Central European University

Department of Mathematics and its Application

Heegaard Floer Invariants

of Smooth and Legendrian Links
in
Rational Homology 3-spheres
by

Alberto Cavallo

Supervisor
CEU eTD Collection

Prof. András I. Stipsicz

B UDAPEST, H UNGARY
2018
CEU eTD Collection
 Declaration

I hereby declare that this dissertation contains no materials accepted for any other de-
grees in any other institutions.
Moreover, the dissertation contains no materials previously written and/or published
by another person, except where appropriate acknowledgment is made in the form of
bibliographical reference.

Budapest, April 10 2018 Alberto Cavallo

CEU eTD Collection

iii
CEU eTD Collection
 Abstract

Our main goal in this thesis is to describe the most important results on smooth and
Legendrian knots obtained from Heegaard Floer homology in the last ten years and gen-
eralize some of them to links.
Our first result is a link version of the Thurston-Bennequin inequality. This inequality,
in its usual formulation, gives an upper bound for the maximal Thurston-Bennequin and
self-linking numbers of a knot K in a tight contact 3-manifolds ( M, ξ ), in terms of the Euler
characteristic of a Seifert surface for K. We generalize the inequality to every link L, where
the resulting upper bound involves the Thurston norm of L. This is a rational number
extracted from the semi-norm, introduced by Thurston, on the relative second homology
group of a 3-manifold with toric boundary.
We say that two n-component links in S3 are strongly concordant if there is a cobordism
between them consisting of n disjoint annuli, each one realizing a knot concordance.
 Then, 
starting from a grid diagram D of a link L, we define a filtered chain complex GC c ( D ), b
∂
and we prove that its homology, denoted with HF \ L( L), is a strong concordance invariant.
Since we can prove that HF L( L) has dimension one in Maslov grading zero, we also
\
extract a numerical invariant from the homology group that we call τ ( L). The τ-invariant
gives a lower bound for the slice genus g4 ( L), which is the minimum genus of an oriented,
compact surface properly embedded in D4 and whose boundary is L. We also show that
there is a strict relation between the filtration levels of HF
\ L( L) and the Alexander grad-
−
ing of the torsion-free quotient of cHFL ( L), a different bigraded version of link Floer
homology.
Furthermore, we define an invariant of Legendrian links by using open book decom-
positions. This is done by describing a suitable condition for an open book decomposition
( B, π, A) to be adapted to a Legendrian link L, where A is a system of generators of the
first relative homology group of π −1 (1). At this point, we define a special Heegaard dia-
CEU eTD Collection

gram D for the link L in the 3-manifold − M, given by reversing the orientation on M; we
have that D is obtained up to isotopy from ( B, π, A) and then, when L is zero in homology,
the invariant L( L, M, ξ ) is the isomorphism class of a distinguished cycle L( D ) in the link
Floer complex cCFL− ( D, tξ ).
We also give some results on quasi-positive links in S3 . In particular, we introduce
the subfamily of connected transverse C-links: these are links such that the surface Σ B ,
associated to a quasi-positive braid B for L, is connected. We show that, for this kind of
links, the slice genus g4 is determined by τ. This allows us to prove that the slice genus is
additive under connected sums of connected transverse C-links.

v
CEU eTD Collection
 Contents

Introduction 3

Acknowledgements 7

1 Preliminaries 9
1.1 Knots and links in 3-manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.1 Definition of links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.2 Connected sum and mirror image of links . . . . . . . . . . . . . . . . 10
1.2 Links in the 3-sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.1 Diagrams for links in S3 . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.2 The Seifert algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.3 Operations on links and planar diagrams . . . . . . . . . . . . . . . . 14
1.3 Cobordisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 Definition of cobordism and the slice genus . . . . . . . . . . . . . . . 15
1.3.2 Link concordance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 Filtered chain complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4.1 Filtered spaces ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4.2 ... and filtered maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5 Tensor product of modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5.1 General properties and vector spaces . . . . . . . . . . . . . . . . . . 21
1.5.2 Finitely generated K[ x ]-modules . . . . . . . . . . . . . . . . . . . . . 22

2 The Thurston norm 25

2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
CEU eTD Collection

3 Contact structures and Legendrian links 31

3.1 Contact structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 Type of structures and convex surfaces . . . . . . . . . . . . . . . . . 31
3.1.2 Classification theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.3 Fillings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Legendrian links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3 Transverse links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Legendrian and transverse links in (S3 , ξ st ) . . . . . . . . . . . . . . . . . . . 37
3.4.1 Front projections of Legendrian links . . . . . . . . . . . . . . . . . . 37
3.4.2 Braid presentation for transverse links . . . . . . . . . . . . . . . . . . 38
3.5 Connected sum and disjoint union in the contact setting . . . . . . . . . . . . 40
3.6 The Thurston-Bennequin inequality . . . . . . . . . . . . . . . . . . . . . . . 42

vii
 4 Representations of links and 3-manifolds 45
4.1 Heegaard diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.1.1 3-manifolds and Spinc structures . . . . . . . . . . . . . . . . . . . . . 45
4.1.2 Links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Open book decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.2.1 Adapted open book decompositions for links in 3-manifolds . . . . . 50
4.2.2 Abstract open books . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5 Heegaard Floer homology 57

5.1 Link Floer homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.1.1 Heegaard diagrams and J-holomorphic curves . . . . . . . . . . . . . 57
5.1.2 Gradings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.1.3 Differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1.4 Homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.2 Heegaard Floer homology of 3-manifolds . . . . . . . . . . . . . . . . . . . . 64
5.3 Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.3.1 Choice of the almost-complex structures . . . . . . . . . . . . . . . . 66
5.3.2 Isotopies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3.3 Handleslides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.4 Stabilizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.4 Multiplication by U + 1 in the link Floer homology group . . . . . . . . . . . 70

6 Legendrian and contact invariants from open book decompositions 73

6.1 Legendrian Heegaard diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.1.1 Heegaard diagrams through abstract open books . . . . . . . . . . . 73
6.1.2 Global isotopies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.1.3 Positive stabilizations . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
6.1.4 Admissible arc slides . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.2 The Legendrian and transverse invariants . . . . . . . . . . . . . . . . . . . . 80
6.3 Proof of the invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.3.1 Global isotopies and admissible arc slides . . . . . . . . . . . . . . . . 81
6.3.2 Invariance under positive stabilizations . . . . . . . . . . . . . . . . . 82
6.3.3 Invariance theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.4 A remark on the Ozsváth-Szabó contact invariant . . . . . . . . . . . . . . . 84
6.5 Properties of L and T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6.6 Connected sums and invariance of T . . . . . . . . . . . . . . . . . . . . . . . 88
CEU eTD Collection

6.6.1 Behaviour of the invariants under connected sum . . . . . . . . . . . 88

6.6.2 Naturality and a refinement for L . . . . . . . . . . . . . . . . . . . . 92

7 Loose and non-loose links 95

7.1 Legendrian links in overtwisted structures . . . . . . . . . . . . . . . . . . . 95
7.2 A classification theorem for loose knots . . . . . . . . . . . . . . . . . . . . . 97
7.3 Existence of non-loose links with loose components . . . . . . . . . . . . . . 99
7.4 Non-simple link types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

viii
 8 Link Floer homology and grid diagrams 105
8.1 Filtered simply blocked link grid homology . . . . . . . . . . . . . . . . . . . 105
8.1.1 The chain complex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
8.1.2 The differential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8.1.3 The homology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8.1.4 Equivalence between GH d and HFL d . . . . . . . . . . . . . . . . . . . 108
8.2 The invariant tau in the filtered theory . . . . . . . . . . . . . . . . . . . . . . 109
8.2.1 Definition of the invariant . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.2.2 Dropping the special O-markings . . . . . . . . . . . . . . . . . . . . 111
8.2.3 Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
8.3 Cobordisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
8.3.1 Induced maps and degree shift . . . . . . . . . . . . . . . . . . . . . . 115
8.3.2 Strong concordance invariance . . . . . . . . . . . . . . . . . . . . . . 120
8.3.3 A lower bound for the slice genus . . . . . . . . . . . . . . . . . . . . 121
8.4 The HF L− version of the homology . . . . . . . . . . . . . . . . . . . . . . . 123
8.4.1 A different point of view . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.4.2 Filtered link homology and the τ-set . . . . . . . . . . . . . . . . . . . 124
8.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.5.1 Computation for some specific links . . . . . . . . . . . . . . . . . . . 126
8.5.2 Torus links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.5.3 Applications to Legendrian invariants . . . . . . . . . . . . . . . . . . 128

9 Quasi-positive links in the 3-sphere 131

9.1 Maximal self-linking number and the slice genus . . . . . . . . . . . . . . . . 131
9.2 Strongly quasi-positive links . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
9.3 The case of positive links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

Bibliography 141

Index 146
CEU eTD Collection

ix
CEU eTD Collection
 List of Figures

1.1 Meridian curve along a knot . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.2 Elementary tangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Positive trefoil knot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Crossing signs in a link diagram . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Reidemeister moves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 Orientation preserving resolutions of crossings . . . . . . . . . . . . . . . . . 13
1.7 Oriented resolutions of link diagrams . . . . . . . . . . . . . . . . . . . . . . 13
1.8 Connected sum of knots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.9 Figure-eight knot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.10 Elementary cobordisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.11 Morse moves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.12 Strong cobordism between the Whitehead link and the unlink . . . . . . . . 16
1.13 Positive Hopf link . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1 2-handle attachment on genus 1 handlebody . . . . . . . . . . . . . . . . . . 28

3.1 Standard Legendrian unknot in (S3 , ξ st ) . . . . . . . . . . . . . . . . . . . . . 38

3.2 Legendrian Reidemeister moves . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3 A generator of the group of k-braids . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Trivial elementary Legendrian tangle . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Stabilizations of the standard Legendrian unknot in (S3 , ξ st ) . . . . . . . . . 41
3.6 Disjoint union of Legendrian links . . . . . . . . . . . . . . . . . . . . . . . . 41

4.1 Genus one Heegaard diagram for S3 . . . . . . . . . . . . . . . . . . . . . . . 45

4.2 Handleslide move on Heegaard diagrams . . . . . . . . . . . . . . . . . . . . 46
4.3 Stabilization move on Heegaard diagrams . . . . . . . . . . . . . . . . . . . . 46
4.4 Genus zero Heegaard diagram for the unlink 2 in S3 . . . . . . . . . . . . 48
CEU eTD Collection

4.5 System of generators on a planar surface adapted to a link . . . . . . . . . . 51

4.6 Separating arc on a page of an open book decomposition-1 . . . . . . . . . . 52
4.7 Separating arc on a page of an open book decomposition-2 . . . . . . . . . . 53
4.8 A strip on a page of an open book decomposition . . . . . . . . . . . . . . . 54
4.9 Basepoints on a page of an open book decomposition . . . . . . . . . . . . . 54
4.10 Orientation induced by the basepoints . . . . . . . . . . . . . . . . . . . . . . 55

5.1 Domains in the underlying surface of a Heegaard diagram . . . . . . . . . . 58

5.2 Stabilization with an unknotted component . . . . . . . . . . . . . . . . . . . 65
5.3 Isotopy on a Heegaard diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 68

6.1 Legendrian Heegaard diagram of a Legendrian unlink . . . . . . . . . . . . 73

xi
 6.2 Positive stabilization of an abstract open book . . . . . . . . . . . . . . . . . 75
6.3 Reducing the twisting of ξ along a curve in the page of an open book de-
composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.4 Splitting a 2-cell in a contact cell decomposition . . . . . . . . . . . . . . . . 76
6.5 Admissible arc slide on the page of an abstract open book . . . . . . . . . . . 77
6.6 Separating arcs-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.7 Separating arcs-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.8 Separating arcs-3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.9 A stabilized Legendrian Heegaard diagram for a contact 3-manifold . . . . 86
6.10 Murasugi sum of abstract open books-1 . . . . . . . . . . . . . . . . . . . . . 89
6.11 Murasugi sum of abstract open books-2 . . . . . . . . . . . . . . . . . . . . . 89

7.1 A non-loose Legendrian link with non-loose components . . . . . . . . . . . 96

7.2 Loose Legendrian knots and overtwisted disks . . . . . . . . . . . . . . . . . 98
7.3 Legendrian knots with non-zero Legendrian invariant in contact 3-spheres-1 100
7.4 Legendrian knots with non-zero Legendrian invariant in contact 3-spheres-2 101
7.5 A connected sum of multiple Legendrian Hopf links . . . . . . . . . . . . . . 102

8.1 Grid diagram for the positive trefoil knot . . . . . . . . . . . . . . . . . . . . 106

8.2 Adding a disjoint unknot to a link . . . . . . . . . . . . . . . . . . . . . . . . 110
8.3 Cobordism with no critical points . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.4 Cobordisms with an index one critical point . . . . . . . . . . . . . . . . . . . 116
8.5 Band move in a grid diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.6 Cobordism with two index one critical points . . . . . . . . . . . . . . . . . . 117
8.7 Cobordism with an index zero critical point . . . . . . . . . . . . . . . . . . . 118
8.8 Cobordism with an index zero and an index one critical points . . . . . . . . 118
8.9 Birth move in a grid diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.10 Snail-like domains in a grid diagram . . . . . . . . . . . . . . . . . . . . . . . 119
8.11 Cobordism with an index one and an index two critical points . . . . . . . . 120
8.12 Canonical form of a cobordism between two links . . . . . . . . . . . . . . . 121
8.13 O-markings in a grid diagram-1 . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.14 O-markings in a grid diagram-2 . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.15 Grid diagram for a positive torus link . . . . . . . . . . . . . . . . . . . . . . 127
8.16 Link L919n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.17 Link L940a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
CEU eTD Collection

9.1 Surface associated to a quasi-positive braid . . . . . . . . . . . . . . . . . . . 131

9.2 Quasi-positive surface (4, (σ13 , σ24 , σ13 , σ24 )) . . . . . . . . . . . . . . . . . . . 134
9.3 Transverse realization of a strongly quasi-positive link . . . . . . . . . . . . . 135
9.4 Legendrianization of a positive link . . . . . . . . . . . . . . . . . . . . . . . 138

xii
CEU eTD Collection
CEU eTD Collection
 ( ... ) while a cloud of smoke
settled heavily over the
battlements in the distinct
colossal figure of-a horse.
E. A. Poe
CEU eTD Collection
CEU eTD Collection
 Introduction

In the early years of the 21st century, Ozsváth and Szabó introduced an invariant [68],
that they decided to call Heegaard Floer homology, which had a great impact in the study
of low dimensional topology. This is an invariant of 3-manifolds up to diffeomorphism; it
is obtained from a particular presentation of the 3-manifold M, called a Heegaard diagram,
consisting of a closed, oriented, genus g surface Σ, together with two sets α and β of g
independent, disjoint, simple closed curves and a basepoint w in Σ.
A chain complex over F, the field with two elements, can be associated to a Heegaard
diagram, where the generators are combinatorially defined from α and β. The differential
is gotten by counting J-holomorphic disks in the g-fold symmetric power of Σ, equipped
with an almost-complex structure J, as we explain in Subsection 4.1.1. The Heegaard Floer
homology group is the homology of this chain complex. There exist many different ver-
sions of the homology, depending on how many J-holomorphic disks we take into account,
but all of them are invariants of M. One of the most useful, that we define in Section 5.2, is
denoted with d HF ( M).
Heegaard Floer homology groups possess an F-splitting induced by Spinc structures.
Using a definition of Turaev in [88], we say that a Spinc structure on M is the homotopy
class, away from a point, of a 2-plane field on M. This means that we can write a summand
HF ( M) as d
of d HF ( M, t), where t is a Spinc structure on M.
Moreover, we have an additional grading, called Maslov grading, in the case when
some conditions are satisfied; more specifically, we require the first Chern class of t to be
a torsion class. This always happens if M is a rational homology 3-sphere, which is a
3-manifold with rational homology isomorphic to the one of the 3-sphere S3 ; in fact, in
this case the second integer cohomology group is a finite group. For this reason, and the
fact that many proofs are easier in this setting, we suppose that 3-manifolds are rational
homology spheres unless the converse is explicitly written.
Ozsváth and Szabó found a way to extend Heegaard Floer homology to an invariant
CEU eTD Collection

of links in 3-manifolds [69], in this setting called link Floer homology. This was discovered
independently by Rasmussen in his doctoral thesis [78].
Moreover, in [70] they use it to define an invariant of contact 3-manifolds. A contact
structure ξ on M is a cooriented 2-plane field, given as the kernel of a 1-form α on M, such
that α ∧ dα is a volume form for M. We say that ξ is overtwisted if there is an embedded
disk E in M whose boundary is a Legendrian knot, which means that at each point the
tangent vector is contained in the contact structure, and the contact framing of ∂E has
zero twisting along E. The last sentence can be rephrased by saying that the Thurston-
Bennequin number of ∂E is equals to zero, see Chapter 3. We say that the structure ξ is
tight if it is not overtwisted.
In [70] the contact invariant cb( M, ξ ) is defined. This is done by identifying the isomor-
phism class of an element in the group d HF ( M, tξ ), where tξ is the Spinc structure induced

3
 by ξ. The class cb( M, ξ ) is a contact isotopy invariant of ( M, ξ ). One of its applications
consists of a simpler proof of a well-known result of Eliashberg and Gromov in [25], which
says that a symplectically fillable contact structure on a rational homology sphere is always
tight.
A contact strucutre ξ is weakly symplectically fillable if there is a compact symplectic
4-manifold ( X, ω ) such that ∂X = M and ω |ξ > 0. It follows from [70] that cb( M, ξ ) vanishes
if ξ is overtwisted, while the contact invariant is non-trivial when the contact structure is
symplectically fillable.
Our main goal in this thesis is to describe, in the best way we can, the most important
results on smooth and Legendrian knots obtained from Heegaard Floer homology in the
last ten years and generalize some of them to links.
After giving some background on smooth links in 3-manifolds and contact topology
in Chapters 1 and 3 respectively, we prove our first result: a link version of the Thurston-
Bennequin inequality. This inequality, in its usual formulation, gives an upper bound for
the maximal Thurston-Bennequin and self-linking numbers of a knot K in a tight contact
3-manifold ( M, ξ ), in terms of the Euler characteristic of a Seifert surface for K. A proof of
the Thurston-Bennequin inequality for null-homologous knots was given by Eliashberg in
[23]. We apply the same strategy, already used by Baker and Etnyre [2], to generalize the
inequality to every link L, see Section 3.6, where the resulting upper bound involves the
Thurston norm of L.
We define the Thurston norm k Lk T for every link in a rational homology sphere in
Chapter 2; it is a rational number extracted from the semi-norm, introduced by Thurston
in [86], on the relative second homology group of a 3-manifold with toric boundary. A
result of Ni [62] and Ozsváth and Szabó [73] implies that link Floer homology detects
k LkT . The results on the Thurston-Bennequin inequality and the Thurston norm appear in
[12].
The computation of link Floer homology in the J-holomorphic setting happens to be
very hard. To solve this problem, in [58] Manolescu, Ozsváth, Szabó and Thurston find a
combinatorial reformulation of the homology, in terms of grid diagrams. These are grids
of squares, that were already known in the 19th century, which can also be used to present
links in S3 .
We say that two n-component links in S3 are strongly concordant if there is a cobordism
between them consisting of n disjoint annuli, each one realizing a knot concordance. Then,
starting from a grid diagram D of a link L, we define a filtered chain complex GC ( D ), ∂
c b

and we prove in Section 8.3 that its homology, denoted with HF \ L( L), is a strong concor-
CEU eTD Collection

dance invariant. Our claim is similar to a result of Pardon [74] about Lee homology.
Since we can prove that HF
\ L( L) has dimension one in Maslov grading zero, we also
extract a numerical invariant from the homology group that we call τ ( L) because, in the
case of knots, it coincides with the concordance invariant introduced by Ozsváth and Sz-
abó in [69]. The τ-invariant gives a lower bound for the slice genus g4 ( L), which is the
minimum genus of an oriented, compact surface properly embedded in D4 and whose
boundary is L; see Subsection 8.3.3.
In Chapter 8 we also show that there is a strict relation between the filtration levels of
HF
\ L( L) and the Alexander grading of the torsion-free quotient of cHFL− ( L), a different
bigraded version of link Floer homology introduced in [67, 72]. The results about the τ-
invariant and link Floer homology appear in [9].

4
 In Chapter 4 we talk in detail about Heegaard diagrams. More specifically, we describe
how to construct such a diagram starting from a link L in a 3-manifold M and we show
how to relate two diagrams representing smoothly isotopic links in M. As we remarked
before, Heegaard diagrams play a crucial role in the definition of Heegaard and link Floer
homology, see also Chapter 5, but in the case that M is equipped with a contact structure ξ
and L is Legendrian, we need to use a different presentation, which takes into account the
contact information.
For this reason we introduce open book decompositions. Given a contact 3-manifold
( M, ξ ), this is a pair ( B, π ) where B is a smooth link in M and π is a locally trivial fibration
of M \ B onto S1 , such that the closures of the fibers have B as boundary. When some
compatibility conditions, described in Section 4.2, are satisfied, we say that ( B, π ) supports
the structure ξ.
In [57] Lisca, Ozsváth, Stipsicz and Szabó define open book decompositions ( B, π, A),
adapted to a Legendrian knot K, by also requiring that K is contained in S1 = π −1 (1)
and A is an appropriate basis of the relative first homology group of S1 . In this way they
construct the Legendrian knot invariant L(K, M, ξ ) by applying a result in [47].
We extend this invariant to every Legendrian link L in Section 6.1. This is done by
describing a suitable condition for an open book decomposition ( B, π, A) to be adapted to
L; this time the set A is not a basis anymore, but only a system of generators, the details of
the construction are given in Subsection 4.2.1. At this point, we define a special Heegaard
diagram D, that we call a Legendrian Heegaard diagram, for the link L in the 3-manifold
− M, given by reversing the orientation on M; we have that D is obtained up to isotopy
from ( B, π, A) and then, when L is zero in homology, the invariant L( L, M, ξ ) is the iso-
morphism class of a distinguished cycle L( D ) in the link Floer complex cCFL− ( D, tξ ). In
particular, the invariant L( L, M, ξ ) can be seen as an isomorphism class in the homology
group cHFL− (− M, L, tξ ). All the details can be found in Chapter 6.
We have that L( L, M, ξ ) is a Legendrian link invariant in the following sense. If L1 and
L2 are Legendrian isotopic links in ( M, ξ ) then, given Legendrian Heegaard diagrams Di
representing Li for i = 1, 2, we can find a chain map between cCFL− ( D1 ) and cCFL− ( D2 )
which sends L( D1 ) into L( D2 ), see Section 6.3.
In [3] Baldwin, Vela-Vick and Vértesi, using a different construction, introduce another
invariant of Legendrian links in contact 3-manifolds which generalizes L in the case of
knots in the standard 3-sphere. The same argument in [3] implies that this invariant coin-
cides with our L for every Legendrian link in (S3 , ξ st ).
Furthermore, our invariant has the same property of its knot version in [57] regard-
ing loose and non-loose links. We say that a Legendrian link L in an overtwisted contact
CEU eTD Collection

structure is non-loose if its complement is tight, while it is loose if its complement is over-
twisted. We prove in Section 7.1 that L( L, M, ξ ) vanishes when L is loose. This result gives
a sufficient condition for L to be non-loose. In fact, in Chapter 7 it allows us to prove
the existence of a non-split, non-loose n-component Legendrian link in every overtwisted
structure on S3 . Moreover, using a naturality property of connected sums, we are also able
to distinguish two non-loose Legendrian links L1 and L2 with the same classical invariants
(link type and Thurston-Bennequin and rotation numbers) and Legendrian isotopic com-
ponents. In other words, we say that the link type of L1 and L2 is non-loose non-simple.
The results about the invariant L and adapted open book decompositions appear in [10].
The invariant L seems to give less information in the case of loose links. In fact, we have
some results that go in the opposite direction. One of these is a theorem of Dymara, see

5
 [17], which says that two loose knots L1 and L2 in ( M, ξ ), with the same classical invariants,
such that there is an overtwisted disk in M \ ( L1 t L2 ) are always Legendrian isotopic. In
Section 7.2 we generalize this theorem as follows. We relax the condition of having an
overtwisted disk disjoint from both knots, but we claim that, in order to have a Legendrian
isotopy between L1 and L2 , we only need the existence of two disjoint overtwisted disks E1
and E2 such that Ei is in the complement of Li for i = 1, 2. An example where we can apply
our result, but not Dymara’s theorem, is the case of a disjoint union of two non-loose knots;
in other words, when M is a contact connected sum of M1 and M2 and Li is a non-loose
knot in Mi for i = 1, 2. The proofs of these results appear in [11].
Finally, in Chapter 9 we give some results on quasi-positive links in S3 . These are links
which are obtained as closures of some particular braids, called quasi-positive braids. In
particular, we show that we can always compute the maximal self-linking number and the
invariant τ of a quasi-positive link.
Moreover, in Section 9.1 we introduce the subfamily of connected transverse C-links:
these are links such that the surface Σ B , associated to a quasi-positive braid B for L, is
connected; see Section 9.1. We show that, for this kind of links, the slice genus g4 is deter-
mined by τ. This allows us to prove that the slice genus is additive under connected sums
of connected transverse C-links; a result which follows from [44] for knots, but appears to
be new in the case of links.

Budapest, March 14 2018

CEU eTD Collection

6
 Acknowledgements

I would like to thank Paolo Lisca and András Stipsicz for their advice, the many helpful
conversations and the lots of time spent meeting and discussing mathematics.
I want to thank John Etnyre and Brendan Owens for their responsiveness and the help
given in these years.
I am grateful to Irena Matkovič for all the time spent together.
After a long time, and maybe a lifetime again, I say thank you to my family.

Budapest, March 21 2018

CEU eTD Collection

7
CEU eTD Collection
 Chapter 1

Preliminaries

1.1 Knots and links in 3-manifolds

1.1.1 Definition of links
We start with the definition of knots and links. Let us consider a connected, compact,
oriented 3-manifold M. Then an n-component link is the image of a smooth embedding of a
collection of n disjoint circles into M. We usually denote a link with L ,→ M. If we orient
all the circles then the link L also inherits an orientation. In this thesis we always assume
that a link is oriented, even when this is not explicitly stated. We call a knot a link with one
component.
Let L1 and L2 be two links in a 3-manifold M. Then we say that L1 is smoothly isotopic
to L2 if there exists a smooth map H : M × I → M such that

1. H (·, t) is a diffeomorphism for every t ∈ I,

2. H (·, 0) = Id M ,

3. H ( L1 , 1) = L2 .

When two links are smoothly isotopic we also say that they are equivalent. The trivial knot,
which is called the unknot and is denoted with the symbol , is a knot in M that bounds
an embedded disk. The unknot is well-defined in every 3-manifold M, in the sense that all
such knots are equivalent. See [79].
If a link L is null-homologous in M, which means that the homology class [ L] in
CEU eTD Collection

H1 ( M; Z) is zero, then it bounds a compact, oriented embedded surface F inside M. It

is easy to see that such a surface can be made connected by adding some tubes and in this
case we call F a Seifert surface for L. We denote with g3 ( L) the minimum genus among
all the Seifert surfaces for the link L. Then the following corollary is an immediate conse-
quence of this definition.

Corollary 1.1.1 The unknot is the only knot K, up to isotopy, such that g3 (K ) = 0.

We say that a link L ,→ M is split if L is the disjoint union of L1 with L2 and there is a
separating, embedded 2-sphere S in M such that S gives the connected sum decomposition
M = M1 #S M2 and Li ,→ Mi for i = 1, 2. When this happens we write L = L1 t L2 .
Otherwise L is called non-split. Every link L can be written as L1 t ... t Lk where each Li is
a non-split link in M. The Li ’s are called split components of L.

9
 We recall that a k-manifold Y is a rational homology sphere if Hi (Y; Q) ∼
= Hi (Sk ; Z) for
every i ∈ Z. Let us take two knots K1 and K2 in M and suppose that the 3-manifold M is
a rational homology 3-sphere. Then we can define the rational linking number between K1
and K2 as the rational number
F · K2 | F t K2 |
lkQ (K1 , K2 ) = = ;
t t
where F is a compact, oriented surface, properly embedded in MK1 = M \ ν(K ˚ 1 ) and such
that F · µ = PD[µ][ F ] = t, while t is the order of [K1 ] in H1 ( M; Z). Moreover, here [µ] and
[K2 ] are elements of H1 ( MK1 ; Q) ∼
= Q. The curve µ represents the meridian of the knot K1

K1

Figure 1.1: The meridian µ is oriented accordingly to the knot K1 .

as shown in Figure 1.1.

Proposition 1.1.2 The linking number is symmetric. Moreover, if K1 is null-homologous in M
then lkQ (K1 , K2 ) is an integer.
Proof. The symmetry follows from Poincaré-Lefschetz duality. See [43] for details. On
the other hand, the second claim is implied by the fact that, if t = 1, the linking number
coincides with the algebraic intersection of a Seifert surface for K1 with the knot K2 ; which
is clearly an integer.

In the case when J = J1 ∪ ... ∪ Jn and L = L1 ∪ ... ∪ Lm are two links in a rational
homology sphere M, we define the linking number as
n m
lkQ ( J, L) = ∑ ∑ lkQ ( Ji , L j ) . (1.1)
i =1 j =1
CEU eTD Collection

We observe that the linking number between two components of a link is invariant un-
der smooth isotopies. Moreover, if J and L are both null-homologous then we omit the
subscript Q; thus, the linking number is denoted with lk( J, L) and it is an integer from
Proposition 1.1.2.

1.1.2 Connected sum and mirror image of links

We define an elementary tangle as a pair ( D3 , α) such that α ∼ = I = [0, 1] and ∂D3 ∩ α =
∂α. See Figure 1.2. An elementary tangle is called trivial if it is isotopic, rel boundary, to
( D2 × I, {0} × I ).
Given two links Li ,→ Mi for i = 1, 2 we define a connected sum L1 # L2 as follows. Let
D1 , D2 be two balls such that ( Di , Di ∩ Li ) are trivial elementary tangles. Then we glue the

10
 α

Figure 1.2: Examples of elementary tangles. The one on the right is trivial.

complements of these elementary tangles in Mi using an orientation reversing diffeomor-

phism Φ : ∂D1 → ∂D2 in the way that D1 ∩ L1 is identified with D2 ∩ L2 , respecting the
orientations of the links. The result of this operation is a pair ( M1 # M2 , L1 # L2 ), where
L1 # L2 is a connected sum of L1 and L2 .
The link L1 # L2 is well-defined up to the choice of the components of Li that intersect
the trivial elementary tangles in the definition. This means that, if we fix a component Ki
in Li for i = 1, 2, the link L1 #(K1 ,K2 ) L2 is determined up to smooth isotopy. Moreover, if
L1 and L2 are oriented then L1 # L2 is also oriented and, from [55], we have the following
properties.
Proposition 1.1.3 Once we fix the components of L1 and L2 where we sum the links, the connected
sum is commutative and associative.
Starting from a link L ,→ M, we can always reverse the orientation of the 3-manifold
M and then L defines a new link when seen as embedded in − M. Now, suppose that the
manifold M admits a diffeomorphism i− : M → M which reverses the orientation. Then
in this case the oriented link L∗ = i− ( L) is called the mirror image of L.

1.2 Links in the 3-sphere

1.2.1 Diagrams for links in S3
Every link embedded in S3 can be represented by a planar diagram. In fact, first we
CEU eTD Collection

Figure 1.3: An oriented diagram for the positive trefoil knot.

observe that a link L ,→ S3 is also embedded in R3 ; then we choose an affine 2-subspace

H ⊂ R3 , which is diffeomorphic to R2 , such that for the corresponding orthogonal projec-
tion p H : R3 → H is
• | p− 1
H ( x ) ∩ L | 6 1 for every x ∈ H except a finite number of points, called singular
values, where this number is equal to 2;
• the lines tangent to L in the singular values never project to the same line in H.

11
 A proof of the fact that the plane H exists for every link L in S3 can be found in [55]. At this
point we define a diagram D for L as the image of L under the projection p H . Moreover,

+ −
Figure 1.4: The signs of a crossing in a diagram.

we say that

• a crossing in D is a singular value;

• an arc in D is a line which connects two crossings, possibly the same, one as starting
point and the other one as end point;

• each crossing contains the information of which arcs are overcrossings or undercross-
ings.

When the link is oriented, a sign can be assigned to each crossing in the diagram D, as
shown in Figure 1.4.
A theorem of Reidemeister assures us that two links are equivalent if and only their
diagrams are related by some moves, which are called Reidemeister moves. There are

Type 1 Type 2 Type 3

Figure 1.5: The three Reidemister moves. For each move, we have to consider all the
possible combinations of reflections and orientations.
CEU eTD Collection

three types of these moves and they are pictured in Figure 1.5.

Theorem 1.2.1 (Reidemeister) Two links L1 and L2 in S3 are smoothly isotopic if and only if,
given D1 , D2 diagrams for L1 , L2 respectively, the two diagrams D1 and D2 differ by a finite se-
quence of Reidemeister moves and planar isotopies.

A proof of this theorem is found in [55].

1.2.2 The Seifert algorithm

In the 3-sphere we have a method which allows us to determine a Seifert surface for a
given oriented link L, starting from a diagram D. We define the oriented resolution of D as

12
 the collection of circles in R2 obtained by resolving all the crossings in D preserving the
orientation. It is easy to see that, for each crossing, this can be done only in one possible
way, as shown in Figure 1.6. The circles have the orientation induced on them by D.
From the oriented resolution we can construct a compact, oriented surface F in the
3-space which is bounded by the link L. We start by pushing all the circles up, starting
from the innermost ones, until they are all on different levels and then we take the disks
that they bound. So we now have a collection of disjoint disks in R3 . We connect these

Figure 1.6: Examples of orientation preserving resolutions.

disks by attaching negative (positive) bands in correspondence with the positive (negative)
crossings in D. The orientation on F is defined in the following way: it coincides with the
one induced by R2 on the circles oriented counter-clockwise, while it is the opposite on the
circles oriented clockwise. It is easy to check that such orientation can be extended to the
whole surface F.
Since we can suppose that the resolutions are always performed in a very small neigh-
borhood of the crossing, we can connect two circles, in the oriented resolution of D, by
adding lines in correspondence with each crossings. In this way we define a planar sub-
space D associated to the oriented resolution, as in Figure 1.7. The subspace D is homo-
topy equivalent to the 4-valent graph given by D when we remove the under and over-
crossings. Moreover, the surface F is connected if and only if its associated planar subspace
is connected. The latter property, in knot theory, is the definition of non-split diagram of a

D0 D0

F0
CEU eTD Collection

Figure 1.7: A diagram and its associated oriented resolution. The surface F 0 is obtained by
applying the given algorithm to D 0 .

link; moreover, it implies the following theorem.

Theorem 1.2.2 (Seifert algorithm) Every diagram D of a link in S3 gives a Seifert surface by
applying the procedure described above.
Proof. When D is non-split, the surface F determined by the oriented resolution of the
diagram is a Seifert surface for L for what we say before. If D is split then we need to
connect the components of F together by removing small disks and inserting long, thin
tubes; at this point the claim follows easily.

13
 1.2.3 Operations on links and planar diagrams
In Subsection 1.1.1 we defined the linking number between two knots. In S3 such

Figure 1.8: A diagram for the connected sum between a negative trefoil knot and a figure-
eight knot.

number is always an integer, since the group H1 (S3 ; Z) is trivial, and it is computed easily
using knot diagrams .

