strong>of</strong> curves :<br />

˜vn : R × [−π/2, π/2] → R × M<br />

Rn<br />

z ↦→ (b(Sk + z) − b(Sk), v(Sk + z)).<br />

We have then that |∇˜vn| ≤ C < ∞. Thus, there exists a Type (II) pseudoholomorphic<br />

curve in the sense <strong>of</strong> Definition 4.3.3 ˜w, so that ˜vn → ˜w in C∞ loc . We<br />

observe that<br />

<br />

w<br />

[−R,R]×[−π/2,π/2]<br />

∗ <br />

dλ = lim<br />

vn<br />

n→∞<br />

[−R,R]×[−π/2,π/2]<br />

∗ dλ<br />

<br />

= lim<br />

v<br />

n→∞<br />

∗ dλ<br />

= 0<br />

[Sn−R,Sn+R]×[−π/2,π/2]<br />

and thus we obtain a finite energy strip with no contact area and bounded<br />

gradient.<br />

62<br />

)


Furthermore, we have that this strip ˜w has boundary in R × L. Indeed, for<br />

any s, we have :<br />

w(s ± iπ/2) = lim<br />

n→∞ vn(s ± iπ/2)<br />

= lim<br />

n→∞ v(Sn + s ± iπ/2)<br />

= p ∈ L by the asymp<strong>to</strong>tic boundary condition on v.<br />

It then follows from Proposition 4.4.8 that ˜w = (cs + d, γ(ct)), where γ is a<br />

homoclinic orbit and c and d are constants.<br />

From this, we now compute that for any φ : R → [0, 1] with φ ′ ≥ 0 and<br />

φ(−∞) = 0 and φ(+∞) = 1, we have :<br />

<br />

T =<br />

˜w<br />

R×[−π/2,π/2]<br />

∗ d(φλ)<br />

<br />

=<br />

φ ′ (cs + d)c 2 ds ∧ dt<br />

= cπ<br />

R×[−π/2,π/2]<br />

Hence, we have that w(Sk + it) → γ(T t/π) as Sk → ∞, and the convergence is<br />

in C ∞ ([−π/2, π/2], M).<br />

4.4.2 Exponential rate <strong>of</strong> convergence<br />

Let us assume we have a pseudoholomorphic half-plane, ũ : H → R × M,<br />

satisfying the hypotheses <strong>of</strong> Theorem 4.4.1. Furthermore, let us assume that<br />

for some sequence Rk → ∞, u(Rke it ) converges <strong>to</strong> a non-degenerate homoclinic<br />

orbit. Then, we may conclude that u(re it ) has a limit as r → ∞. Furthermore,<br />

we have that u(re it ) approaches the asymp<strong>to</strong>tic limit exponentially fast.<br />

Finally, we will show that the component <strong>of</strong> u transversal <strong>to</strong> its asymp<strong>to</strong>tic<br />

limit will approach an eigenvec<strong>to</strong>r <strong>of</strong> a suitable unbounded opera<strong>to</strong>r, at an<br />

exponential rate governed by the corresponding eigenvalue. These results are<br />

similar in flavour <strong>to</strong> results in [2], [24] and [35].<br />

This result is in keeping with the Floer theoretic analogy between pseudoholomorphic<br />

curves and gradient flow lines. The underlying idea is that the<br />

pseudoholomorphic curve equation is the gradient flow equation for an action<br />

functional on the loop space. This action functional is not Morse-Smale, and<br />

so the gradient flow is not well posed. However, we may formally write the<br />

gradient flow as the non-linear Cauchy-Riemann equations. The analogy proves<br />

63


accurate in many cases — this exponential rate <strong>of</strong> convergence being one <strong>of</strong><br />

them.<br />

The first result is that a pseudoholomorphic half-plane, if it is asymp<strong>to</strong>tic<br />

as in Theorem 4.4.1 <strong>to</strong> a non-degenerate homoclinic orbit, then the asymp<strong>to</strong>tic<br />

limit is the same no matter what sequence <strong>of</strong> Rk we use.<br />

Theorem 4.4.11. Let M, λ and ˜ J as in Definition 4.3.1. Suppose that ũ :<br />

H → R × M is Type (II) ˜ J holomorphic, in the sense <strong>of</strong> Definition 4.3.3, with<br />

finite energy E(ũ) = T < ∞ and boundary condition u(∂H) ∈ L. Suppose that<br />

γ : [0, T ] → M is a non-degenerate homoclinic orbit so that for some sequence<br />

Rk → ∞<br />

u(Rke it ) → x(T t/π).<br />

Then, for any sequence R ′ k<br />

→ ∞,<br />

u(R ′ ke it ) → x(T t/π).<br />

Pro<strong>of</strong>. Let ℓu and ℓs be unit unstable and stable Legendrians <strong>of</strong> action δ, for<br />

some δ > 0, small, as given by Lemma 3.2.2. Then, we have x(δ) ∈ ℓu and x(T −<br />

δ) ∈ ℓs. We recall that one <strong>of</strong> the equivalent formulations <strong>of</strong> non-degeneracy<br />

for a homoclinic orbit is that the image <strong>of</strong> [0, T − δ] × ℓu → M : (t, p) ↦→ φt(p)<br />

intersect ℓs transversely at x(T − δ). It follows then that the homoclinic orbit x<br />

is isolated among homoclinic orbits <strong>of</strong> nearby action. Let U be a neighbourhood<br />

<strong>of</strong> x in the space <strong>of</strong> paths in M, with respect <strong>to</strong> the C ∞ <strong>to</strong>pology so that there<br />

are no other homoclinics <strong>of</strong> nearby action in U. Let V ⊂ U be a smaller<br />

neighbourhood <strong>of</strong> x.<br />

We will proceed by contradiction. Suppose that u(R ′ k eit ) does not converge<br />

<strong>to</strong> x(T t/π). By taking a subsequence, we may assume that u(R ′ k eit is bounded<br />

away from x(T t/π) in the C ∞ <strong>to</strong>pology on paths in M. Then, by Theorem<br />

4.4.1, R ′ k<br />

has a subsequence R′′<br />

k<br />

so that u(R′′<br />

k eit ) converges <strong>to</strong> a y(T t/π). We<br />

must have y = x and thus y /∈ U. We observe that the map S ↦→ u(Se it ) is<br />

a continuous map in the C ∞ <strong>to</strong>pology on paths in M. Thus, we must have<br />

infinitely many S so that u(Se it ) /∈ V but u(Se it ) ∈ U. By Theorem 4.4.1,<br />

we then have a subsequence Sk along which u(Ske it ) → z(T t/π), where z is a<br />

homoclinic orbit. As z /∈ V , z = x. But there are no homoclinic orbits <strong>of</strong> action<br />

T in U. The result now follows from this contradiction.<br />

Theorem 4.4.12. Let M, λ and ˜ J as in Definition 4.3.1. Let ũ be a pseudoholomorphic<br />

strip, as in Theorem 4.4.11, and γ be its asymp<strong>to</strong>tic limit. Suppose<br />

ı : Dr ↩→ M \ L is an embedded disk, transverse <strong>to</strong> the Reeb vec<strong>to</strong>r field, so that<br />

ı(0) is a point on γ, and so that ı ∗ dλ = dx ∧ dy.<br />

64


Let x(s, t) and y(s, t) be the projections <strong>of</strong> u(s, t) on <strong>to</strong> Dr by means <strong>of</strong> the<br />

Reeb flow, where these are defined. Let z(s, t) so that φz(s,t)(ı(x(s, t), y(s, t)) =<br />

u(s, t). (We note this is not a priori defined for all s, t.)<br />

Then, there exist constants a0 ∈ R, r > 0, so that for each k, l ∈ N, and<br />

for each 0 < ρ < min{r/2, 1/T }, there exist constants ck,l so that for all ɛ > 0,<br />

there exists Sɛ so that for all s ≥ Sɛ, x(s, t), y(s, t) and z(s, t) are defined for<br />

s ≥ Sɛ and we have<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l tx(s, t)| ≤ ck,le −rs<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l ty(s, t)| ≤ ck,le −rs<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l t(z(s, t) − t)| ≤ ck,le −ρs<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l t(a(s, t) − a0 − s)| ≤ ck,le −ρs .<br />

We may say a bit more, in fact : we may find a representation <strong>of</strong> ζ(s, t) :=<br />

(x(s, t), y(s, t)) in terms <strong>of</strong> an eigenvec<strong>to</strong>r <strong>of</strong> an unbounded opera<strong>to</strong>r. To make<br />

this precise, we will put very specific symplectic coordinates on the disk, Dr.<br />

Let ℓu and ℓs be the unit unstable and stable Legendrians, respectively. Then,<br />

projecting them on Dr by means <strong>of</strong> the Reeb flow, we have intersecting Lagrangians.<br />

Since γ was assumed non-degenerate, they intersect transversally.<br />

We may make a symplectic change <strong>of</strong> variables so they are given by R and<br />

iR respectively. We let (x, y) denote the corresponding coordinates. We let<br />

ζ(s, t) = (x(s, t), y(s, t)), the projection <strong>of</strong> u(s, t) <strong>to</strong> the disk by following the<br />

Reeb flow.<br />

We now introduce a family <strong>of</strong> almost complex structures on Dr by using<br />

the projection along the Reeb flow : for each z, we define M(x, y, z) <strong>to</strong> be<br />

the almost complex structure on Dr given by projecting the almost complex<br />

structure ˜ J|ξ using the Reeb flow <strong>to</strong> time z. For each z, M(·, ·, z) is an almost<br />

complex structure on Dr, compatible with dx ∧ dy.<br />

We have then that ζ satisfies the following non-linear Cauchy–Riemann equation<br />

:<br />

∂<br />

∂<br />

ζ + M(ζ, z(s, t)) ζ = 0.<br />

∂s ∂t<br />

Let M0(t) = M(0, 0, t). We introduce the function space<br />

W 1,2<br />

ℓ ([0, T ], R 2 ) = {f ∈ W 1,2 ([0, T ], R 2 ) | f(0) ∈ R · e1 and f(1) ∈ R · e2}<br />

and we equip L 2 ([0, T ], R 2 ) with the inner product T<br />

0 ( dx ∧ dy)(·, M0(t)·) dt.<br />

(This gives a norm equivalent <strong>to</strong> the standard norm.)<br />

65


We then define an unbounded linear opera<strong>to</strong>r by :<br />

We now have :<br />

A∞ : L 2 ([0, T ], R 2 ) ⊃ W 1,2<br />

ℓ ([0, T ], R 2 ) −→ L 2 ([0, T ], R 2 )<br />

γ ↦→ −M0(·) ˙γ(·).<br />

Theorem 4.4.13. If ζ is not identically zero,<br />

where<br />

R s<br />

s Λ(τ) dτ<br />

ζ(s, t) = e 0 (e(t) + r(s, t))<br />

A∞ is self-adjoint (with respect <strong>to</strong> the new inner product),<br />

e ∈ W 1,2<br />

ℓ ([0, T ], R 2 ) is an eigenvec<strong>to</strong>r <strong>of</strong> A∞ corresponding <strong>to</strong> an eigenvalue<br />

λ < 0,<br />

Λ(s) : [s0, ∞) → R is a smooth function with Λ(s) → λ as s → ∞,<br />

r : [s0, ∞) × [0, T ] → R decays <strong>to</strong> zero uniformly in t as s → ∞, <strong>to</strong>gether with<br />

all <strong>of</strong> its derivatives.<br />

In order <strong>to</strong> prove the asymp<strong>to</strong>tic convergence results, we will use work done<br />

by Abbas in [2]. In [2], he considers a (smooth) three dimensional contact<br />

manifold with contact form λ and a Legendrian knot ℓ. He then considers the<br />

boundary value problem <strong>of</strong> considering a pseudoholomorphic half-plane ũ : H →<br />