Proposition 1.2.3 Let D1 , D2 be diagrams for the oriented knots K1 , K2 respectively. Then

lk(K1 , K2 ) = ∑ ep ,
p | D1 ↑ D2

where the symbol ↑ means that, at the crossing p, the arc in D1 overcrosses the one in D2 and e is
the sign of p, according to Figure 1.4.

Proof. It follows easily from the definition of linking number; see [79].

Given a diagram D, we can also define the writhe of D, and we denote it with wr( D ),
as
wr( D ) = ∑ ep ,
p

where e and p are as in Proposition 1.2.3. Here the sum is taken over all the crossings in D.
The writhe is not a link invariant, but it has the following property.
CEU eTD Collection

Figure 1.9: A diagram for the Figure-eight knot and its mirror image. It is easy to see that
the two knots are isotopic.

Proposition 1.2.4 The writhe of the diagram D is invariant under Reidemeister 2 and 3 moves,
while it changes by ±1 under Reidemeister 1 moves.

Proof. It is enough to check that the claim holds for the moves in Figure 1.5.

14
 Figure 1.8 shows how the connected sum between two links can be seen from planar
diagrams . On the other hand, a diagram for the mirror image of L is gotten from a diagram
of L by changing the sign of all the crossings. See Figure 1.9.

1.3 Cobordisms
1.3.1 Definition of cobordism and the slice genus
We start this section by giving the definition of cobordism for links in S3 . A genus g

Figure 1.10: The standard cobordisms in S3 × I. Each one contains at most one critical
point.

cobordism, between links L1 and L2 in the 3-sphere, is the image of a smooth embedding
f : Σ g → S3 × I, where Σ g is a compact, oriented surface of genus g; more precisely, the
surface Σ g has connected components Σ g1 , ..., Σ g J and g = g1 + ... + g J . Furthermore, we

Birth

Band

Death
CEU eTD Collection

Figure 1.11: The three Morse moves. Birth and Death moves consist of creating and delet-
ing disjoint unknots.

have that Σ g satisfies the following properties:

1. f (∂Σ g ) = (− L1 ) × {0} t L2 × {1};

2. f (Σ g \ ∂Σ g ) ⊂ S3 × (0, 1);

3. every connected component of Σ g has boundary in both L1 and L2 .

In all the figures in this section cobordisms are drawn as standard surfaces in S3 , but they
can be knotted in S3 × I.

15
 It is a standard result in Morse theory, see [59], that a link cobordism can be decom-
posed into five standard cobordisms, which appear in Figure 1.10. Then this gives a gen-
eralization of Reidemeister theorem.
Theorem 1.3.1 Let D1 and D2 be two diagrams for the links L1 and L2 in S3 . Then for every cobor-
dism Σ, between L1 and L2 , the diagram D2 is obtained from D1 by a finite number of Reidemeister
and Morse moves, where the latter ones are pictured in Figure 1.11.
In particular, there is a cobordism between L1 and L2 in S3 that does not have any
critical point if and only if L1 is smoothly isotopic to L2 .
We say that a cobordism is a strong cobordism if the links have the same number of
components and, together with the properties stated above, the surface Σ g is such that each
connected component Σ gi determines a knot cobordism. Figure 1.12 shows an example of

W
W

Band moves

Figure 1.12: The Whitehead link W can be changed into the 2-component unlink by per-
forming two Band moves. Note that one component of W remains untouched.

this type of cobordism.

If there is a genus zero cobordism between a link J and the unknot then J is called
weakly slice. This is equivalent to say that J bounds a properly embedded punctured disk
in D4 . Since H1 ( D4 , S3 ; Z) is trivial, we conclude that there is always a compact, oriented
surface, which is properly embedded in D4 and bounds a given link L. We define the slice
genus of L as the minimum genus of a surface of this kind and we denote it with g4 ( L).
Obviously, it follows that g4 ( L) 6 g3 ( L).
CEU eTD Collection

1.3.2 Link concordance

Two n-component links L1 and L2 are concordant, or strongly concordant, if there exists
a genus zero strong cobordism between them. In other words, when the cobordism Σ
consists of n disjoint annuli and the boundary of each of these is the union of a component
of L1 and one of L2 . Theorem 1.3.1 implies the following corollary.
Corollary 1.3.2 Suppose that L1 is smoothly isotopic to L2 as a link in S3 . Then L1 and L2 are
concordant.
If a link L is concordant to the n-component unlink then we say that L is slice, or strongly
slice. In particular, the link L is slice if and only if it bounds n disjoint disks, properly
embedded in D4 .

16
 D

Figure 1.13: A diagram for the positive Hopf link.

In the case of knots, we immediately observe that the notions of weakly and strongly
slice coincide. This is no longer true when we consider links with at least two components.
In fact, let us take the positive Hopf link H+ , which is represented by a the diagram D in
Figure 1.13. Then H+ is weakly slice, because the surface obtained by applying the Seifert
algorithm to D is diffeomorphic to an annulus. On the other hand, it cannot be strongly
slice since, say K1 and K2 are its components, we have that lk(K1 , K2 ) = 1 and this would
contradict the following proposition.

Proposition 1.3.3 If a link L = L1 ∪ ... ∪ Ln is strongly cobordant to an unlink then lk( Li , L j ) = 0

for every i 6= j. As a consequence, the linking numbers between the components of a link are
concordance link invariants.

Proof. The claim can be proved by using Proposition 1.2.3 and Theorem 1.3.1.

We saw in Subsection 1.1.2 that the connected sum is a well-defined, associative and
commutative operation on the set of knots, up to isotopy. However, this operation does
not give rise to a group because a non-trivial knot cannot be inverted. This changes when
we consider the relation of knot concordance. In fact, we now have the following theorem.

Theorem 1.3.4 The set of all the knots in S3 up to concordance, equipped with the operation given
by the connected sum, is an abelian group that it is called smooth knot concordance group and it is
denoted with C1 .

Proof. We have that the connected sum is well-defined up to concordance; this follows
from the properties of the connected sum and the fact that if K and L are two concordant
knots then K # − L∗ is slice, which is proved in [55], where − L∗ denotes the mirror image
of the knot obtained from L by reversing the orientation.
Therefore, if [K ] is the concordance class of K, we can define
CEU eTD Collection

[ K ] + [ L ] = [ K # L ], −[K ] = [−K ∗ ] and 0 = [ ]

for every knot K and L and where is the unknot. It is now easy to check that C1 is indeed
an abelian group.

The algebraic structure of the group C1 has not been determined completely, but there
are some results in this direction. In [34] Fox and Milnor prove that C1 has 2-torsion, by
showing that the Figure-eight knot, see Figure 1.9, which is isotopic, and then concordant,
to its reverse mirror image is not a slice knot. Moreover, the work of Levine in [53, 54] tells
us that the knot concordance group has a direct summand isomorphic to Z∞ and then it is
not finitely generated.

17
 1.4 Filtered chain complexes
1.4.1 Filtered spaces ...
For the purposes of this background section, we fix a field K. Let us consider a graded
chain complex C = (C, ∂), where C is a K-vector space and ∂ : C → C is a linear map,
satisfying the following properties:

• there is a splitting of C as a K-vector space C =

M
Cd . In other words, the space C
d ∈Z
is graded;

• the differential ∂ is such that ∂ ◦ ∂ = 0 and it is compatible with the grading, in the
sense that ∂(Cd ) ⊂ Cd−1 .
We equip C with a sequence of K-subspaces F s C ⊂ C with F s C ⊂ F s+1 C for every s ∈ Z,
which is such that
F sC = C .
[

s ∈Z

This collection of subspaces is called a filtration on C if the following conditions hold:

F s Cd ;
M
• if we define F s Cd = (F s C ) ∩ Cd then it is F s C =
d ∈Z

• the filtration is compatible with the differential, which means that ∂(F s C ) ⊂ F s C;

• for every d ∈ Z there is an sd such that F sd Cd = {0}.

A complex C as above is called filtered chain complex. Moreover, the filtration level of a non-
zero element x ∈ C is the minimal s for which x ∈ F s C ⊂ C.
We now define the graded
object associated to a filtered complex C to be the bigraded
chain complex gr(C ), gr(∂) , where

F s Cd
gr(C )d,s =
F s−1 Cd
and gr(∂) is the map induced by ∂ on gr(C ). We denote the associated graded object with
gr(C).
The filtered chain complex C and its associated graded object, which is a bigraded chain
complex, gr(C) give us two homology groups. In the second case the space is constructed
CEU eTD Collection

easily. In fact, we say that

M Ker gr(∂)d,s
H∗,∗ (gr(C)) = Hd,s (gr(C)) =

d,s∈Z
Im gr(∂)d+1,s

and the result is a bigraded space.

On the other hand, the complex C is graded and then we can define the homology
group
M Ker ∂d
H∗ (C) = Hd (C) = ,
d ∈Z
Im ∂ d +1
which is a graded K-vector space. Moreover, since C is also equipped with a filtration F ,
we can find a way to induce F on H∗ (C). More specifically, we introduce the subspaces

18
 F s Hd (C) as follows: consider the projection πd : Ker ∂d → Hd (C). Denote with Ker ∂d,s the
subspace Ker ∂d ∩ F s Cd ; we say that

F s Hd (C) = πd (Ker ∂d,s )

for every s ∈ Z. Thus, the fact that Ker ∂d,s ⊂ Ker ∂d,s+1 implies that the filtration F
descends to homology. We can extend the filtration F on the total homology H∗ (C) by
taking
F s H∗ (C) = F s Hd (C) .
M

d ∈Z

1.4.2 ... and filtered maps

Fix two filtered, graded chain complexes C = (C, ∂) and C 0 = (C 0 , ∂0 ) over K. A chain
map f : C → C 0 is a linear map such that ∂0 ◦ f = f ◦ ∂. A chain map preserves the grading
if f (Cd ) ⊂ Cd0 and it is filtered of degree t if f (F s C ) ⊂ F s+t C 0 for every d, s ∈ Z. A filtered
chain map induces a map in homology that is filtered of the same degree. This means that
f induces a map f ∗ : H∗ (C) → H∗ (C 0 ) such that f ∗ (F s H∗ (C )) ⊂ F s+t H∗ (C 0 ) for every
s ∈ Z.
Now, given two chain maps f , g : C → C 0 filtered of degree t, such that they preserve
the gradings, a filtered chain homotopy from g to f is a map H : C → C 0 that satisfies the
following properties:

• H maps F s Cd into F s+t Cd0 +1 ;

• H satisfies the homotopy relation

∂0 ◦ H + H ◦ ∂ = f − g ; (1.2)

• H is a linear map.

Two maps as the ones above are said to be filtered chain homotopic.
Two chain complexes C and C 0 are filtered chain homotopy equivalent if there are maps
f : C → C 0 and g : C 0 → C, which preserve the gradings and are filtered of degree zero,
with the property that the maps f ◦ g and g ◦ f are filtered chain homotopic to the respec-
tive identity maps. In this case, the map f is called a filtered chain homotopy equivalence;
moreover, the maps f and g are said to be filtered chain homotopy inverses of one another.
CEU eTD Collection

A filtered chain map f naturally induces a chain map gr( f ) : gr(C ) → gr(C 0 ) between
the associated graded objects. It is easy to see that, if f : C → C 0 is a filtered chain homo-
topy equivalence, the map gr( f ) is also a chain homotopy equivalence.
Furthermore, we call filtered quasi-isomorphism a chain map f : C → C 0 , which is again
filtered of degree zero and preserves the grading, whose associated graded map gr( f ) in-
duces an isomorphism between the homology groups H∗,∗ (gr(C)) and H∗,∗ (gr(C 0 )). We
say that C and C 0 are filtered quasi-isomorphic if there exists a third complex C 00 and filtered
quasi-isomorphisms from C 00 to C and from C 00 to C 0 . Then, from [67], the following propo-
sition holds.

Proposition 1.4.1 Two filtered, graded chain complexes C and C 0 over a field K are filtered quasi-
isomorphic if and only if they are filtered chain homotopy equivalent.

19
 Because of this result, we always write that two filtered chain complexes are filtered
chain homotopy equivalent, instead of saying that they are quasi-isomorphic.
We say that a linear map F : H∗ (C) → H∗ (C 0 ) is a filtered isomorphism if F and its in-
verse are both filtered of degree zero. In general, we also require that such isomorphism
preserves the grading; we explicitly write when this does not happen. The following corol-
lary follows immediately from this definition.
Corollary 1.4.2 The map F as above is a filtered isomorphism if and only if
F s Hd (C) ∼ =K F s Hd (C 0 )
for every d, s ∈ Z.
Furthermore, we can prove the following proposition.
Proposition 1.4.3 If C and C 0 are filtered chain homotopy equivalent complexes over K then
H∗,∗ (gr(C)) ∼
= H∗,∗ (gr(C 0 )), as bigraded K-vector spaces, and H∗ (C) is filtered isomorphic to
H∗ (C 0 ).
Proof. The first implication follows from Proposition 1.4.1 and the definition of quasi-
isomorphism. For the second implication, say φ is a filtered chain homotopy equivalence
between C and C 0 , we notice that φ has a filtered chain homotopy inverse, that we call ψ,
and, since the homotopy condition in Equation (1.2) holds, we have that the induced maps
in homology φ∗ and ψ∗ are inverses one of the other. Now, we just need to observe that φ
and ψ are both filtered of degree zero and then this holds for φ∗ and ψ∗ too. This implies
precisely that φ∗ is a filtered isomorphism.
Let f : C → C 0 be a chain map, which is filtered of degree t and preserves the grading.
We can define its associated mapping cone , that is denoted with Cone( f : C → C 0 ), whose
underlying filtered chain complex is exactly
F s (C ⊕ C 0 )d = F s−t Cd−1 ⊕ F s Cd0 ;
while the differential is given by
∂Cone : (C ⊕ C 0 )d → (C ⊕ C 0 )d−1
∂Cone ( x, y) = (−∂( x ), ∂(y) + f ( x )) .
Then we have the following result about maps between mapping cones, whose proof ap-
pears in [67].
Lemma 1.4.4 Let C , C 0 , E , E 0 be four filtered, graded chain complexes over K. Suppose that there
are filtered chain maps f and g as in the following square.
CEU eTD Collection

f
C C0
φ φ0
g
E E0

Moreover, we can find two filtered chain homotopy equivalences φ and φ0 such that φ0 ◦ f is
filtered chain homotopic to g ◦ φ. Then Cone( f ) and Cone( g) are filtered homotopy equivalent.
This lemma will be useful in the remaining of the thesis. In particular, we use it to
prove a result in Chapter 8.

20
 1.5 Tensor product of modules
1.5.1 General properties and vector spaces
In this section we recall the definition of tensor product of R-modules, where R is a
commutative ring. More details can be found in [1].
Let M and N be two R-modules over the commutative ring R. Then the tensor product
of M with N over R is an R-module M ⊗ R N, together with the canonical R-bilinear map

⊗ : M ⊕ N −→ M ⊗ R N ,

which is universal in the following sense: for every R-module L and every R-bilinear map
f : M ⊕ N → L, there is a unique R-linear map

φ : M ⊗ R N −→ L

such that φ ◦ ⊗ = f ; which means that the following diagram commutes.

⊗
M⊕N M ⊗R N

φ
f
L

The fact that the R-module M ⊗ R N exists and it is unique, up to a canonical isomor-
phism, is proved in [1]. Moreover, we have the following canonical isomorphisms:

1. identity:
R ⊗R M = M ; (1.3)

2. associativity:
( M ⊗ R N ) ⊗ R P = M ⊗ R ( N ⊗ R P) ; (1.4)

3. symmetry:
M ⊗R N = N ⊗R M ; (1.5)
CEU eTD Collection

4. distributivity:
M ⊗ R ( N ⊕ P) = ( M ⊗ R N ) ⊕ ( M ⊗ R P) . (1.6)

The following proposition can also be found in [1].

Proposition 1.5.1 (Right exactness) If the following sequence of R-modules

f g
0 −→ N1 −→ N −→ N2 −→ 0

is exact, then
1⊗ f 1⊗ g
M ⊗ R N1 −−→ M ⊗ R N −−→ M ⊗ R N2 −→ 0
is also an exact sequence, where (1 ⊗ f )( x ⊗ y) = x ⊗ f (y).

21
 From these properties we can immediately compute the tensor product of finite dimen-
sional vector spaces over the field K.
Proposition 1.5.2 If V and W are K-vector spaces, whose dimensions are equal to n and m re-
spectively, then we have that V ⊗ W is isomorphic to Knm . Furthermore, if B1 = {v1 , ..., vn } and
B2 = {w1 , ..., wm } are basis for V and W then {v1 ⊗ w1 , ..., vn ⊗ wm } is a basis for V ⊗ W.
Proof. We have that V ∼
=K Kn and W ∼
=K Km . Then, using Equations (1.4), (1.5) and (1.6),
we obtain
n M m
V ⊗W ∼ (K ⊗ K) ,
M
=K
i =1 j =1

which is isomorphic to Knm by Equation (1.3).

1.5.2 Finitely generated K[ x ]-modules

Suppose now that R = K[ x ], where K is a field. Since K[ x ] is a principal ideal domain,
from [1] we know that it is in particular a Dedekind domain and this implies that a K[ x ]-
module is the sum of cyclic submodules. More specifically, if M is a module over K[ x ] and
it is finitely generated then we can write
!
n
K [ x ]
M∼= K[ x ] K [ x ] r ⊕
M
;
f ( x )K[ x ]
i =1 i

where each f i ( x ) is a non-zero polynomial in K[ x ] and r, n > 0. Therefore, we are able to

prove the following result .
Proposition 1.5.3 Let us consider M and N two finitely generated K[ x ]-modules presented as
follows:
n m
K[ x ] K[ x ]
M∼ and N ∼
M M
= K[ x ] = K[ x ] ,
f ( x )K[ x ]
i =1 i
g ( x )K[ x ]
j =1 j

where f i ( x ), g j ( x ) ∈ K[ x ] for every i = 1, ..., n and j = 1, ..., m. Then the tensor product of M
with N over K[ x ] is given by
n M
m
K[ x ]
M ⊗ K[ x ] N ∼
M
= K[ x ] .
i =1 j =1
gcd( f i ( x ), g j ( x ))K[ x ]
CEU eTD Collection

Before starting the proof, we need a lemma.

Lemma 1.5.4 If I and J are two ideals of R then
R ⊗ R = R
I R J I+J.
Proof. Tensoring with an R-module M and using Proposition 1.5.1, the exact sequence

0 −→ I −→ R −→ RI −→ 0

gives the exact sequence

f
I ⊗ R M −→ R ⊗ R M = M −→ RI ⊗ R M −→ 0 ,

22
 where f is given by i ⊗ x 7→ ix. From the fact that the image of f is the submodule I M, we
then obtain that
R ⊗ M = M .
I R IM
Hence, the claim follows once we put R/J in place of M.

Now we can go back to Proposition 1.5.3.

Proof of Proposition 1.5.3. We apply the properties of the tensor product that we stated in
Subsection 1.5.1 and we obtain that, in order to prove the statement, it is enough to show
that the following relation

K[ x ] K[ x ] ∼ K[ x ]
⊗ =
f ( x )K[ x ] K[ x] g( x )K[ x ] K[ x] gcd( f ( x ), g( x ))K[ x ]

holds for every f ( x ), g( x ) ∈ K[ x ]. This is done easily from Lemma 1.5.4 by taking R =
K[ x ], I = ( f ( x )) and J = ( g( x )).
CEU eTD Collection

23
CEU eTD Collection
 Chapter 2

The Thurston norm

2.1 Definition
Thurston in [86] introduced a semi-norm on the homology of some 3-manifolds. In this
chapter we recall the construction in the specific case of rational homology spheres.
Let us consider a compact, connected, oriented 3-manifold Y such that its boundary
consists of some tori. We call the complexity of a compact, oriented surface F properly
embedded in (Y, ∂Y ), the integer
m
χ− ( F ) = − ∑ χ( Fi ) ,
i =1

where F1 , ..., Fm are the connected components of F which are not closed and not diffeo-
morphic to disks. If m happens to be zero then we say that the complexity is also zero.
Then we define the function

k·kY : H2 (Y, ∂Y; Z) −→ Z>0

by requiring that
k akY = min {χ− ( F )} ,
where F is a surface as above that represents the relative homology class a ∈ H2 (Y, ∂Y; Z).
It can be shown [7] that
1. klakY = |l | · k akY for any l ∈ Z and a ∈ H2 (Y, ∂Y; Z),
CEU eTD Collection

2. kla + mbkY 6 |l | · k akY + |m| · kbkY for any l, m ∈ Z and a, b ∈ H2 (Y, ∂Y; Z).
Suppose from now on that M is a rational homology 3-sphere and L is a smooth link in
M. This implies that H2 ( ML , ∂ML ; Q) ∼
= Qn , where ML = M \ ν(˚L) and n is the number of
components of L. Then the Thurston semi-norm

k·k ML : H2 ( ML , ∂ML ; Q) −→ Q>0

is defined as before for integer homology classes and extended to the whole Qn by saying
that klak ML = |l | · k ak ML for every l ∈ Q and a ∈ H2 ( ML , ∂ML ; Q).
If L is a null-homologous n-component link in M then we can easily extract a number
from k·k ML . Namely, we define the Thurston norm of L as the integer

k LkT = k[ F ]k ML ,

25
 where F is a Seifert surface for L. See Subsection 1.1.1 for the definition of Seifert surfaces.
While k·k ML is effectively a semi-norm, the integer k·k T is just a numerical invariant of the
link L.
Lemma 2.1.1 In a rational homology sphere M, all the Seifert surfaces for a null-homologous
n-component link L represent the same relative homology class in ML .
Proof. From Poincaré-Lefschetz duality, see [43], we have that H2 ( ML , ∂ML ; Z) ∼ =
H 1 ( ML ; Z) and the latter group is isomorphic to Zn , since M is a rational homology
sphere. Moreover, duality gives an identification between the Z-vector ( a1 , ..., an ) and all
the compact, oriented, properly embedded surfaces F in ML such that F · µi = ai for ev-
ery i = 1, ..., n, where µi is the meridian of the i-th component of L, see Figure 1.1. This
happens because L is null-homologous in M.
The meaning of our claim is that such surfaces represent the same relative homology
class. Therefore, since in the case when F is a Seifert surface we have that F · µi = 1 for each
i, the proof is completed because Seifert surfaces satisfy by definition all the properties that
we required.

It is clear that this lemma implies that the Thurston norm k Lk T is well-defined; in
fact, the value of the semi-norm k·k ML depends only on the relative homology class of the
surface.
In [73] Ozsváth and Szabó prove that the link Floer homology group HFL[ ( L) of a link
3
in S , introduced in [69, 72], detects the Thurston norm. This result has been generalized
to null-homologous links in rational homology spheres by Ni [62].
Theorem 2.1.2 Suppose that L is a null-homologous n-component link in a rational homology
3-sphere M. Then we have that

[ ∗,s ( L) 6= {0} = k LkT − o ( L) + n ,

n o
max s ∈ Q | HFL
2
where o ( L) denotes the number of disjoint unknots in L.
The definition of the Thurston norm k·k T does not immediately extend to links that are
not null-homologous; in fact, these links do not admit Seifert surfaces. In order to avoid
this problem we need to introduce rational Seifert surfaces, see also [2].
Let us consider an n-component link L with order t in M. This means that [ L] has order
t in the group H1 ( M; Z), which is finite because M is a rational homology sphere. Then
F is rationally bounded by L if there is a map j : Σ → M, where Σ is a compact, oriented
surface with no closed components, such that
CEU eTD Collection

• j(Σ) = F;
• j|Σ̊ is an embedding of the interior of Σ in M \ L;
• j|∂Σ : ∂Σ → L is a t-fold cover of all the components of L.
Moreover, if F is also connected then we call it a rational Seifert surface for L.
Let us consider again the meridian curves {µ1 , ..., µn } of L as in Figure 1.1; where µi
is embedded in ∂ν( Li ) and Li is the i-th component of L. As we remark in the proof of
Lemma 2.1.1, Poincaré-Lefschetz duality gives that two properly embedded surfaces F1
and F2 in ML represent the same relative homology class if and only if F1 · µi = F2 · µi for
every i = 1, ..., n, where as before we mean the algebraic intersection of the surface Fj with
the curve µi in the 3-manifold ML .

26
 Lemma 2.1.3 Suppose that F is a compact, oriented surface properly embedded in ML = M \ ν(˚L)
with F · µi = t for every i = 1, ..., n, where M is a rational homology sphere and L ,→ M is an
n-component link with order t.
Then there exists an F 0 in M which is rationally bounded by L and it is such that
χ− ( F 0 ∩ M L ) 6 χ− ( F )
and the surface F 0 ∩ ML represents the same relative homology class of F.
Proof. First, we observe that trivial properly embedded disks in ML , which are connected
components of F, do not increase the complexity χ− ( F ); then we can just delete them.
Moreover, suppose that there are other boundary components of F, whose algebraic in-
tersection with µi is zero for every i = 1, ..., n. Then either those components are parallel
to the meridians, but this cannot happen because F · µi = t for every i, either they are
homologically trivial in the tori ∂ν( L), which means that they are circles which bound a
disk. We can then push these disks slightly out of ∂ν( L), starting from the innermost ones,
and then cap off the surface. After removing all the closed components we obtain a new
surface whose complexity is smaller or equal to the one of F.
In the second step we show that we can take F 0 such that F 0 ∩ ∂ν( L) consists of essential,
parallel, simple closed curves all oriented in the same direction. Therefore, suppose there
are two components C1 and C2 of F 0 ∩ ∂ν( Li ) with opposite orientations; then there is an
innermost pair of such components, say precisely C1 and C2 without loss of generality,
such that C1 ∪ C2 is the boundary of an annulus A ⊂ ∂ν( Li ) with Ci ∩ Å empty. We can
alter the surface F 0 by attaching a copy of the annulus A to F 0 and pushing it into the
interior of ML . This operation does not change the relative homology class of F 0 nor its
complexity.
We have now obtained that F 0 is a compact, oriented surface properly embedded in
ML , representing the same relative homology class of F, such that χ− ( F 0 ) 6 χ− ( F ) and
F 0 ∩ ∂ν( Li ) is a link with slope (t, s), where s ∈ Z, in the torus ∂ν( Li ) for every i = 1, ..., n.
To conclude we need to extend F 0 inside ν( Li ) in a way that it is rationally bounded by L.
This can always be done and it is proved in [2]. This completes the proof.
From Lemma 2.1.3 and [86] we have that rational Seifert surfaces exist for every link in
a rational homology sphere. Hence, we say that the Thurston norm of a link L with order t
in M is the rational number
[ F]
k LkT = ,
t ML
where F is a rational Seifert surface for L. In the same way as for null-homologous links,
CEU eTD Collection

the following lemma implies that k Lk T is well-defined.

Lemma 2.1.4 In a rational homology sphere M, all the rational Seifert surfaces for an n-component
link L, with order t in M, represent the same relative homology class in ML .
Proof. We reason in the same way as in the proof of Lemma 2.1.1. In fact, we observe that
the intersection of each rational Seifert surface for L with ML is clearly a compact, oriented
and properly embedded surface F; but more importantly, every such surface is such that
F · µi = t whenever i = 1, ..., n. Then the claim follows again from Poincaré-Lefschetz
duality.
We also notice that if L is a null-homologous link then the two definitions coincide. In
fact, in this case it is t = 1 and rational Seifert surfaces correspond precisely to genuine
Seifert surfaces.

27
 2.2 Results
In Subsection 1.1.1 we define the disjoint union of two links. We now show that the
Thurston norm is additive under disjoint unions. First, we need the following definition.
A component K of a link L in a rational homology sphere M is called compressible if K
is rationally bounded by an F, disjoint from L \ K, that is the image of a map j : D ,→ M,
where D is diffeomorphic to a disk. We prove that compressible knots do not appear in

B
T

β = ∂F

K β

Figure 2.1: The ball B is attached to the solid torus T along the curve β.

every 3-manifold; more specifically, we show that we can find them only when M has a
lens space as a connect summand.
Lemma 2.2.1 Let us consider K, L and M as before and denote with L0 the link L \ K. Suppose
that K rationally bounds F, where F is determined by j : D → M with D a disk and j| D a p-fold
cover of K. Moreover, suppose also that F is disjoint from L0 . Then K is an order p knot in a lens
space L( p, q) such that M = L( p, q) # M0 and L0 ,→ M0 .

Proof. Let T = ν(K ) and B be the tubular neighborhoods of K and F \ ν(˚K ) respectively.
Clearly, T is a solid torus and B is a 3-ball; moreover, we can suppose that T ∪ B is disjoint
from L0 .
Denote with β the simple, closed curve in the torus ∂T given by ∂T ∩ F. Then we can
see β as the attaching sphere of the 3-dimensional 2-handle B on T, which in turn can be
thought as a lower handlebody. This implies that T ∪ B is diffeomorphic to a lens space
minus a 3-ball. See Figure 2.1. Those are standard results in the theory of 3-manifolds and
more details are in [41].
In particular, we have that T ∪ B ∼ = L( p, q) \ {3-ball} for some integer q, such that
gcd( p, q) = 1, and where p appears because j| D is a p-fold cover of K. Hence, it is M =
L( p, q) # M0 and the link L0 is contained in M0 . The fact that K is an order p knot in L( p, q) is
CEU eTD Collection

also straightforward, because K is the core of the handlebody T and it is easy to check that,
for this reason, its homology class is the generator of H1 ( L( p, q); Z) which is isomorphic
to Z/pZ.

We recall that a surface S is compressible in Y if a non-trivial circle in S bounds a disk

in Y \ S. Hence, we can also say that K is compressible if it is embedded in a 3-manifold N,
which is a connect summand of M, and ∂ν(K ) is a compressible torus in N. Later in this
section, we prove that these two definitions are equivalent.
Proposition 2.2.2 Let us consider a link L in M such that L = L1 t L2 , where each Li is embedded
in Mi with M = M1 # M2 . Then we have that

k L k T = k L1 k T + k L2 k T

28
 and the order of L in M is t = lcm(t1 , t2 ), where ti is the order of Li in Mi for i = 1, 2.

Proof. We suppose first that L1 and L2 have no compressible components. Let us consider a
surface F which is rationally bounded by L. Fix a separating 2-sphere S ,→ M, that gives a
connected sum decomposition of M, in a way that S intersects F transversely in a collection
of circles S . Moreover, we suppose that there exist neighborhoods for L1 and L2 which are
disjoint from S, each one lying in one component of M \ S. Take the map j : Σ ,→ M which
defines F; if the neighborhoods are chosen small enough then we can also suppose that
they intersect F only in the image of a neighborhood of ∂Σ.
Now each circle in S separates S into two disks. Let C ⊂ S be a circle that is innermost
on F. This means that C bounds a disk D in S, the interior of which misses F. Now use D
to do surgery on F in the following way: create a new surface Fb from F by deleting a small
annular neighborhood of C and replacing it by two disks, each a “parallel” copy of D, one
on either side of D. We then perform the surgery on F described before on all the circles
in S . At this point Fb may have closed components, but in this case we just delete them.
Therefore, we are left with a disconnected surface whose connected components, that are
no longer closed, stay in Mi according wether they bound a component of Li . We call F1
the surface given by the union of the components of Fb of the first type and F2 the other one.
We clearly have that F1 and F2 are disjoint, they lie in M1 and M2 respectively and that
they do not intersect S; moreover, they are such that

• F1 and F2 have no closed components;

• ML ∩ Fi is a proper submanifold of ML with no disk components, while Fi coincide

with F in a small neighborhood of L for i = 1, 2;

• χ( F1 ) + χ( F2 ) = χ( F1 t F2 ) > χ( F ).

Since F is rationally bounded by L and F1 t F2 coincide with F in a neighborhood of L, we

have that Fi has Li as boundary and then [ Li ] = [0] in the group H∗ ( M; Q). This implies
that t is a multiple of both t1 and t2 . This proves that t = lcm(t1 , t2 ).
We now want to show that the Thurston norm is additive. The fact that k L1 t L2 k T 6
k L1 kT + k L2 kT follows immediately from Property 2 in Section 2.1. Then we suppose that
the inequality is strict: this means that

χ( F1 ) χ( F2 ) χ( F )
− − 6− < k L1 k T + k L2 k T .
t t t
CEU eTD Collection

In particular, at least one of the two surfaces, say F1 , is such that −t−1 χ( F1 ) < k L1 k T . Then
the claim follows from the fact that by construction, if t = at1 , the class [ F1 ] coincide with
a times the class represented by a rational Seifert surface of L1 in H2 ( ML , ∂ML ; Q); which
gives a contradiction.
To conclude we need to prove that, if L is the disjoint union of a link L0 with a compress-
ible knot K, it is k Lk T = k L0 k T + kK k T . This is done in the same way as the previous case,
but we take into account the fact that disks do not increase the complexity of a surface.

The proof of this proposition also implies the following corollary.

Corollary 2.2.3 The Thurston norm of a link L in M1 coincides with the one obtained when L is
seen as a link in M, where M = M1 # M2 . Furthermore, the order of L in M1 concides with the
order of L in M.

29
 In light of Corollary 2.2.3, we use the symbol k Lk T not only when L is seen as a link
in M, but also in the case when L lies inside a connect summand of M. Moreover, we can
now prove the following proposition.

Proposition 2.2.4 The component K of a link L in M is compressible if and only if L = K t

L0 , where K ,→ N with M = N # M0 , and the boundary of the tubular neighborhood of K is a
compressible torus in N.

Proof. Let us start to prove the only if implication. It follows easily from the definition of
compressibility and Corollary 2.2.3 that K rationally bounds F, the image of a disk D under
the map j, in N; where here j is such that j| D is a t-fold cover of K. It remains to show that
t is actually the order of K in N.
We move to the if implication for the moment. In this case we have to show that the
knot K is embedded in a connect summand N of M and L0 = L \ K does not intersects N.
Then we can conclude by reasoning as in the proof of Proposition 2.2.2.
Knowing this, if we use Lemma 2.2.1 then we immediately prove both the previous
claims. This completes the proof.

We observe that if K is a null-homologous knot then being compressible is equivalent

to say that K is an unknot disjoint from L0 .
We define the compressibility term of a link L as the rational number

o ( L)
1
o( L ) = ∑ ti
,
i =1

where o ( L) is the number of compressible components in L and t1 , ..., to( L) are the orders of
such components. Then, in order to prove our main result in Chapter 3 we need another
lemma.

Lemma 2.2.5 Suppose that L is a non-split link in a rational homology 3-sphere M. Then we have
that
−χ( F )
 
k LkT − o( L) = min ,
t
where F is rationally bounded by L and t is the order of L in M.

Proof. If L is not a compressible knot then o( L) = 0, because otherwise L would be split

from Proposition 2.2.4, and so the claim follows from the definition of the Thurston norm
CEU eTD Collection

and Lemma 2.1.3.

On the other hand, if the boundary of a tubular neighborhood of L is a compressible
torus in M then F is rationally bounded by L, where F is the image of a map j : D ,→ M
with D diffeomorphic to a disk, and o( L) = t−1 . Therefore, its Thurston norm k Lk T is
equal to zero and the equality in the statement is given by F, since χ( F ) = 1.