R×M with Lagrangian boundary condition u(0+it) ∈ ℓ. He shows that a finite<br />

energy half-plane <strong>of</strong> this kind is asymp<strong>to</strong>tic <strong>to</strong> a Reeb chord. Furthermore, if the<br />

Reeb chord is non-degenerate (analogous <strong>to</strong> our condition in Definition 4.4.2),<br />

he proves versions <strong>of</strong> our theorems 4.4.12 and 4.4.13. Since these analytical<br />

questions are local in nature, we may use our local model <strong>of</strong> the neighbourhood<br />

<strong>of</strong> a homoclinic orbit, Proposition 4.2.6, in order <strong>to</strong> reduce our problem <strong>to</strong> the<br />

problem studied by Abbas.<br />

We cite the following results from Abbas, in [2]. The original statements <strong>of</strong><br />

these theorems are in a more general setting, but his pro<strong>of</strong>s reduce the problem<br />

<strong>to</strong> the local situation described here.<br />

In the following two theorems, we consider a neighbourhood, V , in R 3 <strong>of</strong><br />

{(0, 0)} × [0, T ], for some T > 0. We have two Legendrian curves in V , ℓ0<br />

and ℓ1, with ℓ0 = R × {(0, 0)} ∩ V and with ℓ1 = {(x(τ), τ, z(τ) | τ small}<br />

and x(0) = 0 and z(0) = T . This is the local model <strong>of</strong> any non-degenerate<br />

Legendrian chord, in the sense <strong>of</strong> [2]. In our case, this is the image <strong>of</strong> the local<br />

diffeomorphism given by Proposition 4.2.6.<br />

66


Theorem 4.4.14 ([2, Theorem 1.3]). Suppose ũ : [s0, ∞) × [0, T ] → R × V<br />

is pseudoholomorphic with respect <strong>to</strong> a smooth almost complex structure ˜ J, adjusted<br />

<strong>to</strong> λ, with finite energy, and satisfies u(S + it) → (0, 0, t) as S → ∞. Let<br />

us write u(s, t) = (x(s, t), y(s, t), z(s, t)). Then, there exist constants a0 ∈ R,<br />

r > 0, S ≥ s0 so that for each k, l ∈ N, and for each 0 < ρ < min{r/2, 1/T },<br />

there exist constants ck,l so that for all s ≥ S, we have :<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l tx(s, t)| ≤ ck,le −rs<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l ty(s, t)| ≤ ck,le −rs<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l t(z(s, t) − t)| ≤ ck,le −ρs<br />

sup |∂<br />

0≤t≤T<br />

k s ∂ l t(a(s, t) − a0 − s)| ≤ ck,le −ρs<br />

This result is extended <strong>to</strong> a representation formula for the component <strong>of</strong> u<br />

transverse <strong>to</strong> the Reeb vec<strong>to</strong>r field. First, we consider the symplectic surface<br />

Σ := {z = 0}. This is transverse <strong>to</strong> the Reeb vec<strong>to</strong>r field, so we may use<br />

the Reeb flow <strong>to</strong> map any point (x, y, z) <strong>to</strong> (x, y, 0). Furthermore, by projecting<br />

along ∂<br />

∂z , we may find an isomorphism between ξ(x,y,0 and T(x,y,0)Σ. This induces<br />

a symplectic form on the surface, which in this case is dx∧ dy. Furthermore, We<br />

see that ℓ0 and ℓ1 are Lagrangian curves in Σ. Thus, we may find a symplectic<br />

change <strong>of</strong> variables so that ℓ0 = R · ∂<br />

∂x ′ and ℓ1 = R · ∂<br />

∂y ′ . For p ∈ V , let ρp<br />

denote the induced map from ξp <strong>to</strong> R 2 . Then, we introduce the almost complex<br />

structure M(x, y, z) by ρx,y,z ◦ J(x, y, z) = M(x, y, z) ◦ ρp. For each fixed z,<br />

this is an almost complex structure on Σ compatible with the symplectic form<br />

dx ∧ dy.<br />

Let ζ(s, t) be the representation <strong>of</strong> (x(s, t), y(s, t)) in the x ′ , y ′ coordinates<br />

introduced on Σ. We have then that ζ satisfies the following non-linear Cauchy–<br />

Riemann equation :<br />

∂<br />

∂<br />

ζ + M(ζ, z(s, t)) ζ = 0.<br />

∂s ∂t<br />

Let M0(t) = M(0, 0, t). We introduce the function space<br />

W 1,2<br />

ℓ ([0, T ], R 2 ) = {f ∈ W 1,2 ([0, T ], R 2 ) | f(0) ∈ R · e1 and f(1) ∈ R · e2}<br />

and we equip L 2 ([0, T ], R 2 ) with the inner product T<br />

0 ( dx ∧ dy)(·, M0(t)·) dt.<br />

(This gives a norm equivalent <strong>to</strong> the standard norm.)<br />

We then define an unbounded linear opera<strong>to</strong>r by :<br />

A∞ : L 2 ([0, T ], R 2 ) ⊃ W 1,2<br />

ℓ ([0, T ], R 2 ) −→ L 2 ([0, T ], R 2 )<br />

γ ↦→ −M0(·) ˙γ(·).<br />

67


Then, the result <strong>of</strong> Abbas gives :<br />

Theorem 4.4.15 ([2, Theorem 1.4]). If ζ is not identically zero,<br />

where<br />

ζ(s, t) = e<br />

R s<br />

s Λ(τ) dτ<br />

0 (e(t) + r(s, t))<br />

A∞ is self-adjoint (with respect <strong>to</strong> the new inner product),<br />

e ∈ W 1,2<br />

ℓ ([0, T ], R 2 ) is an eigenvec<strong>to</strong>r <strong>of</strong> A∞ corresponding <strong>to</strong> an eigenvalue<br />

λ < 0,<br />

Λ(s) : [s0, ∞) → R is a smooth function with Λ(s) → λ as s → ∞,<br />

r : [s0, ∞) × [0, T ] → R decays <strong>to</strong> zero uniformly in t as s → ∞, <strong>to</strong>gether with<br />

all <strong>of</strong> its derivatives.<br />

Remark 4.4.16. Robbin and Salamon, in [35], provide a summary <strong>of</strong> the key<br />

ideas <strong>of</strong> the pro<strong>of</strong>s <strong>of</strong> these results, which are carried out carefully in [2].<br />

Now, finally, we are able <strong>to</strong> prove our versions <strong>of</strong> these theorems. Our pro<strong>of</strong>s<br />

are straightforward combinations <strong>of</strong> Abbas’s work with our local model theorem.<br />

Pro<strong>of</strong> <strong>of</strong> Theorems 4.4.12 and 4.4.13. Denote by Φ the local diffeomorphism<br />

given by Proposition 4.2.6, and U, the neighbourhood <strong>of</strong> γ on which it is defined,<br />

and V ⊂ R 3 , its image. We denote by ℓ0 and ℓ1, the images <strong>of</strong> L at γ(0)<br />

and γ(T ). We may take Φ so that Φ is a diffeomorphism in a neighbourhood <strong>of</strong><br />

the image <strong>of</strong> Dr.<br />

By Theorem 4.4.11, u(s, t) has image in U for s sufficiently large. Let<br />

˜v = (a, Φ(u). We have that ˜v is a pseudoholomorphic curve in<strong>to</strong> R × V, with<br />

boundary in R × ℓi, i = 0, 1. We write v(s, t) = (x ′ (s, t), y ′ (s, t), z ′ (s, t)) where<br />

(x ′ , y ′ ) are the coordinates as in the result <strong>of</strong> Abbas (see Theorem 4.4.14).<br />

We observe that where the coordinates x, y and z are defined (as in the<br />

theorem statement), Φ(x, y, z) = (x ′ , y ′ , z ′ ).<br />

We may now apply the result <strong>of</strong> Abbas, Theorem 4.4.14. By virtue <strong>of</strong><br />

the exponential convergence <strong>of</strong> z(s, t) <strong>to</strong> t, we have that for each t in the<br />

interior <strong>of</strong> (0, T ), for s large enough, z ′ (s, t) > 0. It then follows that<br />

(x ′ (s, t), y ′ (s, t), z ′ (s, t)) have a unique lift, and Theorem 4.4.12 now follows.<br />

Similarly, we apply Theorem 4.4.15 <strong>to</strong> ˜v. Again, we have that for each<br />

compact subset <strong>of</strong> (0, T ), for s sufficiently large, there is a unique lift, and thus<br />

Φ tells us the whole s<strong>to</strong>ry.<br />

Remark 4.4.17. We note that in this pro<strong>of</strong>, we have a good estimate on how<br />

large s must be. An avenue for future work is <strong>to</strong> study how the strip winds<br />

about the singular Legendrian. A good control on this will allow for a stronger<br />

asymp<strong>to</strong>tic statement.<br />

68


4.5 Future work<br />

This work with singular almost complex structures raises many questions. The<br />

first question is about the relationship between Type (I) and Type (II) pseudoholomorphic<br />

curves. This could potentially lead <strong>to</strong> a new generalization <strong>of</strong> the<br />

Gromov removal <strong>of</strong> singularities theorem. Another problem that needs addressing<br />

is the compactification <strong>of</strong> the space <strong>of</strong> Type (II) pseudoholomorphic curves.<br />

And fundamental is the question <strong>of</strong> the relationship between these curves and<br />

the (standard) pseudoholomorphic curves on nearby regular energy levels.<br />

All <strong>of</strong> these questions come out <strong>of</strong> the larger background project <strong>to</strong> understand<br />

the change in contact homology caused by Legendrian surgery. Ultimately,<br />

the goal is <strong>to</strong> understand the relationship between the various relevant symplectic<br />

and contact invariants : contact homology for M ± , symplectic field theory<br />

for the symplectic cobordism given by W , and the yet-<strong>to</strong>-be-developed homoclinic<br />

contact homology. In turn, this should give a lot <strong>of</strong> information about<br />

Hamil<strong>to</strong>nian dynamics in relation <strong>to</strong> these invariants.<br />

69


Chapter 5<br />

Energy quantization for<br />

pseudoholomorphic curves with<br />

a relaxed area bound<br />

5.1 Overview<br />

We are interested in studying the behaviour <strong>of</strong> a pseudoholomorphic plane<br />

ũ : C → R × M<br />

with merely a bound on the contact area :<br />

<br />

0 < u ∗ dλ = C < ∞. (5.1)<br />

We note that such a plane is not, a priori, a finite energy curve. We recall that<br />

a pseudoholomorphic curve ũ is a finite energy curve if<br />

<br />

sup{ ũ ∗ d(φλ) | φ : R → [0, 1] with φ ′ ≥ 0} < ∞.<br />

We denote the left hand side <strong>of</strong> this inequality by E(ũ), and call it the E-energy<br />

<strong>of</strong> the curve ũ. This energy was introduced by H<strong>of</strong>er in [21]. Finiteness <strong>of</strong> this<br />

energy is necessary for all <strong>of</strong> the existing work on compactness for pseudoholomorphic<br />

curves in symplectizations. Infinite energy curves can be very badly<br />

behaved (as our example below will show) — in particular, such a map will not<br />

be proper!<br />

We will prove the following result :<br />

70


Theorem 5.1.1. Suppose ũ : C → R × M is pseudoholomorphic with respect <strong>to</strong><br />

an almost complex structure ˜ J, adjusted <strong>to</strong> the contact form λ, and has finite<br />

contact area :<br />

<br />

0 < u ∗ dλ = C < ∞.<br />

C<br />

We also impose the technical condition that πλ T u never vanishes.<br />

Then, there exists a periodic orbit <strong>of</strong> the Reeb vec<strong>to</strong>r field with action 0 < T ≤ C.<br />

This result establishes an energy threshold for the contact area <strong>of</strong> a pseudoholomorphic<br />

curve. This paves the way for establishing an energy quantization<br />

result. This is part <strong>of</strong> an ongoing project with Abbas and H<strong>of</strong>er <strong>to</strong> understand<br />

the behaviour <strong>of</strong> these curves with infinite E-energy, but finite dλ contact area.<br />

The theory <strong>of</strong> pseudoholomorphic curves in<strong>to</strong> symplectizations with an Eenergy<br />

bound is very well developed, with a good understanding <strong>of</strong> the compactification<br />

<strong>of</strong> the space <strong>of</strong> such curves [6] and <strong>of</strong> how they may be perturbed<br />

[25]. The study <strong>of</strong> them has led <strong>to</strong> some very interesting results in dynamical<br />

systems, for instance [27], and <strong>to</strong> the pro<strong>of</strong> <strong>of</strong> the Weinstein conjecture for many<br />

classes <strong>of</strong> contact manifolds.<br />

In order <strong>to</strong> solve the Weinstein conjecture for all contact 3-manifolds, H<strong>of</strong>er<br />

has introduced a programme <strong>to</strong> study a generalization <strong>of</strong> the pseudoholomorphic<br />

curve equation. The goal is <strong>to</strong> exploit a result <strong>of</strong> Giroux’s in order <strong>to</strong> obtain<br />

existence for a finite energy curve. Giroux proved that every contact structure<br />

on a 3–manifold is supported by an open book [18]. Furthermore, Giroux’s construction<br />

exhibits a “nice” contact form generating the contact structure. The<br />

idea, then, is <strong>to</strong> take such an open book, with its contact form, and construct<br />

a pseudoholomorphic curve by a construction on a leaf <strong>of</strong> the open book. This<br />

exploits the special structure <strong>of</strong> the Giroux contact form. This construction is<br />

done by Abbas, in [1]. One then takes a homo<strong>to</strong>py from the Giroux contact form<br />

<strong>to</strong> the contact form <strong>of</strong> interest. The goal is <strong>to</strong> show that a pseudoholomorphic<br />

curve persists through the homo<strong>to</strong>py.<br />

The case in which the leaves <strong>of</strong> the Giroux foliation have no genus is taken<br />

care <strong>of</strong> in [3]. The case in which the leaves have genus is more subtle. The<br />

Cauchy–Riemann opera<strong>to</strong>r is Fredholm, but with negative index. Thus, there<br />

is no way <strong>to</strong> obtain transversality. Instead, as discussed in [3], H<strong>of</strong>er has introduced<br />

a twisted pseudoholomorphic curve equation.<br />

We let (Σ, j) be a Riemann surface, and J an almost complex structure on<br />

the contact structure, compatible with the symplectic form dλ|ξ. We have a<br />

finite set <strong>of</strong> punctures Γ ⊂ Σ and denote ˙ Σ := Σ\Γ. Let πλ be the projection <strong>to</strong><br />

ξ = ker λ along the Reeb vec<strong>to</strong>r field. The goal is <strong>to</strong> show existence <strong>of</strong> a tuple<br />

71


(Σ, j, Γ, ũ, γ), where<br />

ũ = (a, u) : ˙ Σ −→ R × M<br />

πλ T u ◦ j = J ◦ πλ T u on ˙ Σ<br />

(u ∗ λ) ◦ j = da + γ on ˙ Σ<br />

and dγ = 0 = d(γ ◦ j) on Σ.<br />

This PDE reduces <strong>to</strong> the standard Cauchy–Riemann equation if γ = 0. The<br />

energy used for this generalized Cauchy–Riemann equation remains the same<br />

as before :<br />

<br />

E(ũ) = sup{<br />

ũ ∗ d(φλ) | φ : R → [0, 1] and φ ′ ≥ 0}.<br />

The key point being that this energy does not keep track <strong>of</strong> the harmonic form<br />

γ.<br />

This new problem is Fredholm, <strong>of</strong> the correct index and can be made surjective.<br />

However, the existing compactness theory is no longer applicable. In<br />

particular, one <strong>of</strong> the objects which may bubble <strong>of</strong>f during the homo<strong>to</strong>py is a<br />

pseudoholomorphic plane with no bound on the E-energy. The contact area,<br />

however, will be non-zero, and bounded. This is the problem that leads us <strong>to</strong><br />

study planes with control only on the contact area (5.1).<br />

Remark 5.1.2. There exist pseudoholomorphic curves with (non-zero) finite contact<br />

area (5.1), but infinite E-energy. Let (M, ξ) be a contact manifold with<br />

contact form λ. Suppose there exists an adjusted almost complex structure ˜ J<br />

on R×M, invariant under the Reeb flow (this is a strong assumption). Suppose<br />

also that there exists a finite energy plane ũ0 = (a0, u0) : C → R × M. We now<br />

observe that if f + ig : C → C is holomorphic (classically) then<br />

˜v(z) := (a0(z) + f(z), φf(z) ◦ u0(z))<br />

is also pseudoholomorphic.<br />

We may take f + ig so that |∇f| is bounded from below, for |z| ≥ R (for<br />

instance, f + ig = zk ). We observe that<br />

<br />

E(˜v) ≥ sup{ φ ′ (b(s, t))|∇b(s, t)| 2 ds dt | φ : R → [0, 1] and φ ′ ≥ 0}.<br />

C<br />

We have |∇b(s, t)| 2 ≥ |∇f(s, t)| 2 ≥ c > 0 for |z| ≥ R. Let us also restrict<br />

our attention <strong>to</strong> φ for which φ ′ has compact support. Then, we have (where µ<br />

72


denotes the Lebesgue measure on R2 ):<br />

<br />

φ ′ (b(s, t))|∇b(s, t)| 2 <br />

ds dt ≥ c φ ′ (b(s, t)) ds dt<br />

C<br />

<br />

= c<br />

<br />

= c<br />

|z|≥R<br />

|z|≥R<br />

b(s,t)<br />

−∞<br />

φ ′′ (τ) dτ ds dt<br />

φ<br />

R<br />

′′ (τ)µ{(s, t) | b(s, t) ≥ τ and s 2 + t 2 ≥ R 2 } dτ.<br />

From the asymp<strong>to</strong>tic formula for a(s, t), as derived in [24], a(z) is asymp<strong>to</strong>tic<br />

<strong>to</strong> c1 log(|z| 2 )+c2, where c1 > 0. Thus, {(s, t) | b(s, t) ≥ τ} has infinite measure.<br />

We may choose a φ for which φ ′′ is positive on an interval. It follows that the<br />

E energy <strong>of</strong> v is now infinite. It also follows by S<strong>to</strong>kes theorem that the contact<br />

area <strong>of</strong> ˜v is the same as the contact area <strong>of</strong> ũ. Thus, ˜v has infinite E-energy,<br />

but finite contact area.<br />

An example <strong>of</strong> a manifold and contact form admitting an almost complex<br />

structure invariant under the Reeb flow is S 3 with the standard contact form<br />

(corresponding <strong>to</strong> the Hopf fibration). The symplectization R × S 3 may be<br />

identified with C 2 \ {0}. The standard complex structure on C 2 is adjusted <strong>to</strong><br />

the contact form. This complex structure is invariant under the action <strong>of</strong> the<br />

Reeb flow.<br />

5.2 “Renormalization” by the Reeb flow<br />

The pro<strong>of</strong> <strong>of</strong> Theorem 5.1.1 uses a key construction, which we call renormalization.<br />

It is inspired by the example in Remark 5.1.2, where the infinite energy<br />

pseudoholomorphic curve is a graph over a finite energy curve, by means <strong>of</strong> the<br />

Reeb flow.<br />

We let<br />

<br />

0 < C = u ∗ dλ < ∞.<br />

C<br />

We will consider our original infinite E-energy curve, ũ, restricted <strong>to</strong> a disk<br />

<strong>of</strong> radius R. We now consider the immersed cylinder L obtained by flowing the<br />

image <strong>of</strong> the boundary, u(∂DR), by means <strong>of</strong> the Reeb flow. Our goal is <strong>to</strong><br />

find a pseudoholomorphic disk ˜v = (b, v) for which b(∂DR) = const and with<br />

boundary in L (<strong>of</strong> degree 1 on the S1 fac<strong>to</strong>r). In order <strong>to</strong> construct this disk, we<br />

will consider a homo<strong>to</strong>py through pseudoholomorphic curves with boundary in<br />

the cylinder L. By S<strong>to</strong>kes theorem, we have that the dλ area <strong>of</strong> any curve with<br />

(degree 1) boundary in L is the same as the dλ area <strong>of</strong> ũ|DR . Furthermore, by<br />

73


the boundary behaviour <strong>of</strong> ˜v and S<strong>to</strong>kes theorem, we have that<br />

<br />

E(˜v) = v ∗ <br />

dλ = u ∗ dλ ≤ C < ∞.<br />

Thus, we have an a priori bound on the E-energy <strong>of</strong> the curve ˜v obtained in<br />

this way. This new curve, ˜v, represents a renormalization <strong>of</strong> ũ|DR by means <strong>of</strong><br />

the Reeb flow.<br />

In order <strong>to</strong> prove Theorem 5.1.1, we describe a boundary value problem<br />

corresponding <strong>to</strong> finding a homo<strong>to</strong>py from our original curve ũ|DR <strong>to</strong> the renormalized<br />

˜v :<br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

(b, v) : DR → R × M<br />

πλ T v ◦ i = J(v) ◦ πλ T v<br />

v ∗ λ ◦ i = db<br />

b(z) = τ a(β(z)) for z ∈ ∂DR<br />

where β : ∂DR → ∂DR is <strong>of</strong> degree 1<br />

v(z) = φf(z)(u(β(z))) for z ∈ ∂DR<br />

where f : ∂DR → R, and f(R) = 0.<br />

DR<br />

(5.2)<br />

Our original curve corresponds <strong>to</strong> the homo<strong>to</strong>py parameter τ = 1. The<br />

renormalized curve corresponds <strong>to</strong> τ = 0.<br />

We prove that if we have a solution for a value τ0 > 0, then there exist<br />

solutions for τ sufficiently close <strong>to</strong> τ0. This follows by a straightforward application<br />

<strong>of</strong> well-known properties <strong>of</strong> the non-linear Cauchy-Riemann opera<strong>to</strong>r. We<br />

discuss this in section 5.3.3.<br />

We also prove that non-compactness <strong>of</strong> the solution space gives us one <strong>of</strong><br />

two outcomes : either there exists a periodic Reeb orbit <strong>of</strong> period T ≤ C, or<br />

we “short-circuit” the homo<strong>to</strong>py and there exists a curve ˜v : DR → R × M,<br />

satisfying (5.2) for τ = 0. This is proved in section 5.3.2.<br />

Thus, we reduce the problem <strong>to</strong> the case that for any R > 0, we may find a<br />

curve ˜vR that renormalizes ũ|DR .<br />

We now take a sequence <strong>of</strong> Rn → ∞. We consider then our original curve<br />

restricted <strong>to</strong> the disks <strong>of</strong> radius Rn, ũn := ũ|Rn. This is now a sequence <strong>of</strong><br />

pseudoholomorphic disks with a uniform bound on the contact area, but no<br />

bound on the E energy.<br />

By means <strong>of</strong> the renormalization, for each Rn, we obtain a curve ˜vn. These<br />

curves all have uniformly bounded E-energy. Thus, we either have gradient<br />

bounds, and thus they converge <strong>to</strong> a pseudoholomorphic plane with finite Eenergy,<br />

or they bubble <strong>of</strong>f a pseudoholomorphic plane with finite E-energy.<br />