30
 Chapter 3

Contact structures and Legendrian

links

3.1 Contact structures

3.1.1 Type of structures and convex surfaces
We recall that a contact structure ξ on an oriented 3-manifold M is a 2-plane field on M
such that ξ = Ker α, where α is a 1-form on M and α ∧ dα is a volume 3-form for M. We
say that ξ is cooriented by the form α. Moreover, we call a contact 3-manifold a pair ( M, ξ )
as before.
We define two relations between contact 3-manifolds. The first has the name of contac-
tomorphism and it is given as follows. Consider ( M1 , ξ 1 ) and ( M2 , ξ 2 ) as above; they are
contactomorphic if there exists an F : M1 → M2 , which is a diffeomorphism and it is such
that F∗ (ξ 1 ) = ξ 2 .
On the other hand, given a 3-manifold M, we say that ( M, ξ 1 ) is contact isotopic to
( M, ξ 2 ) if we can find a map F : M × I → M that satisfies the following properties:
• F (·, t) is a contactomorphism for every t ∈ I;

• F (·, 0) = Id M ;

• F (·, 1) sends ξ 1 into ξ 2 .

We observe that contactomorphisms give a relation between different contact 3-manifolds,

CEU eTD Collection

while contact isotopies only relate contact structures on a fixed 3-manifold. Furthermore,
the following corollary follows immediately from the previous definitions.

Corollary 3.1.1 Suppose ξ 1 and ξ 2 are contact isotopic structures on a 3-manifold M. Then
( M, ξ 1 ) and ( M, ξ 2 ) are contactomorphic.
An embedded disk E, in a contact manifold ( M, ξ ), is an overtwisted disk if its boundary
∂E is such that TE ⊂ ξ and E · eξ = 0, where eξ is the contact framing given by ξ. See [27].
In Section 3.2, we see that these conditions are equivalent to say that E is bounded by a
Legendrian unknot in M and tbE (∂E) = 0, where tb is called Thurston-Bennequin number
and it is also defined later.
A contact structure ξ on a 3-manifold M is an overtwisted structure if ( M, ξ ) contains an
overtwisted disk. While, we say that ξ is a tight structure if it is not overtwisted.

31
 We now need to recall the definition of convex surface in a contact 3-manifold and
dividing set. We use [27] as a reference. We start by saying that a surface S in ( M, ξ ) is
called convex if there is a vector field v, transverse to S, whose flow preserves ξ; such a
vector field is called a contact vector field. We have the following approximation result.
Proposition 3.1.2 Every closed surface Σ can be smoothly perturbed into a convex surface. Fur-
thermore, in the case Σ has the link L = L1 ∪ ... ∪ Ln as boundary, if L is such that TL ⊂ ξ and
Σ · Li,ξ 6 0 for i = 1, ..., n, where Li,ξ is the contact framing of Li , then Σ can be again smoothly
perturbed into a convex surface, fixing the boundary.
Proof. The closed case is done in [37], while the proof for surfaces with boundary can be
found in [50].

If v is a contact vector field in ( M, ξ ) transverse to S then the set

ΓS = { x ∈ S | v( x ) ∈ ξ x }

is a collection of embedded curves, and arcs if the surface has non-empty boundary, in S
and is called the dividing set of S in ( M, ξ ).
We have that the relative homotopy type of ΓS is invariant under contact isotopy. More-
over, if the knot K is a component of the boundary of S, and it is Legendrian, which means
such that TK ⊂ ξ, then
1
S · Kξ = − · | K ∩ ΓS | ,
2
where Kξ is the contact framing of K. More details can be found in [65].

3.1.2 Classification theorems

In this subsection we briefly state the most important results about existence and
uniqueness of tight and overtwisted structures on a given 3-manifold.
Let us start with the overtwisted case. The major theorems in this settings are due to
Eliashberg [18, 22].
Theorem 3.1.3 (Eliashberg) The homotopy classes of 2-plane fields on an oriented, closed 3-
manifold M are in bijection with overtwisted contact structures on M up to isotopy.
This means that every 2-plane field on M is homotopic to an overtwisted contact struc-
ture and that two overtwisted structures, which are homotopic as 2-plane fields, are actu-
ally contact isotopic. In particular, overtwisted structures exist for every M.
CEU eTD Collection

In the specific case of S3 , all of the overtwisted structures can be explicitly determined.
Let us recall that there is an isomorphism d3 : π3 (S2 ) → Z and a generator of π3 (S2 ) is
the Hopf fibration, introduced in [48]. The map d3 is called the Hopf invariant. Then the
following result holds.
Corollary 3.1.4 The space of overtwisted structures on S3 is in bijection with Z and then such
structures are classified by the Hopf invariant d3 .
Proof. From Eliashberg’s Theorem 3.1.3 we have that an overtwisted structure ξ on S3 is
determined by the homotopy type of ξ as 2-plane field.
Since contact structures are cooriented, there is a vector field v such that v( x ) is orthog-
onal to ξ x for every x ∈ S3 and it is positively oriented. This means that ξ induces a map
from S3 to S2 and then its homotopy type is an element of π3 (S2 ).

32
 From now on, we denote the overtwisted structures on S3 with ξ n , where d3 (ξ n ) = n
and ξ 0 is homotopic to the Hopf fibration.
We now suppose that M is non-compact. In this case, an overtwisted contact structure
is called tight at infinity if it is tight outside a compact set. Otherwise, we say that it is
overtwisted at infinity. We state the version of Theorem 3.1.3 for non-compact 3-manifold.

Theorem 3.1.5 (Eliashberg) The classification of contact structures overtwisted at infinity coin-
cides with the homotopical classification of tangent 2-plane distributions.

On the other hand, we do not have a complete classification of tight contact structures,
but Ding and Geiges in [16] proved that there exists a prime decomposition under contact
connected sum. We say that a contact manifold (Y, ζ ) is prime if (Y, ζ ) = (Y1 , ζ 1 ) # (Y2 , ζ 2 )
implies that at least one of the two summands is contactomorphic to (S3 , ξ st ), which is the
unique tight structure on S3 [21].

Theorem 3.1.6 (Ding and Geiges) Every tight contact 3-manifold ( M, ξ ), that is not (S3 , ξ st ),
is contactomorphic to a connected sum

( M1 , ξ 1 ) # ... # ( Mk , ξ k )

of finitely many prime tight contact 3-manifolds. The summands are unique up to order and con-
tactomorphism.

More details on the connected sum of contact 3-manifolds can be found in [13].

3.1.3 Fillings
We recall that a smooth 4-manifold is symplectic if it is equipped with a closed non-
degenerate differential 2-form ω, called a symplectic form. A contact manifold ( M, ξ ) is
called weakly symplectically fillable if there is a compact symplectic 4-manifold ( X, ω ) such
that ∂X = M and ω |ξ > 0.
We also say that ( M, ξ ) is strongly symplectically filled by X if there is a vector field v,
transversely pointing out of X along M, such that the Lie derivative Lv ω is a positive
multiple of ω in every point and the contraction ι v ω is a contact 1-form for ξ.
It is clear from the definition that a strong filling is also a weak filling. Moreover, the
two notions coincide on rational homology spheres. See [20, 25, 63].
An important result of Eliashberg and Gromov is that simplectically fillable contact
structures are always tight.
CEU eTD Collection

Theorem 3.1.7 (Eliashberg and Gromov) Suppose that ( M, ξ ) is a contact 3-manifold which
admits a weak symplectic filling. Then the structure ξ is tight.

This theorem was reproved by Ozsváth and Szabó, using Heegaard Floer homology, in
[70].
Another type of filling that is interesting to study consists of the family of Stein fillings.
A Stein manifold is a triple ( X, J, ψ), where J is a complex structure on X, there is a Morse
function ψ from X to R and ωψ (v, w) = −d(dψ ◦ J )(v, w) is non-degenerate.
A contact manifold ( M, ξ ) is called Stein fillable if there is a Stein manifold ( X, J, ψ) such
that

• the map ψ is bounded from below;

33
 • the manifold M is a non-critical level of ψ;
• −(dψ ◦ J ) is a contact form for ξ.
The following corollary follows easily from the definition of Stein filling.
Corollary 3.1.8 A Stein fillable contact 3-manifold is strongly symplectically fillable
The converse of this result is not true as we know from Ghiggini [36].
We can now say a few words on contact surgeries. More details are in [27]. Let K be
a knot in a contact 3-manifold ( M, ξ ) such that TK ⊂ ξ (a Legendrian knot). It is known,
see [35], that K has a neighborhood ν(K ) that is contactomorphic to a neighborhood of the
x-axis in (R3 , Ker (dz − ydx ))/ ∼, where ∼ identifies ( x, y, z) with ( x + 1, y, z).
With respect to these coordinates on ν(K ), we can remove ν(K ) from M and topologi-
cally glue it back by performing a Dehn surgery along K, with framing the contact framing
±1. Denote the resulting manifold with M(K,±1) . There is a unique way, up to isotopy,
to extend ξ | M\ν(K) to a contact structure ξ (K,±1) over all of M(K,±1) so that ξ (K,±1) |ν(K) is
tight, see [46]. The new contact manifold ( M, ξ )(K,±1) is said to be obtained from ( M, ξ ) by
±1-contact surgery along K.
We can iterate this procedure multiple times on different Legendrian knots in M. Then
in this case we say that we are doing a contact surgery on ( M, ξ ) along a Legendrian link.
We recall that every ( M, ξ ) is obtained from (S3 , ξ st ) by contact surgery [15]. We conclude
this subsection with the following theorem.
Theorem 3.1.9 Suppose that a contact manifold ( M, ξ ) admits a contact surgery presentation
which contains only −1-surgeries. Then ( M, ξ ) is Stein fillable.
This fact is proved in [19, 40].

3.2 Legendrian links

A link L in a contact 3-manifold ( M, ξ ) is called Legendrian if TL ⊂ ξ, which means
that Tp L ⊂ ξ p for every p ∈ L.
Two Legendrian links L1 and L2 are Legendrian isotopic if there is a contact isotopy G :
M × I → M that sends L1 into L2 .
A number or an element of a group is a Legendrian invariant if it remains unchanged
under Legendrian isotopy. In the following thesis, we call classical invariants of Legen-
drian links the link type, the Thurston-Bennequin number and the rotation number. The
definitions of these invariants, that are given for links in this section, coincide with the
CEU eTD Collection

ones in [2] for knots.

The link type is the most obvious Legendrian invariant. In fact a Legendrian isotopy of
links is also a smooth isotopy. The second one is the Thurston-Bennequin number, which
we denote with tb and it is defined as follows. Let us consider a Legendrian link L, with
non-split link type for now, in a rational homology contact 3-sphere ( M, ξ ). If we call N ( L)
the normal bundle over L then we have that, since L is Legendrian, N ( L) ∩ ξ determines a
non-zero vector field v along L, which is uniquely determined after normalization. This is
the definition of the contact framing of L in ( M, ξ ).
Let Lξ be the link obtained by pushing L along the contact framing. We extract a ratio-
nal number from Lξ by saying that
F · Lξ
tb( L) = lkQ ( L, Lξ ) = ;
t

34
 where F is a rational Seifert surface for L in M, as defined in Section 2.1, and t is the order
of L in M.
We define the Thurston-Bennequin number of every Legendrian link L by taking tb( L) =
tb( L1 ) + ... + tb( Lr ), where the Li ’s correspond to the split components of the link type of
L. It follows from the properties of the linking number in Subsection 1.1.1 that tb( L) does
not depend on the choice of the rational Seifert surface. Moreover, it is also easy to check
that this number is a Legendrian invariant.
The Thurston-Bennequin number satisfies the properties stated in the following propo-
sition.
Proposition 3.2.1 Suppose that L is a Legendrian link in a rational homology contact 3-sphere
( M, ξ ). Then we have that
1. if L is null-homologous then tb( L) is an integer;
2. if we reverse the orientation of L then tb(− L) = tb( L).
Proof. The link L is null-homologous in M if and only if Li is null-homologous in M for
every i = 1, ..., r, where the Li ’s correspond to the split components of the link type as
before, because of Corollary 2.2.3. Then the first claim follows from Proposition 1.1.2.
Now we observe that − F represents a rational Seifert surface for − L and − Lξ is its
contact framing. Then this clearly means that nothing changes in the definition of tb(− L).

We note that, when M is not a rational homology sphere, the Thurston-Bennequin

number may depend on the choice of the surface F, even if L is null-homologous; in par-
ticular, it depends on the relative homology class of the surface we take. Therefore, in this
case we define the invariant as
tbΣ ( L) = F · Lξ ,
where F is a Seifert surface for L representing the class Σ ∈ H2 ( M, ML ; Z).
The third classical invariant of L is the rotation number rot( L). As before, we take M
as a rational homology sphere. Let F be a rational Seifert surface for L. Recall that we have
an inclusion j : Σ ,→ M, which is an embedding on the interior of Σ and an r-fold cover
∂Σ → L. Then we can trivialize the pull-back i∗ ξ on Σ, because a surface with boundary
retracts on a bouquet of circles.
This means that there exist two sections s1 , s2 : Σ → i∗ ξ such that {s1 ( p), s2 ( p)} is a
basis of (i∗ ξ ) p for every p ∈ ∂Σ. Therefore, if u is the unit vector field tangent to L and
oriented coherently, we have that (i∗ u) p ∈ (i∗ ξ ) p , since L is Legendrian, and
CEU eTD Collection

(i∗ u) p = α( p)s1 ( p) + β( p)s2 ( p) for any p ∈ ∂Σ ;

where α and β are real-valued functions.
From this we define a map γ in the following way
γ : ∂Σ −→ S1 t ... t S1
(α( p), β( p))
p 7−→
|| (α( p), β( p)) ||
and we call rotation number of L the rational number
1
rot(K ) = · ∑ deg(γi ) ,
t i

35
 where γi is the restriction of γ to the i-th component of ∂Σ and t is the order of L in M. In
the same way as before, the value of rot L does not depend on the choice of F nor of the si ’s
and it is a Legendrian invariant.
The following proposition is the version of Proposition 3.2.1 for rot L.
Proposition 3.2.2 Suppose that L is a Legendrian link in a rational homology contact 3-sphere
( M, ξ ). Then we have that
1. if L is null-homologous then rot( L) is an integer;
2. if we reverse the orientation of L then rot(− L) = − rot( L).
Proof. Property 1 is trivial. In order to prove Property 2, we take − F as rational Seifert
surface and −u as unit tangent vector field, but we can still take s1 and s2 as sections. This
means that, if γ0 is the map for − L, it is deg(γi0 ) = −deg(γi ) for every i.

3.3 Transverse links

A smooth link T ,→ ( M, ξ ) is called transverse if Tp T ⊕ ξ p = Tp M for every p ∈ T and
α| T is a volume 1-form for T. This definition implies that transverse links always come
with a natural orientation, induced by ξ.
As for Legendrian links, we say that two transverse links T1 and T2 are transverse isotopic
if there is a contact isotopy G : M × I → M that sends T1 into T2 .
A Legendrian link L ,→ ( M, ξ ) always determines two special transverse links named
positive and negative transverse push-off of L, that we denote with TL+ and − TL− , where the
minus sign appears because transverse links need to be oriented accordingly to the contact
structure.
We briefly recall the construction of TL± . Let Ai = S1 × [−1, 1] be a collection of em-
bedded annuli in M such that S1 × {0} = Li and Ai is transverse to ξ for every i = 1, ..., n,
where n is the number of components of L. If the Ai ’s are sufficiently thin, see [26], then
TL± is the link with components S1 × {± 21 }, which is transverse up to orientation. It is easy
to check that any two positive (or negative) transverse push-offs are transversely isotopic
and then TL± is uniquely defined, up to transverse isotopy. Moreover, if we reverse the
orientation of L then we have that T−±L = − TL∓ .
Since we are working with rational homology spheres, we can define the self-linking
number sl( T ) of a transverse link T with order t in ( M, ξ ). The self-linking number is one
of the two classical invariants of transverse links, together with the link type. Suppose
CEU eTD Collection

for now that T is non-split; take a surface F which is rationally bounded by T and the
map j : Σ → M that defines F, as described before in Section 2.1. Consider the pull-back
bundle j∗ ξ on Σ. Then again we have that j∗ ξ is trivial. Let v be a non-zero section of j∗ ξ.
Normalize v so that v|∂Σ defines a link Tξ in ∂ν( T ). We define the self-linking number of T
to be
1 F · Tξ
sl( T ) = · lkQ ( T, Tξ ) = .
t t2
This definition coincides with the one given by Baker and Etnyre in [2] for transverse knots.
In the case that T has split link smooth type, we consider T1 , ..., Tr the connected com-
ponents of T as a smooth link. Then we say that
r
sl( T ) = ∑ sl(Ti ) .
i =1

36
 The fact that M is a rational homology sphere tells us that sl( T ) is independent of the
choice of the surface F. Furthermore, the self-linking number is a transverse invariant and,
as in the Legendrian case, it has the following property.
Corollary 3.3.1 Suppose that T is a null-homologous transverse link in a rational homology con-
tact 3-sphere ( M, ξ ). Then the invariant sl( T ) is an integer.
We can find the value of the self-linking number of the transverse push-offs of a Leg-
endrian link L, starting from its Thurston-Bennequin and rotation numbers.
Proposition 3.3.2 Let L be an n-component Legendrian link in the rational homology contact
3-sphere ( M, ξ ). Then we have that
sl T±+L = tb( L) ∓ rot( L) .

Proof. The proof is done in [2] for Legendrian knots. Hence, the claim follows by observing
that
n n
tb( L) ∓ rot( L) = ∑ tb( Li ) + 2 ∑ lkQ ( Li , L j ) ∓ ∑ rot( Li ) =
i =1 i6= j i =1
n
= ∑ sl(T±+L ) + 2 ∑ lkQ ( Li , L j ) = sl(T±+L ) ,
i
i =1 i6= j

where the first and the last equality are consequences of Equation (1.1).

3.4 Legendrian and transverse links in (S3 , ξ st )

3.4.1 Front projections of Legendrian links
In Subsection 1.2.1 we saw that smooth links in S3 can be represented by planar dia-
grams; moreover, two such diagrams, corresponding to isotopic links, differ by a sequence
of Reidemeister moves (Figure 1.5). We now describe a similar construction for Legendrian
links in (S3 , ξ st ).
Since a link in S3 is embedded in R3 and ξ st |R3 is still tight, we can work in (R3 , ξ ),
where ξ = Ker(dz − ydx ). In fact, there is a unique tight contact structure on R3 up to con-
tact isotopy [21]. Hence, we can define the front projection of an n-component Legendrian
link L. This is the map
n
S1 −→ R2
G

i =1
CEU eTD Collection

θ 7−→ ( x (θ ), z(θ ))
where θ = (θ1 , ..., θn ) and ( x (θ ), y(θ ), z(θ )) is the parametrization of L.
A generic front projection of L is a special diagram for L with no vertical tangencies,
that are replaced by cusps, and at each crossing, the slope of the overcrossing is smaller
than the one of the undercrossing. An example of front projection is shown in Figure
3.1: we picture a diagram of the standard Legendrian unknot O , which is the unique, up to
Legendrian isotopy, Legendrian unknot such that tb(O) = −1. This is proved in [24].
There exist moves on Legendrian front projections, called the Legendrian Reidemeister
moves and they are illustrated in Figure 3.2. A further move is the Legendrian planar
isotopy; such a move is an isotopy of the front projection that does not introduce vertical
tangencies. These moves appear in the following Legendrian analogue of Reidemeister’s
Theorem 1.2.1. A proof can be found in [85].

37
 Figure 3.1: A front projection of the standard Legendrian unknot.

Theorem 3.4.1 (Światkowski)

˛ Two front projections correspond to Legendrian isotopic Legen-
drian links if and only if the projections can be connected by Legendrian planar isotopies and by
Legendrian Reidemeister moves.
The classical invariants of an oriented Legendrian link L can be easily computed from
its oriented front projection P. We recall that we define the writhe in Subsection 1.1.1. Then
we have the following formulas:

Type 1 Type 2 Type 3

Figure 3.2: The three Legendrian Reidemeister moves. For each move, we consider all
the possible reflections and orientations, provided that they satisfy the conditions in the
definition of front projection.

1
tb( L) = wr( P) − · |cusps in P| ; (3.1)
2
CEU eTD Collection

1  
rot( L) = · |down-ward cusps in P| − |up-ward cusps in P| . (3.2)
2

3.4.2 Braid presentation for transverse links

Transverse links can be studied by using closed braids. We recall that a closed braid is
an n-component link in R3 , where here we use cylindrical coordinates (r, ψ, z), that can be
parametrized by a map
n
S1 −→ R3
G

i =1

θ 7−→ (r (θ ), ψ(θ ), z(θ ))

38
 for which θ = (θ1 , ..., θn ), r (θ ) 6= 0 and ψ0 (θ ) > 0 for every θ. More details on braids and
closed braids can be found in [5].
To see the connection between braids and transverse
links we use the contact structure
ξ sym on R3 , which is given by ξ sym = Ker dz + r2 dψ . Since this structure is contact
isotopic to the standard one, we can transfer any question about ξ st to ξ sym . Now given
a closed braid B, we can isotopy it through closed braid so that it is enough far from the
z-axis that the planes that make up ξ sym are almost vertical, which means they are close to
∂
the planes spanned by ∂z and ∂r∂ . Thus the closed braid type B represents a transverse link.
Bennequin proved the opposite assertion in [4].

Theorem 3.4.2 (Bennequin) Every transverse link in (R3 , ξ sym ) is transversely isotopic to a
closed braid.

Proof. The proof of the theorem follows the same strategy of the one that smooth links can
be braided. See [5]. We just need to check that such procedure can be done in a transverse
way. For details we refer to [4, 64].

Let us fix k points pi in a disk D2 , for i = 1, ..., k. Then we recall that a k-braid is an
embedding of k arcs γi : [0, 1] → D2 × [0, 1] so that γi (t) ∈ D2 × {t} and the endpoints of
the γi ’s, corresponding to 0 (or 1), map as a set to pi in D2 × {0} (or D2 × {1}). It is easy to
σi

i+1

Figure 3.3: The generator σi for the braid group Bk .

see that braids and closed braids are in one-to-one correspondence.

The set of all the k-braids Bk forms a group. Such group is generated by σi for i =
1, ..., k − 1, where σi is as in Figure 3.3. Given a braid b in Bk , then the positive stabilization
of b is bσk in Bk+1 . Moreover, a conjugation of b is βbβ−1 for some β ∈ Bk . Then from
[64, 91] we have that braid presentations of transversely isotopic links are related by those
CEU eTD Collection

two moves.

Theorem 3.4.3 (Orevkov and Shevchishin; Wrinkle) Two braids represent the same transverse
link if and only if they are related by a finite sequence of positive stabilizations (and their inverse)
and conjugations in the braid group.

From [4] we also have that the self-linking number of a transverse link T in (S3 , ξ st ),
represented by the closed k-braid B, is computed as follows:

sl( T ) = wr( B) − k , (3.3)

where wr( B) is the writhe of B seen as a link diagram for T.

39
 3.5 Connected sum and disjoint union in the contact setting
The definition of connected sum of two 3-manifolds can be easily given also when
they come together with a contact structure. Let us take two connected contact manifolds
( M1 , ξ 1 ) and ( M2 , ξ 2 ); we call Mi0 (for i = 1, 2) the manifolds obtained from Mi by removing
an open Darboux ball, that is a 3-ball with the its unique tight contact structure [21], and we
define ( M1 # M2 , ξ 1 # ξ 2 ) the contact manifold which is gotten by gluing together M10 and
M20 . The structure ξ 1 # ξ 2 is well-defined because it is always possible to glue two contact
structures along the boundary of a Darboux ball. Moreover, the result is independent of
the choice of the balls themselves and we have the following proposition.
Proposition 3.5.1 For every Darboux ball B in the overtwisted manifold ( M, ξ ) there exists at
least one overtwisted disk disjoint from B. In particular, if ( M, ξ ) = ( M1 # M2 , ξ 1 # ξ 2 ) and a
summand Mi is overtwisted then we can always find overtwisted disks which are contained entirely
in Mi0 ⊂ M1 # M2 .
Proof. From Eliashberg’s classification, the connected sum of ( M, ξ ) with (S3 , ξ 0 ) is always
contact isotopic to ( M, ξ ). Then the first statement can be proved by noting that we can
find two disjoint Darboux balls in ( M, ξ ) and, since ξ is overtwisted, one can be replaced
with (S3 , ξ 0 ) minus a Darboux ball. At this point, the second statement is an immediate
consequence of the first one.

Clearly if ( M1 , ξ 1 ) is overtwisted then the connected sum is still overtwisted, but if both
the summands are tight then it is important to cite the following result of Colin [13].
Theorem 3.5.2 The connected sum of two contact 3-manifolds ( M1 # M2 , ξ 1 # ξ 2 ) is tight if and
only if ( M1 , ξ 1 ) and ( M2 , ξ 2 ) are both tight.
If a sphere S separates a contact manifold ( M, ξ ) into two components M1 and M2 such
that smoothly it is M = M1 # M2 then we can ask whether ξ also splits accordingly. Then
we can state the following lemma.
Lemma 3.5.3 A convex separating sphere S in ( M, ξ ) gives a connected sum decomposition
( M, ξ ) = ( M1 #S M2 , ξ 1 #S ξ 2 ) if and only if the dividing set ΓS is trivial, which means that
consists of a unique simple, closed curve.
Proof. The claim follows from the fact that a contact manifold, whose boundary is a convex
sphere, can be glued together with a Darboux ball if and only if its dividing set is trivial as
shown in [27].
CEU eTD Collection

Etnyre and Honda in [31] extended the definition of connected sum to Legendrian
links. Suppose that we have two Legendrian links L1 and L2 in ( M1 , ξ 1 ) and ( M2 , ξ 2 )
respectively. We take two Darboux balls D1 and D2 as before, but with the condition that
Di ∩ Li is a Legendrian arc αi where αi ∩ ∂Di consists of two points and its front projection
is isotopic to the one in Figure 3.4; we call this the Legendrian elementary tangle. In this way,
the link L1 # L2 in ( M1 # M2 , ξ 1 # ξ 2 ) is Legendrian and it does not depend on the choice of
the Darboux balls Di , but only of which component of Li contains the arc αi .
We observe that we have:
• tb( L1 # L2 ) = tb( L1 ) + tb( L2 ) + 1;

• rot( L1 # L2 ) = rot( L1 ) + rot( L2 ).

40
 αi

Di

Figure 3.4: Front projection of αi in a Darboux ball.

We define positive (negative) stabilization of a Legendrian link L in (S3 , ξ st ), with front

projection P, the Legendrian link L± represented by the front projection P± ; which is ob-
tained by adding two consecutive down-ward (up-ward) cusps to P. Stabilizations are
well-defined, in the sense that they do not depend on the choice of the point of P where
we add the new cusps.
At this point it is easy to define stabilizations in every contact manifold. In fact we
say that L± is the positive (negative) stabilization of L, a Legendrian link in ( M, ξ ), if
L± = L # O ± ; where O ± is the positive (negative) stabilization of the standard Legendrian
unknot O in (S3 , ξ st ). The Legendrian knots O ± are shown as front projections in Figure

Figure 3.5: Front projection of O + (left) and O − (right).

3.5. From Equations (3.1) and (3.2) it results that:

• tb( L± ) = tb( L) − 1;

• rot( L± ) = rot( L) ± 1.

Now, consider Li Legendrian links in ( Mi , ξ i ) for i = 1, 2. We define the disjoint union

CEU eTD Collection

Σ = S2 × {0}

L1

L2

Tight S2 × I

( M1 , ξ 1 ) ( M2 , ξ 2 )
Figure 3.6: Disjoint union of L1 and L2 .

41
 L1 t L2 in ( M1 # M2 , ξ 1 # ξ 2 ) as the Legendrian link given by a particular connected sum
L1 # O2 # L2 ; where O2 is the standard Legendrian unlink with 2 components in (S3 , ξ st ).
The connected sum is such that one component of O2 is summed to L1 and the other one
to L2 as shown in Figure 3.6. Note that L1 and L2 are Legendrian links in ( M1 # M2 , ξ 1 # ξ 2 ).
Using the same criterion of the smooth case in Subsection 1.1.1, we say that a Legen-
drian link L, in a contact 3-manifold ( M, ξ ), is split if there exist Li ,→ ( Mi , ξ i ) for i = 1, 2
such that ( M, ξ ) = ( M1 , ξ 1 ) # ( M2 , ξ 2 ) and L is Legendrian isotopic to the disjoint union of
L1 and L2 . Otherwise, we say that L is non-split.
Finally, we note that as before:
• tb( L1 t L2 ) = tb( L1 ) + tb( L2 );
• rot( L1 t L2 ) = rot( L1 ) + rot( L2 ).

3.6 The Thurston-Bennequin inequality

The Thurston-Bennequin inequality is a very powerful result in contact topology. One
of its most important implications is that, provided the structure is tight, the Thurston-
Bennequin numbers of all the Legendrian knots, with the same smooth knot type, are
bounded from above. This property actually characterizes tight contact structures on 3-
manifolds; in fact, when the structure is overtwisted, it is always possible to increase the
Thurston-Bennequin number indefinitely, without changing the smooth isotopy class of
the knot. This is done by connect summing with the boundary of an overtwisted disk.
A proof for a version of the Thurston-Bennequin inequality was given by Eliashberg in
[23]. In the following thesis we show that an analogous inequality, involving the Thurston
norm k Lk T of a link defined in Chapter 2, holds for Legendrian and transverse links in
every rational homology contact 3-sphere, equipped with a tight structure. For Legen-
drian links in (S3 , ξ st ) this result, and the resulting upper bound for the maximal Thurston-
Bennequin number, was found first by Dasbach and Mangum in [14]. On the other hand,
if we consider only Legendrian and transverse knots then our bounds coincide with the
ones of Baker and Etnyre [2].
We recall that if Σ is a surface in ( M, ξ ) then, at each point x ∈ Σ, we have that lx =
ξ x ∩ Tx Σ is a singular line field on Σ. This line field can be integrated to give a singular
foliation on Σ. This singular foliation is called the characteristic foliation and is denoted with
Σξ . More details on foliations are found in [90].
In order to prove our result, we use the same strategy of [2, 23] and then the proof
CEU eTD Collection

follows from the following lemma.

Lemma 3.6.1 Suppose that T is a non-split transverse link in a rational homology tight 3-sphere
( M, ξ ) and take an F in M that is rationally bounded by T. Then we can perturb F in a way that
T 0 = F ∩ ∂ν( T ) is still transverse and sl( T 0 ) = t · sl( T ).
Furthermore, we have that
F · Tξ0 6 −χ( F ) ,
where Tξ0 is the framing defined as in Section 3.3.
Proof. Consider the map j : Σ → M which determines F. Let us take a small neighborhood
ν( T ) of T such that it intersects F only in the image of a neighborhood of ∂Σ, as in the
proof of Proposition 2.2.2. Denote the properly embedded surface ML ∩ F with F 0 , where
the manifold ML is M \ ν(˚T ).

42
 We know from [2] that we can modify F 0 in a way that ∂F 0 = T 0 is as wanted and the
characteristic foliation Fξ0 is generic. In particular, we can assume that all the singularities
are isolated elliptic or hyperbolic points. Moreover, each singularity has a sign depending
on whether the orientation of ξ and TF 0 agree at the singularity. Let us denote with e± and
h± the number of such singularities.
Since we can interpret F 0 · Tξ as a relative Euler class, in other words F 0 · Tξ is the ob-
struction to extending the framing Tξ to a non-zero vector field on F 0 , we have that
F · Tξ0 = F 0 · Tξ0 = (e− − h− ) − (e+ − h+ ) ,
as in [2]. Moreover, a simple computation gives that
χ ( F ) = ( e+ + e− ) − ( h + + h − ) ;
thus it is
F · Tξ0 + χ( F ) = 2(e− − h− ) . (3.4)
At this point, since ξ is tight, every negative elliptic point is connected to a negative hyper-
bolic point, otherwise we could find overtwisted disks in ( M, ξ ), and then we can cancel
this pair using Giroux’s elimination lemma [38]. This means that we can isotope F 0 so
that the characteristic foliation is such that e− = 0 and then the claim follows easily from
Equation (3.4).
We can now prove the main result of the chapter.
Theorem 3.6.2 Suppose that L is a Legendrian link in a tight contact 3-manifold ( M, ξ ) and M is
a rational homology sphere. Then we have that
tb( L) + | rot( L)| 6 k Lk T − o( L) .
Furthermore, suppose that T is a transverse link in ( M, ξ ). Then we have that
sl( T ) 6 k T k T − o( T ) ,
where the Thurston norm and the compressibility term o are defined in Chapter 2.
Proof. Suppose first that T is a non-split transverse link of order t in ( M, ξ ). Take an F that
is rationally bounded by T and gives the equality in Lemma 2.2.5. Then from Lemma 3.6.1
we have that
sl( T 0 ) F · Tξ0 χ( F )
sl( T ) = = 6− = k T k T − o( T ) .
t t t
For the second equality we also use that T 0 is null-homologous.
On the other hand, if T has split smooth link type then we reason as in Section 3.3 and
CEU eTD Collection

we write T1 , ..., Tr , which are the split components of T. Hence, this time we obtain
r r
sl( T ) = ∑ sl(Ti ) 6 ∑ kTi kT − o(Ti ) = kT kT − o(T ) ,
i =1 i =1
because of Proposition 2.2.2 and the additivity of the Thurston norm and the compressibil-
ity term.
Now suppose that L ,→ ( M, ξ ) is a Legendrian link. Then from Proposition 3.3.2 we
have that
tb( L) ∓ rot( L) = sl T±+L 6 k Lk T − o( L) ;

in fact, L and its transverse push-offs are all smoothly isotopic, up to orientation, and then
they have same Thurston norm and number of disjoint unknots.

43
CEU eTD Collection
 Chapter 4

Representations of links and

3-manifolds

4.1 Heegaard diagrams

4.1.1 3-manifolds and Spinc structures
In this chapter we describe two ways of presenting links in 3-manifolds. Heegaard
diagrams are crucial for the construction of Heegaard Floer homology, both the version for
manifolds than the one for links, while in Chapter 6 we use open book decompositions in
order to define some contact and Legendrian invariants.
Hereby, we use terminology from [41, 67, 72]. An oriented handlebody is an oriented
3-manifold with boundary obtained by attaching 3-dimensional 1-handles to a 3-ball. It is
a classical result on 3-manifolds that any connected, oriented, closed 3-manifold M can be
decomposed, along a separating surface Σ, as the union of two handlebodies [84]. This is
called a Heegaard decomposition of M.
Let Σ be an oriented, closed, genus g surface, which means a connected sum of g tori,
and fix a g-tuple of homologically linearly independent, mutually disjoint curves γ =
{γ1 , ..., γg }; this is equivalent to say that Σ \ (γ1 ∪ ... ∪ γg ) is a connected, planar surface.
We call such a collection of curves a system of attaching circles. These curves always specify

Σ β1
CEU eTD Collection

α1

Figure 4.1: A Heegaard diagram for the 3-sphere. The torus Σ splits S3 into two solid tori,
which are precisely the handlebodies Uα and −Uβ .

a handlebody with boundary Σ, uniquely up to diffeomorphism inducing the identity on

the boundary; moreover, the γi ’s bound disjoint, embedded disks Di in the handlebody.
A Heegaard diagram is a triple (Σ, α, β), consisting of an oriented surface Σ equipped
with two systems of attaching circles α = {α1 , ..., α g } and β = { β 1 , ..., β g }, which inter-

45
 sect transversely. The associated 3-manifold M and its Heegaard splitting consist of the
handlebodies Uα and −Uβ , specified by the g-tuples α and β, glued to Σ. See Figure 4.1.
We now define an enriched version of Heegaard diagrams. More specifically, a pointed
Heegaard diagram is a quadruple H = (Σ, α, β, w) given by a triple (Σ, α, β) as before and a
point w ∈ Σ \ (α ∪ β).
Now, fix a connected, oriented 3-manifold M, and a generic self-indexing Morse func-
tion on M, which has one index zero and one index three critical point. Fix also a generic

α2 α2

α1
α1 + α2
Figure 4.2: The curve α1 + α2 is the handleslide of α1 over α2 .

metric g together with a choice of a gradient flowline, connecting the index zero and three
critical points.
Then, there is an associated pointed Heegaard diagram H for M, whose surface Σ is
the mid-level of the Morse function; the curves in α are the locus of points on Σ where the
gradient flowline leaving the index one critical point meets Σ. Similarly, the curves in β
are the one which flow from the index two critical point. Finally, w is the point on Σ that
lies on the flowline.
If H is obtained in this manner, from a Morse function f , then we call f a Morse function
compatible with H. Given a pointed Heegaard diagram for M, it is easy to construct a

Σg Σ g +1
β g +1

α g +1

Figure 4.3: The new curves α g+1 and β g+1 are in standard position. This move corresponds
CEU eTD Collection

to a connected sum with S3 .

compatible Morse function f . Moreover, we have the following result.

Theorem 4.1.1 (Ozsváth and Szabó) Any two pointed Heegaard diagrams for M can be con-
nected by a finite sequence of the following moves:

1. isotopies and handleslides of the curves in α (or β), supported in the complement of w, see
Figure 4.2;

2. stabilizations (and their inverse), see Figure 4.3.

Proof. The proof is sketched in [72]. The fact that every two (unpointed) Heegaard dia-
grams can be connected by such moves follows from standard Morse theory. This can be

46
 done in the complement of a single basepoint because we can trade an isotopy across the
basepoint for a sequence of handleslides in the opposite direction [68].

Using the basepoint we obtain additional information when the 3-manifold is equipped
with a Spinc structure. In this thesis, since we are working on 3-manifolds, the definition
of Spinc structure will be the one given by Turaev in [88]: an isotopy class, away from a
point, of nowhere vanishing vector fields on the manifold.
We define the g-th symmetric power of a genus g surface Σ as the manifold
g times
z }| {
Σ × ... × Σ
Sym (Σ) =
g  
( x1 , ..., x g ) = xσ(1) , ..., xσ( g)

where σ ∈ S g , the group of permutations of g elements. Consider the g-dimensional tori

Tα = α1 × ... × α g and Tβ = β 1 × ... × β g

in Symg (Σ).
Given a pointed Heegaard diagram (Σ, α, β, w) for the 3-manifold M, we have a well-
defined map
sw : Tα ∩ Tβ −→ Spinc ( M)
which sends an intersection point x ∈ Tα ∩ Tβ into the Spinc structure sw ( x ) on M, as
described in [72]. The map is defined as follows. An intersection point x corresponds to a
g-tuple of gradient flowlines which connect all the index one and two critical points. Let
γx be the union of gradient flowlines passing through each xi ∈ x, and γw be the union of
gradient flowlines passing through each wi ∈ w.
The closure of γx ∪ γw is a collection of arcs whose boundaries consist of all the critical
points of f , which is a compatible Morse function. Moreover, each component contains
a pair of critical points whose indices have opposite parities. Thus, we can modify the
gradient vector field in an arbitrarily small neighborhood of γx ∪ γw to obtain a new vector
field which vanishes nowhere in M. We call this vector field πw ( x ) and its homotopy class,
in the sense of [88], sw ( x ).
The space Spinc ( M) is (not canonically) identified with H 2 ( M; Z). For this reason we
think of sw ( x ) as a 2-cohomology class in M.
CEU eTD Collection

4.1.2 Links
In the case of links, we need (balanced) multi-pointed Heegaard diagrams. These diagrams
are defined as a quintuple (Σ, α, β, w, z), but the objects are slightly different. In fact, now
the sets α and β consist of g + n − 1 curves, where n > 1 is an integer, and they are such
that each one spans a g-dimensional subspace of H1 (Σ; Z). Moreover, say S1 , ..., Sn are the
connected components of Σ \ α and T1 , ..., Tn are the ones of Σ \ β, we have that w and z are
two sets of n basepoints in Σ such that wi , zi ∈ Si ∩ Ti for every i = 1, ..., n. See Figure 4.4.
A multi-pointed Heegaard diagram gives rise to a 3-manifold M together with an ori-
ented link L in M as follows. Start from [−1, 1] × Σ and attach 3-dimensional 1-handles
along copies of the α-curves in {−1} × Σ; then attach three-dimensional 2-handles along
the copies of the β-curves in {1} × Σ. The result has 2n copies of S2 as boundary, which
we can close off with 2n 3-handles to get the closed 3-manifold M.

47
 When we connect the basepoints in the components Si and Ti , in the handlebodies
given by the α-curves and β-curves, with unknotted and unlinked arcs, we get an embed-
ded closed 1-manifold L in M. Next, we orient the link L so that the portions of the link

w1 Σ
β
z1

α
L
w2
z2

Figure 4.4: A very simple multi-pointed Heegaard diagram for the 2-component unlink in
S3 .

in the α-handlebody point away from the w-basepoints and into the z-basepoints. The
resulting link L has n components.
Indeed, any oriented link in any 3-manifold can be presented by an appropriate multi-
pointed Heegaard diagram. Given an n-component link, one can find a self-indexing
Morse function f : M → R with n index zero and three critical points, and g + n − 1
index one and two critical points; with the additional property that there are two n-tuples
of flowlines γw and γz , connecting all the index three and index zero critical points, so that
our link L can be realized as the difference γz − γw . Such a Morse function gives rise to a
multi-pointed Heegaard diagram for L in M.

Theorem 4.1.2 (Ozsváth and Szabó) Every two multi-pointed Heegaard diagrams for the same
n-component link L in M can be connected by a finite sequence of moves of the following types:

1. isotopies and handleslides of the curves in α (or β ), supported in the complement of w and z;

2. stabilizations (and their inverse).

Proof. The claim follows from Morse theory in the usual manner.

In the previous subsection we saw that, starting from (Σ, α, β, w, z), we define two maps
sw and sz from the set of the intersection points Tα ∩ Tβ to the one of Spinc structures over
CEU eTD Collection

M. The relation between these two maps is given in the following proposition.

Proposition 4.1.3 Let (Σ, α, β, w, z) be a multi-pointed Heegaard diagram for a link L in a 3-

manifold M. Then we have that
sw ( x ) − sz ( x ) = PD[ L]
for every x ∈ Tα ∩ Tβ .

Proof. It is proved in [68] that sw ( x ) − sz ( x ) coincides with the Poincaré dual of [γ], where
γ is a curve in Σ such that αi · γ = 1 for every i = 1, ..., g. The link L respects such
conditions.

As in Section 2.1, let us call ML the 3-manifold with boundary M \ ν(˚L). When L has
n components we have that ∂ML consists of n disjoint tori. On this kind of 3-manifold we

48
 define a relative Spinc structure as in [72]: the isotopy class, away from a point, of a nowhere
vanishing vector field such that the restriction on each boundary torus coincides with the
canonical one. This is the vector field induced by the Spin structure on the boundary tori
of ML ; each structure is given by the product of the extendable Spin structure on the circle
(see [72, 88]).
We denote the set of the relative Spinc structures on ML by Spinc ( M, L); then we have
an identification of Spinc ( M, L) with the relative cohomology group H 2 ( ML , ∂ν( L); Z).
Moreover, from [72] we know that, given a multi-pointed Heegaard diagram, we construct
a map
sw,z : Tα ∩ Tβ −→ Spinc ( M, L)
applying the following procedure.
Let γ be a gradient flowline connecting an index zero and index three critical point,
and let ν(γ) denote a neighborhood of this flowline. One can construct a nowhere vanish-
ing vector field v over ν(γ), which has an integral flowline P which enters ν(γ) from its
boundary, contains γ as a subset and then exits ν(γ).
Let f be an orientation-preserving involution of ν(γ). We can arrange for −v|∂ν(γ) to
agree with f ∗ (v|∂(ν(γ)∪ P ). Indeed, we can construct v in the way that the difference v −
f ∗ (v) is the Poincaré dual of a meridian for γ, thought of as an element of H1 (ν(γ) \ P; Z).
Armed with this vector field, we can now define the map sw,z . Fix a Morse function f ,
compatible with the Heegaard diagram. Given x ∈ Tα ∩ Tβ , consider the flowlines γx , γw
and γz . We replace the gradient vector field in a neighborhood of γx so as not to vanish
there. Similarly, we do the same in a neighborhood of γw , using v. In fact, arranging for P
to consist of arcs on γz ∪ γw , we obtain in this manner a vector field on M which contains
L as a closed orbit. It is easy to see that this is equivalent to a vector field on ML which is a
standard non-vanishing vector field on the boundary tori.
Clearly the relative Spinc structure sw,z ( x ) extends to the actual Spinc structure sw ( x ).
Furthermore, Poincaré duality gives that

H 2 ( ML , ∂ν( L); Z) ∼
= H1 ( ML ; Z) ∼
= H1 ( L; Z) ⊕ H1 ( M; Z) ∼
= Zn ⊕ H 2 ( M; Z)

and if M is a rational homology sphere then H 2 ( M; Z) is a finite group. A basis of the

image of H1 ( L; Z) ∼
= Zn in H1 ( ML ; Z) is given by the homology classes {[µi ]}i=1,...,n ;
where µi is the meridian of the i-th component of L (Figure 1.1). Hence, the isomorphism
between Spinc ( M, L) and H 2 ( ML , ∂ν( L); Z) is given by
n
sw,z ←→ c1 (sw,z ) − ∑ PD[µi ] ,
CEU eTD Collection

i =1

where c1 (·) denotes the first Chern class, see [60]. From now on, we think of sw,z as a
relative 2-cohomology class under this identification. In particular, when M is a rational
homology sphere we can write
n
sw,z ( x ) = ∑ ci · PD[µi ] + sw (x) for any x ∈ Tα ∩ Tβ ,
i =1

where ci is an integer for i = 1, ..., n.

Let us consider a multi-pointed Heegaard diagram H = (Σ, α, β, w, z), which represents
an n-component link L in a rational homology sphere M. We easily obtain two diagrams

49
 for the link − L, that is gotten from L by reversing the orientation. The first diagram is
H1 = (−Σ, β, α, w, z), while the second one is H2 = (Σ, α, β, z, w). Let us denote with
s0w,z ( x ) the relative Spinc structure induced by x in H1 and with sz,w ( x ) the corresponding
one in H2 . Then we can prove the following proposition.

Proposition 4.1.4 Let H = (Σ, α, β, w, z) be a multi-pointed Heegaard diagram for L in M as

above. Then we have that

s0w,z ( x ) = −sw,z ( x ) and sz,w ( x ) = sw,z ( x ) − PD[ L]

for every x ∈ Tα ∩ Tβ .

Proof. For the first statement, note that if f is a Morse function compatible with H then
− f is a Morse function compatible with H1 . It is now a straightforward consequence of its
definition that if sw,z is represented by v then −v represents sw,z . Thus, the claim follows.
For the second observation, note that the vector fields v and w, representing the non-
torsion parts of sw,z ( x ) and sz,w ( x ) respectively, differ only in a collar neighborhood of the
boundary. In fact, one can see that in this neighborhood, they can be made isotopic away
from a neighborhood of the meridian, where they point in opposite directions. in this way,
we have proved that
sz,w ( x ) − sz ( x ) = sw,z ( x ) − sw ( x )
and then we conclude by applying Proposition 4.1.3.

We briefly recall that we can define an unbalanced multi-pointed Heegaard diagram as a

multi-pointed Heegaard diagram (Σ, α, β, w, z) where we drop the condition that both wi
and zi are in the intersection between the i-th connected component of Σ \ α and the one
of Σ \ β; instead we only require that there is exactly one element from w and z inside each
component of Σ \ α and Σ \ β. For this kind of diagrams we have that the resulting link
has at most n components; the other results in this section still apply.

4.2 Open book decompositions

4.2.1 Adapted open book decompositions for links in 3-manifolds
Let us start this section by taking a 3-manifold M. We say that an open book decomposition
for M is a pair ( B, π ) where
CEU eTD Collection

• the binding B is a smooth link in M;

• the map π : M \ B → S1 is a locally trivial fibration such that π −1 (θ ) = Sθ is a

compact surface with ∂Sθ = B for every θ ∈ S1 . The surfaces Sθ are called pages of
the open book.

Moreover, in the case that M comes equipped with a contact structure ξ, we say that the
pair ( B, π ) supports ξ if we can find a 1-form α for ξ such that

• the 2-form dα is a positive area form on each page Sθ ;

• we have α > 0 on the binding B.

50
 Let us consider the page S1 = π −1 (1); that is a connected, oriented, compact surface of
genus g and with l boundary components. Suppose that an n-component Legendrian
link L in ( M, ξ ) sits inside S1 and its components represent n independent elements in
H1 (S1 ; F), where we denote by F the field with two elements. We say that a collection of
disjoint, simple arcs A = { a1 , ..., a2g+l +n−2 } is a system of generators for S1 adapted to L if the
following conditions hold:
1. the subset A1 t A2 ⊂ A, where A1 = { a1 , ..., an } and A2 = { an+1 , ..., a2g+l −1 }, is a
basis of H1 (S1 , ∂S1 ; F) ∼
= F2g+l −1 ;
2. we have that L t ai = {1 pt} for every ai ∈ A1 and L ∩ ai = ∅ for i > n. The arcs in
the subset A1 are called distinguished arcs;
3. the subset A3 = { a2g+l , ..., a2g+l +n−2 } is such that the surface S1 \ ( A2 t A3 ) has n
connected components; each one containing exactly one component of L.
The arcs in A2 t A3 that appear twice on the same component of the boundary of
S1 \ ( A2 t A3 ) are called dead arcs; the others separating arcs;
4. the elements in A3 are separating arcs such that the disk D = S1 \ ( A1 t A2 ) is dis-
connected into n disks and every arc in A3 separates a unique pair of components of
D.
An example of adapted system of generators is shown in Figure 4.5. With this definition in

S1

L a3

a1 a2

Figure 4.5: A system of generators adapted to the link L.

place, we say that the triple ( B, π, A) is an open book decomposition adapted to the Legendrian
link L if
• the pair ( B, π ) is compatible with M and supports ξ;
CEU eTD Collection

• the link L is contained in the page S1 ;

• the n components of L represent independent elements in H1 (S1 ; Z);
• the set A is a system of generators for S1 adapted to L.
In this case we also say that the adapted open book decomposition ( B, π, A) is compatible
with the triple ( L, M, ξ ). It is important to observe that, since the components of L are
required to be independent in homology, we only consider open book decompositions
with pages not diffeomorphic to a disk.
We can prove that open book decompositions adapted to Legendrian links always exist.
In order to do this we recall the definition of contact cell decomposition (of a contact 3-
manifold) and ribbon of a Legendrian graph. A contact cell decomposition of ( M, ξ ), see [27],
is a finite CW-decomposition of M such that

51
 1. the 1-skeleton is a connected Legendrian graph;

2. each 2-cell E satisfies tb(∂E) = −1;

3. the contact structure ξ is tight when restricted to each 3-cell.

Moreover, if we have a Legendrian link L ,→ ( M, ξ ) then we also suppose that
4. the 1-skeleton contains L.
Denote the 1-skeleton of a contact cell decomposition of ( M, ξ ) with G. Then G is a Legen-
drian graph and its ribbon is a compact surface SG satisfying:
• SG retracts onto G;

• Tp SG = ξ p for every p ∈ G;

• Tp SG 6= ξ p for every p ∈ S \ G.
We say that an adapted open book decomposition ( B, π, A), compatible with the triple
( L, M, ξ ), comes from a contact cell decomposition if S = π −1 (1) is a ribbon of the 1-
skeleton of ( M, ξ ).
Theorem 4.2.1 Every Legendrian link L in a contact 3-manifold ( M, ξ ) admits an adapted open
book decomposition ( B, π, A), which is compatible with the triple ( L, M, ξ ). Moreover, the contact
framing of L coincides with the framing induced on L by the page S1 .
Proof. In order to find an open book decomposition ( B, π ) which comes from a contact cell
decomposition of ( M, ξ ), such that the link L is contained in S1 , we simply include the

L a2g+l

ai

Figure 4.6: We add a new separating arc which is parallel to Li except near the distin-
guished arc.
CEU eTD Collection

Legendrian link L in the 1-skeleton of the contact cell decomposition. Hence, it results that
the page S1 is precisely a ribbon of the 1-skeleton of ( M, ξ ). This argument also gives that
the two framings of L agree.
The components of L are independent because it is easy to see from the construction
that there is a collection of disjoint, properly embedded arcs { a1 , ..., an } in S1 such that
 

aj  = ∅
[
Li t ai = {1 pt} and Li ∩ 
j 6 =i

for every i. To conclude we only need to show that there exists a system of generators
A = { a1 , ..., a2g+l +n−2 } for S1 which is adapted to L.

52
 The arcs a1 , ..., an are taken as before. If we complete L to a basis of H1 (S; F) then
Poincaré-Lefschetz duality gives a basis { a1 , ..., a2g+l −1 } with the same property. We define
a2g+l , ..., a2g+l +n−2 in the following way. Each new separating arc is parallel to Li , extended
by following the distinguished arc to the boundary of S1 as in Figure 4.6. Clearly, it dis-
connects the surface, because the first 2g + l − 1 arcs are already a basis of H1 (S, ∂S; F). If
one of the components contains no distinguished arcs, like in Figure 4.7, then we choose
the other endpoint of ai to extend the arc.

In this thesis we use adapted open book decompositions to present Legendrian links
in contact 3-manifolds. Moreover, we study how to relate two different open book decom-

a2g+l
L

ai

Figure 4.7: The picture appears similar to Figure 4.6, but this time the new arc follows the
distinguished arc in the opposite direction.

positions representing isotopic Legendrian links; this is done in Chapter 6.

4.2.2 Abstract open books

An abstract open book is a quintuple (S, Φ, A, z, w) defined as follows. We start with the
pair (S, Φ). We have that S = Sg,l is an oriented, connected, compact surface of genus g
and with l boundary components, not diffeomorphic to a disk. While Φ is the isotopy class
of a diffeomorphism of S into itself which is the identity on ∂S. The class Φ is called the
monodromy.
Theorem 4.2.2 (Thurston and Winkelnkemper) The pair (S, Φ) determines a contact 3-
manifold.
Proof. We begin by observing that, starting from (S, Φ), we get a 3-manifold MΦ as follows:
!
CEU eTD Collection

MΦ = S Φ ∪ ψ ∏ S1 × D 2 ,
l

where l is the number of boundary components of S and SΦ is the mapping torus of Φ. By

this we mean
S × [0, 1] ,
∼
where ∼ is the equivalence relation (φ( x ), 0) ∼ ( x, 1) for every x ∈ S. Finally, ∪ψ means
that the diffeomorphism ψ is used to identify the boundaries of the two manifolds.
For each boundary component Si of S, the map ψ : ∂(S1 × D2 ) → Si × S1 ⊂ SΦ is
defined to be the unique, up to isotopy, diffeomorphism that takes S1 × { p} to Si , where
p ∈ ∂D2 , and {q} × ∂D2 to
{q0 } × [0, 1] = S1 ,
∼

53
 where q ∈ S1 and q0 ∈ ∂S.
The fact that (S, Φ) also determines a contact structure on MΦ was proved in [87].

The set A consists of two collections of properly embedded arcs, B = {b1 , ..., b2g+l +n−2 }
and C = {c1 , ..., c2g+l +n−2 } in S with n > 1, such that all the arcs in B are disjoint, all the

bi ci
Figure 4.8: Two arcs in strip position.

arcs in C are disjoint and each pair bi , ci appears as in Figure 4.8. We suppose that each
strip, the grey area between bi and ci , is disjoint from the others. We also want B and C to
represent two systems of generators for the relative homology group H1 (S, ∂S; F). In this
way, if we name the strips Ai , we have that S \ bi , S \ ci and S \ Ai have exactly n
S S S

connected components.
Finally, z and w are two sets of basepoints: w = {w1 , ..., wn } and z = {z1 , ..., zn }. We
require these sets to have the following properties:

• there is a zi in each component of S \ Ai , with the condition that every component

S

contains exactly one element of z;

• each wi is inside one of the strips Ai , between the arcs bi and ci , with the property
that every strip contains at most one element of w. See Figure 4.9;

wi wi
Si zi zj Si Si
Sj zi
b c b c
Figure 4.9: On the left Si and S j are different components of S \ Ai .
S
CEU eTD Collection

• we choose z and w in a way that each component of S \ B and S \ C contains exactly

one element of z and one element of w.

We can draw an n-component link inside S using the following procedure: we go from
the z’s to the w’s by crossing B and from w to z by crossing C, as shown in Figure 4.10.
Moreover, we observe that the components of the link are independent in S.
Every abstract open book ( f (S), f ◦ Φ ◦ f −1 , f (A), f (z), f (w)), obtained from B =
(S, Φ, A, z, w) by applying a diffeomorphism f of S into itself, is said to be equivalent to
B. Then Theorem 4.2.2 gives the following corollary.

Corollary 4.2.3 Two equivalent pairs (S, φ) and ( f (S), f ◦ Φ ◦ f −1 ) determine contactomorphic
contact 3-manifolds.

54
 w
z
w z
b c b c
Figure 4.10: The link is oriented accordingly to the basepoints.

An abstract open book B as above also determines a multi-pointed Heegaard diagram

(S ∪ −S, b ∪ b, c ∪ c, z, w) and then a link in a 3-manifold up to (smooth) isotopy. Here
the arcs b and c are defined as h(b) and (h ◦ Φ)(c) respectively, where h : S → −S is the
Identity, seen as an unoriented diffeomorphism. Using this procedure we described we
can prove the following theorem.
Theorem 4.2.4 (Legendrian realization theorem) Every abstract open book (S, Φ, A, z, w) de-
termines a Legendrian link L in a contact 3-manifold ( M, ξ ). Furthermore, equivalent abstract
open books give rise to contactomorphic links.
Proof. The contact manifold ( M, ξ ) is obtained from the pair (S, Φ) as in Theorem 4.2.2. It
is easy to check that its diffeomorphism type is the same of the manifold obtained from the
Heegaard diagram (S ∪ −S, b ∪ b, c ∪ c, z, w). This gives us a smooth link L ⊂ S ,→ M.
The Legendrian realization L is obtained by applying the procedure described in [57]
to every component of L. We only need to be careful that the new components are actually
disjoint.

We are now interested in proving that an adapted open book decomposition ( B, π, A)

always determines an abstract open book.
Proposition 4.2.5 We can associate to an adapted open book decomposition ( B, π, A), compatible
with the triple ( L, M, ξ ), an abstact open book (S, Φ, A, z, w) up to isotopy.

Proof. The surface S is obviously the page π −1 (1). Now consider the subsets of unit com-
plex numbers I± ⊂ S1 ⊂ C with non-negative and non-positive imaginary part. Since they
are contractible, we have that π |π −1 ( I± ) is a trivial bundle. This gives two diffeomorphisms
between the pages S1 and S−1 . The monodromy Φ is precisely the isotopy class of the
composition of first diffeomorphism with the inverse of the second.
At this point, we want to define the strips A. Hence, we need the collections of arcs
B and C: starting from the system of generators A, which is adapted to L ⊂ S, we take
them to be both isotopic to A, in “strip position” like in Figure 4.8 and such that L does not
CEU eTD Collection

cross the intersections of the arcs in B with the ones in C. We only have an ambiguity on,
following the orientation of L, which is the first arc intersected by L. To solve this problem
we have to follow the rule that we fixed in Figure 4.10.
Now we need to fix the basepoints. We put the z’s on L; exactly one on each component
of L \ ( L ∩ A). The points in z on different components of L stay in different domains
because of Condition 3 in the definition of adapted system of generators. Then S \ A has
n connected components, since the components of L are independent, and each of these
contains exactly one element of z. Since the z’s are outside of the strips then we have that
each component of S \ B and S \ C also contains only one zi .
The w’s are still put on L, but inside the strips containing the n distinguished arcs. The
points w1 , ..., wn correspond to A1 , ..., An .

55
CEU eTD Collection
 Chapter 5

Heegaard Floer homology

5.1 Link Floer homology

5.1.1 Heegaard diagrams and J-holomorphic curves
Link Floer homology is an invariant for knots and links in 3-manifolds, discovered in
2003 by Ozsváth and Szabó [69] and independently by Rasmussen [78], in his PhD thesis.
It is constructed, here in the case of rational homology 3-spheres, by associating some
chain complexes to a Heegaard diagram, with the help of Lagrangian Floer homology, a
construction from symplectic geometry given by Andreas Floer in [32]. In this thesis we
only discuss two of these complexes, that we define in this chapter.
Starting from a genus g multi-pointed Heegaard diagram H = (Σ, α, β, w, z) for an n-
component link L in M, consider the ( g + n − 1)-fold symmetric product of Σ, that we
defined in Subsection 4.1.1, and the ( g + n − 1)-dimensional tori Tα and Tβ , which are
submanifolds of Symg+n−1 (Σ).
The surface Σ can be equipped with a complex structure. Then Symg+n−1 (Σ) inherits
a complex structure from the direct product of Σ with itself. In particular, this means that
we can fix an open set of almost-complex structures on Symg+n−1 (Σ) compatible with an
appropriate symplectic structure, which exists from results in [75].
Consider the set { J (s)}, where s ∈ [0, 1], a path of such almost-complex structures and
define a J-holomorphic strip, connecting x, y ∈ Tα ∩ Tβ , as a continuous map

u : [0, 1] × R −→ Symg+n−1 (Σ)

CEU eTD Collection

satisfying the following conditions:

1. u maps {0} × R into Tβ and {1} × R into Tα ;

2. the 1-parameter family of paths ut = u|[0,1]×{t} converges uniformly to the constant

path at x (or y) when t goes to −∞ (or +∞);

3. u satisfies
∂u ∂u
+ J (s)u(s+it) = 0.
∂s ∂t
Note that the strip [0, 1] × R is conformally equivalent to the unit disk. This means that
we can use this notion in place of the one of J-holomorphic disk. Details can be found in
[67, 68, 69, 72].

57
 We define the set π2 ( x, y), where x, y ∈ Tα ∩ Tβ , as the set of Z-linear combinations of
the domains {Dk }, which are the components of Σ \ α ∪ β, such that
! !
1 if p ∈ x −1 if p ∈ y
a+b = c+d+ + ;
0 otherwise 0 otherwise

where a, b, c, d are multiplicities of domains which appear like in Figure 5.1. For every
φ ∈ π2 ( x, y) we call
nzi (φ) = φ ∩ {zi } × Symg+n−2 (Σ)
and
nwi (φ) = φ ∩ {wi } × Symg+n−2 (Σ) ;
where here we mean algebraic intersection. Moreover, we say that
n n
nz (φ) = ∑ n zi ( φ ) and nw (φ) = ∑ nw (φ) .
i
i =1 i =1

We also define M(φ) as the moduli space of holomorphic strips that connects x to y. The
αi
a c
p
βi
d b

Figure 5.1: Corner point between four domains.

formal dimension of M(φ), denoted by µ(φ), is the Maslov index; moreover, we call M(
c φ)
the quotient M(φ)/R given by vertical translation of the strip.
For every Spinc structure t on M we define the group cCFL− ( H, t) as the free F[U ]-
module over
Tα ∩ Tβ |t = x ∈ Tα ∩ Tβ such that sw ( x ) = t .


Moreover, we obtain another group by taking

−
d ( H, t) = cCFL ( H, t) ;
CEU eTD Collection

CFL
U=0
this group is an F-vector space, where F is again the field with two elements.

5.1.2 Gradings
We can put a bigrading on cCFL− ( H, t), where H is a Heeegaard diagram as in the
previous subsection, if we keep the condition that the 3-manifold M is a rational homology
3-sphere.
For every x ∈ Tα ∩ Tβ |t we denote with (Wx , Jx ) an almost-complex 4-manifold, with
M as boundary, such that the vector field π ( x ), induced on M by Jx , is isotopic to the one
given by x (and w), that we called πw ( x ) in Subsection 4.1.1. Such 4-manifolds always exist

58
 [60], thus we define the Maslov grading as the absolute Q-grading given by generalizing
the Hopf invariant, defined in Subsection 3.1.2, as follows

1 2 
M ( x ) = d3 ( M, πw ( x )) = c1 (Wx , Jx )[Wx , ∂Wx ] − 3σ(Wx ) − 2χ(Wx ) + 2
4
and we extend it to the whole cCFL− ( H, t) by saying that

M (U p ) = M ( p ) − 2 where p ∈ cCFL− ( H, t) is homogeneous .

The Maslov grading is independent of the choice of (Wx , Jx ).

Lemma 5.1.1 If (Wx0 , Jx0 ) is another almost-complex 4-manifold with the same properties of
(Wx , Jx ) then d3 ( M, πw ( x )) does not change.
Proof. First, we have that t is torsion, because M is a rational homology sphere. This im-
plies that c21 (Wx , Jx )[Wx , ∂Wx ] is well-defined. Now we can find an almost-complex man-
ifold (W, J ) such that ∂W = − M and the 2-plane field induced by J on − M is isotopic to
πw ( x ). This means that we can glue (W, J ) to both our manifolds and obtain two closed
almost-complex 4-manifolds. Since the evaluation on the square of the first Chern class,
the signature and the Euler characteristic are all additive under gluing, the Hirzebruch’s
signature theorem gives that

c21 (Wx , Jx )[Wx , ∂Wx ] − 3σ(Wx ) − 2χ(Wx ) =

−c21 (W, J )[W, ∂W ] + 3σ(W ) + 2χ(W ) =
c21 (Wx0 , Jx0 )[Wx0 , ∂Wx0 ] − 3σ(Wx0 ) − 2χ(Wx0 )

and the claim follows.

The difference between the Maslov grading of two intersection points can be expressed
easily.

Proposition 5.1.2 Given φ ∈ π2 ( x, y) we have that

M( x ) − M (y) = µ(φ) − 2nw (φ)

which is an integer.

Proof. From [72] we know that

CEU eTD Collection

µ(φ) − µ(ψ) = 2(nw (φ) − nw (ψ))

for every φ, ψ ∈ π2 ( x, y). Since we can always find ψ such that nw (ψ) = 0, we just apply
the Atiyah-Singer index theorem and we obtain that

M( x ) − M (y) = µ(ψ) = µ(φ) − 2nw (φ) .

The Maslov grading gives an F-splitting

cCFL− ( H, t) = cCFL−
M
d ( H, t) ;
d ∈Q

59
 moreover, the same holds for CFL d ( H, t).
The definition of M( x ) does not require that there is a link L → M. The grading that
we are going to define now, on the other hand, strictly requires that we take L, which has
n components, into account. We call the Alexander multi-grading of x ∈ Tα ∩ Tβ |t as the
n-tuple (s1 , ..., sn ) such that
n
sw,z ( x ) = ∑ 2si · PD[µi ] + sw (x) .
i =1

Since L always admits a rational Seifert surface F, see Section 2.1, if t is the order of the
Z
link in M then we define the Alexander absolute 2t -grading as follows:
n
sw,z ( x )[ F ]
A( x ) = ∑ si = 2t
i =1

and we extend it on the whole cCFL− ( H, t) by saying that

A (U p ) = A ( p ) − 1 where p ∈ cCFL− ( H, t) is homogeneous .

The Alexander grading is independent of the choice of the surface F.

Lemma 5.1.3 If F 0 is another rational Seifert surface for L then the value of A( x ) does not change.
Proof. The gluing F ∪ − F 0 defines a closed surface and then a 2-homology class in ML =
M \ ν(˚L) and M. Since H2 ( M; Z) is trivial, then [ F ∪ − F 0 ] is null-homologous in it and
also in H2 ( ML ; Z). This means that

0 = sw,z ( x ) F ∪ − F 0 = sw,z ( x )[ F ] − sw,z ( x ) F 0 .

   

Like for the Maslov grading, we can compute the difference between the Alexander
grading of two intersection points.
Proposition 5.1.4 Given φ ∈ π2 ( x, y) we have that

A( x ) − A(y) = nz (φ) − nw (φ)

which is again an integer.

CEU eTD Collection

Proof. From [72] we know that

n  
sw,z ( x ) − sw,z (y) = 2 · ∑ nzi (φ) − nwi (φ) · PD[µi ] .
i =1

Then the claim follows by evaluating both sides of this equation with [ F ], where F is a
rational Seifert surface for L.

We have another F-splitting

cCFL− ( H, t) = cCFL−
M
d,s ( H, t)
d,s∈Q

d ( H, t).
and then again the same is true for CFL

60
 5.1.3 Differential
In the previous subsection we introduced the underlying space of our chain complexes;
now we want to construct the differentials. We define ∂− in the following way

∂− x = ∑ ∑ c φ ) · U nw (φ) y
M(
y ∈Tα ∩T β |t φ∈π2 ( x,y),
µ(φ)=1,nz (φ)=0

and
∂− (Ux ) = U · ∂− x
for every x ∈ Tα ∩ Tβ |t .
We note that, since we are interested in φ’s that are image of some J-holomorphic disks,
we have that nzi (φ), nwi (φ) > 0 [68]. This means that nz (φ) = nw (φ) = 0 if and only if
nzi (φ) = nwi (φ) = 0 for every i = 1, ..., n.
The map ∂− is well-defined if the diagram D is admissible, which means that every
φ ∈ π2 ( x, x ) with nw (φ) = 0, representing a non-trivial homology class, has both positive
and negative local multiplicities.

Lemma 5.1.5 If M is a rational homology 3-sphere then, given a multi-pointed Heegaard diagram
(Σ, α, β, w, z), we can always achieve admissibility through isotopies.
Proof. The proof is taken from [72]. We embed in Σ a tree Γ, whose n vertices are the
basepoints in w. We claim that admissibility can be achieved by isotoping some of the
β-curves in a regular neighborhood of Γ. More specifically, if γ is an arc connecting w1 to
w2 in Γ then we perform an isotopy of the β’s in a regular neighborhood of γ, in such a
manner that there is a pair of arcs δ1 and δ2 so that δ1 ∪ δ2 is isotopic to γ as an arc from w1
to w2 , but δ1 is disjoint from the β-circles, while δ2 is disjoint from the α-circles.
Moreover, we find another pair of arcs δ10 and δ20 so that δ10 ∪ δ20 is isotopic to γ, only this
time δ10 is disjoint from the α-circles, while δ20 is disjoint from the β-circles. Isotoping the β’s
in a regular neighborhood of all the edges in Γ as above, we obtain a Heegaard diagram
that we claim is weakly admissible.
Suppose that φ ∈ π2 ( x, x ), which decomposes as φ = A + B where A has boundary on
Tα and B has boundary on Tβ , has the property that the oriented intersection number of
∂A with γ is non-zero. Then, at the intermediate endpoint of δ1 , we see that A + B has local
multiplicity given by ∂A ∩ γ, while at the intermediate endpoint of δ10 we have that A + B
has local multiplicity given by ∂B ∩ γ = −∂A ∩ γ, since φ ∈ π2 ( x, x ) [68]. Thus, if for some
CEU eTD Collection

edge in Γ it is ∂A ∩ γ 6= 0, then φ = A + B has both positive and negative coefficients.