Then, by the usual compactness results for finite energy pseudoholomorphic<br />

curves, there must exist a Reeb orbit <strong>of</strong> period T ≤ C.<br />

74


5.3 Pro<strong>of</strong> <strong>of</strong> the result<br />

In the following, M will denote a three dimensional manifold equipped with a<br />

contact form λ, generating the contact structure ξ = ker λ. We will denote by<br />

πλ : T M → ker λ the projection along the Reeb vec<strong>to</strong>r field Xλ. Let J : ker λ →<br />

ker λ be a complex structure compatible with dλ. We then define ˜ J <strong>to</strong> be the<br />

almost complex structure on R × M with ˜ J ∂<br />

∂a = Xλ and ˜ J|ξ = J. We will now<br />

prove<br />

Theorem 5.1.1. Assume we have a non constant solution ũ = (a, u) <strong>of</strong><br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

ũ : C −→ R × M<br />

πλ T u ◦ i = J(u) ◦ πλ T u<br />

u ∗ λ ◦ i = da<br />

0 < <br />

C u∗ dλ < ∞<br />

We also assume that πλ T u never vanishes.<br />

Then there exists a periodic trajec<strong>to</strong>ry x <strong>of</strong> the Reeb vec<strong>to</strong>r field such that<br />

<br />

u ∗ <br />

dλ ≥ x ∗ λ.<br />

C<br />

S 1<br />

(5.3)<br />

Our pro<strong>of</strong> will be by contradiction. We assume that no such periodic orbit<br />

exists. We consider then ũ restricted <strong>to</strong> a disk <strong>of</strong> radius R. As discussed above,<br />

we will find a homo<strong>to</strong>py between this and a pseudoholomorphic ˜v = (b, v) on<br />

the disk, with, at the boundary, v = φf ◦ u (where φt denotes the Reeb flow)<br />

and b = constant. The map ˜v will then have both dλ and E energy equal <strong>to</strong><br />

the dλ-energy <strong>of</strong> u|DR .<br />

In this and in the following, we denote DR = {z ∈ C | |z| ≤ R}, the closed<br />

disk <strong>of</strong> radius R. We will also write D for the closed disk <strong>of</strong> radius 1.<br />

After repeating this construction for a sequence <strong>of</strong> R → ∞, we obtain a<br />

sequence <strong>of</strong> pseudoholomorphic curves with uniformly bounded energy. A subsequence<br />

then converges <strong>to</strong> a finite energy plane asymp<strong>to</strong>tic <strong>to</strong> a periodic orbit<br />

<strong>of</strong> action no greater than <br />

C u∗ dλ. This contradiction then establishes the result.<br />

First, we will describe the boundary value problem that will give us the<br />

homo<strong>to</strong>py. Then, we will prove a compactness result for the solutions <strong>of</strong> this<br />

boundary value problem. In particular, we will show that non-compactness<br />

implies the existence <strong>of</strong> a curve ˜v, which was the “goal” <strong>of</strong> the homo<strong>to</strong>py. Finally,<br />

we will prove that this problem is elliptic and satisfies a Fredholm theory.<br />

5.3.1 The boundary value problem<br />

75


We recall that we are interested in studying solutions (˜v, τ) <strong>to</strong> the boundary<br />

value problem (5.2) :<br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

(b, v) : DR → R × M<br />

πλ T v ◦ i = J(v) ◦ πλ T v<br />

v ∗ λ ◦ i = db<br />

b(z) = τ a(β(z)) for z ∈ ∂DR<br />

where β : ∂DR → ∂DR is <strong>of</strong> degree 1<br />

v(z) = φf(z)(u(β(z))) for z ∈ ∂DR<br />

where f : ∂DR → R, and f(R) = 0.<br />

(5.2)<br />

We note that for each τ, the boundary condition is <strong>to</strong>tally real. We denote<br />

L := {φt(u(z)) | z ∈ ∂DR , t ∈ R}.<br />

We observe that {0} × L is an immersed Lagrangian submanifold <strong>of</strong> (R ×<br />

M, d(etλ)). For each τ, we will take the boundary condition <strong>to</strong> lie in the immersed,<br />

<strong>to</strong>tally real submanifold Lτ := {(τa(z), φt(u(z)) | z ∈ ∂DR , t ∈ R}.<br />

This immersed submanifold may be made Lagrangian by twisting the symplectic<br />

structure on R × M. In this sense, we have a homo<strong>to</strong>py <strong>of</strong> Lagrangian<br />

boundary conditions.<br />

We also observe that for τ = 1, ˜v = ũ|DR<br />

is a solution.<br />

We will prove that assuming no periodic orbit exists, for each R > 0, there<br />

must exist a disk ˜v that satisfies (5.2) for τ = 0.<br />

In order <strong>to</strong> do this, we will first show that if we have a sequence <strong>of</strong> solutions<br />

(˜vτ, τ), for τ ∈ (0, 1], then either we have a subsequence that converges <strong>to</strong> a<br />

solution <strong>to</strong> (5.2), or we “short-circuit” the homo<strong>to</strong>py and bubble <strong>of</strong>f a curve ˜v<br />

that solves (5.2) for τ = 0. This part involves some careful a priori estimates<br />

and the introduction <strong>of</strong> a variant on H<strong>of</strong>er’s E energy.<br />

We will then show that the set <strong>of</strong> τ for which a solution exists, is open.<br />

We will suppose we have a solution for some τ0 and show a solution exists for<br />

nearby τ. In order <strong>to</strong> do this, we will view solutions <strong>to</strong> the PDE problem (5.2)<br />

as zeros <strong>of</strong> a section <strong>of</strong> a Banach space bundle. We will linearize the section at<br />

our solution, and then show that the linearized opera<strong>to</strong>r is a surjective Fredholm<br />

opera<strong>to</strong>r. This part will be developed in Section 5.3.3. This uses well established<br />

methods, as in for instance [33, 4].<br />

Finally, we will conclude that the existence, for each R, <strong>of</strong> a solution <strong>to</strong> (5.2)<br />

for τ = 0 gives us Theorem 5.1.1.<br />

5.3.2 Compactness <strong>of</strong> disks<br />

76


We will now study compactness <strong>of</strong> solutions <strong>to</strong> the problem (5.2). This boundary<br />

value problem features some important non-compactness phenomena. We recall<br />

that pseudoholomorphic curves satisfy an interior regularity estimate so that (in-<br />

terior) gradient bounds gives (interior) C ∞ loc<br />

bounds. Furthermore, they satisfy<br />

a form <strong>of</strong> Gromov compactness : essentially, non-compactness (gradient blow<br />

up) can be explained by the non-compactness <strong>of</strong> the space <strong>of</strong> bi-holomorphic<br />

maps on the domain.<br />

The most difficult non-compactness problem comes from the boundary condition.<br />

We require our curves <strong>to</strong> have boundary in an immersed Lagrangian<br />

whose image in the manifold can be quite wild. In order <strong>to</strong> deal with this, we<br />

will establish an a priori bound on how far the boundary condition can flow.<br />

In other words, we will show that the solutions <strong>to</strong> (5.2) only “see” a finite piece<br />

<strong>of</strong> the boundary condition L, an immersed [−T, T ] × S1 for some T > 0, large.<br />

The easiest problem <strong>to</strong> deal with is interior bubbling. In this case, we conclude<br />

that interior bubbling gives the existence <strong>of</strong> a finite energy plane. This,<br />

by [21], is then asymp<strong>to</strong>tic <strong>to</strong> a periodic orbit whose action equals the dλ area<br />

<strong>of</strong> the plane. If this were <strong>to</strong> occur, we would then be done.<br />

Finally, we deal with the problem <strong>of</strong> boundary bubbling. In this case, by<br />

virtue <strong>of</strong> the a priori bound that forces our curves <strong>to</strong> have boundary in a<br />

finite portion <strong>of</strong> L, we have that if we have boundary bubbling, after rescaling,<br />

we obtain a curve that satisfies (5.2) for τ = 0. In a sense then, boundary<br />

bubbling immediately “short-circuits” the renormalization process and gives us<br />

the renormalized curve without going through the whole homo<strong>to</strong>py. In this<br />

case, we don’t obtain compactness. Instead, we have that non-compactness is<br />

as good as compactness for our purposes.<br />

Let us now make this precise. Suppose we have a sequence <strong>of</strong> τn → τ ≥ 0<br />

<strong>to</strong>gether with curves ˜vn : DR → R × M satisfying the boundary value problem<br />

(5.2). We need <strong>to</strong> establish this sequence must have uniformly bounded gradient<br />

and hence a subsequence converges in C ∞ loc<br />

on the interior <strong>of</strong> the disk. We also<br />

need <strong>to</strong> establish a bound on ||fn||L∞ <strong>to</strong> show that the limiting disk satisfies the<br />

boundary condition.<br />

A priori bounds on |f|.<br />

We will first establish an a priori bound on the amount by which we flow the<br />

boundary in (5.2). We will establish this by establishing that the f we obtain is<br />

<strong>of</strong> bounded variation. This, combined with the constraint that f(R) = 0 gives<br />

us an L ∞ bound on f. We note that this result relies on the fact that we may<br />

estimate the Maslov index <strong>of</strong> the boundary condition in terms <strong>of</strong> the orders <strong>of</strong><br />

the zeros <strong>of</strong> πλ T u on DR.<br />

77


Lemma 5.3.1. Let (˜v, τ) be a solution <strong>to</strong> (5.2) for some value <strong>of</strong> 0 < τ ≤ 1, with<br />

˜v = (b, v). We lift β : ∂DR → ∂DR <strong>to</strong> a map β : R → R with β(t+1) = β(t)+1.<br />

Then,<br />

• The function β satisfies ˙ β > 0,<br />

• There is an a priori bound on the L ∞ norm <strong>of</strong> f, independent <strong>of</strong> τ, b and<br />

v. This means that any solution <strong>of</strong> (5.2) has its boundary lying a finite<br />

part {φT (u(z)) | z ∈ ∂DR , |T | ≤ C} ⊂ L<br />

Pro<strong>of</strong>. The first item follows immediately since the Maslov index <strong>of</strong> L is given<br />

by twice the sum <strong>of</strong> the order <strong>of</strong> interior zeros <strong>of</strong> πλ T v plus the sum <strong>of</strong> the orders<br />

<strong>of</strong> the boundary zeros. Now, πλ T v satisfies a first order Cauchy-Riemann type<br />

PDE [23]. Thus, its zeros occur with positive multiplicity. For our original<br />

curve, πλ T u is non-vanishing, so it follows that πλ T v must be non-vanishing.<br />

In particular, πλ T v cannot have a zero at the boundary. Hence, ˙ β > 0.<br />

The second result follows since b(z) is subharmonic. Indeed, we have that if<br />

f : Σ → R is a smooth function and j a complex structure on Σ inducing the<br />

orientation ds ∧ dt on Σ, then we define ∆jf by<br />

−d(df ◦ j) = ∆jf ds ∧ dt.<br />

The opera<strong>to</strong>r ∆j is a uniformly elliptic opera<strong>to</strong>r without any zero order term.<br />

In particular, the weak and strong maximum principles apply. Our case is<br />

particularly simple since the complex structure j = i on the disk. Then, ∆i =<br />