However, if φ = A + B as above and ∂A has algebraic intersection number equal to
zero with each edge in Γ then, after subtracting off nw1 ( A) of copies of Σ, we can write φ =
A0 + B0 , where A0 and B0 again have boundaries on Tα and Tβ and nwi ( A) = nwi ( B) = 0.
Hence, a simple computation gives that φ = 0.

We now prove that ∂− ◦ ∂− = 0. See [72] for more details.

Theorem 5.1.6 (Ozsváth and Szabó) We have that ∂− (∂− x ) is zero for every x ∈ Tα ∩ Tβ |t and
every Spinc structure t on M, which means that ∂− is a differential.

Proof. We consider the moduli space M(c φ), where φ ∈ π2 ( x, z) is non-negative with
µ(φ) = 2, x and z are intersection points and nz (φ) = 0. We use a structure theorem

61
 in [33, 69] which proves that there is a compact 1-dimensional manifold with boundary,
whose interior is M(
c φ), and its boundary is identified with
[ [
M(
c φ1 ) × M(
c φ2 ) .
y φ1 ∈π2 ( x,y),φ2 ∈π2 (y,z)
µ(φ1 )=µ(φ2 )=1,φ1 ∗φ2 =φ
nz (φ1 )=nz (φ2 )=0

Since a compact 1-manifold has an even number of boundary points, we can conclude that

∑ ∑ M( c φ2 ) · U nw (φ) = 0 .
c φ1 ) · M(
y φ1 ∈π2 ( x,y),φ2 ∈π2 (y,z)
µ(φ1 )=µ(φ2 )=1,φ1 ∗φ2 =φ
nz (φ1 )=nz (φ2 )=0

Adding up the left-hand-side over all φ ∈ π2 ( x, z) with µ(φ) = 2 computes the z compo-
nent of ∂− (∂− x ), then ∂− ◦ ∂− = 0.

Using Lemma 5.1.5 and Theorem 5.1.6, we obtain the chain complexes
 
cCFL− ( H, t), ∂− ∂ = ∂ − |U =0 .


and d ( H, t), b
CFL

Moreover, Proposition 5.1.2 leads to the following result.

Proposition 5.1.7 For every intersection point x ∈ Tα ∩ Tβ |t the element ∂− x is homogeneous

with respect to the Maslov grading and M (∂− x ) = M ( x ) − 1.

Proof. If U nw (φ) y, where φ ∈ π2 ( x, y) and y ∈ Tα ∩ Tβ |t , is a monomial of ∂− x then we

have that  
M U nw (φ) y = M (y) − 2nw (φ) = M ( x ) − µ(φ) = M( x ) − 1 .

The last equality holds because in the differential we only consider φ’s with Maslov index
equal to 1.

This implies that

∂− − −
d : cCFLd ( H, t) −→ cCFLd−1 ( H, t)

is the restriction of the differential to cCFL−

d ( H, t).
In the same way, Proposition 5.1.4 immediately gives a similar relation for the Alexan-
der grading.
CEU eTD Collection

Proposition 5.1.8 For every intersection point x ∈ Tα ∩ Tβ |t the element ∂− x is homogeneous

with respect to the Alexander grading and A(∂− x ) = A( x ).

Proof. If U nw (φ) y, where φ ∈ π2 ( x, y) and y ∈ Tα ∩ Tβ |t , is a monomial of ∂− x then we

have that  
A U nw (φ) y = A ( y ) − n w ( φ ) = A ( x ) − n z ( φ ) = A ( x ) .

The last equality holds because in the differential we only consider φ’s with nz (φ) = 0.

Hence, the map

∂− − −
d,s : cCFLd,s ( H, t) −→ cCFLd−1,s ( H, t)

is the restriction of the differential to cCFL−

d,s ( H, t).

62
 5.1.4 Homology
In this section we define two versions of link Floer homology, corresponding to the
chain complexes that we defined in the previous subsection. The first group is called col-
lapsed link Floer homology and it is defined by

cHFL− ( H, t) = cHFL−
M
d,s ( H, t) ,
d,s∈Q

where
Ker ∂−
d,s
cHFL−
d,s ( H, t) = .
Im ∂−
d+1,s

The group cHFL− ( H, t) is a bigraded F-vector space with a structure of finite dimensional
F[U ]-module.

Theorem 5.1.9 (Ozsváth and Szabó) The isomorphism type of the group cHFL− ( H, t) is invari-
ant under smooth isotopy of the link L ,→ M. Which means that, if H 0 is another multi-pointed
Heegaard diagram for a link isotopic to L in M, it is

cHFL− ∼ − 0
d,s ( H, t) =F cHFLd,s ( H , t)

for every d, s ∈ Q and t ∈ Spinc ( M ).

Hence, we can denote cHFL− ( H, t) with cHFL− ( M, L, t). We leave the proof of this
theorem for Section 5.3.
The second version that we want to study is the hat link Floer homology group
M
[ ( H, t) =
HFL [ d,s ( H, t) ;
HFL
d,s∈Q

given by
[ d,s ( H, t) = Ker ∂d,s .
b
HFL
Im b
∂d+1,s
[ ( H, t) is a finite dimensional, bigraded F-vector space. Moreover, as the
The group HFL
collapsed minus version, it is also a smooth isotopy invariant of links in 3-manifolds. In
fact, we have an analogous theorem in this case.
CEU eTD Collection

Theorem 5.1.10 (Ozsváth and Szabó) The isomorphism type of the group HFL[ ( H, t) is invari-
0
ant under smooth isotopy of the link L ,→ M. Which means that, if H is another multi-pointed
Heegaard diagram for a link isotopic to L in M, it is

[ d,s ( H, t) ∼
HFL [ d,s ( H 0 , t)
=F HFL

for every d, s ∈ Q and t ∈ Spinc ( M).

Hence, we can denote HFL[ ( H, t) with HFL

[ ( M, L, t). Theorems 5.1.9 and 5.1.10 have
been proved in [69] for knots and generalized to links in [72].
The group HFL
[ has the following interesting symmetry property.

63
 Proposition 5.1.11 Given a smooth link L in a rational homology 3-sphere M, we have that
[ d,s ( M, L, t) ∼
HFL [ d−2s,−s ( M, L, −t + PD[ L])
=F HFL
for every d, s ∈ Q and t ∈ Spinc ( M).
Proof. It follows from the relation between the two relative Spinc structures obtained
by reversing the orientation of L, see Subsection 4.1.2 before. More specifically, given
(Σ, α, β, w, z) a multi-pointed Heegaard diagram for L in M, we call H1 = (Σ, β, α, w, z)
the first diagram for − L. Then Proposition 4.1.4 gives us the isomorphism

Φ1 : HFL
[ d,s ( M, L, t) −→ HFL
[ d,s ( M, − L, −t) .
In the same way, we call H2 = (Σ, α, β, z, w), which also represents the link − L. Now from
Propositions 4.1.3 and 4.1.4 we obtain the map

Φ2 : HFL
[ d,s ( M, − L, −t) −→ HFL
[ d−2s,−s ( M, L, −t + PD[ L])
and the proof is completed by taking the composition of Φ1 and Φ2 .
We can use this property to normalize the relative grading a( x, y) = nz (φ) − nw (φ),
where φ ∈ π2 ( x, y), given in Proposition 5.1.4. Hence, this can be seen as an equivalent
definition of the Alexander grading; in fact, we have that A( x ) is the only normalization
of a( x, y) which respects the symmetry in Proposition 5.1.11. This property proves the
following corollary.
Corollary 5.1.12 For every multi-pointed Heegaard diagram (Σ, α, β, w, z), and x ∈ Tα ∩ Tβ , we
have that
2A( x ) = M ( x ) − Mz ( x ) ,
where A( x ) and M ( x ) are the Alexander and Maslov grading of x and Mz ( x ) = d3 ( M, πz ( x )).
Proof. From Propositions 5.1.2 and 5.1.4 we have that the relative grading
b( x, y) = ( M ( x ) − M(y)) − ( Mz ( x ) − Mz (y))
coincides with 2a( x, y) = 2( A( x ) − A(y)). Therefore, we just need to show that b( x, y)
satisfies the symmetries in Proposition 5.1.11 and this can be done easily.

5.2 Heegaard Floer homology of 3-manifolds

Let us take a multi-pointed Heegaard diagram H = (Σ, α, β, w), where the set w con-
CEU eTD Collection

tains n basepoints. We can change the differential b

∂ to b
∂w as follows:

∂w x =
b ∑ ∑ M(
c φ) · y
y ∈Tα ∩T β |t φ∈π2 ( x,y),
µ(φ)=1,nw (φ)=0
 
for every x ∈ Tα ∩ Tβ |t and b c ( H, t), b
∂w is U-equivariant. The new complex CF ∂w is well-
defined. Moreover, its homology
M
HF ( H, t) =
d HF d ( H, t)
d
d ∈Q

is an invariant of the manifold, once we fix the number of basepoints in H, as stated in the
following theorem.

64
 Theorem 5.2.1 (Ozsváth and Szabó) The isomorphism type of the group d HF ( H, t) is invariant
under diffeomorphism of the Spinc 3-manifold ( M, t). In other words, if H 0 is another Heegaard
diagram for a manifold diffeomorphic to ( M, t), and with the same number of basepoints in H, then
it is
HF d ( H, t) ∼
d =F dHF d ( H 0 , t)
for every d ∈ Q and t ∈ Spinc ( M).

When the cardinality of w is one we denote d HF ( H, t) with d

HF ( M, t). The proof of this
theorem follows the same techniques of the ones of Theorems 5.1.9 and 5.1.10, see [68].
It is easy to observe that if H w,z = (Σ, α, β, w, z) is a multi-pointed Heegaard diagram,
where each zi and wi are in the same component of Σ \ α ∪ β, then the differential b ∂ coin-
w,z
cides with ∂w , which is defined from the diagram H, obtained from H by dropping the
b
basepoints z. Hence, we have the following proposition.

b a

w1
wn

Figure 5.2: This stabilization represents a disjoint union of a link in a 3-manifold with an
unknot.

Lemma 5.2.2 Given a rational homology 3-sphere ( M, t), we have that

  N ( n −1)
HFL
[ ( M, ∼
n , t) = HF ( M, t) ⊗ F(−1) ⊕ F(0)
d

and the isomorphism preserves the Maslov grading. Here, the symbol n denotes the n component
unlink in M.

Proof. We reason by induction on n. If n = 1 then the claim follows immediately from

the previous observation. For the general case, we use Theorem 5.2.1 and we obtain that,
after some handleslides and stabilizations, a multi-pointed Heegaard diagram for n in
CEU eTD Collection

M appears as in Figure 5.2. Then, using the inductive hypothesis, we obtain that we just
have to prove that changing a Heegaard diagram H into H 0 by this type of stabilization
corresponds to tensor product with F(−1) ⊕ F(0) in the homology group.
To do this we observe that an explicit computation gives that M( x a ) = M( x ) and
M ( xb ) = M( x ) − 1, where x a (or xb ) is given by adding the point a (or b) to the inter-
section point x; see Figure 5.2. Since every intersection point in the new complex needs to
have one between a or b as a coordinate, we have an isomorphism of chain complexes
 
d ( H 0 , t) −→ CFL
CFL d ( H, t) ⊗ F(−1) ⊕ F(0) .

Therefore, we conclude by noting that the differential is zero on F(−1) ⊕ F(0) and then the
isomorphism descends in homology.

65
 This lemma implies the following theorem.

Theorem 5.2.3 (Ozsváth and Szabó) If the set w in the Heegaard diagram H contains n base-
points then we have that
  N ( n −1)
d ∼
HF ( H, t) =F HF ( M, t) ⊗ F(−1) ⊕ F(0)
d . (5.1)

More explicitly, it is
n −1 
n−1 d

HF d ( H, t) ∼
M
d =F HF d+i ( M, t)
i =0
i

for every d ∈ Q and t ∈ Spinc ( M).

From Propositions 5.1.4 and 5.1.11 we see that the Alexander grading is always zero in
the case of unlinks. Hence, we can express the hat link Floer homology of n entirely in
terms of the homology of M:
 n −1 
M n−1

HF d+i ( M, t) if s = 0


∼
d
HFL
[ d,s ( M, n , t) = i = 0
i .


{0} otherwise

A particular case is given by the 2-component unlink 2 in S3 . In fact, we have that

[ ( 2) ∼
HFL = F(−1,0) ⊕ F(0,0) ; note that we do not specify the manifold when we work in
the 3-sphere.
We conclude this section by computing the homology of the unlink also in the col-
lapsed version. We observe that, for the same argument used to prove Lemma 5.2.2, the
differential ∂− coincides with b ∂ too if we choose an appropriate diagram. Moreover, the
generators of cCFL− ( H, t) and CF c ( H, t) are the same; with the difference that the first
space is an F[U ]-module. This gives the following result.

Corollary 5.2.4 The homology group cHFL− ( M, n , t), seen as an F[U ]-module, is the tensor
[ ( M, n , t) with (the module) F[U ] over F.
product of HFL

Using Lemma 5.2.2 we can restate this corollary by saying that

 n −1 
M n−1

CEU eTD Collection

HF d−2s+i ( M, t) if s 6 0


∼
d
cHFL−
d,s ( M, n , t) = i = 0
i .


{0} otherwise

5.3 Invariance
5.3.1 Choice of the almost-complex structures
In this section we give ideas for the proofs that link Floer homology is invariant under
the moves in Section 4.1, that relates two multi-pointed Heegaard diagrams represent-
ing the same link L in the Spinc 3-manifold ( M, t), and the analytic input of the path of

66
 almost-complex structures J (s) on the symmetric power and the complex structure on Σ,
see Subsection 5.1.1. We always suppose that M is a rational homology 3-sphere.
Suppose that H = (Σ, α, β, w, z) is a multi-pointed Heegaard diagram for an n-
component link in M. First, we argue that if we fix a complex structure on Σ then the link
Floer homology groups are independent of the choice of J (s) on Symg+n−1 (Σ). Take two
paths J0 (s) and J1 (s) on Symg+n−1 (Σ); since the symmetric power is simply-connected, we
can connect J0 (s) and J1 (s) with a 2-parameter family J : [0, 1] × [0, 1] → Symg+n−1 (Σ),
thought of as a 1-parameter family of paths indexed by t ∈ [0, 1] where Jt (s) is the path
obtained by fixing t. In fact, we can arrange that Jt (s) is independent of t for t’s near to
zero or one, so that Jt (s) can be naturally extended to the whole R.
Then, as it is familiar in Floer theory, we can define an associated chain map
   
Φ−Jt (s)
: cCFL −
( H, t ) , ∂ −
J (s)
−→ cCFL −
( H, t ) , ∂ −
J (s)
0 1

by
Φ−
Jt (s)
x= ∑ ∑ c J (s) ( φ ) · U nw (φ) y ,
M t
y φ∈π2 ( x,y)
µ(φ)=1,nz (φ)=0

where M c J (s) (φ) denotes the moduli space of holomorphic strips with a time-dependent
t
complex structure on the target.
The usual arguments from Floer theory then apply to show that Φ− Jt (s)
is a chain map
which induces an isomorphism in homology. We give a sketch of the proof of this fact. The
details can be found in [68].
A transversality theorem in [68] shows that, for a generic path Jt (s), the zero-
dimensional components of the moduli spaces M c J (s) (φ) are smoothly cut out and com-
t
−
pact. Thus, the map Φ Jt (s) is well-defined. To show that it is a chain map, we consider
the ends of the 1-dimensional moduli spaces M c J (s) (ψ), where µ(ψ) = 2. Then the same
t
argument in the proof of Theorem 5.1.6 and Gromov’s compactness [42] give that Φ− Jt (s)
is
also a chain map.
The fact that Φ−
Jt (s)
induces an isomorphism in homology is done by showing that the
composition with Φ− J1−t (s)
is chain homotopic to the identity; we recall that we said what
is a chain homotopy in Subsection 1.4.2. The chain homotopy is constructed by using a
homotopy Jt,τ (s) between two 2-parameter families of complex structures. We can define
a moduli space
CEU eTD Collection

[
M
c J (s) ( φ ) =
t,τ
Mc J (s)
t,η
η ∈[0,1]

for each φ ∈ π2 ( x, y). For a generic Jt,τ (s) this manifold has dimension µ(φ) + 1. We define
a map
H− x = ∑
Jt,τ (s) ∑ Mc J (s) ( φ ) · U nw (φ) y .
t,τ
y φ∈π2 ( x,y)
µ(φ)=0,nz (φ)=0

To see then that H J−t,τ (s) is the chain homotopy between Φ−

Jt (s)
◦ Φ−
J1−t (s)
and the identity map,
we reason in the same way as before [68].
Next, we see that the chain complex is independent of the complex structure J on the
surface Σ. To this end, we observe that the chain complexes remain unchanged under

67
 small perturbations of the path of almost-complex structures J (s). Furthermore, we can
approximate J (s) with a symmetric path J 0 (s), induced by a complex structure J0 close to
J. This shows that the Floer homology is also independent of the choice of J, since the
space of allowed complex structures over Σ is connected, because it is obtained from the
space of all complex structures by removing a codimension two subset, see [68].

5.3.2 Isotopies
Consider two multi-pointed Heegaard diagram H = (Σ, α, β, w, z) and H 0 =
(Σ, α0 , β0 , w, z) which differ by an isotopy. First of all, isotopies which preserve the con-
dition that the α are transverse to the β can be thought of as variations in the metric, or
equivalently the complex structure on Σ. Hence, the invariance of the Floer homology
under such isotopies follows from Subsection 5.3.1.
It suffices then to show that the homology remains unchanged when a pair of canceling
intersection points between, α1 and β 1 is introduced. See Figure 5.3 for an illustration.
Such an isotopy can be realized by moving α by an exact Hamiltonian diffeomorphism of
Σ. We recall that, on a symplectic manifold, a 1-parameter family of real-valued functions
Ht naturally gives rise to a unique 1-parameter family of Hamiltonian vector fields Xt ,
specified by
ω ( Xt , ·) = dHt ,
where the left-hand side denotes the contraction of the symplectic form ω with the vector
field. A 1-parameter family of diffeomorphisms Ψt is said to be an exact Hamiltonian isotopy
if it is obtained by integrating a Hamiltonian vector field, which means when

• Ψ0 is the Identity;
dΨt
• = Xt .
dt
By taking a positive bump function h, supported in a neighborhood of a point which lies
on α1 , and letting f : R → [0, 1] be a non-negative smooth function whose support is (0, 1),
we can consider the Hamiltonian Ht = f (t)h. The corresponding diffeomorphism moves
the curve α1 slightly, without moving any of the other curves in α.

H
β1
β1
α10 H0
CEU eTD Collection

α1
Figure 5.3: By moving the curve α1 to α10 through a Hamiltonian isotopy, we introduce a
pair of canceling intersection points.

We assume that the exact Hamiltonian is supported on Σ \ (α2 ∪ ... ∪ α g+n−1 ∪ w ∪ z)

and that the Hamiltonian is supported in [0, 1]. The isotopy Ψt induces an isotopy of Tα .
We need to show that this isotopy gives a map on Floer homology, by imitating the usual
constructions from Lagrangian Floer theory.
The map is induced by counting points in the zero-dimensional components of the
moduli spaces with a time-dependent constraint. To be precise, here, we fix a 1-parameter
family J (s) of almost-complex structures on the symmetric power of Σ. Now, we have

68
 homotopy classes of holomorphic strips π2Ψt ( x, y), where x is an intersection point and
y ∈ Ψ1 (Tα ) ∩ Tβ , which denote homotopy classes of maps satisfying the condition in
Subsection 5.1.1 and such that

u(1 + it) ∈ Ψ1 (Tα ) and u(0 + it) ∈ Tβ

for every t ∈ [0, 1].

Now, we define the map associated to the isotopy

Γ− − − 0
Ψt : cCFL ( H ) −→ cCFL ( H )

by the formula
Γ−
Ψt x = ∑ ∑ cΨ t ( φ ) · U n w ( φ ) y .
M
y Ψ
φ∈π2 t ( x,y)
µ(φ)=1,nz (φ)=0

The important observation is that the moduli spaces considered have Gromov compactifi-
cations. This is shown in [68]. Then, the fact that Γ−
Ψt induces an isomorphism in homology
can be proved in the same way as in Subsection 5.3.1.

5.3.3 Handleslides
Consider a multi-pointed Heegaard diagram H 0 = (Σ, α, γ, w, z) which is obtained
from (Σ, α, β, w, z) by a single handleslide on a curve in β. Then there is a chain map

Ψαβγ : cCFL− ( H, t) −→ cCFL− ( H, t0 )

defined by counting holomorphic triangles as follows

Ψαβγ x = ∑ ∑ c φ ) · U nw (φ) y ,
M(
y φ∈π2 ( x,Θ βγ ,y)
µ(φ)=1,nz (φ)=0

where Θ βγ is the canonical top-generator of the homology of the chain complex

CF − (Σ, β, γ, w), ∂−


w , whose definition is in [72] together with all the details.
Let us denote with Ψ∗αβγ the induced map in homology, then we note that
 
sw,z ( x ) = s0w,z Ψ∗αβγ ( x ) ,
CEU eTD Collection

where here s0w,z is the map associated to the Heegaard diagram H 0 , since the triangles used
in the definition of Ψ∗ are disjoint from w and z.
Now, associativity of the triangle maps [72] gives that the map (Ψαγβ0 ◦ Ψαβγ )∗ is an
isomorphism, where

Ψαγβ0 (Ψαβγ ( x )) = ∑ ∑ c φ ) · U nw (φ) y

M(
y φ∈π2 ( x,Θ ββ0 ,y)
µ(φ)=1,nz (φ)=0

and β0 is a small perturbation of β such that each β0i intersects β i in precisely two intersec-
tion points. Therefore, we have that the map Ψ∗αβγ is also an isomorphism.

69
 5.3.4 Stabilizations
Finally, suppose that H 0 = (Σ0 , α0 , β0 , w0 , z0 ) is obtained from H = (Σ, α, β, w, z) by a
stabilization. Then we immediately observe that, since we perform the stabilization away
from w and z, the assignment to relative Spinc structures is unaffected. Denote with c the
intersection point that appears in the stabilization. Then it is easy to see that the intersec-
tion points in H 0 correspond to T0α ∩ T0β = Tα ∩ Tβ × {c}.
Now fix basepoint wi in the same path-component of Σ \ (α ∪ β) as p ∈ Σ, the point
that we use to perform the connected sum that leads to Σ0 . Let wi0 be the corresponding
basepoint in Σ0 . If x ∈ Tα ∩ Tβ and x 0 ∈ T0α ∩ T0β is the new intersection point x × {c} then
the induced Spinc structures sw and sw0 agree, since the corresponding vector fields agree
away from the 3-ball where the stabilization occurs.
Consider x and y intersection points and take φ ∈ π2 ( x, y) such that nz (φ) = 0. More-
over, consider x 0 = x × {c} and y0 = y × {c} and take φ0 ∈ π2 ( x 0 , y0 ) such that nz0 (φ0 ) = 0.
Hence, using a result in [68] we have that, for certain special paths of almost-complex struc-
tures, the moduli space M(φ0 ) is identified with M(φ) × {c}; together with its deforma-
tion theory, including the determinant line bundles. This implies that the chain complexes
are identical and then the homology groups also coincide.

5.4 Multiplication by U + 1 in the link Floer homology group

Take a multi-pointed Heegaard diagram H = (Σ, α, β, w, z) for an n-component link L
in the Spinc 3-manifold ( M, t), where M is a rational homology 3-sphere. Consider also
the diagram H z = (Σ, α, β, z). We have the following surjective F-linear map:
U =1 c
F : cCFL− ( H, t + PD[ L]) −−→ CF ( H z , t) ;
which is given by setting U equal to 1.
The map F clearly commutes with the differentials; it is well-defined and surjective,
because every intersection point x is such that F ( x ) = x. Moreover, if sz ( x ) = t then
sw ( x ) = t + PD[ L] from Proposition 4.1.3.
Lemma 5.4.1 The map F sends an element with bigrading (d, s) into an element with Maslov
grading d − 2s.
Proof. We have that

Mz ( F (U k x )) = Mz ( x ) = M ( x ) − 2A( x ) =
CEU eTD Collection

= ( M( x ) − 2k) − 2( A( x ) − k) = M(U k x ) − 2A(U k x ) ,

where the second equality follows from Corollary 5.1.12.

The map F induces a map F∗ in homology:

  N ( n −1)
− U =1
F∗ : cHFL ( M, L, t + PD[ L]) −−→ HF ( M, t) ⊗ F(−1) ⊕ F(0)
d .

We now recall that, since the link Floer homology group is an F[U ]-module and F is a field,
it is
cHFL− M, L, t0 ∼ = F [U ] r ⊕ T ,

where r is an integer and T is the torsion F[U ]-module; see Subsection 1.5.2.

70
 Lemma 5.4.2 The following two statements hold:

1. we have that F ( x ) = 0 and x is homogeneous with respect to the Alexander grading if and
only if x = 0;

2. we have that F∗ [ x ] = [0] and [ xi ] is homogeneous with respect to the Alexander grading, for
every i = 1, ..., r, if and only if [ x ] is torsion. Here the [ xi ]’s are a decomposition of [ x ] in the
torsion-free quotient of cHFL− ( M, L, t + PD[ L]).

Proof. 1. The if implication is trivial. For the only if, suppose that F ( x ) = 0; this gives
that
x = (1 + U )λ1 (U )y1 + ... + (1 + U )λt (U )yt ,
where λi (U ) ∈ F[U ] for every i = 1, ..., t and y1 , ..., yt are all the intersection points
that induce the Spinc structure t.
Since each yi is homogeneous and the monomial U drops the Alexander grading by
one, we have that λi (U ) = 0 for every i = 1, ..., t and then x = 0.

2. Again the if implication is trivial. Now we have that

r
[x] = ∑ [ xi ] + [ x ] T ,
i =1

where [ x ] T is the projection of [ x ] on the torsion submodule T. Since F∗ [ x ] = [0], it is

[ x ] = (1 + U )[z] = (1 + U )λ10 (U )[z1 ] + ... + (1 + U )λr0 (U )[zr ] + [ x ]T ;

where [ xi ] = (1 + U )λi0 (U )[zi ] for every i = 1, ..., r and the [zi ]0 s are a homogeneous
basis of the torsion-free quotient.
The same argument of 1 implies that the polynomials λi0 (U ) are all zero and then
[ x] = [ x]T .

Now we use Lemma 5.4.2 to show that there is a correspondence between the torsion-
free quotient of the link Floer homology group and the multi-pointed hat Heegaard Floer
homology.

Theorem 5.4.3 If L is an n-component link in a rational homology 3-sphere M then there exists
an isomorphism of F[U ]-modules
CEU eTD Collection

cHFL− ( M, L, t + PD[ L])

−→
T
    N ( n −1)
HF ( M, t) ⊗F F[U ] ⊗F[U ] F[U ](−1) ⊕ F[U ](0)
d ;

which sends a homology class of bigrading (d, s) into one of Maslov grading d − 2s.

Proof. We just have to show that F∗ sends {[z1 ], ..., [zr ]}, a homogeneous F[U ]-basis of
the torsion-freeNquotient of cHFL− ( M, L, t + PD[ L]), into an F-basis of X = d HF ( M, t) ⊗
  ( n −1)
F(−1) ⊕ F(0) . Statement 1 in Lemma 5.4.2 tells us that F∗ is surjective. In fact, if
[y] ∈ X then it is 0 = b ∂z y = F (∂− x ); where F ( x ) = y. We apply Lemma 5.4.2 to ∂− ( x ),

71
 since we can suppose that both x and y are homogeneous, and we obtain that ∂− ( x ) = 0;
then [ x ] is a homology class. At this point, it is easy to see that the set { F∗ [z1 ], ..., F∗ [zr ]} is
a system of generators of X .
In order to provethat F∗ [z1 ], ..., F∗ [zr ] are also linearly independent in X we suppose
that there is a subset F∗ [zi1 ], ..., F∗ [zik ] such that

F∗ [zi1 ] + ... + F∗ [zik ] = F∗ [zi1 + ... + zik ] = [0] .

Then we apply Statement 2 of Lemma 5.4.2 to [zi1 + ... + zik ] and we obtain that it is a
torsion class. This is a contradiction, because [zi1 ], ..., [zik ] are part of an F[U ]-basis of a
torsion-free F[U ]-module.
CEU eTD Collection

72
 Chapter 6

Legendrian and contact invariants

from open book decompositions

6.1 Legendrian Heegaard diagrams

6.1.1 Heegaard diagrams through abstract open books
In this chapter we introduce a specific type of Heegaard diagrams, obtained from
adapted open book decompositions, that we call Legendrian Heegaard diagrams. Given
an adapted open book decomposition ( B, π, A), compatible with the triple ( L, M, ξ ), see
Subsection 4.2.1, a Legendrian Heegaard diagram is a multi-pointed Heegaard diagram
D = (Σ, α, β, w, z) which is obtained from the abstract open book associated to ( B, π, A),
defined in Subsection 4.2.2.
Denote such abstract open book with (S1 , Φ, A, z, w). We have that S±1 = π −1 (±1).
The sets of arcs in S1 are A = { a1 , ..., a2g+l +n−2 } and B = {b1 , ..., b2g+l +n−2 }, as in Figure

S ∪ −S

z2

z1 α3
β3 w2

w1 α2
β2
CEU eTD Collection

β1 α1
Figure 6.1: A diagram for the standard Legendrian 2-unlink in (S3 , ξ st ), obtained from an
open book decomposition with page as in Figure 4.5.

4.8. We recall that g is the genus of S1 , while l is the number of its boundary components
and n is the number of components of L. We define ai and bi for i = 1, ..., 2g + l + n − 2 as
in Subsection 4.2.2.
Then D is gotten as follows:

• the surface Σ is S1 ∪ −S−1 ; since π is a locally trivial fibration the pages S1 and S−1
are diffeomorphic, but we reverse the orientation of the second one when we glue
them together;

73
 • the curves are given by αi = bi ∪ bi and β i = ai ∪ ai for every i = 1, ..., 2g + l + n − 2;

• the z’s and the w’s are the set of baspoints that we introduced in Section 4.2.
In the settings of Chapter 4 the Legendrian Heegaard diagram (Σ, α, β, w, z) is a diagram
for the (smooth) link L in the 3-manifold − M, see [72] and Subsection 4.2.2 before, that is
M considered with the opposite orientation. We remark that here α and β are swapped
with respect to the convention of Ozsváth and Szabó, which is the one we used in Chapter
4. See Figure 6.1 for an example.
We observe that, given ( B, π, A), the only freedom in the choice of the Legendrian Hee-
gaard diagram is in the arcs b1 , ..., b2g+l +n−2 inside S−1 , which can be taken in the isotopy
class of Φ(b1 ), ...Φ(b2g+l +n−2 ).
Proposition 6.1.1 Suppose D = (Σ, α, β, w, z) is a Legendrian Heegaard diagram given by an
adapted open book decomposition compatible with ( L, M, ξ ), where M is a rational homology 3-
sphere. Then D is always admissible up to isotopy of the arcs in B.
Proof. It follows from Lemma 5.1.5 and [77].

Our strategy is to study how Legendrian isotopic triples are related to one another,
which means to define a finite set of moves between two adapted open book decomposi-
tions. Then find chain maps Ψ( D1 ,D2 ) for each move, that induce isomorphisms in homol-
ogy, which will be particular cases of the maps in Section 5.3.

6.1.2 Global isotopies

We want to show that, given two Legendrian isotopic links L1 , L2 ,→ ( M, ξ ), two open
book decompositions ( Bi , πi , Ai ), compatible with the triples ( Li , M, ξ ), are related by a
finite sequence of moves. The first lemma follows easily.
Lemma 6.1.2 Let us consider an adapted open book decomposition ( B1 , π1 , A1 ), compatible with
the triple ( L1 , M, ξ ), and suppose that there is a contact isotopy of ( M, ξ ), sending L1 into L2 .
Then the time-1 map of the isotopy is a diffeomorphism F : M → M such that F ( B1 ), π1 ◦
F −1 , F ( A1 ) is an adapted open book decomposition, compatible with ( L2 , M, ξ ).

Proof. The proof of the statement can be checked without effort.

This lemma says that, up to global contact isotopies, we can consider ( Bi , πi , Ai ) to be

both compatible with a triple ( L, M, ξ ), where the link L is Legendrian isotopic to Li for i =
CEU eTD Collection

1, 2. In other words, we can just study the relation between two open book decompositions
compatible with a single triple ( L, M, ξ ).

6.1.3 Positive stabilizations

 
Let us start with a pair (S, Φ). A positive stabilization of (S, Φ) is the pair eΦ
S, e
obtained in the following way:
• the surface Se is given by adding a 1-handle H to S;

• the monodromy Φ e is isotopic to Φ0 ◦ Dγ . The map Φ0 concides with Φ on S and it is

the Identity on H. While Dγ is the right-handed Dehn twist along a curve γ; which
intersects S ∩ H transversely precisely in the attaching sphere of H. See Figure 6.2.

74
 H
a

Se γ
S

Figure 6.2: The arc γ ∩ S̊ is a generic arc in the interior of S.

We say that ( B0 , π 0 , A0 ) is a positive stabilization of ( B, π, A) if

• The pair (S0 , Φ0 ), obtained from ( B0 , π 0 ), is a positive stabilization of (S, Φ), the one
coming from ( B, π ).

• The system of generators A0 is isotopic to A ∪ { a}, where a is the cocore of H as in

Figure 6.2.

We call a positive stabilization L-elementary if the curve γ as above, see Figure 6.2, inter-
sects the link L, that sits in S and then also in S,
e in at most one point transversely. Then we
prove the following theorem, a generalization of a result of Giroux [39]. More details can
be found in [27].

Theorem 6.1.3 (Giroux) If ( Bi , πi ) are two open book decompositions, compatible with the triple
( L, M, ξ ), then they admit isotopic L-elementary stabilizations. Namely, there is another compatible
open book ( B, π ) which is isotopic to both ( B1 , π1 ) and ( B2 , π2 ), after an appropriate sequence of
L-elementary stabilizations.

The proof of Theorem 6.1.3 requires many steps. We recall that the definitions of contact
cell decomposition and ribbon of a Legendrian graph can be found in Subsection 4.2.1.

Lemma 6.1.4 Every open book compatible with ( L, M, ξ ), after possibly positively stabilizing,
comes from a contact cell decomposition.

Proof. Let S be the page of the open book decomposition and G be the core of S. That
is, G is a graph embedded in S onto which S retracts. From an observation in [27] we
can Legendrian realize G. Therefore, we note that S is the ribbon of G. Let ν(S) be a
neighborhood of S such that ∂S ⊂ ∂ν(S).
We now follow the proof in [27]. Let ai be a collection of properly embedded
 arcs onS
CEU eTD Collection

that cut S into a disk. Let Ai be ai × [0, 1] in M \ S = S × [0, 1] and Ai0 = Ai ∩ M \ ν(S) .

S0
S
C
∂Ai0
∂Ai
Figure 6.3: Abstract picture of the stabilization on S.