∆ = ∂2<br />

∂s2 + ∂2<br />

∂t2 , the standard Laplacian. We note that b is subharmonic since<br />

−d(db ◦ jρ) = v∗dλ ≥ 0.<br />

We will show an a priori bound on the normal derivative <strong>of</strong> b at the boundary.<br />

In order <strong>to</strong> do this, we need <strong>to</strong> have a zero boundary condition, which b<br />

does not satisfy. To get around this, we will use the following well–known PDE<br />

trick (pointed out <strong>to</strong> me by Casim Abbas).<br />

We solve the following boundary value problem<br />

<br />

◦<br />

∆δ = 0 on DR<br />

δ(z) = a(β(z)) on ∂DR<br />

Consider now the function b − τ δ which satisfies<br />

<br />

◦<br />

DR<br />

∆(b − τ δ) ≥ 0 on<br />

b − τ δ = 0 on ∂DR<br />

78


With ν denoting the outer normal vec<strong>to</strong>r along ∂DR we obtain<br />

<br />

∂ <br />

(b − τ δ) > 0.<br />

∂ν<br />

Hence, for any point z ∈ ∂DR<br />

∂DR<br />

∂b ∂<br />

(z) ≥ (τδ)(z) ≥ κ<br />

∂ν ∂ν<br />

for a suitable constant κ (possibly negative). By virtue <strong>of</strong> Poisson’s formula, we<br />

see that the normal derivative <strong>of</strong> δ may be bounded from below by a constant<br />

(depending only on R) times ||a||L ∞ (∂DR).<br />

If we write z = e s+it , we obtain that z ∈ ∂DR is <strong>of</strong> the form z = Re 2πit . In<br />

this case, up <strong>to</strong> a constant fac<strong>to</strong>r, ∂<br />

∂ν<br />

From this, we obtain :<br />

= ∂<br />

∂s .<br />

∂b<br />

(z)<br />

∂ν<br />

=<br />

∂<br />

∂s b(z)<br />

= λv(z)[ ∂<br />

∂t v(z)]<br />

= ft(z) + λu(β(z))[ut(β(z))]βt(z)<br />

= ft(z) − as(β(z))βt(z)<br />

ft(z) = ∂<br />

b(z) + as(β(z))βt(z)<br />

∂ν<br />

≥ κ − βt(z)||a(z)||L∞ This uses the fact that βt > 0.<br />

Let now<br />

S± := {t ∈ S 1 | ± ∂tf ≥ 0}.<br />

Then, 0 = <br />

∂DR ∂tf dt = <br />

S+ ∂tf + <br />

S− ∂tf.<br />

We obtain<br />

<br />

0 ≥ ft(t)dt ≥ κ|S−| − ||a||L∞ <br />

βt(t) ≥ κ − ||a||L∞ S−<br />

because β is mono<strong>to</strong>ne increasing. From this we conclude that f has bounded<br />

variation. Since f(R) = 0, we obtain a C 0 bound.<br />

This bound depends only on R and ||a||L ∞ (DR).<br />

79


Interior gradient bounds<br />

We have now established an a priori bound on the amount we flow the boundary.<br />

Thus, a sequence <strong>of</strong> solutions <strong>to</strong> the boundary value problem (5.2) that converge<br />

in C∞ loc on the interior <strong>of</strong> the disk will have a subsequence converging <strong>to</strong> a solution<br />

<strong>of</strong> the boundary value problem (5.2).<br />

We now suppose we have a sequence <strong>of</strong> solutions <strong>to</strong> (5.2), ˜vn, corresponding<br />

<strong>to</strong> τn → τ. If we have gradient bounds, a subsequence converges <strong>to</strong> a solution<br />

<strong>to</strong> (5.2) for τ. If we do not have gradient bounds, we can find a sequence <strong>of</strong><br />

points zn so that |∇˜vn(zn)| =: Rn → ∞.<br />

We introduce a slight variant <strong>of</strong> the E energy introduced by H<strong>of</strong>er [21] <strong>to</strong><br />

study pseudoholomorphic curves in symplectizations <strong>of</strong> contact manifolds. We<br />

recall that H<strong>of</strong>er’s energy is given by<br />

<br />

E(ũ) = sup{ ũ ∗ d(φλ)|φ : R → [0, 1] with φ′ ≥ 0}.<br />

In the following analysis, we will use a variant <strong>of</strong> this energy in which we restrict<br />

the class <strong>of</strong> φ that we allow. We will take φ so that<br />

φ ′ = 0 on ∪n∈Z [(4n − 1)||a||L ∞ (∂DR), (4n + 1)||a||L ∞ (∂DR)].<br />

We observe that this energy is essentially the same energy as used by H<strong>of</strong>er,<br />

but instead <strong>of</strong> being invariant under the action <strong>of</strong> R translation, it is invariant<br />

under a discrete subgroup <strong>of</strong> translations. Nevertheless, invariance under<br />

this discrete subgroup is sufficient <strong>to</strong> establish all <strong>of</strong> the existing results about<br />

finite energy curves. In particular, finiteness <strong>of</strong> this energy implies the same<br />

compactness results as the finiteness <strong>of</strong> the standard energy. In the following,<br />

we will denote this variant <strong>of</strong> the energy by ER, <strong>to</strong> emphasize the fact that the<br />

definition <strong>of</strong> the energy depends on R, the radius <strong>of</strong> the disk we take.<br />

For any curve that satisfies our boundary condition on the disk, we have<br />

that ER(˜v) ≤ <br />

∂DR v∗dλ by S<strong>to</strong>kes theorem and the fact that the φ in the<br />

definition <strong>of</strong> the energy vanish on τa(z)|∂DR for τ ∈ [0, 1]. Since this energy is<br />

finite, any interior bubbling gives a finite energy plane, which is then asymp<strong>to</strong>tic<br />

<strong>to</strong> a periodic orbit <strong>of</strong> action no greater than <br />

C u∗dλ, in contradiction <strong>to</strong> our<br />

assumption.<br />

Boundary bubbling<br />

We are now left with the case <strong>of</strong> bubbling at the boundary. We will show that<br />

this implies that we bubble <strong>of</strong>f a solution <strong>to</strong> our boundary value problem (5.2)<br />

with τ = 0. Since proving the existence <strong>of</strong> such a curve was the goal <strong>of</strong> our<br />

80


construction, this “short-circuits” the homo<strong>to</strong>py. The pro<strong>of</strong> <strong>of</strong> this result relies<br />

on our a priori bound on how far we flow in the immersed Lagrangian L.<br />

Suppose ˜vn : DR → R × M are a sequence <strong>of</strong> solutions <strong>to</strong> (5.2) so that<br />

we have a sequence <strong>of</strong> points zn → ∂DR so that Rn := |∇˜vn(zn)| → ∞. By<br />

considering a subsequence, we may assume that zn → z∞ ∈ ∂DR.<br />

Then, for n large, |zn − z∞| < 1/4. Let us now introduce a biholomorphic<br />

change <strong>of</strong> coordinates :<br />

We observe that<br />

1<br />

z − z∞<br />

˜wn := ˜vn(− ).<br />

z + z∞<br />

where z ′ n = −z∞ ·<br />

2 |∇˜vn(zn)| ≤ |∇ ˜wn(z ′ n)| ≤ 2|∇˜vn(zn)|<br />

zn−1<br />

zn+1<br />

<br />

. Thus we may assume that we are working with<br />

pseudoholomorphic half-planes on H = {z ∈ C | ℜ(z) ≥ 0} and a sequence <strong>of</strong><br />

points zn → ∂H, with |zn| ≤ 1.<br />

By H<strong>of</strong>er’s Lemma, we have a sequence ɛn > 0, with ɛn → 0 so that ɛnRn →<br />

∞ and so that for |z − zn| < ɛn, we have |∇ũn(zn)| ≤ 2Rn. Let δn = d(zn, ∂H).<br />

Then, we must have δnRn → c < ∞. Indeed, if δnRn → ∞, then we could<br />

rescale along the disks D(zn, δn/2) and obtain interior bubbling.<br />

Let ζn ∈ D(zn, ɛn) ∩ iR. Let ρn = ɛn − δn. Then, D(ζn, ρn) ⊂ D(zn, ɛn). We<br />

may take a subsequence along which ɛn > 2δn. Then, zn ∈ D(ζn, ρn).<br />

We consider now the following rescaling :<br />

˜Wn := ˜wn( z<br />

Rn<br />

+ ζn) for |z| ≤ ρnRn and ℜ(z) ≥ 0<br />

then we obtain that |∇ ˜ Wn(z)| ≤ 2 for |z| ≤ ρnRn. Furthermore,<br />

|∇ ˜ Wn(Rn(zn − ζn))| = 1.<br />

Since |Rn(zn − ζn)| ≤ Rnδn → c < ∞, we may take a subsequence along<br />

which Rn(zn − ζn) converges. Let us denote the limit by ζ∞. We note that<br />

|ζ∞| ≤ c.<br />

By construction, the sequence ˜ Wn has a uniform gradient bound. Thus, a<br />

subsequence converges <strong>to</strong> a pseudoholomorphic half-plane ˜ W = (C, W ) : H →<br />

R × M.<br />

Since the sequence ˜ Wn has a uniform gradient bound, for each k ≥ 1 and<br />

for each compact K ⊂ H, we have uniform C k bounds. Let b be the bound for<br />

C 2 (D(ζ∞, 3c)). We then have<br />

|∇ ˜ Wn(ζ∞)| ≥ |∇ ˜ Wn(Rn(zn − ζn))| − sup(D 2 ˜ W )|ζ∞ − Rn(zn − ζn)|<br />

≥ 1 − b|ζ∞ − Rn(zn − ζn)|,<br />

81


Thus, |∇ ˜ W (ζ∞)| ≥ 1, so the curve is non-constant.<br />

Let us write ˜vn = (bn, vn), ˜wn = (cn, wn), ˜ Wn = (C, W ) and ˜ W = (C, W ).<br />

Then, we verify the boundary condition satisfied by ˜ W . For any point it ∈ ∂H,<br />

we have<br />

C(it) = lim Cn(it) = lim cn( it<br />

Rn<br />

+ ζn)<br />

1 − it/Rn − ζn<br />

= lim bn(z∞<br />

)<br />

1 + it/Rn + ζn<br />

1 − it/Rn − ζn<br />

= lim τna(z∞<br />

)<br />

1 + it/Rn + ζn<br />

1 − ζ∞<br />

= τ∞a(z∞ )<br />

1 + ζ∞<br />

= constant.<br />

By shifting our curve by means <strong>of</strong> the R action, we may take the constant <strong>to</strong><br />

be zero.<br />

We will now show that ˜ W , defined on a half-plane, has a removable singularity<br />

at infinity. This will then show that ˜ W may be extended <strong>to</strong> a smooth<br />

map on the disk, with boundary in L. This resulting disk will be non-constant<br />

and thus will be the renormalized disk we are seeking.<br />

Let us consider ˜ W with the half-plane re-parametrized as a strip. Thus, we<br />

let ˜ W ′ (z) = ˜ W (ie 2π(s+it) ). Even though we had a bound on the gradient <strong>of</strong> ˜ W ,<br />

we no longer have a bound on the gradient <strong>of</strong> ˜ W ′ . Let us suppose that the<br />

gradient <strong>of</strong> ˜ W ′ is unbounded. Then, there is a sequence <strong>of</strong> points (sk, tk) in the<br />

strip, along which |∇ ˜ W ′ | is unbounded. We have that sk → ∞ and may take a<br />

subsequence along which tk → t∞. As before, we have no interior bubbling since<br />

the resulting pseudoholomorphic plane would be asymp<strong>to</strong>tic <strong>to</strong> a periodic orbit<br />

we have assumed does not exist. Thus, we may only have boundary bubbling.<br />

We consider now the case <strong>of</strong> boundary bubbling. After re-scaling, we obtain<br />

a non-constant pseudoholomorphic half-plane, ˜ W ′′ = (B ′′ , W ′′ ). This half-plane<br />

maps the boundary ∂H <strong>to</strong> {0}×L and has bounded gradient. Furthermore, the<br />

half-plane satisfies <br />

H W ′′∗ dλ = 0. Thus, B ′′ is harmonic on the half-plane with<br />

bounded gradient. We obtain, therefore, that B ′′ (s, t) = cs for some c. Hence,<br />

W ′′ (s, t) = x(ct), where x(t) is a trajec<strong>to</strong>ry <strong>of</strong> the Reeb vec<strong>to</strong>r field. We know,<br />

however, that W ′′ has finite energy (in our modified sense) and thus c = 0 and<br />

W ′′ is constant. This contradiction establishes that our original strip W ′ has<br />

bounded gradient.<br />

We will now show that ˜ W ′ has a removable singularity at +∞. Let us<br />

introduce ˜ W ′ k (z) := ˜ W ′ (z − sk) for some sequence sk → ∞. This sequence<br />

has uniform gradient bounds since W ′ does. Hence, ˜ W ′ k converge in C∞ loc <strong>to</strong><br />

82


a pseudoholomorphic strip, ˜ W ′′′ . Observe that ˜ W ′′′ = (B ′′′ , W ′′′ ) maps the<br />

boundary <strong>of</strong> the strip <strong>to</strong> {0} × L and has 0 dλ area. Hence, B ′′′ is harmonic.<br />

Furthermore, B ′′′ has bounded gradient. Hence, B ′′′ (s, t) ≡ 0. Thus ˜ W ′′′ is<br />

constant.<br />

We let<br />

˜V (z) = (B(z), V (z)) : {z ∈ C | |z| ≤ 1 and z = 1} −→ R × M<br />

z ↦→ ˜ <br />

1 + z<br />

W .<br />

1 − z<br />

By our previous work, we have that B(z) is bounded along the boundary, with<br />

boundary in a Lagrangian submanifold, {0} × L. Thus, by Gromov’s removable<br />

singularity theorem, it follows that ˜ V may be extended <strong>to</strong> a smooth pseudoholomorphic<br />

map with boundary in {0} × L.<br />

We now observe that<br />

<br />

V ∗ <br />

dλ ≤ u ∗ dλ.<br />

D<br />

DR<br />

By S<strong>to</strong>kes theorem, since V (∂D) ∈ L, we have<br />

<br />

V ∗ <br />

dλ = deg(V |∂D) ·<br />

DR<br />

u ∗ dλ.<br />

Finally, since B = 0 on ∂D, E( ˜ V ) = <br />

D V ∗ dλ (this energy is the standard,<br />

H<strong>of</strong>er energy). The map ˜ V is non-constant, so this energy must be positive.<br />

Thus we must have deg(V |∂D) = 1. This shows that ˜ V is a solution <strong>to</strong> the<br />

boundary value problem (5.2), for τ = 0. Thus, if we have boundary bubbling<br />

along our sequence, we immediately obtain a renormalized solution.<br />

Conclusion <strong>of</strong> the pro<strong>of</strong><br />

Let us assume for now that we have proved the local existence result, that if we<br />

have a solution <strong>of</strong> (5.2) for τ = τ0, then there exist nearby solutions for τ close<br />

<strong>to</strong> τ0. (We prove this in the following section, 5.3.3).<br />

We then have that for any disk DR, we may find a renormalized solution :<br />

⎧<br />

⎪⎨<br />

⎪⎩<br />

˜vR = (bR, vR) : DR → R × M<br />

πλ T v ◦ i = J(v) ◦ πλ T v<br />

v ∗ λ ◦ i = db<br />

b(z) = 0 for z ∈ ∂DR<br />

v(z) = φf(z)(u(β(z))) for z ∈ ∂DR<br />

where f : ∂DR → R<br />

and β : ∂DR → ∂DR <strong>of</strong> degree 1<br />

with the constraint, f(R) = 0<br />

83<br />

(5.4)


We observe now that the standard H<strong>of</strong>er energy, E(˜vR) = <br />

BR u∗d λ, by<br />

S<strong>to</strong>kes Theorem.<br />

We take a sequence Rk → ∞. Let ˜vk = ˜vRk . For each k, E(˜vk) ≤ C. We<br />

have one <strong>of</strong> two possibilities, after taking subsequences as necessary :<br />

(a) For each n, there exists a bound Mn > 0 so that for all k and for all<br />

|z| ≤ min{Rn, Rk}, we have |∇vk(z)| ≤ Mn. In this case, after a diagonal<br />

subsequence argument, we obtain that, after translation in the symplecti-<br />

zation direction, ˜vk converge in C ∞ loc<br />

<strong>to</strong> a finite energy pseudoholomorphic<br />

plane. This plane must then be asymp<strong>to</strong>tic <strong>to</strong> a periodic Reeb orbit <strong>of</strong><br />

period no greater than C. This contradicts our assumption.<br />

(b) There exists a sequence <strong>of</strong> points zk so that |∇˜vk(zk)| → ∞ as k → ∞.<br />

We may now apply the usual bubbling <strong>of</strong>f argument <strong>to</strong> obtain a finite<br />

energy plane, ˜w, with E( ˜w) ≤ C. This is again a contradiction <strong>to</strong> our<br />

assumption.<br />

Theorem 5.1.1 follows from the contradictions in (a) and (b).<br />

5.3.3 Fredholm theory<br />

In this section, we will establish the following result :<br />

Proposition 5.3.2. Suppose the boundary value problem (5.2) has a solution<br />

for some τ0 > 0. Then there exist solutions for τ sufficiently close <strong>to</strong> τ0.<br />

We first make the observation that it suffices <strong>to</strong> consider τ0 = 1. Indeed, if<br />

we have a solution ˜vτ0, for τ0, then we may “re-start” our homo<strong>to</strong>py by taking<br />

ũ0 = ˜vτ0.<br />

Thus, in this section we will introduce a suitable functional analytic set-up,<br />

and prove that our problem is Fredholm and surjective at the initial solution<br />

for τ = 1, ũ. We will also show that the kernel <strong>of</strong> <strong>of</strong> the linearized opera<strong>to</strong>r<br />

includes an element which projects non-trivially on ∂ . Thus, we will conclude<br />

∂τ<br />

that the solutions exist for nearby values <strong>of</strong> τ.<br />

This work is very similar <strong>to</strong> the case in a closed symplectic manifold (as in<br />

[33]). We wish <strong>to</strong> consider all nearby curves that satisfy the boundary condition<br />

as elements <strong>of</strong> a Banach manifold. Such curves may be represented as graphs<br />

over ũ0 by using the flow <strong>of</strong> ∂t (the genera<strong>to</strong>r <strong>of</strong> the symplectization direction)<br />

and the flow <strong>of</strong> X (the Reeb vec<strong>to</strong>r field), <strong>to</strong>gether with a re-parametrization<br />

<strong>of</strong> the domain. This allows us <strong>to</strong> identify a sufficiently small neighbourhood <strong>of</strong><br />

our original curve with a small neighbourhood <strong>of</strong> the zero section in the bundle<br />

E := C ⊕ T D → D. We may now pull the almost complex structure ˜ J back <strong>to</strong><br />

84


a complex structure on E, ¯ J. We define the Sobolev space W 1,p (E) <strong>of</strong> sections<br />

<strong>of</strong> this bundle. We now define a Banach manifold<br />

B := {u ∈ W 1,p (E) | u satisfies the boundary conditions}.<br />

Over this Banach manifold, we put a Banach space bundle E, whose fibre over<br />

a section v is given by<br />

Ev = Hom 0,1 ( T D, v ∗ E)<br />

where Hom 0,1 refers <strong>to</strong> the bundle <strong>of</strong> i– ¯ J antiholomorphic real linear homomorphisms.<br />

The (non-linear) Cauchy–Riemann opera<strong>to</strong>r,<br />

v ↦→ dv − ¯ J ◦ dv ◦ i<br />

is a smooth section <strong>of</strong> this Banach space bundle. Its linearization at the zero<br />

section will turn out <strong>to</strong> be a linear Cauchy–Riemann opera<strong>to</strong>r. This is then<br />

Fredholm, with the index determined by the boundary data (as given by the<br />

Riemann-Roch theorem).<br />

Finding holomorphic sections <strong>of</strong> a vec<strong>to</strong>r bundle<br />

Our goal is <strong>to</strong> show that our problem may be identified with a non-linear problem<br />

whose linearization at our initial curve is <strong>of</strong> linear Cauchy-Riemann type. We<br />

will thus construct a Banach manifold consisting <strong>of</strong> maps from the disk that<br />

satisfy the boundary condition. On this, we will put the Banach space bundle<br />

whose fibre over a maps v consists <strong>of</strong> antiholomorphic sections in v∗T W (where<br />

W is an almost complex manifold we will specify shortly). The non-linear<br />

opera<strong>to</strong>r will then be a smooth section <strong>of</strong> this bundle.<br />

For clarity <strong>of</strong> notation, we will work with R = 1. This covers the general<br />

case, since we may precompose ũ with the map z ↦→ z/R.<br />

Our first step is <strong>to</strong> introduce convenient coordinates. To this end, we will<br />

observe that since πλ T u = 0 on D, we have that ∂ and X span the normal<br />

∂a<br />

bundle <strong>of</strong> ũ in ũ∗ (R × M). Notice that we have that ũ is immersed. Thus, we<br />

may write a C1 close curve as a graph in the normal bundle <strong>of</strong> ũ, with possibly<br />

a different parametrization <strong>of</strong> the domain. This means we then have<br />

Lemma 5.3.3. A solution ˜v <strong>to</strong> (5.2), sufficiently C 1 close <strong>to</strong> ũ, may be written<br />

as v(z) = φf(z) ◦ u(β(z)), where f : D → R and β : D → D. Moreover, β<br />

restricts <strong>to</strong> a degree 1 map on the boundary and β(z) = z for all z ∈ Γ.<br />

This lemma allows us <strong>to</strong> introduce the coordinates we will use for our Fredholm<br />

theory :<br />

Ψ : R × R × D → R × M<br />

(ρ, σ, ξ, η) ↦→ (ρ + a(ξ, η), φσ ◦ u(ξ, η))<br />

85


We note that Ψ is an immersion.<br />

Let us introduce ¯ J on R × R × D by T Ψ ◦ ¯ J = ˜ J ◦ T Ψ. We thus transformed<br />

our problem (5.2) in<strong>to</strong> a problem <strong>of</strong> finding pseudoholomorphic disks in<strong>to</strong> the<br />

almost complex manifold W = R × R × D with the almost complex structure<br />

¯J : (in the following, we identify C with R 2 )<br />

w = (f, h) : D → C × D,<br />

T w ◦ i = ¯ J ◦ T w on D<br />

h|∂D : ∂D → ∂D is a degree 1 map,<br />

f1(z) = (τ − 1)a(z) for z ∈ ∂D<br />

f2(1) = 0<br />

(5.5)<br />

We observe that for τ = 1, we have the solution (w, c) given by w : z ↦→<br />

(0, 0, z).<br />

We now put then a metric on D for which the boundary is <strong>to</strong>tally geodesic.<br />