We observe that each Ai intersects ∂S on ∂ν(S) exactly twice. Thus, if we extend the Ai ’s

75
 into ν(S) so that their boundaries lie on G then the twisting of S, and hence ξ, along ∂Ai
with respect to Ai is -1,0 or 1. If the value of all the twisting is -1 then we have a contact
cell decomposition, because the contact structure restricted to the complement of S is tight.
Then we just need to see how to reduce the twisting of ξ along ∂Ai .
Suppose ∂Ai has twisting 0. Then positively stabilize S as shown in Figure 6.3. Note
that the curve C can be assumed to be Legendrian and bounds a disk E in M. Now isotope
G across E to get a new Legendrian graph with all the A j ’s unchanged except that Ai is
replace with the disk Ai0 obtained from Ai by isotoping across E. We also add C to G and
add E to the 2-skeleton.
Clearly the twisting of ξ along E is -1. Thus we can reduce the twisting of ξ along ∂Ai
as needed and after sufficiently many positive stabilizations we have an open book that
comes from a contact cell decomposition.

We observe that, in the proof of Lemma 6.1.4, we only use L-elementary stabilizations.
We apply Lemma 6.1.4 to both ( Bi , πi ). Obviously, we can also suppose that the 1-skeleton
of ( M, ξ ) that we have in the beginning is a subgraph of the one after the stabilizations. At
this point, we need to know how to relate two different contact cell decompositions for the
same triple ( L, M, ξ ). To do this we use the following proposition.

Proposition 6.1.5 Given two contact cell decompositions, for the same contact manifold, we have
that they are related by a finite sequence of the following moves:

1. a subdivision of a 2-cell by a Legendrian arc, intersecting the dividing set in one point;

2. add a 1-cell c0 and a 2-cell E so that ∂E = c0 ∪ c; where c is part of the original 1-skeleton
and tb(∂E) = −1;

3. add a 2-cell E whose boundary is already in the 1-skeleton and tb(∂E) = −1.

Proof. The proof follows from standard results in contact topology.

The definition of dividing set of a convex surface is in Subsection 3.1.1. Clearly, we

c0
ΓE
E
c10 c20
c
CEU eTD Collection

L
Figure 6.4: The arc Γ E is the dividing set of the 2-cell E.

have that Move 1 corresponds to an L-elementary stabilization of ( B1 , π1 ), while Move 3

does not change the open book at all. For Move 2 on the other hand we have to reason
more. In fact it can happen that the arc c in the original 1-skeleton intersects the link L
transversely in more than one point. We solve this problem by applying other appropriate
Moves 1 as in Figure 6.4. We observe that we obtain the same contact cell decomposition
if we apply a sequence of Moves 2, using the arcs ci0 , which now will give L-elementary
stabilizations. To complete the proof we only need the following lemma.

76
 Lemma 6.1.6 Suppose ( Bi , πi ) are two open books, compatible with ( L, M, ξ ), which come from
the same contact cell decomposition. Then we have that ( B1 , π1 ) is isotopic to ( B2 , π2 ).

Proof. First we have that the pages π1−1 (1) and π2−1 (1) are isotopic, because it is easy to
see that the ribbon of the 1-skeleton is unique up to isotopy. Then we can suppose that
the bindings coincide and S = πi−1 (1) for i = 1, 2. Now, we have that M \ S fibers along
the interval (0, 1) in two different ways. At this point, since (0, 1) is contractible the two
fibrations result to be trivial and then isotopic. It remains to see what happens when we
glue S back.
The two open books induce monodromies Φ1 and Φ2 on S and they are conjugate
because otherwise, from Corollary 4.2.3, the resulting open books would be compatible
with non contactomorphic contact manifolds; but this is impossible. Then the fact that
the monodromies are in the same conjugacy class tells us precisely that we can extend the
isotopy on S and this completes the proof.

In other words, we may need to stabilize both open book decompositions

  many times
and eventually we obtain other two open book decompositions B, π, Ai , both compati-
e
ble with L ,→ ( M, ξ ); which is contact isotopic to ( Li , M, ξ i ) for i = 1, 2.

6.1.4 Admissible arc slides

In this subsection we use the terminology introduced in Subsection 4.2.1. Take an
adapted system of generators A for an n-component link L, lying inside a surface S.
We define admissible arc slide, a move that change A = { a1 , ..., ai , ..., a j , a2g+l +k−2 } into

S S
aj aj
L
ai
L
ai + a j
Figure 6.5: The arc ai + a j is obtained by sliding a j over ai .

A0 = {..., ai + a j , ..., a j , ...}; where a j is not a distinguished arc and one of the endpoints of
ai and a j are adjacent, like in Figure 6.5. We can prove the following proposition.

Proposition 6.1.7 Let us consider two open book decompositions ( B, π, Ai ), compatible with the
CEU eTD Collection

Legendrian link L ,→ ( M, ξ ). Then, after a finite number of admissible arc slides and isotopies on
Ai , the open books coincide.

We need two preliminary lemmas.

Lemma 6.1.8 An admissible arc slide, from A to A0 , can be inverted. In the sense that we can
perform another admissible arc slide, now from A0 to A00 , such that A00 is isotopic to A.

Proof. If the arc slide changes the arc ai into ai + a j then it is easy to see that we can just
slide ai + a j over an arc a0j , isotopic to a j ; in a way that ai + a j + a0j is isotopic to ai .

Lemma 6.1.9 Suppose that we perform an admissible arc slide that changes ai into ai + a j . Then
we have the following facts:

77
 a) The set { a1 , ..., a2g+l −1 } is a basis of H1 (S, ∂S; F) and i 6 2g + l − 1 if and only if { a1 , ..., ai +
a j , ..., a2g+l −1 } is a basis of H1 (S, ∂S; F),

b) The arc ai is distinguished if and only if ai + a j is a distinguished arc,

c) The arc ai is separating if and only if ai + a j is a separating arc,

d) The arc ai is dead if and only if ai + a j is a dead arc.

Proof. a) Obviously, the arc ai + a j represents the sum of the relative homology classes of
ai and a j . At this point we just need basic linear algebra.

b) This claim is trivial.

c) We take { a1 , ..., a2g+l −1 } as a basis of the first relative homology group of S. We can
suppose that i > 2g + l. This is because if ai is the only separating arc for a component
of S \ { a1 , ..., a2g+l −1 } then ai is homologically trivial; otherwise, we can switch ai with
another separating arc for the same component.
Now, if ai + a j is not a separating arc then S \ { a1 , ..., a2g+l −1 , ai + a j } should remain
connected, but this is impossible because { a1 , ..., a2g+l −1 } is a basis.
The other implication follows easily from Lemma 6.1.8.

d) It follows from b) and c).

The strategy of the proof of Proposition 6.1.7 is, say S1 , ..., Sn and S10 , ..., Sn0 are the com-
ponents of S minus the separating arcs of A1 and A2 respectively, we modify all the sep-
arating arcs, in both A1 and A2 , with admissible arc slides; in order for Si to concide with
Si0 for every i = 1, ..., n. Moreover, we also want that each separating arc in A1 becomes
isotopic to a separating arc in A2 .
At the end, the components of S minus the separating arcs will be n surfaces, each one
containing a component of L, a distinguished arc and the same number of dead arcs. We
a2g+l
S1 a2g+l
S1
L1
a1 a1
L1
Dead arcs
CEU eTD Collection

a1 a1 ai

Figure 6.6: Case 1 is on the left. The figure shows a portion of S \ A1 .

conclude applying n times a result in [57], which proves the claim in the case of Legendrian
knots.

Proof of Proposition 6.1.7. We want to prove that there is an adapted system of generators A
for L in S1 = π −1 (1) such that A is obtained, from A1 and A2 , by a sequence of admissible
arc slides (and isotopies).
We start from the component S1 . We can suppose that ∂S1 contains the separating arc
a2g+l ⊂ A31 , the set of separating arcs in ⊂ A1 , and the distinguished arc a1 ⊂ A11 , the set
of distinguished arcs in A1 , with almost adjacent endpoints. We can have two cases: in

78
 the first one there are no other separating arcs in the portion of ∂S1 on the opposite side
of a2g+l . In the second case they appear; possibly more than one, but we can suppose that
there is exactly one of them. See Figure 6.6.

Case 1. When we consider S10 , which contains the same component of L that is in S1 , after
some arc slides we have that it appears as in Figure 6.6 (left). This is because in the

0
a2g a2g+l
+l
S1
L1

a10

a1 Dead arcs a1

Figure 6.7: There are no other separating arcs, except for a2g+l in S1 .

same figure we see that S1 is split in two pieces by L1 and the innermost one is not
connected in any way to other components of S; in fact there are no separating arcs
on that side. This means that the same holds for S10 too. At this point it is easy to
see that S1 can be modified to be like in Figure 6.7; more explicitly, the separating
arcs are parallel and the distinguished arcs lie in the region where L1 is.

Case 2. As before we have that also S10 appears like in the right part of Figure 6.6 (always
after some arc slides). The reason is the same of previous case. Hence, now we can
modify S1 to be like in Figure 6.8; just in the same way as we did in Case 1.

We have obtained that the separating arcs are fixed on S1 and S10 and then the surfaces
now have isotopic boundaries. Hence, we can move to another component S2 , whose
0
a2g +l

a2g+l
CEU eTD Collection

S1

a10
a1
a1
L1

ai

ai0

Figure 6.8: There are exactly two separating arcs, namely a2g+l and ai in S1 .

79
 boundary contains a separating arc that has not yet been fixed, and we repeat the same
process described before. We may need to slide some separating arcs over the ones that we
already fixed in the previous step, but this is not a problem. We just iterate this procedure
until all the Si coincides with Si0 and this completes the proof.

The results in this section prove the following theorem.

Theorem 6.1.10 If L1 , L2 ,→ ( M, ξ ) are Legendrian isotopic links then the open book decom-
positions ( Bi , πi , Ai ), which are compatible with the corresponding triples, are related by a finite
sequence of global contact isotopies, positive stabilizations and admissible arc slides.

Proof. It follows immediately from Propositions 6.1.1 and 6.1.7, Lemma 6.1.2 and Theorem
6.1.3.

Though they are easy to deal with, we must not forget that, when we define the cor-
responding Legendrian Heegaard diagrams, we need to consider the choices of the mon-
odromy and the families of arcs and basepoints inside their isotopy classes.

6.2 The Legendrian and transverse invariants

In this subsection we identify a cycle in the link Floer chain complexes that we pre-
viously introduced. The corresponding homology class will be our Legendrian invariant.
Let us consider the only intersection point of D = (Σ, α, β, w, z) which lies in the page S1 .
We recall that, in general, the intersection points live in the space Tα ∩ Tβ , but they can be
represented inside Σ. We denote this element with L( D ).

 The intersection
Proposition 6.2.1  point L( D ) is such that ∂− L( D ) = 0 and then L( D ) is a
cycle in cCFL− D, tL( D) ; where tL( D) is the Spinc structure that it induces on − M.

Proof. Every φ ∈ π2 (L( D ), y) such that µ(φ) = 1, where y ∈ Tα ∩ Tβ , has the property
that nz (φ) > 0. The claim follows easily from the definition of the differential.

At the end of the chapter we show that more can be said on the Spinc structure tL( D) .
Now we want to show
 that the isomorphism
 class of the element [L( D )] inside the homol-
ogy group cHFL− − M, L, tL( D) is a Legendrian invariant of the triple ( L, M, ξ ).
Let us be more specific: we consider a Legendrian link L ,→ ( M, ξ ) in a rational ho-
mology contact 3-sphere; we associate two open book decompositions compatible with
CEU eTD Collection

( L, M, ξ ), say ( B1 , π1 , A1 ) and ( B2 , π2 , A2 ), and these determine (up to isotopy) two Legen-

drian Heegaard diagrams, that we call D1 = (Σ1 , α1 , β 1 , w1 , z1 ) and D2 = (Σ2 , α2 , β 2 , w2 , z2 )
respectively. Then we want to find a chain map

Ψ( D1 ,D2 ) : cCFL− ( D1 , t) −→ cCFL− ( D2 , t) , (6.1)

that induces an isomorphism in homology, preserves the bigrading and it is such that

Ψ( D1 ,D2 ) (L( D1 )) = L( D2 ) ,

where t = tL( D1 ) = tL( D2 ) ∈ Spinc (− M ). This is done in order to prove the following
theorem.

80
 Theorem 6.2.2 Let us consider a Legendrian Heegaard diagram D, given by an open book com-
patible with a triple ( L, M, ξ ), where M is a rational homology 3-sphere and L is an n-component
Legendrian link.
Let us take the cycle L( D ) ∈ cCFL− ( D, tL( D) ). Then, the isomorphism class of [L( D )] in
cHFL− − M, L, tξ is a Legendrian invariant of ( L, M, ξ ) and we denote it with L( L, M, ξ ).

Furthermore, the Spinc structure tL( D) coincides with tξ + PD[ L].

Here we denote with tξ the Spinc structure induced on M by ξ.
Proposition 6.2.3 Given a contact 3-manifold (Y, ζ ) we have that ζ induces a Spinc structure tζ
on Y.
Proof. The contact structure ζ gives a cooriented 2-plane field π on Y. Hence, if we take
the isotopy class of the vector field orthogonal to π and oriented coherently with Y then,
according to Turaev definition, we obtain a Spinc structure on Y that we denote with tζ .
 
In Section 5.1 we define another chain complex CFL ∂ for every Spinc structure
d ( D, t), b
t on − M, whose homology is
[ (− M, L, t) .
HFL
The intersection point L( D ) is a cycle also in CFL d ( D, t) and it also determines the Spinc
structure tL( D) . Then the same procedure will show that the homology class [L( D )] in
[ (− M, L, t) is also a Legendrian invariant of L.
HFL


Theorem 6.2.4 The isomorphism class of [L( D )] in HFL [ − M, L, tξ + PD[ L] is a Legendrian
invariant of ( L, M, ξ ) and we denote it with L
b ( L, M, ξ ). Moreover, if L
b ( L, M, ξ ) is non-zero then
L( L, M, ξ ) is also non-zero.
We recall that in Section 3.3 we give the definition of the transverse push-off TL of a
Legendrian link L. In the same way, if T is a transverse link then we can define a Legen-
drian approximation L T of T. The procedure is described in [26]. The Legendrian link L T
is not well-defined up to Legendrian isotopy, but only up to negative stabilizations, see
Section 3.5. Then we can also define transverse invariants from these link Floer homology
groups by taking
T( T, M, ξ ) = L( L T , M, ξ ) and b ( T, M, ξ ) = L
T b ( L T , M, ξ ) ,

where L T is a Legendrian approximation of T.

Theorem 6.2.5 The isomorphism classes T( T, M, ξ ) in cHFL− − M, T, tξ + PD[ T ] , and


CEU eTD Collection

[ − M, T, tξ + PD[ T ] , are transverse link invariants. Furthermore, we have
b ( T, M, ξ ) in HFL
T
that Tb ( T, M, ξ ) has the same non-vanishing property of L.
b

In the case of knots the invariants L and T have been introduced first in [57].

6.3 Proof of the invariance

6.3.1 Global isotopies and admissible arc slides
If two open book decompositions are related by a global isotopy then it easy to see that
the induced abstract open book coincide, up to conjugation of the monodromy and isotopy
of the curves and the basepoints in the diagrams.

81
 Hence, let us consider an abstract open book (S, Φ, A, z, w) and recall that in S we have
two set of arcs A and B, as explained in Subsection 4.2.2. The first check is easy: in fact if we
perturb the basepoints inside the corresponding components of S \ ( A ∪ B) then even the
complex (cCFL− ( D, t), ∂− ) does not change; where D is a Legendrian Heegaard diagram
obtained from (S, Φ, A, z, w). The same is true for an isotopy of S.
Now for what it concerns S we do not have problems, but when we define the chain
complex we also consider the closed surface Σ, obtained by gluing together S and −S. We
still have some choices on −S, in fact by Proposition 6.1.1 we may need to modify the arcs
B in their isotopy classes to achieve admissibility for D = (Σ, α, β, w, z). Then the proof
rests on the following proposition.
Proposition 6.3.1 Suppose that two curves in a Heegaard diagram are related by the move shown
in Figure 5.3. Then we can find a map Ψ( D1 ,D2 ) as in Equation (6.1).
Proof. The map Ψ( D1 ,D2 ) is constructed using a Hamiltonian diffeomorphism of the surface,
as described in Subsection 5.3.2. Since the new disks appear in −S, we have that L( D1 ) is
sent to L( D2 ) which both lie in S.

An arc slide {..., ai , ..., a j , ...} → {..., ai + a j , ..., a j , ...} in S corresponds to a handleslide
{..., αi , ..., α j , ...} → {..., αi0 , ..., α j , ...} in Σ, where αi0 = ai + a j ∪ ai + a j ⊂ Σ. Thus, a chain
map Ψ( D1 ,D2 ) , which induces an isomorphism in homology, is obtained by counting holo-
morphic triangles, as explained in Subsection 5.3.3. The admissibility of the arc slide is
required only to avoid crossing a basepoint in w.
Remember that for every arc slide we actually have two handleslides. In fact we need
to slide both the α and the β curves.
The fact that Ψ( D1 ,D2 ) (L( D1 )) = L( D2 ) follows from [47]; where the arc slides invari-
ance is proved in the open books setting.

6.3.2 Invariance under positive stabilizations

At this point, in order to complete the proof of the invariance of [L( D )], it would be
enough to define Ψ( D,D+ ) , such that Ψ( D,D+ ) (L( D )) = L( D + ), in the case where D and D +
are obtained from an open book and one of its positive stabilizations.
Since we have already proved invariance under admissible arc slides, we can complete
the open books ( B, π ) and ( B+ , π + ) with every possible adapted system of generators and
then eventually define the map Ψ( D,D+ ) .

Proposition 6.3.2 Let us call S = Sg,l = π −1 (1) ⊃ L the page over 1 of ( B, π ). Then we can
CEU eTD Collection

always find A, an adapted system of generators for L in S, with the property that A is disjoint from
the arc γ0 = γ ∩ S; where γ is the curve that we used to perform the L-elementary stabilization.
Proof. We have to study four cases:
a) The arc γ0 intersects L (transversely in one point),
b) The intersection γ0 ∩ L is empty and γ0 does not disconnect S,
c) The intersection γ0 ∩ L is empty, the arc γ0 disconnects S and L is not contained in one
of the two resulting connected components of S,
d) The intersection γ0 ∩ L is empty, the arc γ0 disconnects S and the link L lies in one of the
resulting connected components of S.

82
 Let us start with Case a). We observe that γ0 does not disconnect S; in fact if this was the
case then a component of L would be split inside the two resulting components of S, thus
we would have at least another intersection point between L and γ0 ; which is forbidden.
We define A = { a1 , ..., a2g+l +n−2 } as follows. Take a1 as a push-off of γ0 ; clearly a1 is a
distinguished arc, because it intersects L1 , a component of L, in one point. Now call a0 the
arc given by taking a push-off of L1 and extend it through a1 , on one side of L1 ; this is
the same procedure described in the proof of Theorem 4.2.1. If a0 disconnects S into S1 , S2
and there are components of L that lie in both Si then we take a0 as a separating arc; thus
we extend { a1 , a0 } to an adapted system of generators A by using Theorem 4.2.1. On the
other hand if L is contained in S2 and S1 is empty then we consider { a1 , a00 }; where a00
is another push-off of L1 , this time extended through a1 on the other side of L1 . We still
have problems when a0 does not disconnect S. We can fix this by taking { a1 , a0 , a00 }; which
together disconnect S into two connected components, one of them containing only L1 and
the other one L \ L1 . Again we extend the set { a1 , a0 , a00 } to A applying Theorem 4.2.1.
The other three cases are easier. In Case b) we just denote with an+1 the push-off of γ0
and we complete it to A.
In Case c) we denote with a2g+l our push-off of γ0 and we take it as a separating arc.
Then we can complete it to A.
Finally, Case d) is as follows. Since in this case the push-off is trivial in homology, and
it bounds a surface disjoint from L, we can actually ignore it and easily find a set A which
never intersects γ0 .

Now we have the abstract open book (S, Φ, A, z, w), obtained from ( B, π, A). Denote
with (S+ , Φ+ , A+ , z, w) the one coming from ( B+ , π + , A+ ), where S+ = (π + )−1 (1) is S
with a 1-handle attached; Φ+ = Φ0 ◦ Dγ where Φ0 coincides with Φ on S, extended with
the Identity on the new 1-handle; and A+ = A ∪ { a} with a being the cocore of the new
1-handle (see Figure 6.2). Then we call D and D + the corresponding Legendrian Heegaard
diagrams.
We define Ψ( D,D+ ) in the following way. For every x ∈ Tα ∩ Tβ |tL(D) it is

Ψ( D,D+ ) ( x ) = x 0 ;

where x 0 = x ∪ { a ∩ b} with b being the arc in strip position with a, as in Figure 4.8.
It results that Ψ( D,D+ ) is a chain map because the curve α = b ∪ b only intersects the
curve β = a ∪ a, and moreover it is α ∩ β = {1 pt}, since we choose A+ in a way that every
arc in it is disjoint from γ0 . We have that Ψ( D,D+ ) induces an isomorphism in homology,
CEU eTD Collection

because it is an isomorphism also on the level of chain complexes, and sends L( D ) into
L( D + ).

6.3.3 Invariance theorems

We prove our main results in this chapter, which shows the invariance of L and T. The
claim about the Spinc structure tL( D) is studied in Section 6.4.

Proof of Theorem 6.2.2. We proved that L( D ) is a cycle in Proposition 6.2.1. The invariance
follows from the results obtained in this section: in fact, we proved that if D1 and D2 are
Legendrian Heegaard diagrams, representing Legendrian isotopic links, then we have a
chain map Ψ( D1 ,D2 ) that preserves the bigrading and the Spinc structure and sends L( D1 )
to L( D2 ).

83
  
We note that the invariant can be a U-torsion class in the group cHFL− − M, L, tL( D) ,
which means that there is a k > 0 such that U k · L( L, M, ξ ) = [0]. Moreover, the cycle
L( D ) possesses a bigrading (d, s); such bigrading is induced on the invariant L( L, M, ξ ),
because all the maps Ψ defined in this section preserves both the Maslov and the Alexander
grading.

Proof of Theorem 6.2.4. The proof of this theorem is the same as the one of Theorems
6.2.2 and 6.2.5, except for the non-vanishing property, which follows from the fact that
d ( D, tL( D) ) is a quotient of cCFL− ( D, tL( D) ).
CFL

6.4 A remark on the Ozsváth-Szabó contact invariant

Now we spend a few words about the Ozsváth-Szabó contact invariant c(ξ ), intro-
duced in [70].
Given a Legendrian Heegaard diagram D. Let us call c( D ) the onlyintersection point 
c ( D, tc( D) ), b
which lies on the page S1 as before, but now considered as an element in CF ∂z .
The proof of Proposition 6.2.1 says that c( D ) is also a cycle. Moreover, we have the follow-
ing theorem.

Theorem 6.4.1 (Ozsváth and Szabó) Let us consider a Legendrian Heegaard diagram D with
a single basepoint z, given by an open book compatible with a pair ( M, ξ ), where M is a rational
homology 3-sphere. Let us take the cycle c( D ) ∈ CF
c ( D, tc( D) ).
 
Then the isomorphism class of [c( D )] ∈ d
HF − M, tc( D) is a contact invariant of ( M, ξ ) and
we denote it with cb( M, ξ ). Furthermore, we have the following properties:

• the Spinc structure tc( D) coincides with tξ ;

• the Maslov grading of c( D ) is given by Mz (c( D )) = −d3 ( M, ξ ).

The proof of this theorem comes from [70], where Ozsváth and Szabó first define the
invariant cb( M, ξ ), and [47], where Honda, Kazez and Matić give the reformulation using
open book decompositions that we use in this thesis.
Furthermore, now we can see which is the Spinc structure induced by L( D ).

Corollary 6.4.2 The intersection point L( D ) induces the Spinc structure tξ + PD[ L].
CEU eTD Collection

This completes the proof of Theorem 6.2.2.

Proof of Corollary 6.4.2. We recall that

sw (L( D )) − sz (L( D )) = PD[ L]

from Proposition 4.1.3. Then we have that

tL( D) = sw (L( D )) = sz (L( D )) + PD[ L] = tc( D) + PD[ L] = tξ + PD[ L] ,

from Theorem 6.4.1.

84
 We show that the argument used for the invariance of L also proves a generalization of
Theorem 6.4.1, which involves Legendrian Heegaard diagrams with more than  one base- 
c ( D ), b
point. This is done considering the fact that we can define the chain complex CF ∂z
starting from multi-pointed Heegaard diagrams; as we saw in Theorem 5.2.1.
Theorem 6.4.3 Let us consider a Legendrian Heegaard diagram D, given by an open book com-
patible with a pair ( M, ξ ), where M is a rational homology 3-sphere. Let us take the cycle
c( D ) ∈ CF
c ( D, tc( D) ).
    N ( n −1)
Then the isomorphism class of [c( D )] ∈ HF − M, tc( D) ⊗ F(−1) ⊕ F(0)
d is a con-
tact invariant of ( M, ξ ) and we denote it with cb( M, ξ; n).
Proof. As in the proof of Theorem 6.1.10 we claim that two open book decompositions,
compatible with contactomorphic manifolds ( M1 , ξ 1 ) and ( M2 , ξ 2 ), are related by a finite
sequence of isotopies, arc slides and positive stabilizations.
First we observe that if ( B1 , π1 ,
A1 ) is the open book decomposition compatbile with
( M1 , ξ 1 ) then F ( M1 ), F (ξ 2 ), F ( A1 ) is compatible with ( M2 , ξ 2 ), where F is the contacto-
morphism between the two manifolds.
At this point we suppose that ( Bi , πi ) for i = 1, 2 are compatible with a contact manifold
( M, ξ ). We argue that we can postively stabilize both many times until the two open book
decompositions are isotopic to ( B, π ), which is also compatible with ( M, ξ ). This is done
in the same way as in Subsection 6.1.3, using a weaker version of Giroux’s Theorem 6.1.3.
We fix ( B, π ) compatible with ( M, ξ ). It only remains to show that, say A1 and A2 are
two system of generators as in the definition of abstract open book in Subsection 4.2.2,
there is a sequence of arc slides which changes A1 into A2 . This follows by adapting the
proof of Proposition 6.1.7.
At this point, we consider the Legendrian Heegaard diagrams D1 and D2 , coming from
open book decompositions compatible with contactomorphic manifolds. We define the
chain maps Ψ( D1 ,D2 ) exactly as in the proof of Theorem 6.2.2; in fact, there are even less
restrictions than before: the arc slides do not need to be admissible, because we do not
consider the basepoints in w; moreover, the positive stabilizations do not need to be L-
elementary, since there is no link on the page of ( B, π ). Then we conclude in the same way
as in Theorem 6.2.2.

We observe that cb( M, ξ; n) is not a new invariant; in fact, we immediately have the
following corollary.
CEU eTD Collection

Corollary 6.4.4 Suppose that ( M, ξ ) is a rational homology contact 3-sphere. Then we have that
  N ( n −1)
cb( M, ξ; n) = cb( M, ξ ) ⊗ F(−1) ⊕ F(0) .

Furthermore, if D is a Legendrian Heegaard diagram for ( M, ξ ) then we have the following prop-
erties:
• the Spinc structure induced by the cycle c( D ), representing cb( M, ξ; n), coincides with tξ ;
• the Maslov grading of c( D ) is given by Mz (c( D )) = −d3 ( M, ξ ) + 1 − n.
Proof. From Theorem 6.4.3 we find a chain map
   
Ψ( D,D0 ) : CF
c ( D ), t −→ CF
c ( D0 ), t

85
 D0

Σ
zn βn
β1
z1 β n −1
α1 α n −1 αn

Figure 6.9: The Legendrian Heegaard diagram D0 is obtained by stabilizing n − 1 times

D 0 , a diagram compatible with ( M, ξ ).

which induces an isomorphism in homology, preserves the grading and sends c( D ) into
c( D0 ), where D0 is the Legendrian Heegaard diagram in Figure 6.9.
The diagram D0 is also compatible with ( M, ξ ) because is gotten by stabilizing D 0 ,
which is a single-pointed Legendrian Heegaard diagram taken from an open book decom-
position compatible with ( M, ξ ).
Therefore, using Theorem 5.2.3, and the argument in its proof, we obtain that
  N ( n −1)
0
c( D0 ) = c( D ) ⊗ F(−1) ⊕ F(0) .

Then the first claim follows from Theorem 6.4.1.

The fact that t coincides with tξ is a direct consequence of Theorem 6.4.1, since sta-
bilizing does not change the Spinc structure. Finally, the cycle c( D0 ) has Maslov grading
c( D 0 ) + 1 − n because, at each stabilization, we add one coordinate in c( D 0 ) which decrease
its Maslov grading by one, as we saw in the proof of Lemma 5.2.2. This means that we can
conclude again by using Theorem 6.4.1.

6.5 Properties of L and T

We recall that in Subsection 5.4 we introduce the surjective F-linear map
CEU eTD Collection

U =1
F : cCFL− ( D, tξ + PD[ L]) −−→ CF
c ( D z , tξ )

which induces the map F∗ in homology:

( n −1)
N
− U =1
 
HF − M, tξ ⊗ F(−1) ⊕ F(0)


F∗ : cHFL − M, L, tξ + PD[ L] −−→ d ,

where D is a Legendrian Heegaard diagram coming from an open book decomposition

compatible with ( M, ξ ).
Then we can prove the following lemma.
( n −1) ,
N
Lemma 6.5.1 The map F as above is such that F∗ (L( L, M, ξ )) = cb( M, ξ ) ⊗ (e−1 ) where
e−1 is the generator of F(−1) .

86
 Proof. Since L( D ) and c( D ) both correspond to the unique intersection point on the page
of the open book decomposition, we immediately see that F (L( D )) = c( D ). Then the
claim follows from Theorems 6.2.2 and 6.4.3.

Lemma 6.5.1 implies the following important result.

Corollary 6.5.2 The Legendrian invariant L( L, M, ξ ) is non-torsion if and only if the contact
invariant cb( M, ξ ) is non-zero.
Furthermore, if D is a Legendrian Heegaard diagram compatible with ( L, M, ξ ) then the Maslov
and Alexander grading of the element L( D ) are related by the following equality:

M(L( D )) = −d3 ( M, ξ ) + 2A(L( D )) + 1 − n ;

where n is the number of component of L.

Proof. It follows from Lemma 5.4.1 and Theorem 6.4.3.

In particular, this corollary says that the Legendrian link invariant L is always a torsion
class if ξ is overtwisted and always non-torsion if ( M, ξ ) is symplectically fillable. In fact,
from [70] we know that cb( M, ξ ) is zero in the first case and non-zero in the second one.
We proved that the isomorphism class L( L, M, ξ ) is a Legendrian invariant. This is true
also for the Alexander (and Maslov) grading of the element L( D ) ∈ cCFL− ( D, tξ ), that we
denote with A(L( D )).
From Corollary 6.5.2 we know that cb( M, ξ ) 6= [0] implies that L( L, M, ξ ) is non-torsion,
hence it determines A(L( D )). On the other hand, if cb( M, ξ ) is zero then a priori the grad-
ings of the element L( D ) could give more information.
Starting from these observations, we want to express the value of A(L( D )) in terms of
the other known invariants of the Legendrian link L. Note that Corollary 6.5.2 also tells us
that the Maslov grading of L( D ) is determined, once we know A(L( D )).
First we recall that, from the definitions of Thurston-Bennequin and rotation number
in Chapter 3, we have
n
tb( L) = ∑ tbi ( L) , where tbi ( L) = tb( Li ) + lkQ ( Li , L \ Li )
i =1

and
n
rot( L) = ∑ roti ( L) , where roti ( L) = rot( Li ) .
CEU eTD Collection

i =1

Theorem 6.5.3 Consider L ,→ ( M, ξ ) an n-component Legendrian link in a rational homology

contact 3-sphere and D a Legendrian Heegaard diagram, that comes from an open book decomposi-
tion compatible with ( L, M, ξ ). Then we have that

tb( L) − rot( L) + n
A(L( D )) = and M(L( D )) = −d3 ( M, ξ ) + tb( L) − rot( L) + 1 .
2
Proof. If L is a null-homologous knot then the claim has been proved by Ozsváth and Stip-
sicz in [66]; moreover, the proof can be generalized to work with every knot in a rational
homology 3-sphere.
At this point, in order to obtain the claim for links, we need to relate A(L( D )) with the
Alexander grading of the Legendrian invariants of the components Li of L.

87
 A Legendrian Heegaard diagram Di for the knot Li is easily gotten from D by removing
some curves and basepoints. We denote the intersection point representing the Legendrian
invariant of Li with L( Di ). Then we have
n  
1
A(L( D )) = ∑ A (L( Di )) + lkQ ( Li , L \ Li ) =
i =1
2
n n
tb( Li ) − rot( Li ) + 1 lkQ ( Li , L \ Li ) tb ( L) − roti ( L) + 1
 
=∑ + =∑ i =
i =1
2 2 i =1
2
tb( L) − rot( L) + n
=
2
and the claim follows.

This property holds also for the invariant T and we state it in the following corollary.

Corollary 6.5.4 Suppose T ,→ ( M, ξ ) is an n-component transverse link in a rational homology

contact 3-sphere with cb( M, ξ ) 6= [0]. Then we have that the bigrading of T is

sl( T ) + n
A(T( T, M, ξ )) = and M(T( T, M, ξ )) = −d3 ( M, ξ ) + sl( T ) + 1 .
2
Proof. It follows from Theorems 6.2.5 and 6.5.3, Proposition 3.3.2 and Corollary 6.5.2.

6.6 Connected sums and invariance of T

6.6.1 Behaviour of the invariants under connected sum
Let us consider two adapted open book decompositions ( Bi , πi , Ai ), compatible with
the triples ( Li , Mi , ξ i ). We can define a third open book ( B, π, A), for the manifold M1 # M2 ,
with the property that π −1 (1) is a Murasugi sum , see [27], of the pages π1−1 (1) and π2−1 (1),
which is defined as follows.
Given two abstract open books (Si , Φi , Ai , zi , wi ) for i = 1, 2, let ci be an arc properly
embedded in Si and Ri a rectangular neighborhood of ci such that Ri = ci × [−1, 1]. Then
the Murasugi sum of (S1 , Φ1 ) and (S2 , Φ2 ) is the pair (S1 , Φ1 ) ∗ (S2 , Φ2 ) with page S1 ∗ S2 =
S1 ∪ R1 = R2 S2 , where R1 and R2 are identified so that ci × {−1, 1} = (∂ci+1 ) × [−1, 1], and
the monodromy is Φ1 ◦ Φ2 . Since we want to define an abstract open book, we also need
CEU eTD Collection

to say what happens to the system of generators and the basepoints.

We may assume that suitable portions of the open books and the adapted system of
generators appear as in the left in Figure 6.10. We take the Murasugi sum of the two open
books in the domains containing the basepoints z1 and w1 , which correspond to the compo-
nents of L1 and L2 that we want to merge, and then we drop these two basepoints. We can
make sure that the resulting oriented link L1 # L2 is smoothly determined on a page of the
resulting open book. The arcs can also be arranged to be the same as the ones illustrated in
Figure 6.10. The resulting open book is compatible with the triple ( L1 # L2 , M1 # M2 , ξ 1 # ξ 2 ).