To do this, we consider a metric that near the boundary is <strong>of</strong> the form ds 2 ⊕ dt 2<br />

where z = e 2π(s+it) . Outside <strong>of</strong> a neighbourhood <strong>of</strong> the boundary, we take the<br />

metric <strong>to</strong> be the standard metric on the disk.<br />

We take now the exponential map from this metric. Any map h : D → D,<br />

mapping boundary <strong>to</strong> boundary, sufficiently C 0 close <strong>to</strong> the identity map may<br />

be represented as h(z) = exp z(H(z)), where H : D → T D is a vec<strong>to</strong>r field on<br />

D, with H(z) ∈ izR for z ∈ ∂D.<br />

Let us now define the bundle<br />

E = C ⊕ T D → D.<br />

We fix a one parameter family <strong>of</strong> <strong>to</strong>tally real subbundles at the boundary given<br />

by<br />

ℓτ|z = ((1 − τ)a(z) + iR, izR) for |z| = 1.<br />

To a section <strong>of</strong> this bundle, we may define a Sobolev norm. Indeed, we<br />

may identify a section <strong>of</strong> E with a map (s1, s2) : D → C ⊕ C. On this, we<br />

put the standard Sobolev norms W 1,p . We denote the space <strong>of</strong> these sections<br />

by W 1,p (E). We observe that any two choices <strong>of</strong> trivialization <strong>of</strong> E induce<br />

equivalent norms.<br />

We are primarily interested in certain special subspaces <strong>of</strong> W 1,p , representing<br />

variations <strong>of</strong> curves that satisfy our boundary conditions. We note that we<br />

have two different types <strong>of</strong> boundary condition : we have a one parameter<br />

family <strong>of</strong> <strong>to</strong>tally real boundary conditions, and we have a pointwise constraint.<br />

We will thus find it useful <strong>to</strong> introduce four spaces : we will consider the one<br />

parameter family <strong>of</strong> <strong>to</strong>tally real boundary conditions, with and without the<br />

pointwise constraint; we will also consider a boundary condition in a fixed <strong>to</strong>tally<br />

86


eal boundary condition (at τ = 1) with and without the pointwise constraint.<br />

The first such space we denote by W 1,p<br />

ℓτ . This is the space <strong>of</strong> W 1,p sections s such<br />

that there exists a τ = τ(s) so that s|∂D ∈ ℓτ. The next such space we denote by<br />

W 1,p<br />

1,p<br />

ℓτ ,pt . This is the co-dimension one subspace <strong>of</strong> Wℓτ such that a section (s1, s2)<br />

has Im(s1)(1) = 0. The third space we introduce is another codimension one<br />

subspace <strong>of</strong> W 1,p<br />

, corresponding <strong>to</strong> the sections for which τ = 1. We denote this<br />

ℓτ<br />

by W 1,p<br />

ℓ , and is the space <strong>of</strong> sections s with s|∂D ∈ iR⊕izR. Finally, we consider<br />

the space W 1,p<br />

ℓ,pt<br />

, consisting <strong>of</strong> all sections in W 1,p<br />

ℓτ ,pt such that s(z) ∈ ℓ1 = iR⊕izR<br />

for z ∈ ∂D. This corresponds <strong>to</strong> the space <strong>of</strong> sections that satisfy our pointwise<br />

constraint.<br />

We have that for ||s|| W 1,p < ɛ0 sufficiently small, we have that ||s||L∞ is<br />

small, thus the map (s1, s2) ↦→ (s1, expz s2) is injective.<br />

We define now the following Banach manifold :<br />

Bℓτ = {z ↦→ (s1(z), expz s2(z)) | s ∈ W 1,p<br />

(E), with ||s|| < ɛ0}.<br />

ℓτ<br />

and similarly Bℓ, Bℓτ ,pt and Bℓ,pt. We note that each <strong>of</strong> these is a submanifold<br />

<strong>of</strong> Bℓτ .<br />

Over this large Banach manifold, we now put the Banach space bundle<br />

F → Bℓτ so that the fibre over any u ∈ Bℓτ is given by<br />

Fu = Hom 0,1 ( T D, u ∗ T W ).<br />

On this bundle, we may put the L p norm, similarly <strong>to</strong> our previous construction.<br />

We define now the non-linear opera<strong>to</strong>r, ¯ ∂ ¯ J, whose zeros are pseudoholomorphic<br />

curves.<br />

¯∂ ¯ J : Bℓτ −→ E<br />

u ↦→ du − ¯ J ◦ du ◦ i.<br />

(5.6)<br />

We recall that ¯ ∂ ¯ Ju0 = 0, where u0(z) = (0, 0, z). We note that the zeros <strong>of</strong><br />

¯∂ ¯ J restricted <strong>to</strong> the space Bℓτ ,pt are precisely the solutions <strong>of</strong> (5.2) near <strong>to</strong> our<br />

original curve.<br />

The fact that the various B are Banach manifolds and that F is a Banach<br />

space bundle over them follows from the work <strong>of</strong> Eliasson [14]. We note that ¯ J<br />

is smooth, and thus it follows (again by [14]) that ¯ ∂ ¯ J is a smooth section <strong>of</strong> F.<br />

In particular, then, it is C 1 .<br />

We consider now the problem <strong>of</strong> finding zeros <strong>of</strong> the opera<strong>to</strong>r restricted <strong>to</strong><br />

the space Bℓ. In other words, we are working over the space <strong>of</strong> maps that<br />

satisfy the boundary condition for a fixed <strong>to</strong>tally real boundary condition. By<br />

[33, Prop. 3.1.1.], the linearization <strong>of</strong> ¯ ∂ ¯ J at a map v, which we denote Lv, is a<br />

Cauchy-Riemann type opera<strong>to</strong>r.<br />

Lv : W 1,p<br />

ℓ (v ∗ E) −→ L p (Hom 0,1 ( T ˙ Σ, v ∗ E)).<br />

87


In particular, when we linearize the opera<strong>to</strong>r at the zero section, we obtain<br />

a Cauchy-Riemann type opera<strong>to</strong>r L0 := D( ¯ ∂ ¯ J)(0, z)| W 1,p<br />

ℓ (E)<br />

L0 : W 1,p<br />

ℓ (E) −→ L p (Hom 0,1 ( T D, E)).<br />

By the Riemann-Roch theorem [33, Theorem C.1.10], we have then that this<br />

opera<strong>to</strong>r is Fredholm, and its index is given by :<br />

Ind(L0) = 2 + µ(ℓ) = 2 + 2 = 4.<br />

We have that L0 is the restriction <strong>of</strong> D( ¯ ∂J)(0, ¯ z) <strong>to</strong> the space W 1,p<br />

ℓ from the<br />

space W 1,p<br />

1,p 1,p<br />

. We have that W ℓτ ℓ ⊂ W has co-dimension one. Thus, the full<br />

ℓτ<br />

linearization <strong>of</strong> ¯ ∂ ¯ J at the zero section,<br />

D( ¯ ∂ ¯ J)(0, z) : W 1,p<br />

ℓτ (E) −→ Lp (Hom 0,1 ( T D, E))<br />

is Fredholm, and its index is Ind(L0) + 1 = 5.<br />

Finally, we wish <strong>to</strong> consider D( ¯ ∂J)(0, ¯ z), restricted <strong>to</strong> the subspace corresponding<br />

<strong>to</strong> subspace enforcing the pointwise constraint, W 1,p<br />

ℓτ ,pt . Let us call this<br />

restricted opera<strong>to</strong>r L0. This is then L0 : W 1,p<br />

ℓτ ,pt (E) −→ Lp (E). By virtue <strong>of</strong> the<br />

fact that W 1,p<br />

ℓτ ,pt<br />

and its index is given by<br />

(E) is codimension one in W 1,p<br />

ℓτ (E), we have that L0 is Fredholm,<br />

Ind(L0) = Ind(D( ¯ ∂ ¯ J)(0, z)) − 1 = 4.<br />

The same argument applies <strong>to</strong> the restriction <strong>of</strong> D( ¯ ∂J)(0, ¯ z) <strong>to</strong> W 1,p<br />

ℓ,pt , where<br />

the index is then 3.<br />

Transversality<br />

We now must prove transversality for L0. In other words, we must show that<br />

the kernel <strong>of</strong> L0 is four dimensional. Furthermore, <strong>to</strong> show that we have a family<br />

that depends on the parameter τ, we must show that the kernel <strong>of</strong> L0 projects<br />

non-trivially on the τ direction.<br />

Once this is established, by the implicit function theorem, we will have a<br />

four parameter family <strong>of</strong> solutions <strong>to</strong> the problem (5.5) near our original curve.<br />

Furthermore, we will have that one <strong>of</strong> these parameters may be taken <strong>to</strong> be τ.<br />

This will establish then Lemma 5.3.3.<br />

In order <strong>to</strong> establish this result, we will show that<br />

L0 := D( ¯ ∂ ¯ J)(0, z)| W 1,p<br />

ℓ,pt (E)<br />

88


is surjective. The result will then follow, since L must then be surjective, and<br />

thus its kernel is four dimensional. This then must project non-trivially on the<br />

τ direction.<br />

In order <strong>to</strong> show that we have transversality for L0, we must pay more<br />

attention <strong>to</strong> the form <strong>of</strong> the linearization <strong>of</strong> the opera<strong>to</strong>r. To this end, we will<br />

use the fact that we have global holomorphic coordinates on D ⊂ C, given by<br />

z = x + iy, and we also have an isomorphism from Hom 0,1 ( T D, E) <strong>to</strong> E by the<br />

map φ ↦→ φ ∂<br />

∂x .<br />

We also must study the behaviour <strong>of</strong> our almost complex structure ¯ J. Let<br />

us put coordinates on R × R × D by (ρ, σ, x, y). (We recall that we let a be the<br />

coordinate on the R fac<strong>to</strong>r in R × M. We also have ũ = (a, u) is our original<br />

curve, and J is the complex structure on ker λ ⊂ T M.) We obtain :<br />

where<br />

Let<br />

Ψ∗<br />

Ψ∗<br />

Ψ∗<br />

∂<br />

∂ρ<br />

∂<br />

∂x<br />

∂<br />

∂y<br />

=<br />

∂<br />

∂a<br />

∂<br />

Ψ∗<br />

∂σ = X(φσ ◦ u(z))<br />

=<br />

∂<br />

ax<br />

∂a + T φσ ◦ ux<br />

=<br />

∂<br />

ax<br />

∂a + λ(ux)X + ν1<br />

=<br />

∂<br />

ay<br />

∂a + λ(uy)X + ν2.<br />

ν1(σ, z) := T φσ ◦ πλ ux, ν2(σ, z) := T φσ ◦ πλ uy.<br />

B =<br />

ax<br />

<br />

ay<br />

.<br />

λ(ux) λ(uy)<br />

By the fact that ũ is pseudoholomorphic with respect <strong>to</strong> ˜ J, we obtain that B<br />

commutes with J0, the standard complex structure on R 2 identified with C.<br />

Let ˆ J(σ, z) be the matrix representing J(z) with respect <strong>to</strong> the basis ν1(σ, z)<br />

and ν2(σ, z). We note that ˆ J is an almost complex structure on C. We have<br />

that ˆ J(0, z) = J0, the standard complex structure.<br />

Then, we have, with respect <strong>to</strong> the basis ∂<br />

∂ρ<br />

by the matrix ¯ M,<br />

Let N(σ, z) = ˆ J(σ, z) − J0.<br />

, ∂<br />

∂σ<br />

, ∂<br />

∂x<br />

<br />

J0 ¯M(σ,<br />

B J0 −<br />

x, y) =<br />

ˆ <br />

J(σ, z)<br />

.<br />

0 J(σ, ˆ z)<br />

89<br />

, ∂<br />

∂y , that ¯ J is represented


It follows that the non-linear opera<strong>to</strong>r, acting on the section (f, h) =<br />

(f1, f2, h1, h2) : D → C ⊕ T D, may be expressed as :<br />

<br />

∂xf + i∂yf − B(expz h(z))N(f2, expz h(z))∂y(exp<br />

(f, g) ↦→<br />

z(h(z)))<br />

∂xh + i∂yh + N(σ, expz(h(z)))∂y(expz(h(z))) We now linearize this opera<strong>to</strong>r at the zero section. This gives us a linearized<br />

opera<strong>to</strong>r <strong>of</strong> the form<br />

<br />

¯∂J0F + P (z)F<br />

(F, G) ↦→ ¯∂J0H + Q(z)F<br />

where P and Q are matrix valued functions.<br />

We first consider all pairs (0, H) in the kernel <strong>of</strong> this opera<strong>to</strong>r. Hence, we<br />

have that H : D → C is a holomorphic function (classically), with H(z) ∈ izR<br />

for |z| = 1. We then have a three (real) parameter family <strong>of</strong> such H.<br />

We now consider a pair (F, H) in the kernel <strong>of</strong> the opera<strong>to</strong>r. It follows then<br />

that ¯ ∂J0F + P (z)F = 0. Furthermore, F (z) ∈ iR for |z| = 1. This is a <strong>to</strong>tally<br />

real boundary condition <strong>of</strong> Maslov index 0. Thus, if F is not identically 0, F<br />

cannot vanish anywhere. However, we have that F (1) = 0. Thus, F ≡ 0.<br />

We then have that the kernel <strong>of</strong> L is three dimensional. The Fredholm index<br />

<strong>of</strong> L is also three, so it follows that L is surjective. The result now follows.<br />

5.4 Future work<br />

The present work on the behaviour <strong>of</strong> infinite E energy, but finite contact area,<br />

pseudoholomorphic planes is a beginning <strong>of</strong> further work with the compactification<br />

<strong>of</strong> the space <strong>of</strong> solutions <strong>to</strong> the generalized pseudoholomorphic curve<br />

equations with harmonic form as introduced by H<strong>of</strong>er in [3].<br />

The technical constraint that πλ T u never vanish was so that ∂<br />

∂a<br />

and X<br />

spanned the normal bundle <strong>to</strong> the image <strong>of</strong> ũ. In order <strong>to</strong> deal with the zeros<br />

<strong>of</strong> πλ T u, we will study a problem on ˙ D = D \ Γ. We will see this as a Riemann<br />

surface with punctures. The problem will then become more difficult,<br />

analytically speaking.<br />

The current work, along with some examples, indicates that these infinite<br />

energy, finite contact area planes are asymp<strong>to</strong>tic <strong>to</strong> periodic orbits in a suitable<br />

(weak) sense. To study this, we will consider the renormalization <strong>of</strong> long annuli<br />

with contact area below the energy threshold from Theorem 5.1.1. The goal<br />

is <strong>to</strong> develop a theory analogous <strong>to</strong> the work in [26], where a similar theory<br />

is developed for finite energy curves. In the case <strong>of</strong> annuli, the compactness<br />

problem becomes more difficult. This is ongoing project with Casim Abbas<br />

and Helmut H<strong>of</strong>er. The ultimate goal <strong>of</strong> this project is <strong>to</strong> develop a good<br />

compactification for the space <strong>of</strong> H<strong>of</strong>er’s generalized pseudoholomorphic curves.<br />

90


Bibliography<br />

[1] C. Abbas. Holomorphic open book decompositions. (in preparation).<br />

[2] C. Abbas. Finite energy surfaces and the chord problem. Duke Math. J.,<br />

96(2):241–316, 1999.<br />

[3] C. Abbas, K. Cieliebak, and H. H<strong>of</strong>er. The Weinstein conjecture for planar<br />

contact structures in dimension three. <strong>to</strong> appear Math. Helv., 2005.<br />

[4] C. Abbas and H. H<strong>of</strong>er. Holomorphic curves and global questions in contact<br />

geometry. <strong>to</strong> appear in Birkhäuser.<br />

[5] P. Biran and K. Cieliebak. Lagrangian embeddings in<strong>to</strong> subcritical Stein<br />

manifolds. Israel J. Math., 127:221–244, 2002.<br />

[6] F. Bourgeois, Y. Eliashberg, H. H<strong>of</strong>er, K. Wysocki, and E. Zehnder. Compactness<br />

results in symplectic field theory. Geom. Topol., 7:799–888 (electronic),<br />

2003.<br />

[7] Y. V. Chekanov. Lagrangian intersections, symplectic energy, and areas <strong>of</strong><br />

holomorphic curves. Duke Math. J., 95(1):213–226, 1998.<br />

[8] K. Cieliebak. Handle attaching in symplectic homology and the chord<br />

conjecture. J. Eur. Math. Soc. (JEMS), 4(2):115–142, 2002.<br />

[9] F. Ding and H. Geiges. <strong>Symplectic</strong> fillability <strong>of</strong> tight contact structures on<br />

<strong>to</strong>rus bundles. Algebr. Geom. Topol., 1:153–172 (electronic), 2001.<br />

[10] F. Ding and H. Geiges. A Legendrian surgery presentation <strong>of</strong> contact 3manifolds.<br />

Math. Proc. Cambridge Philos. Soc., 136(3):583–598, 2004.<br />

[11] F. Ding, H. Geiges, and A. I. Stipsicz. Surgery diagrams for contact 3manifolds.<br />

Turkish J. Math., 28(1):41–74, 2004.<br />

[12] Y. Eliashberg. Topological characterization <strong>of</strong> Stein manifolds <strong>of</strong> dimension<br />

> 2. Internat. J. Math., 1(1):29–46, 1990.<br />

91


[13] Y. Eliashberg, A. Givental, and H. H<strong>of</strong>er. Introduction <strong>to</strong> symplectic field<br />

theory. Geom. Funct. Anal., (Special Volume, Part II):560–673, 2000.<br />

GAFA 2000 (Tel Aviv, 1999).<br />

[14] H. I. Elĭasson. <strong>Geometry</strong> <strong>of</strong> manifolds <strong>of</strong> maps. J. Differential <strong>Geometry</strong>,<br />

1:169–194, 1967.<br />

[15] A. Floer. Holomorphic curves and a Morse theory for fixed points <strong>of</strong> exact<br />

symplec<strong>to</strong>morphisms. In Aspects dynamiques et <strong>to</strong>pologiques des groupes infinis<br />

de transformation de la mécanique (Lyon, 1986), volume 25 <strong>of</strong> Travaux<br />

en Cours, pages 49–60. Hermann, Paris, 1987.<br />

[16] H. Geiges. Contact geometry. In Handbook <strong>of</strong> differential geometry, Vol. I.<br />

Elsevier, 2006 (<strong>to</strong> appear).<br />

[17] E. Giroux. Convexité en <strong>to</strong>pologie de contact. Comment. Math. Helv.,<br />

(66):637–677, 1991.<br />

[18] E. Giroux. Géométrie de contact: de la dimension trois vers les dimensions<br />

supérieures. In Proceedings <strong>of</strong> the International Congress <strong>of</strong> Mathematicians,<br />

Vol. II (Beijing, 2002), pages 405–414, Beijing, 2002. Higher Ed.<br />

Press.<br />

[19] R. E. Gompf. Handlebody construction <strong>of</strong> Stein surfaces. Ann. <strong>of</strong> Math.<br />

(2), 148(2):619–693, 1998.<br />

[20] M. Gromov. Pseudoholomorphic curves in symplectic manifolds. Invent.<br />

Math., 82(2):307–347, 1985.<br />

[21] H. H<strong>of</strong>er. Pseudoholomorphic curves in symplectizations with applications<br />

<strong>to</strong> the Weinstein conjecture in dimension three. Invent. Math., 114(3):515–<br />

563, 1993.<br />

[22] H. H<strong>of</strong>er and K. Wysocki. First order elliptic systems and the existence<br />

<strong>of</strong> homoclinic orbits in Hamil<strong>to</strong>nian systems. Math. Ann., 288(3):483–503,<br />

1990.<br />

[23] H. H<strong>of</strong>er, K. Wysocki, and E. Zehnder. Properties <strong>of</strong> pseudoholomorphic<br />

curves in symplectisations. II. Embedding controls and algebraic invariants.<br />

Geom. Funct. Anal., 5(2):270–328, 1995.<br />

[24] H. H<strong>of</strong>er, K. Wysocki, and E. Zehnder. Properties <strong>of</strong> pseudoholomorphic<br />

curves in symplectisations. I. Asymp<strong>to</strong>tics. Ann. Inst. H. Poincaré Anal.<br />

Non Linéaire, 13(3):337–379, 1996.<br />

92


[25] H. H<strong>of</strong>er, K. Wysocki, and E. Zehnder. Properties <strong>of</strong> pseudoholomorphic<br />

curves in symplectizations. III. Fredholm theory. In Topics in nonlinear<br />

analysis, volume 35 <strong>of</strong> Progr. Nonlinear Differential Equations Appl., pages<br />

381–475. Birkhäuser, Basel, 1999.<br />

[26] H. H<strong>of</strong>er, K. Wysocki, and E. Zehnder. Finite energy cylinders <strong>of</strong> small<br />

area. Ergodic Theory Dynam. Systems, 22(5):1451–1486, 2002.<br />

[27] H. H<strong>of</strong>er, K. Wysocki, and E. Zehnder. Finite energy foliations <strong>of</strong> tight<br />

three-spheres and Hamil<strong>to</strong>nian dynamics. Ann. <strong>of</strong> Math. (2), 157(1):125–<br />

255, 2003.<br />

[28] K. Honda. On the classification <strong>of</strong> tight contact structures. II. J. Differential<br />

Geom., 55(1):83–143, 2000.<br />

[29] K. Honda. 3-dimensional methods in contact geometry. In Different faces<br />

<strong>of</strong> geometry, Int. Math. Ser. (N. Y.), pages 47–86. Kluwer/Plenum, New<br />

York, 2004.<br />

[30] Y. Kanda. The classification <strong>of</strong> tight contact structures on the 3-<strong>to</strong>rus.<br />

Comm. Anal. Geom., 5(3):413–438, 1997.<br />

[31] G. Liu and G. Tian. Weinstein conjecture and GW-invariants. Commun.<br />

Contemp. Math., 2(4):405–459, 2000.<br />

[32] S. Makar-Limanov. Tight contact structures on solid <strong>to</strong>ri. Trans. Amer.<br />

Math. Soc., 350(3):1013–1044, 1998.<br />

[33] D. McDuff and D. Salamon. J-holomorphic curves and symplectic <strong>to</strong>pology,<br />

volume 52 <strong>of</strong> American Mathematical Society Colloquium Publications.<br />

American Mathematical Society, Providence, RI, 2004.<br />

[34] K. Mohnke. Holomorphic disks and the chord conjecture. Ann. <strong>of</strong> Math.<br />

(2), 154(1):219–222, 2001.<br />

[35] J. W. Robbin and D. A. Salamon. Asymp<strong>to</strong>tic behaviour <strong>of</strong> holomorphic<br />

strips. Ann. Inst. H. Poincaré Anal. Non Linéaire, 18(5):573–612, 2001.<br />

[36] É. Séré. Homoclinic orbits on compact hypersurfaces in R2N , <strong>of</strong> restricted<br />

contact type. Comm. Math. Phys., 172(2):293–316, 1995.<br />

[37] A. Weinstein. Contact surgery and symplectic handlebodies. Hokkaido<br />

Math. J., 20(2):241–251, 1991.<br />

93

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!