Denote with D1 ,D2 and D the Legendrian Heegaard diagrams obtained from the open
book decompositions that we introduced before. Then, we have the following theorem.

88
 b10
S1
b1
w1
a1 a10

b2 z1 b20
a20
S2

a2 S1 ∗ S2

Figure 6.10: On the left we see the starting abstract open books, while on the right there is
the Murasugi sum of the two.

Theorem 6.6.1 For every Spinc structure on M1 and M2 there is a chain map

cCFL− ( D, t1 # t2 ) −→ cCFL− ( D1 , t1 ) ⊗F[U ] cCFL− ( D2 , t2 )

that preserves the bigrading and the element L( D ) is sent into L( D1 ) ⊗ L( D2 ).

Furthermore, this map induces an isomorphism in homology, which means that

cHFL− − M1 # − M2 , L1 # L2 , tξ 1 # ξ 2 + PD[ L1 ] + PD[ L2 ] ∼

=
− −



cHFL − M1 , L1 , tξ 1 + PD[ L1 ] ⊗F[U ] cHFL − M2 , L2 , tξ 2 + PD[ L2 ]

as bigraded F[U ]-modules and

L( L1 # L2 , M1 # M2 , ξ 1 # ξ 2 ) = L( L1 , M1 , ξ 1 ) ⊗ L( L2 , M2 , ξ 2 ) .

Proof. We use the same strategy of the case of knots in [57]. There are three Heegaard

b1

a20
a10
CEU eTD Collection

a1

b2
a2

S1 ∗ S2

Figure 6.11: Intermediate situation between the slidings.

diagrams coming into play: the connected sum diagram, whose α and β-circles are gotten
by doubling the initial bases, an intermediate diagram, shown in Figure 6.11, gotten by
sliding the β’s as dictated by the arcslides, whose attaching circles we denote β0 and α,

89
 and the final one gotten by performing handleslides on the α-circles, as dictated by the arc
slides of the bi as in the right-most diagram in Figure 6.10, whose attaching cicles are β0
and α0 .
From [72] we know that there is a chain map
Φ : cCFL− ( D1 ) ⊗ cCFL− ( D2 ) −→ cCFL− ( D )
which induces an isomorphism in homology and preserves the bigrading and the Spinc
structure. In all three of these diagrams there is a unique intersection point in Tα ∩ Tβ
which is supported in S1 ∗ S2 : the first is x = Φ(L( D1 ) ⊗ L( D2 )), the second one is denoted
with x 0 and the third one, that we call x 00 , clearly determines the Legendrian invariant for
L1 # L2 and then coincides with L( D ).
We claim that the three generators are mapped to one another under the maps induced
by handleslides. We argue first that x 0 is a cycle. We say that, for any y ∈ Tα0 ∩ Tβ , if
φ ∈ π2 ( x 0 , y) has all non-negative local multiplicities, then x 0 = y and and φ is the constant
disk. We need to be careful to the fact that the arcs disconnect S1 ∗ S2 into two regions,
only one of which contains z1 . However, it is still easy to see from Figure 6.11 that any
positive domain flowing out of x 0 , which has positive multiplicity on the other region not
containing z1 , has also to have positive multiplicity at z1 .
We now want to count holomorphic triangles. Consider the Heegaard triple
(Σ, α, β, β0 , w, z), and the diagram (Σ, β, β0 , z) which represents an unknot in the g-fold
connected sum of S2 × S1 . Moreover, in that diagram there is a unique intersection point
Θ ∈ Tβ ∩ Tβ0 , representing the top-dimensional Floer homology class, see also Subsection
5.3.3.
The handleslide map
Ψ1 : cCFL− (Σ, α, β, w, z) −→ cCFL− (Σ, α, β0 , w, z)
is defined by
Ψ1 ( x ) = ∑ ∑ c ψ ) · U nw (ψ) y .
M(
y ψ∈π2 ( x,Θ,y)
µ(ψ)=1,nz (ψ)=0

The claim that Ψ1 ( x ) = x 0 follows from the following two facts:

• there is a triangle ψ0 ∈ π2 ( x, Θ, x 0 ), gotten as a disjoint union of the obvious small
triangles in S1 ∗ S2 ;
• every other triangle ψ ∈ π2 ( x, Θ, y) with nz (ψ) = 0 has a negative local multiplicity
CEU eTD Collection

somewhere.
The first is obtained easily from Figure 6.11. To see the second, we notice that any ho-
mology class ψ ∈ π2 ( x 0 , Θ, y) can be decomposed as ψ0 ∗ φ, where φ ∈ π2 ( x 0 , y). More-
over, if ψ has only positive local multiplicities, then the same follows for φ; also, since
nz (ψ) = 0 = nz (ψ0 ), it follows that nz (φ) = 0 as well. But by our above argument that x 0
is a cycle, it now follows that φ is constant.
Consider next the handleslide map
Ψ2 : (Σ, α, β0 , w, z) −→ (Σ, α0 , β0 , w, z) .
This is defined by pairing with an intersection point Θ0 ∈ Tα0 ∩ Tα , representing the top-
dimensional non-trivial homology. The argument that Ψ2 ( x 0 ) = x 00 follows through a
similar argument.

90
 Finally, we have that the composition Φ = Ψ2 ◦ Ψ1 ◦ Φ now induces an isomorphism
on homology, carrying L( D1 ) ⊗ L( D2 ) into x 00 = L( D ), which completes the proof of the
theorem.

We note that the link Floer homology group of the connected sum does not depend on
the choice of the components. In particular, this means that we can compute the link Floer
homology group and the Legendrian invariant of a disjoint union L1 t L2 , see Section 3.5.
Proposition 6.6.2 If we denote with 2 a smooth 2-component unlink and with O2 the Legen-
drian 2-component unlink in (S3 , ξ st ) such that tb(O2 ) = −2 then we have that

cHFL− − M1 # − M2 , L1 t L2 , tξ 1 # ξ 2 + PD[ L1 ] + PD[ L2 ] ∼

=
− −
cHFL − M1 , L1 , tξ 1 + PD[ L1 ] ⊗F[U ] cHFL − M2 , L2 , tξ 2 + PD[ L2 ] ⊗F[U ] cHFL− ( 2 )

and
L( L1 t L2 , M, ξ ) = L( L1 , M1 , ξ 1 ) ⊗ L( L2 , M2 , ξ 2 ) ⊗ L(O2 ) .
Proof. We just apply Theorem 6.6.1 twice, each time on one of the two components of
O2 .
The homology group cHFL− ( 2 ) is isomorphic, as bigraded F[U ]-module, to
F[U ](−1,0) ⊕ F[U ](0,0) and this can be proved immediately from the diagram in Figure 6.1.
Moreover, Theorem 6.5.3 tells us that

L(O2 ) = e−1,0 ,

that is the generator of F[U ] with bigrading (−1, 0). This means that, if cb( M, ξ ) is non-zero,
we have

M(L( L1 t L2 , M, ξ )) = M (L( L1 , M1 , ξ 1 )) + M(L( L2 , M2 , ξ 2 )) − 1

and
A(L( L1 t L2 , M, ξ )) = A(L( L1 , M1 , ξ 1 )) + A(L( L2 , M2 , ξ 2 )) .
It is easy to check that similar results hold also for the Legendrian invariant L.
b

Corollary 6.6.3 We have that

L b ( L1 , M1 , ξ 1 ) ⊗ L
b ( L1 # L2 , M1 # M2 , ξ 1 # ξ 2 ) = L b ( L2 , M2 , ξ 2 )

and
CEU eTD Collection

b ( L1 t L2 , M, ξ ) = L
L b ( L1 , M1 , ξ 1 ) ⊗ L
b ( L2 , M2 , ξ 2 ) ⊗ L
b (O2 ) .

We can use the connected sum formula in Theorem 6.6.1 to determine how L changes
under stabilizations of Legendrian links, see Section 3.5.
Proposition 6.6.4 For every L ,→ ( M, ξ ) as before we have that

L( L+ , M, ξ ) = U · L( L, M, ξ ) and L( L− , M, ξ ) = L( L, M, ξ )

in cHFL−


− M, L, tξ + PD[ L] . Furthermore, it is
b ( L+ , M, ξ ) = [0]
L and b ( L− , M, ξ ) = L
L b ( L− , M, ξ )

[ (− M, L, tξ + PD[ L]).
in HFL

91
 Proof. It follows from Theorem 6.6.1, which says that we just need to determine L(O ± ),
together with the fact that cHFK − ( ) ∼ = F[U ](0,0) , which says that L(O ± ) is fixed by
the classical invariants of O ± , and Equations (3.1) and (3.2) that give tb(O ± ) = −2 and
rot(O ± ) = ±1.
The claim about L b is done in the same way using Corollary 6.6.3, but now we consider
[( ) ∼
that HFL = F(0,0) .
For the transverse case, we recall the following theorem from [30].

Theorem 6.6.5 (Epstein) Two transverse links in a contact manifold are transversely isotopic if
and only if they admit Legendrian approximations which differ by negative stabilizations.

Hence, we can now prove that T is a transverse invariant of links.

Proof of Theorem 6.2.5. It follows immediately from Proposition 6.6.4 and Theorem 6.6.5.

Moreover, we have that the transverse invariant satisfies the following properties.

Corollary 6.6.6 Take two transverse links T1 and T2 in contact manifolds ( Mi , ξ i ) for i = 1, 2 as
before. Then we have that the invariant of a connected sum is given by

T( T1 # T2 , M1 # M2 , ξ 1 # ξ 2 ) = T( T1 , M1 , ξ 1 ) ⊗ T( T2 , M2 , ξ 2 ) ,

while for the disjoint union we have that

T( T1 t T2 , M, ξ ) = T( T1 , M1 , ξ 1 ) ⊗ T( T2 , M2 , ξ 2 ) ⊗ T( TO2 ) ,

where TO2 denotes the transverse push-off of the Legendrian unknot O2 .

Proof. It follows immediately from the definition of T, Theorem 6.6.1 and Proposition 6.6.2.

6.6.2 Naturality and a refinement for L

The invariant L b ( L, M, ξ ) can be refined using a naturality property of the link Floer
homology group of a connected sum. Suppose that L is a Legendrian link in a contact 3-
sphere (S3 , ξ ) such that Lb ( L, S3 , ξ ) 6= [0]. Let S be a convex, splitting sphere with connected
CEU eTD Collection

dividing set, which intersects L transversely in exactly two points. Such a splitting sphere
expresses L as a connected sum of two links L1 and L2 .
Since L = L1 #S L2 then its hat Heegaard Floer homoloy group admits the splitting

[ ( L∗ ) ∼
HFL [ ( L1∗ ) ⊗F HFL
= HFL [ ( L2∗ ) ,

where the mirror images appear because S3 has a diffeomorphism that reverses the orien-
[ (−S3 , L) with HFL
tation, then we identify HFL [ (S3 , L∗ ) := HFL[ ( L ∗ ).
The Alexander grading of L 3
b ( L, S , ξ ) is well-defined, because we suppose that the in-
variant is non-zero. Moreover, we have that
     
A L b ( L, S3 , ξ ) = A L b ( L1 , S3 , ξ 1 ) + A Lb ( L2 , S3 , ξ 2 ) .

92
 The pair (s1 , s2 ), where  
si = A Lb ( L i , S3 , ξ i )

for i = 1, 2, is called Alexander pair of Lb ( L, S3 , ξ ) respect to S and we denote it with

3
AS ( L, S , ξ ). We have that the Alexander pair is an invariant of L in the sense of the fol-
lowing theorem.
Theorem 6.6.7 Suppose that L is a Legendrian link in (S3 , ξ ) such that L b ( L, S3 , ξ ) is non-zero.
We also assume that there are two convex, splitting spheres S1 and S2 , which decompose L as
Legendrian connected sums, such that we can find a smooth isotopy of M that fix L and sends S1
into S2 .
Then the two Alexander pairs of Lb ( L, S3 , ξ ), respect to S1 and S2 , coincide, which means that
3 3
AS1 ( L, S , ξ ) = AS2 ( L, S , ξ ).
In particular, say L = L1 #S1 L10 = L2 #S2 L20 , the hypothesis tells us that L1 and L10 are
smoothly isotopic to L2 and L20 respectively.

Proof of Theorem 6.6.7. Suppose that D1 and D2 are Legendrian Heegaard diagrams for L1
and L2 , while D10 and D20 are diagrams for L10 and L20 respectively. Moreover, denote with
D1 # D10 and D2 # D20 the Legendrian Heegaard diagrams for L obtained from the corre-
sponding Murasugi sums, as in Subsection 6.6.1.
In [57] it is proved that, given the map Φ in the proof of Theorem 6.6.1, we can find a
map Ψ such that the following diagram commutes.

∗
Φ( D1 ,D10 )
[ ( D10 )
[ ( D1 ) ⊗ HFL
HFL [ ( D1 # D10 )
HFL

Ψ∗( D1 ,D2 ) ⊗ Ψ∗( D0 ,D0 ) Ψ

1 2

∗
Φ( D2 ,D20 )
[ ( D20 )
[ ( D2 ) ⊗ HFL
HFL [ ( D2 # D20 )
HFL

Moreover, it is shown that the map Ψ is an isomorphism which preserves the Alexander
pair. Then this implies that, if the splitting spheres satisfiy the hypothesis of the theorem,
CEU eTD Collection

the homology group HFL [ ( L∗ ) is canonically identified with HFL[ ( L1∗ ) ⊗ HFL
[ ( L2∗ ) and the
Legendrian invariant is given by

b ( L, S3 , ξ ) = L
L b ( L1 , S3 , ξ ) ⊗ L [ ( L∗ ) .
b ( L2 , S3 , ξ ) ∈ HFL

This means precisely that the Alexander pair is well-defined and depends only on the
isotopy class of S rel L.

We observe that this theorem is false if we consider the invariant L instead. In fact, say
K is the standard Legendrian unknot O after one positive stabilization, while denote with
H the Legendrian knot O after two positive stabilizations. We immediately see that K# K
is Legendrian isotopic to O # H, but the corresponding Alexander pairs are different
There is a version of Theorem 6.6.7 for transverse links.

93
 Corollary 6.6.8 Suppose that T is a transverse link in (S3 , ξ ). Assume also that one of its Legen-
drian approximations L T respects the hypothesis of Theorem 6.6.7. Then it is

AS1 ( T, S3 , ξ ) = AS2 ( T, S3 , ξ ) ,

where the Alexander pair is now defined as AS ( T, S3 , ξ ) = AS ( L T , S3 , ξ ).

Proof. It is a consequence of the fact that Legendrian approximations of the same trans-
verse link differ by negative stabilizations. Therefore, the Alexander gradings are the same
because negative stabilizations do not change the invariant, see Proposition 6.6.4.

The Alexander pair can be useful in distinguishing Legendrian and transverse links
that are not isotopic.

Proposition 6.6.9 Suppose that L1 and L2 are smoothly isotopic Legendrian (transverse) links in
(S3 , ξ ) which appear as in one of these two cases:
1. say L1 ≈ L2 = K#S H, where K and H are Legendrian (transverse) knots with prime knot
types;

2. say L1 ≈ L2 is a 2-component Legendrian (transverse) link, obtained from three Legendrian

(transverse) knots K, H and J with prime knot types, defined as follows: take the connected
sum of K#S H with a (standard) positive Hopf link H+ and J, in the way that K#S H is
summed on the first component of H+ and J on the second one.

We have that if AS ( L1 , S3 , ξ ) 6= AS ( L2 , S3 , ξ ) then L1 is not Legendrian (transverse) isotopic to

L2 .

Proof. It follows from Theorem 6.6.7 and Corollary 6.6.8 and the fact that, if there is a
Legendrian (transverse) isotopy F between L1 and L2 , the isotopy F is such that F (S) = S0
and, in both cases, we can smoothly isotope S0 onto S.
CEU eTD Collection

94
 Chapter 7

Loose and non-loose links

7.1 Legendrian links in overtwisted structures

Legendrian links in overtwisted contact 3-manifolds come in two types: loose and non-
loose. A Legendrian link is loose if also its complement is overtwisted, while it is non-loose
if the complement is tight. More explicitely, a Legendrian link is loose if and only if we can
find an overtwisted disk that is disjoint from the link.
We observe that if L1 is loose then L1 # L2 is also loose. This can be stated in an equiv-
alent way by saying that if a Legendrian link L1 # L2 is non-loose then both L1 and L2 are
non-loose or live in a tight contact manifold.
On the other hand, the connected sum of two non-loose Legendrian links can be loose.
In fact, denote with ξ i the overtwisted structure on S3 with Hopf invariant equals to i ∈ Z,
where ξ 0 is homotopic as a plane field to the unique tight structure on S3 , see Subsection
3.1.1. From [24] we have that in (S3 , ξ −1 ) there is a non-loose Legendrian unknot K. Then
the knot K# K is a loose Legendrian unknot in (S3 , ξ −2 ), because in [24] it is also proved
that non-loose unknots in S3 only appear in the structure ξ −1 .
Since cb( M, ξ ) is always zero for overtwisted contact manifold [70], we have that the
Legendrian link invariant L is always torsion in this case. But more can be said for loose
links.

Proposition 7.1.1 Let L be a loose Legendrian link in an overtwisted contact 3-manifold ( M, ξ ).

Then we have that L( L, M, ξ ) = [0].

Proof. The complement of L in M contains an overtwisted disk E. Since E is contractible,

CEU eTD Collection

we can find a ball U such that E ⊂ U ⊂ M \ L. The restriction of ξ to U is obviously

overtwisted, moreover we can choose E such that ∂U has trivial dividing set. Thus we have
that ( M, ξ ) = ( M, ξ 1 )# (S3 , ξ 0 ); where ξ 1 concides with ξ near L and ξ 0 is an overtwisted
structure on S3 .
We now use the fact that the standard Legendrian unknot O is well-defined, up to
Legendrian isotopy, and Theorem 6.6.1 to say that

L( L, M, ξ ) = L( L1 , M, ξ 1 ) ⊗ L(O , S3 , ξ 0 ) .

Since cHFK − ( ) ∼= F[U ](0,0) , and then there is no torsion, we obtain that the Legendrian
invariant of an unknot is zero in overtwisted 3-spheres. Then L( L, M, ξ ) is also zero.

95
 We can use this proposition to say something about stabilizations. In fact, in principle
a stabilization of a non-loose Legendrian link L ,→ ( M, ξ ) can be loose, but if L( L, M, ξ ) is
non zero then all its negative stabilizations are also non-loose.
It is easy to prove that we can always find loose Legendrian links in every overtwisted
contact 3-manifold; in fact Legendrian links inside a Darboux ball need to be loose, see
Proposition 3.5.1, and we have Darboux balls centered at every point. On the other hand,
it is a little harder to show that the same holds for non-loose Legendrian links.

Proposition 7.1.2 In every ( M, ξ ) with ξ overtwisted there exists a non-loose, non-split n-

component Legendrian link such that all the components are also non-loose.

Proof. It is a known theorem in contact topology that every overtwisted contact 3-manifold
is obtained from some −1-surgeries and exactly one +1-surgery on Legendrian knots in
(S3 , ξ st ), see [27].
Take a surgery presentation for ( M, ξ ) and denote by J = J + ∪ J − the corresponding
framed Legendrian link in (S3 , ξ st ); which can be represented by a front projection. The
component J + is the knot where we perform the +1-surgery. We claim that the Legendrian

L2
Ln
L1

+1
J+
Figure 7.1: The knots L1 , L2 , ..., Ln are the components of L.

link L, given by n parallel contact push-offs of J + as shown in Figure 7.1, is what we

wanted. Note that the contact push-off of a Legendrian knot K is a knot K 0 , which is
parallel to K except for a small tangle where it twists along it; the number of twists is equal
to tb(K ).
CEU eTD Collection

We start by proving that each Li is non-loose; this implies that also L is non-loose. Sup-
pose that there is an overtwisted disk E in M \ Li . Thus E will remain unchanged if we ap-
ply a −1-surgery along Li and then the resulting contact manifold ( Mi , ξ i ) is overtwisted.
But it is easy to see that two surgeries, with opposite signs, along contact push-offs cancel
each other and this means that the new manifold has a surgery presentation given by only
−1-surgeries. Now from Theorem 3.1.9 we obtain that ( Mi , ξ i ) is Stein fillable and then
from [70] we know that it has to be tight. This is a contradiction.
We only have to show that L is non-split. If not we can find a connected sum decom-
position ( M, ξ ) = ( M1 , ξ 1 )# ( M2 , ξ 2 ) and say L1 is embedded in M1 and L2 in M2 . Since
we can suppose that ( M1 , ξ 1 ) is overtwisted we have that L2 is loose in ( M, ξ ), but this is
impossible because we already proved before that it is non-loose.

96
 Now take a Darboux ball B in our overtwisted ( M, ξ ). We saw in Proposition 3.5.1 that
we can always find an overtwisted disk in M, which is disjoint from B. We can deform such
an overtwisted disk to intersect B and, since we do not touch the boundary, it remains
an overtwisted disk. This means that every Darboux ball, in an overtwisted contact 3-
manifold, intersects at least one overtwisted disk. More generally, the same argument tells
us that if ( M, ξ ) = ( M1 , ξ 1 )# ( M2 , ξ 2 ) then we always find overtwisted disks that are not
entirely contained in one of the two summands of ( M, ξ ).
The previous observation suggests that the connected sum of a non-loose Legendrian
link with another Legendrian link, which is non-loose or lives in a tight structure, may still
be loose. On the other hand, the following proposition holds for disjoint unions.

Proposition 7.1.3 The disoint union of a non-loose Legendrian link L1 ,→ ( M1 , ξ 1 ) with another
Legendrian link L2 ,→ ( M2 , ξ 2 ), which is non-loose or lives in a tight structure is also non-loose.
Furthermore, if L is non-loose then L is a split Legendrian link if and only if its smooth link
type is split.

Proof. For the first statement we observe that ξ i restricted to Mi \ Li is tight for i = 1, 2.
Then ξ = ξ 1 # ξ 2 restricted to M \ L = M1 # M2 \ ( L1 t L2 ) is also tight. The second statement
follows from Colin’s result in [13].

7.2 A classification theorem for loose knots

While it was known that non-loose Legendrian knots are not classified by their classical
invariants [57], in the case of loose knots such example was found only recently by Vogel
[89]; even though, according to [24], this phenomenon was already known to Chekanov.
Conversely, there were some results that go in the opposite direction.
Etnyre’s coarse classification of loose Legendrian knots [29] is probably the most im-
portant one. It says that loose knots are completely determined by their classical invari-
ants, but only up to contactomorphism. Another result was proved by Dymara in [17]
and it states that two Legendrian knots, with the same classical invariants, such that the
complement of their union contains an overtwisted disk are Legendrian isotopic.

Theorem 7.2.1 (Dymara) Consider a rational homology overtwisted 3-sphere ( M, ξ ). Then two
loose Legendrian knots K1 and K2 in ( M, ξ ) with the same classical invariants and such that K1 ∪
K2 is loose are Legendrian isotopic.

In this thesis we show that Dymara’s result can be strengthened. In fact we prove the
CEU eTD Collection

following theorem.

Theorem 7.2.2 Consider a rational homology contact 3-sphere ( M, ξ ). Suppose that there are two
loose Legendrian knots K1 and K2 in ( M, ξ ) and denote with ( E1 , E2 ) a pair of overtwisted disks,
where Ei is contained in the complement of Ki for i = 1, 2. If we can find E1 and E2 disjoint then
K1 and K2 are Legendrian isotopic if and only if they have the same classical invariants.

Though we still need an assumption on the overtwisted disks, this version can be ap-
plied in many interesting cases like disjoint unions of Legendrian knots.
The proof of Theorem 7.2.2 only requires to show that loose Legendrian knots with the
same classical invariants are Legendrian isotopic.

97
 Proof of Theorem 7.2.2. The idea of the proof is to find another Legendrian knot K, with the
same classical invariants as K1 and K2 , and to show that there are overtwisted disks in the
complement of both K1 ∪ K and K2 ∪ K. Then Theorem 7.2.1 gives that K1 and K2 are both
Legendrian isotopic to K.
If there is an overtwisted disk in the complement of K1 ∪ K2 then the claim follows
immediately from Theorem 7.2.1. Then we suppose that this is not the case.
Since E1 and E2 are disjoint disks we can find two closed balls B1 and B2 such that
Ei ⊂ Bi for i = 1, 2. Moreover, we can suppose that each Bi is disjoint from Ki ; this is
because we start from the assumption that each Ei is in the complement of Ki .
Now we have that the contact manifold ( Bi , ξ | Bi ) is an overtwisted D3 . Then from
Eliashberg’s classification of overtwisted structures [18, 22] we know that ( Bi , ξ | Bi )#(S3 , ξ 0 )
is contact isotopic to ( Bi , ξ | Bi ). This holds since the contact connected sum with (S3 , ξ 0 )

K ( M, ξ )

K2

K1

( D1 , ξ 0 )

( D2 , ξ 0 )

Figure 7.2: Three overtwisted disks are drawn in grey.

does not change the homotopy class of the 2-plane field given by ξ. From this we have
that inside ( Bi , ξ | Bi ) there is a sphere with trivial dividing set which bounds a closed ball
Di such that ξ | Di is contact isotopic to ξ 0 . Moreover, both D̊i and Bi \ Di contains some
overtwisted disks from Proposition 3.5.1.
At this point, we have found two closed balls D1 and D2 in ( M, ξ ) such that each
Di is in the complement of Ki and ξ | Di is contact isotopic to the overtwisted structure
ξ 0 ; furthermore, the contact manifold M \ ( D1 ∪ D2 ) is also overtwisted. This situation
is pictured in Figure 7.2. Let us denote with M0 the overtwisted manifold M \ D2 ; the
boundary of D1 gives a contact connected sum decomposition of M0 where the components
CEU eTD Collection

are M \ ( D1 ∪ D2 ) and D1 . Then the same argument that we applied before tells us that
( M \ ( D1 ∪ D2 ), ξ ) is contact isotopic to ( M0 , ξ ) without a Darboux ball.
Since K2 is a Legendrian knot in M0 and we can suppose that a Darboux ball is missed
by it, we can identify K2 with a Legendrian knot K inside M \ ( D1 ∪ D2 ) with the same
classical invariants as K2 , which are assumed to coincide with the ones of K1 . The Legen-
drian knot K is disjoint from both the overtwisted balls Di and then we find overtwisted
disks in the complement of both Ki ∪ K. This concludes the proof.

The condition on the overtwisted disks in Theorem 7.2.2 cannot be removed. Vogel in
[89] gives an example of two loose Legendrian unknots, both with Thurston-Bennequin
number equal to zero and rotation number equal to one, that are not Legendrian isotopic.

98
 In fact, although we can find overtwisted disks in the complement of both knots such disks
always intersect each other.
An important case where the situation described in Theorem 7.2.2 appears is when we
have a disjoint union of two Legendrian knots.
From Theorem 3.5.2 we observe that if ξ is tight then L is a split Legendrian link if and
only if its smooth link type is split.

Corollary 7.2.3 Suppose K1 and K2 are two loose Legendrian knots in the rational homology
contact sphere ( M, ξ ) such that K1 ∪ K2 is a split Legendrian link. Then they are Legendrian
isotopic if and only if they have the same classical invariants.

Proof. We have that ( M, ξ ) = ( M1 # M2 , ξ 1 # ξ 2 ) and Ki ,→ ( Mi , ξ i ). Hence, if K1 is loose in

( M1 , ξ 1 ) then clearly we can apply Theorem 7.2.1; on the other hand, if ( M1 , ξ 1 ) is tight then
Theorem 3.5.2 gives that K2 is loose in ( M2 , ξ 2 ) and then we can use the same argument.
We only have to consider the case when both Ki are non-loose in ( Mi , ξ i ). Then we apply
Theorem 7.2.2, since we can find overtwisted disks in both summands using Proposition
3.5.1.

We conclude with the following observation. Let us consider two loose Legendrian
knots K1 and K2 in ( M, ξ ) with same classical invariants and such that L = K1 ∪ K2 is
a topologically split 2-component Legendrian link. Corollary 7.2.3 says that if K1 is not
Legendrian isotopic to K2 then L is non-split as a Legendrian link. The example of Vogel
that was mentioned before falls into this case.

7.3 Existence of non-loose links with loose components

In Section 7.1 we found non-loose Legendrian links with non-loose components. A
more interesting result is to show that, under some hypothesis on ( M, ξ ), we can also
find non-split n-component Legendrian and transverse links such that L( L, M, ξ ) 6= [0],
which means that they are non-loose from Proposition 7.1.1, and all of their components
are instead loose. We start by constructing Legendrian knots with non trivial invariant in
all the overtwisted structures on S3 .
Consider the family of Legendrian knots L( j), where j > 1, given by the surgery di-
agram in Figure 7.3. Using Kirby calculus we easily see that L( j) is a positive torus knot
T2,2j+1 in S3 . On the other hand, if we want to determine the Legendrian invariants of L( j)
and the contact structure where it lives we need the following proposition, whose proof is
CEU eTD Collection

found in [57].
Let us introduce the notation. Suppose that the link J = J + ∪ J − ⊂ (S3 , ξ st ) is a contact
(±1)-surgery presentation of a contact 3-manifold ( M, ξ ) and L is a Legendrian knot in
(S3 , ξ st ), disjoint from J. Hence, L represents a Legendrian knot also in ( M, ξ ). Let tb0 and
rot0 denote the Thurston-Bennequin and rotation number of L as a knot in the standard S3 .
Writing J = J1 ∪ ... ∪ Jt , let ai be the integral surgery coefficient on the link component Ji ;
i.e. ai = tb( Ji ) ± 1 if Ji ∈ J ± . Define the linking matrix
t
Q( a0 , a1 , ..., at ) = (qi,j )i,j =0 ;

where (
ai if i = j
qi,j =
lk( Ji , Jj ) if i 6= j ,

99
 −1
j
−1
−1
+1
+1
L( j)

Figure 7.3: Contact surgery presentation for the Legendrian knot L( j).

with the convention that L = J0 and a0 = 0. Similarly, let Q = Q( a1 , ..., at ) denote the
t
matrix (qi,j )i,j =1 .

Proposition 7.3.1 Suppose that J is a surgery presentation of a Legendrian knot L in the contact
manifold ( M, ξ ), as before. Then the Hopf invariant of ξ is determined by the following equation:
     
rot( J1 ) rot( J1 )
1 
d3 (ξ ) = h ...  , Q−1  ... i − 3σ( Q) − 2t + J + .
   
4
rot( Jt ) rot( Jt )

Furthermore, the Thurston-Bennequin and rotation numbers of L can be extracted by the formulae:

det( Q(0, a1 , ..., at ))

tb( L) = tb0 +
det( Q)
and    
rot( J1 ) lk( L, J1 )
rot( L) = rot0 − h ...  , Q−1  ..
i .
   
.
CEU eTD Collection

rot( Jt ) lk( L, Jt )
We apply this proposition to the knots in Figure 7.3 and we immediately obtain that
they are Legendrian knots in S3 , ξ 1−2j and their invariants are:

• tb( L( j)) = 6 + 4( j − 1);

• rot( L( j)) = 7 + 6( j − 1).

b L( j), S3 , ξ 1−2j 6= [0] and then L L( j), S3 , ξ 1−2j



Moreover, from [57] we also know that L
is a non-zero torsion class in HFK − ( T2,−2j−1 ). Both have bigrading (1, 1 − j).
Now we want to consider another family of Legendrian knots: the knots Lk,l , with
k, l > 0, shown in Figure
7.4. From [57] we also know that Lk,l is a negative torus knot
T2,−2k−2l −3 in S3 , ξ 2l +2 and its invariants are:

100
 k l

−1
−1
−1
+1
+1
Lk,l

Figure 7.4: Contact surgery presentation for the Legendrian knot Lk,l .

• tb( Lk,l ) = −6 − 4(k + l );

• rot( Lk,l ) = −7 − 2k − 6l.

b of the Legendrian knots L0,l , L1,1 and
In this case, from [57], we have that the invariant L
L1,2 is non-zero with bigrading (−2k, 1 − k + l ) in the homology group HFK [ ( T2,2k+2l +3 ).
Obviously, the fact that L is non-zero again implies that the same is true for the invariant
b
L.
At this point, we define the Legendrian knots Ki , for every i ∈ Z, in the following way:

L( j)#L(1) if i = −2j < 0




if i = 1 − 2j < 0

 L( j)



Ki = L0,j−1 if i = 2j > 0

L0,j−1 #L(1) if i = 2j − 1 > 0





L0,0 #L(1)2 if i = 0 .

CEU eTD Collection

Then we have the following result.

Proposition 7.3.2 (Lisca, Ozsváth, Stipsicz and Szabó) The Legendrian knot Ki ,→ (S3 , ξ i ) is
such that L(Ki , S3 , ξ i ) 6= [0] and then it is non-loose for every i ∈ Z.

Proof. It follows easily from the previous computation and the connected sum formula.

We can now go back to links. Let us take an overtwisted 3-manifold ( M, ξ ) such that
there exists another contact structure ζ on M with cb( M, ζ ) 6= [0] and tξ = tζ ; in par-
ticular ζ is tight. Consider O the standard Legendrian unknot in ( M, ζ ). We have that
L(O , M, ζ ) coincides with e−d3 ( M,ζ ),0 6= [0] and the invariant is non-torsion; this is because
HFK − − M, O , tζ ∼ = F[U ](−d3 ( M,ζ ),0) and Theorem 6.5.3.

101
 Fix B a Darboux ball in ( M, ζ ), which contains O . Since B is contact isotopic to (R3 , ξ st ),
n −1
we can take the Legendrian link H+ in B, that is defined as n − 1 connected sums of the
standard positive Legendrian Hopf link; see Figure 7.5.
If n = 1 we do not do anything and we keep the the knot O . The invariant

Figure 7.5: Connected sums of n − 1 standard Legendrian positive Hopf links.

n −1
L( H+ , M, ζ ) is the tensor product of L(O , M, ζ ), which as we said is non-zero, with n − 1
times L( H+ ) the invariant in the standard S3 . An easy computation, see also Subsection
1.5.2, shows that L( H+ ) is the only non-torsion element in the group
F [U ]
 
− ∼
cHFL ( H− ) = F[U ](0,0) ⊕ F[U ](1,1) ⊕
U · F[U ] (0,0)
n −1
with bigrading (1, 1). This means that not only L( H+ , M, ζ ) is non-torsion,
 but itis also
n −1
represented by the top generator of one of the F[U ] towers of cHFL− − M, H+ , tζ . Here
the negative Hopf link H− appears because it is the mirror image of H+ and the Legendrian
invariant lives in its homology group, as from the definition in Chapter 6.
n −1
Now we perform a connected sum between H+ and the Legendrian knot L(1), de-
n −1
fined before; where we suppose that L(1) is summed to H+ on the rightmost component
3 n −1
in Figure 7.5. We saw that L(1) is a positive trefoil in (S , ξ −1 ). Thus H+ # L(1) lives in
( M, ξ 0 ); where ξ 0 is an overtwisted structure such that tξ 0 = tζ = tξ and d3 (ξ 0 ) = d3 (ζ ) − 1.
If we do another connected sum with the Legendrian knot Kd3 (ξ )−d3 (ζ )+1 , this time on
n −1
the leftmost component of H+ , we obtain the Legendrian link
n −1
L = Kd3 (ξ )−d3 (ζ )+1 # H+ # L (1) ;
which is a link in M equipped with a contact structure that has the same Hopf invariant
as ( M, ξ ) and induces the same Spinc structure of ξ. From Eliashberg’s classification of
overtwisted structures [18], we conclude that L is a Legendrian n-component link in ( M, ξ ).
We can now prove the following theorem.
CEU eTD Collection

Theorem 7.3.3 In every overtwisted 3-manifold ( M, ξ ), such that there exists a contact structure
ζ with tξ = tζ and cb( M, ζ ) 6= [0], there is a non-split n-component Legendrian link L, for every
n > 1, such that L( L, M, ξ ) is non-zero and all of its components are loose. In particular, L is
non-loose and stays non-loose after a sequence of negative stabilizations.
Furthermore, the transverse link T, obtained as transverse push-off of L, has exactly the same
properties of L.
Proof. We already saw that the link L exists if the hypothesis of the theorem holds. So first
we check that L( L, M, ξ ) is non-zero. In fact, the invariant is represented by the tensor
n −1
product of a non-zero torsion element with L( H+ , M, ζ ). We are working with F[U ]-
modules and we recall, see Subsection 1.5.2, that
F [U ] F [U ]
 
F [ U ] ⊗ F [U ] ∼
=
U · F [U ] U · F [U ]

102
 and, more precisely, in the F[U ] factor only the generator survives. Then, from what we
said before, we have that L( L, M, ξ ) remains non-zero. Note that this is false if instead we
take the negative Hopf link. In this case the invariant L does not lie in the top of an F[U ]
tower of the homology group and then it vanishes after the tensor product.
This immediately implies that T( T, M, ξ ) is also non-zero and then the theorem holds
for transverse links. Moreover, the fact that the components of L (and T) are loose follows
easily from the construction of L. In fact, we performed connected sums of ( M, ζ ) with
two overtwisted S3 ’s and each component lies in one of these three summands. It is only
left to prove that L is non-split.
From [13] we know that the connected sum of two tight contact manifolds is still tight.
This implies that a non-loose Legendrian link is split if and only if its smooth link type is
split. Hence, we just have to show that L is non split as a smooth n-component link. But
L is a connected sum of torus links in a 3-ball inside M and we know that L is non-split as
a link in S3 . Furthermore, if L is split in M then it would be split also in the 3-sphere and
this is a contradiction.

7.4 Non-simple link types

If two Legendrian links L1 and L2 are Legendrian isotopic then they have contacto-
morphic complements. This means that if L1 is loose (we suppose that the Li ’s are in an
overtwisted contact manifold) and L2 is non-loose then they are not Legendrian isotopic.
We recall that the link type L( L) in M, which is the smooth isotopy class of the Leg-
endrian link L in M, is called Legendrian simple if L( L1 ) = L( L2 ) and L1 , L2 have the same
classical Legendrian invariants implies that L1 is Legendrian isotopic to L2 . This means
that finding a pair of Legendrian knots in ( M, ξ ) overtwisted, one loose and the other non-
loose, but with the same classical invariants (knot type, Thurston-Bennequin and rotation
number) proves that they do not have Legendrian simple knot type.
From the work of Eliashberg and Fraser [24] and Etnyre [29] we know that such pairs
really exist.

Proposition 7.4.1 In (S3 , ξ 1 ) the unknot is not a Legendrian simple knot type; where ξ 1 is the
overtwisted structure on the 3-sphere with d3 (ξ 1 ) = 1.

In fact, as we said in Section 7.1, in (S3 , ξ 1 ) there is a non-loose Legendrian unknot K

with tb(K ) = 1 and rot(K ) = 0, while from [29] we also have that, in the same contact
manifold, there is a loose Legendrian unknot K 0 with same classical invariants.
CEU eTD Collection

It is harder to find two non-loose Legendrian links L1 and L2 with same classical in-
variants and such that they are not Legendrian isotopic. The first example of such links
was given by Etnyre in [28] and others were found by Lisca, Ozsváth, Stipsicz and Szabó
in [57]; from which we report the following result.

Proposition 7.4.2 (Lisca, Ozsváth, Stipsicz and Szabó) The knot type T(2,−7) # T(2,−9) in
(S3 , ξ 12 ) contains two non-loose Legendrian knots, with equal Thurston-Bennequin and rotation
numbers, which are not Legendrian isotopic.

The Legendrian isotopy classes of the knots in this proposition are distinguished using
the refinement of the Legendrian invariant Lb for knots.

103
 In the previous subsection we saw that, under some hypothesis, in an overtwisted 3-
manifold ( M, ξ ) we can find non-loose, non-split n-component Legendrian links Ln . Con-
sider the links L0n obtained as the connected sum of Ln with the standard Legendrian un-
knot in (S3 , ξ 0 ), where ξ 0 is the overtwisted S3 with zero Hopf invariant.
Since ( M, ξ ) is already overtwisted, we have that ( M, ξ )# (S3 , ξ 0 ) is contact isotopic to
( M, ξ ). This means that L0n is also a non-split Legendrian link in ( M, ξ ), which is smoothly
isotopic to Ln for every n > 1, but unlike Ln it is clearly loose.
Each component of L0n has the same classical invariants of a component of Ln . More-
over, if n > 2 then there is an overtwisted disk in their complement. From Dymara’s
Theorem 7.2.1, the components of L0n are Legendrian isotopic to the ones of Ln . Hence, we
have the following corollary.

Corollary 7.4.3 The link type of Ln and L0n in M, which is denoted with L, is both Legendrian and
transverse non-simple.

We can also find non-simple link types where the two Legendrian and transverse rep-
resentatives are non-loose.

Proposition 7.4.4 Let us consider the links L1 = ( L0,2 # L1,2 )# H+ # L(1) and L2 =
( L1,1 # L0,3 )# H+ # L(1) in the contact manifold (S3 , ξ 11 ); where in Li the knots on the left are
summed on the first component of H+ and L(1) on the second one. Then L1 and L2 are two non-
loose, non-split 2-component Legendrian links, with the same classical invariants and Legendrian
isotopic components, but that are not Legendrian isotopic.
In the same way, the transverse push offs of L1 and L2 are two non-loose, non-split 2-component
transverse links, with the same classical invariants and transversely isotopic components, but that
are not transversely isotopic.

Proof. We apply Theorem 6.6.7. The Legendrian invariant of L1 and L2 is computed in [57]
b ( L1 , S3 , ξ 11 ) and L
and it is non-zero; moreover, the Alexander pairs of L b ( L2 , S3 , ξ 11 ), respect
to splitting spheres separating L0,2 from L1,2 and L1,1 from L0,3 , are different. The fact that
the components are Legendrian isotopic follows from Theorem 7.2.1. The same argument
proves the theorem in the transverse setting.

Using the same construction, the refined version of L

b and T
b can be applied to find
such examples for links with more than two components in every contact manifold as in
Theorem 7.3.3.

Theorem 7.4.5 Suppose that ( M, ξ ) is an overtwisted 3-manifold as in Theorem 7.3.3. Then

CEU eTD Collection

in ( M, ξ ) there is a pair of non-loose, non-split n-component Legendrian (transverse) links, with

the same classical invariants and Legendrian (transversely) isotopic components, but that are not
Legendrian (transversely) isotopic.

Proof. Let us take a standard Legendrian (transverse) positive Hopf link H+ in (S3 , ξ st ).
On the first component of H+ we perform a connected sum with the knot L0,2 # L1,2 in
one case and L1,1 # L0,3 in the other. While, on the second component of H+ , we sum a
non-split Legendrian (transverse) n-component link in the overtwisted manifold ( M, ξ 0 ),
where d3 (ξ 0 ) = d3 (ξ ) − 12, with non-zero invariants; those links exist as we know from
Theorem 7.3.3. We conclude by applying the same argument of the proof of Proposition
7.4.4.

104
 Chapter 8

Link Floer homology and grid

diagrams

8.1 Filtered simply blocked link grid homology

8.1.1 The chain complex
Grid diagrams are simple combinatorial presentation of links in S3 , dating back to the
19th century. A grid diagram is an l × l grid of squares, l of which are marked with an O
and l of which are marked with an X. Each row and column contains exactly one O and
one X. A projection of a link together with an orientation on it can be associated to a grid
diagram D. These grids can be used to give a simpler reformulation of link Floer homology,
called grid homology. Of course these two homologies are isomorphic, nevertheless grid
homology can be easier to study.
In this thesis we use the same notation of the book “Grid homology for knots and
links” [67]. In this book particular attention is given to two versions of the grid homology
d ( L) and the collapsed unblocked grid homology
of a link L: the simply blocked grid homology GH
−
cGH ( L); both these homology groups are invariant under link equivalence. We study a
d ( L). Let us denote with F the field with two elements; we
slightly different version of GH  
start constructing a filtered F-complex GC c ( D ), b
∂ from a grid diagram D, equipped with
an increasing Z-filtration F and we prove that F induces a filtration in homology, leading
to the filtered homology group GH(d L). The latter is not completely unrelated to GHd ( L) as
we see later in this section.
CEU eTD Collection

We always suppose that a link is oriented. We denote by D a toroidal grid diagram

that represents an n-component link L. The number grd( D ) is the number of rows and
columns in the grid. The orientation in the diagram is taken by going from the X to the
O-markings in the columns and the opposite in the rows. Vertical lines are numbered from
left to right and horizontal lines from bottom to top, as shown in Figure 8.1. We identify
the boundaries of the grid in order to make it a fundamental domain of a torus; then the
lines of the diagram are embedded circles in this torus. Any grd( D )-tuple of points x in
the grid, with the property that each horizontal and vertical circle contains exactly one of
the elements of x, is called a grid state of D.
Consider the set of the O-markings O = {O1 , ..., Ogrd( D) }. We call special O-markings a
non-empty subset sO ⊂ O that contains at most one O-marking from each component of L,
while we call the others normal O-markings. We represent the special ones with a double

105
 (0, 5) (5, 5)
X
X
X
X
X
(0, 0) (5, 0)

Figure 8.1: A grid diagram of the positive trefoil knot.

circle in the grid diagram. We usually consider only the case when |sO| = n, which means
there is exactly one special O-marking on each component. We talk about the general case
in Subsection 8.2.2. From now on there are always n special O-markings in a grid diagram,
unless it is explicitly written differently.
We define the simply blocked complex GC c ( D ) as the free F[V1 , ..., Vgrd( D)−n ]-module,
where F = Z \ 2Z, over the grid states S( D ) = { x1 , ..., xgrd( D)! }.
We associate to every grid state x the integer M ( x ), called the Maslov grading of x,
defined as follows:
M( x ) = MO ( x ) = J ( x − O, x − O) + 1 ; (8.1)
where

∑

J ( P, Q) = ( a, b) ∈ ( P, Q) | b has both coordinates strictly bigger than
a∈ P

the ones of a

with coordinates taken in the interval [0, grd( D )).

Then we have the Maslov F-splitting
M
c (D) =
GC c d (D)
GC
d ∈Z

where GC c d ( D ) is the finite dimensional F-vector space generated by the elements V l1 · ... ·
1
Vmlm x, with x ∈ S( D ) and m = grd( D ) − n, such that
CEU eTD Collection

m
M(V1l1 · ... · Vmlm x ) = M ( x ) − 2 ∑ li = d .
i =1

We define another integer-valued function on grid states, the Alexander grading A( x ), with
the formula
M ( x ) − MX ( x ) grd( D ) − n
A( x ) = − ;
2 2
where MX ( x ) is defined in Equation (8.1), replacing the set O with X. For the proof that
A( x ) is an integer we refer to [67].
Now we introduce an increasing filtration on GC c ( D ), see Subsection 1.4.1 such that

F s GC F s GC
M
c (D) = c d (D)
d ∈Z

106
 c d ( D ) is generated over F by the elements V l1 · ... · Vmlm x with Maslov grad-
and where F s GC 1
ing d and Alexander grading
m
A(V1l1 · ... · Vmlm x ) = A( x ) − ∑ li 6 s .
i =1

8.1.2 The differential

First we take x, y ∈ S( D ). The set Rect( x, y) is defined in the following way: it is always
empty except when x and y differs only by a pair of points, say { a, b} in x and {c, d} in
y; then Rect( x, y) consists of the two rectangles in the torus represented by D that have
bottom-left and top-right vertices in { a, b} and bottom-right and top-left vertices in {c, d}.
We call Rect◦ ( x, y) ⊂ Rect( x, y) the subset of the rectangles which do not contain a point
of x (or y) in their interior, that are called the empty rectangles.
The differential b ∂ is defined as follows:

∑ ∑◦
O1 (r ) O (r )
∂x =
b V1 · ... · Vm m y for any x ∈ S( D )
y∈S( D ) r ∈Rect ( x,y)
r ∩sO=∅
(
1 if Oi ∈ r
where Oi (r ) = . Here {O1 , ..., Om } is the set of the m = grd( D ) − n normal
0 if Oi ∈
/r
O-markings.
We extend b c d ( D ) linearly, and we call it b
∂ to GC ∂d , then again to the whole GC c ( D ) in
the following way: b ∂(Vi x ) = Vi · b
∂x for every i = 1, ..., m and x ∈ S( D ). Since b
∂ keeps the
filtration and drops the Maslov grading by 1, see [67], we have maps

∂d,s : F s GC
b c d ( D ) −→ F s GC
c d −1 ( D )

where b ∂d to the subspace F s GC

∂d,s is the restriction of b c d ( D ) ⊂ GC
c d ( D ). Furthermore, we
have ∂ ◦ ∂ = 0, see [67] fo more details.
b b

8.1.3 The homology

d d ( D ) as the quotient space Ker ∂d . Moreover, we

b
We define the homology group GH
Im b
∂ d +1
s
introduce the subspaces F GHd ( D ) as in Subsection 1.4.1: consider the projection πd :
d
CEU eTD Collection

Ker b∂d → GH
d d ( D ). Since Ker b ∂d ∩ F s GC
∂d,s = Ker b c d ( D ) we say that

F s GH
d d ( D ) = πd (Ker b
∂d,s )

for every s ∈ Z. Ker b

∂d,s ⊂ Ker b∂d,s+1 implies that the filtration F descends to homology.
d d ( D ) is a finite dimensional F-vector space.
We see immediately that each F s GH
We can extend the filtration F on the total homology
M
GH(
d D) = GH
d d (D)
d ∈Z

by taking
F s GH( F s GH
M
d D) = d d (D) .
d ∈Z

107
 From [67] we know that the dimension of F s GH d d ( D ) as an F-vector space is a link
invariant for every d, s ∈ Z, in particular it is independent of the choice of the special O-
markings and the ordering of the markings. Hence we can denote them with F s GH d d ( L ).
Furthermore, [67] also tells us that [Vi p] = [0] for every i = 1, ..., m and [ p] ∈ GH(
d L). This
means that each homology class can be represented by a combination of grid states and
d L) is also a finite dimensional F-vector space.
every level F s GH(
The homology group GH d d,s ( L) of [67] can be recovered from the complex GC c ( D ) in the
following way. We denote the graded
object associated to a filtered complex C , defined in
Subsection 1.4.1, as gr(C), gr(∂) . Then we have that
   
d ∼
GH d,s ( L) =F Hd,s gr GC ( D ) , gr(∂) .
c b

8.1.4 Equivalence between GH

d and HFL
[
 
We want to show that the chain complex GC c ( D ), b
∂1 , obtained from a grid diagram D
 
which represents the link L, can be identified with the chain complex CFL( H ), ∂2 given
d b
by an appropriate Heegaard diagram H.
Let HD be the (unbalanced) multi-pointed Heegaard diagram (Σ, α, β, w, z), see Sub-
section 4.1.2, defined as follows. The surface Σ is the torus which is given by D; in fact,
in Subsection 8.1.1 we defined D precisely as a square representing the fundamental do-
main of a torus. The curves α and β correspond one-to-one with the horizontal and vertical
curves in D. Finally, we take the O-markings as the set of basepoints w and the X-markings
as z. Then we have the following theorem.

Theorem 8.1.1 There is an isomorphism of chain complexes

Ξ : GC
c ( D ) −→ CFL
d ( HD )

which preserves the bigradings.

Proof. We give an idea of the proof, which can be found in [67]. First, we observe that grid
states in D correspond to the intersction points in HD . Now suppose that φ ∈ π2 ( x, y) is
such that its associated domain in D belongs to Rect◦ ( x, y). Then we have that µ(φ) = 1
CEU eTD Collection

and M(c φ) = 1. This is verified using the following interpretation of holomorphic disks
in the symmetric product. A holomorphic disk in Symd (Σ) can be viewed as the following
collection ( F, P, f ) of data:

1. a surface with boundary F, equipped with a complex structure;

2. a degree d holomorphic map P : F → D , where D is the standard disk in C;

3. a holomorphic map f : F → Σ.

In this correspondence, the domain of the holomorphic disk can be thought of as the two-
chain induced by the map f from F to Σ. Then the claim follows from the Riemann map-
ping theorem. See [67] for more details.

108
 We say that a domain is called positive if its local multiplicities are all non-negative,
and at least one of them is positive. Let ψ ∈ π2 ( x, y) be a positive domain. Then, there
is a sequence of grid states { x1 , ..., xk }, with x1 = x and xk = y, and empty rectangles
{ri , ..., rk−1 }, with ri ∈ Rect◦ ( x1 , xi+1 ) for i = 1, ..., k − 1, so that ψ = r1 ∗ ... ∗ rk−1 . The proof
of this fact is also given in [67].
At this point, we can identify the boundary maps as follows. As we said in Chapter 5, if
a domain φ is associated to a holomorphic strip then it has non-negative local multiplicities
everywhere. Domains with zero local multiplicities everywhere correspond to constant
maps, which have µ(φ) = 0, and so they are not counted in the differential. Any other
such domain φ can be factored as φ = φ1 ∗ ... ∗ φm , where each φi is an empty rectangle.
Since µ is additive under composition, and µ(φi ) = 1 for what we said before, we
conclude that µ(φ) = m. Since the holomorphic strips counted in the differential have
Maslov index one, we conclude that only empty rectangles are counted in the differential.
Moreover, each such rectangle really appears in the differential from the first obervation,
completing the identification of b ∂2 with b ∂1 .
Since the Maslov and Alexander gradings transform the same in both theories, when
φ corresponds to an empty rectangle, it follows that the the identification of complexes
respects relative bigradings. In fact, the identification respects absolute bigradings, since
in both theories, the Alexander grading is normalized to be symmetric, and the Maslov
grading is normalized using the total homology of the complexes.
d ( L) is iso-
In particular, this means that the simply blocked grid homology group GH
[ ( L) as bigraded F-vector spaces. In light of this, from now on in this
morphic to HFL  
[ ( L) for the homology of the chain complex GC
chapter we always write HFL c ( D ), b
∂1 .
Furthermore, we denote with HF
\ L( L) the filtered homology group GH(
d L) introduced
in Subsection 8.1.3.

8.2 The invariant tau in the filtered theory

8.2.1 Definition of the invariant
Since F s−1 HF
\ Ld ( L) ⊂ F s HF
\ Ld ( L), and they are finite dimensional vector spaces,
we define the function
F s HF
\ Ld ( L)
TL (d, s) = dimF
F s−1 HF
\ Ld ( L)
CEU eTD Collection

which clearly is still a link invariant.

Our first goal is to see what happens to this function T when we stabilize the link L, in
other words when we add a disjoint unknot to L. Denote the unknot with the symbol .
We claim that

TLt (d, s) = TL (d, s) + TL (d + 1, s) for any d, s ∈ Z . (8.2)

We refer to Subsection 1.4.2 before for some remarks on filtered chain maps. Moreover,
we define the shifted complex C [[ a, b ]] = C 0 as F s Cd0 = F s−b Cd−a . Now, in order to prove
Equation (8.2), we need the following proposition.

109
 Proposition 8.2.1 For any link L we have HF
\ L( L t ) ∼ = HF
\ L( L) ⊗ V, where V is the two
dimensional F-vector space with generators in grading and minimal level (d, s) = (−1, 0) and
(d, s) = (0, 0).
Proof. Take a grid diagram D for L. Then the extended diagram D, obtained from D by
adding one column on the left and one row on the top with a doubly-marked square in the
top left, represents the link L t . The circle in the doubly-marked square is forced to be
a special O-marking and we can also suppose that there is another special O-marking just
below and right of it, as shown in Figure 8.2.

X z

Figure 8.2: We call the top-left special O-marking O0 .

Let I ( D ) denote the set of those generators in S( D ) which have a component at the
lower-right corner of the doubly-marked square, and let N ( D ) be the complement of I ( D )
in S( D ). From the placement of the special O-markings, we see that N ( D ) spans a sub-
complex N in GC c ( D ). Moreover, if b c ( D ) and b
∂1 is the differential in GC c (D)
∂2 the one in GC
we can express the restriction of b ∂1 to the subspace I, spanned by I ( D ), with b ∂2 + b
∂N ; this
because there is a one-to-one correspondence between elements of I ( D ) and grid states in
S( D ). This correspondence induces a filtered quasi-isomorphism i : (I, b ∂2 ) → GC
c ( D ).
Define a map H : N → I by the formula

∑ ∑◦
O1 (r ) O (r )
H (x) = V1 · ... · Vm m y for any x ∈ N ( D ) .
y∈ I ( D ) r ∈Rect ( x,y)
O0 ∈r

We have that H is a filtered chain homotopy equivalence between (N, b ∂1 ) and (I, b
∂2 ), which
increases the Maslov grading by one, and H ◦ b ∂N = 0. To see the first claim, we mark the
b z : I → N, which counts
square just on the right of O0 with z and we define an operator H
only rectangles that contain z. This operator is a homology inverse of H; this and the
second claim can be proved in the same way as in [67]. Those two facts together tell us
CEU eTD Collection

that the following diagram commutes.

∂N
b
I N

i i◦H

c (D) 0 c ( D ) [[ −1, 0 ]]
GC GC

Since H is a filtered chain homotopy equivalence, i ◦ H is a filtered quasi-isomorphism,

just like the map i. Therefore, we can use Lemma 1.4.4 and obtain that the map between

110
 the mapping cones

Cone(b c ( D ) −→ Cone(0) = GC
∂N ) = GC c (D) ⊗ V

is a filtered quasi-isomorphism and so the claim follows easily from Propositions 1.4.1 and
1.4.3. See also [67] for more details.

Equation (8.2) is obtained immediately from Proposition 8.2.1, in fact we have proved
that F s HF
\ Ld ( L t ) ∼ = F s HF
\ Ld ( L) ⊕ F s HF
\ Ld+1 ( L) for every d, s ∈ Z and so it is
enough to apply the definition of T.
Now we are able to do some computations. The homology of the unknot can be easily
computed by taking the grid diagram of dimension one, where the square is marked with
both X and O. The complex has one element of Maslov and Alexander grading 0; then
T (d, s) = 1 if (d, s) = (0, 0) and it is 0 otherwise.
Using Equation (8.2) we get the function T of the n-component unlink n :

 n−1

if (d, s) = (−k, 0), 0 6 k 6 n − 1

T n (d, s) = k .

 0 otherwise

We can also see this directly from Proposition 8.2.1, in fact we have that HF\ L( n ) ∼
=
V ⊗( n − 1 ) .
Now let us consider a grid diagram D of a link L. The Maslov grading of the elements
of S( D ) and the differential b
∂ are independent of the position of the X’s, once we have
fixed the special O-markings. Since we can always change the X-markings to obtain n ,
this means that dimF HF\ Ld ( L) = dimF HF
\ Ld ( n ) for every d ∈ Z and the generators
are the same. In particular
n −1 n −1
HF
\ L( L) ∼
=F HF
\ L( n)
∼
= F F2 and HF
\ Ld ( L) ∼
= F F( − d ) when 1 − n 6 d 6 0 .

From this we have that HF

\ L0 ( L) has always dimension one and then we define τ ( L)
as the only integer s such that TL (0, s) > 0. We remark that for a knot this version of τ
coincides with the one of Ozsváth and Szabó. See the proof of Theorem 8.4.1 in Section 8.4.
We also observe that Equation (8.2) tells that τ ( L t ) = τ ( L).

8.2.2 Dropping the special O-markings

CEU eTD Collection

Inthis subsection
 we study what happens to the homology of the filtered chain com-
plex GC ( D ), ∂ if the grid diagram D has less than n special O-markings.
c b
Let us consider D a grid diagram for an n-component link L. The set sO ⊂ O contains
at most one O-marking from each component of L, but we have that |sO| > 1. Denote
with m = grd( D ) − |sO| the number of normal O-markings in D. Then we can define our
chain complex exactly in the same way as in Section 8.1; on the other hand, the homology
GH(
d D ) is no longer a link invariant, in fact it clearly depends of the choice of the special
O-markings.
Nonetheless, we can show that the F-vector space GH( d D ) is still finite dimensional.
Note that this is not true if instead we consider the bigraded simply blocked grid homology
d ( D ).
group GH

111
 Proposition 8.2.2 The homology group GH(d D ), defined as in Subsection 8.1.3, is F-isomorphic


to GH
d |sO| , the homology of the unlink with | sO| components each containing a special O-
marking. In particular, we have that
|sO| |sO|−1
d D) ∼
GH( d d (D) ∼
=F F2 −1 and GH = F F( − d )

when 1 − |sO| 6 d 6 0.

Proof. As we noted before, the group GH(

d D ) does not depend on the position of the X-
marking. Since we can always change them in a way that D becomes a diagram for an
unlink with a special O-marking on every component, the claim follows from the results
in the previous subsection.

Even though in this case the homology is no longer a link invariant, we can still prove
the following theorem.

Theorem 8.2.3 Let us consider two grid diagrams D1 and D2 representing smoothly isotopic links
L1 and L2 such that all the isotopic components both contain or not contain a special O-marking.
Then we have that GH(
d D1 ) is filtered isomorphic to GH(
d D2 ) and the isomorphism preserves the
Maslov grading.

Proof. From [82] we know that such two grid diagrams differ by a finite sequence of grid
moves: reordering of the O-markings, commutations and stabilizations. Then it is enough
to prove the theorem in the case when D2 is obtained from D1 by one of these three moves.
The maps that we defined in Chapter 5 can be arranged to give filtered quasi-
d D1 ) ∼
isomorphisms for each move. This implies that GH( = GH(
d D2 ). See also the results in
[67] to find the maps in the grids setting.

Hence, we can denote the homology group of an n-component link L with GH d O ( L)

and it depends only on which components of the link contain a special O-marking. We use
the homology groups GHd O ( L) to define some cobordism maps in Section 8.3.

8.2.3 Symmetries
Reversing the orientation
If − L is the link obtained from L by reversing the orientation of all the components
CEU eTD Collection

then
T− L (d, s) = TL (d, s) for any d, s ∈ Z (8.3)
and τ (− L) = τ ( L).
To see this, consider a grid diagram D of L, then it is easy to observe that, if we reflect
D along the diagonal going from the top-left to the bottom-right of the grid, the diagram
D 0 obtained in this way represents − L. Hence, we take the map Φ : S( D ) → S( D 0 ) that
sends a grid state x into its reflection x − and now, from [67], we have that M( x − ) = M ( x )
and A( x − ) = A( x ). This means that Φ is a filtered quasi-isomorphism between GC c (D)
c ( D 0 ), since clearly the differentials commute with Φ.
and GC
This gives another proof that HF \ L(− L) ∼= HF
\ L( L) and then Equation (8.3) follows.

112
 Mirror image
For an n-component link L we have that the function T of the mirror image L∗ is given
by the following equation
TL∗ (d, s) = TL (−d + 1 − n, −s) for any d, s ∈ Z . (8.4)
To see this, we first, given a complex C with a filtration F , introduce the filtered dual
complex C ∗ , equipped with a filtration F ∗ by taking
(F ∗ )s (C ∗ )d = Ann(F −s−1 C−d ) ⊂ (C−d )∗ = (C ∗ )d for any d, s ∈ Z ,
where Ann(F h Ck ) is the subspace of (Ck )∗ consisting of all the linear functionals that are
zero over F h Ck .  
Second, given a grid diagram D of L, we call GC f ( D ), e
∂ the filtered chain complex
 
c ( D ), b
GC ∂
and we also denote with W the two dimensional F-vector space with
V1 = ... = Vm = 0
generators in grading and minimal level (d, s) = (0, 0) and (d, s) = (−1, −1).
Lemma 8.2.4 We have the filtered quasi-isomorphism
f (D) ∼
GC c ( D ) ⊗ W ⊗(grd( D)−n) ,
= GC (8.5)
where D is a grid diagram for an n-component link L.
Proof. Suppose that V1 and V2 are variables which belong to the same component of L.
Then it follows from [67, Lemma 14.1.11] that
c (D)
GC ∼ c (D) ⊗ W .
= GC
V1 = V2
Therefore, we conclude by iterating this procedure.

We want to prove the following proposition.

∗ ∗
Proposition 8.2.5 HF
\ L( L∗ ) ∼
= HF
\ L ( L)J1 − n, 0K, where the filtration on HF
\ L ( L) is F ∗ .
Proof. Let D ∗ be the diagram obtained by reflecting D through a horizontal axis. The di-
agram D ∗ represents L∗ . Reflection induces a bijection x → x ∗ between grid states for D
and those for D ∗ , inducing a bijection between empty rectangles in Rect◦ ( x, y) and empty
rectangles in Rect◦ (y∗ , x ∗ ). Hence, the reflection induces a filtered isomorphism
CEU eTD Collection

f ( D∗ ) ∼
GC f ∗ ( D ) [[ 1 − grd( D ), n − grd( D ) ]] ,
= GC
where the shifts are given by the fact that M( x ∗ ) = − M ( x ) + 1 − grd( D ) and A( x ∗ ) =
− A( x ) + n − grd( D ).
Now Lemma 8.2.4 and observing that

(W ∗ )⊗(grd( D)−n) ∼
= W ⊗(grd( D)−n) [[ grd( D ) − n, grd( D ) − n ]]
lead to the filtered quasi-isomorphism
c ( D∗ ) ∼
GC c ∗ ( D ) [[ 1 − n, 0 ]] .
= GC

113
  ∗
Proposition 8.2.5 says that F s HF
\ Ld ( L∗ ) ∼
= (F ∗ )s HF
\ L ( L) for every d, s ∈ Z.
d −1+ n
Then we can prove Equation (8.4):
 ∗  
(F ∗ )s HF\ L ( L) Ann F −s−1 HF \ L − d +1− n ( L )
d −1+ n
TL∗ (d, s) = dim  ∗ = dim   =
∗
(F ) s − 1 HF L
\ ( L) Ann F − s HF
\ L ( L )
− d +1− n
d −1+ n

F −s HF
\ L − d +1− n ( L )
= dim = TL (−d + 1 − n, −s) .
F − s − 1 HF L−d+1−n ( L)
\
If we define τ ∗ ( L) as the unique integer such that TL (1 − n, τ ∗ ( L)) = 1 then we have
proved that
τ ( L∗ ) = −τ ∗ ( L) . (8.6)
In particular for a knot K, where τ ∗ (K ) = τ (K ), we have τ (K ∗ ) = −τ (K ). Moreover, we
have the following corollary.
Corollary 8.2.6 Suppose that L is smoothly isotopic to L∗ . Then,
TL∗ (d, s) = TL (d, s) for any d, s ∈ Z
1− n


and so Equation (8.4) gives that the function TL has a central symmetry in the point 2 ,0 . In
particular τ ∗ ( L) = −τ ( L) and, for knots, τ (K ) = 0.

Connected sum
Given two links L1 and L2 , the function T of the connected sum L1 # L2 is the convolu-
tion product of the T functions of L1 and L2 ; in other words
TL1 # L2 (d, s) = ∑ TL1 (d1 , s1 ) · TL2 (d2 , s2 ) for any d, s ∈ Z . (8.7)
d = d1 + d2
s = s1 + s2

This equation is very hard to prove in the grid diagram settings, but it can be proved using
the holomorphic definition of link Floer Homology, as we saw in Subsection 6.6.1. See also
[72].
We see immediately that the homology and the T function of L1 # L2 are independent
of the choice of the components used to perform the connected sum; moreover, the τ-
invariant is additive:
τ ( L1 # L2 ) = τ ( L1 ) + τ ( L2 ) .
CEU eTD Collection

Disjoint union
The disjoint union of two links L1 and L2 is equivalent to L1 # ( L2 t ). Thus by Equa-
tions (8.2) and (8.7) we have the following relation:
TL1 t L2 (d, s) = ∑ TL1 (d1 , s1 ) · TL2 (d2 , s2 ) + TL2 (d2 + 1, s2 ) for any d, s ∈ Z , (8.8)


d = d1 + d2
s = s1 + s2

or in other words HF \ ∼ HF
L( L1 t L2 ) = \ L( L1 ) ⊗ HF
\ L( L2 ) ⊗ V, where V is the two di-
mensional F-vector space with generators in grading and minimal level (d, s) = (−1, 0)
and (d, s) = (0, 0). We have immediately that
τ ( L1 t L2 ) = τ ( L1 # L2 ) = τ ( L1 ) + τ ( L2 ) .

114
 Quasi-alternating links
We recall that quasi-alternating links are the smallest set of links Q that satisfies the
two properties:

1. the unknot is in Q;

2. L is in Q if it admits a diagram with a crossing whose two resolutions L0 and L1 are

both in Q, det( Li ) 6= 0 and det( L0 ) + det( L1 ) = det( L).

The above definition and a result in [71] imply that every quasi-alternating link is non-split
and every non-split alternating link is quasi-alternating. Moreover, quasi-alternating links
are both Khovanov and link Floer homology thin, which means that their homologies are
supported in two and one lines respectively, and the homology is completely determined
by the signature and the Jones (Alexander in the hat version of link Floer homology) poly-
nomial. The following proposition says that the same is true in filtered grid homology.

Theorem 8.2.7 If L is an n-component quasi-alternating link then the function TL is supported in

a line; more specifically the following relation holds:

n − 1 − σ( L)
TL (d, s) 6= 0 if and only if s = d + for 1 − n 6 d 6 0
2
where σ( L) is the signature of L.

Proof. We already know that if TL (d, s) 6= 0 then 1 − n 6 d 6 0, so we only have to prove

the alignment part of the statement.
Take a grid diagram
 D for L, then from [67] the claim is true for the bigraded
homology of gr GC c ( D ) , gr(b
∂) ∼
= HFL[ ( L). Since TL (d, s) 6= 0 implies that
   
Hd,s gr GCc ( D ) , gr(b
∂) is non-zero, the theorem follows.

From Theorem 8.2.7 we obtain immediately the following corollary.

n −1− σ ( L )
Corollary 8.2.8 If L is an n-component quasi-alternating link then τ ( L) = 2 and τ ( L∗ ) =
n − 1 − τ ( L ).

8.3 Cobordisms
CEU eTD Collection

8.3.1 Induced maps and degree shift

In this section we study the behaviour of the function T under cobordisms, see Section
1.3. Some of the induced maps that appear in this subsection come from the work of Sarkar
in [82]; though the grading shifts are different, because Sarkar used a different definition
of the Alexander grading, ignoring the number of component of the link.
It is a standard result in Morse theory that a link cobordism can be decomposed into
five standard cobordisms. We find maps in homology for each case. From now on, given
a link Li , we denote with Di one of its grid diagrams.

115
 L1 L2

Figure 8.3: Identity cobordism.

i) Identity cobordism. This cobordism, with no critical points (Figure 8.3), represents a
sequence of Reidemeister moves; in other words L1 and L2 are smoothly isotopic.
At the end of Section 8.1 we remarked that filtered homology is a link invariant; more
precisely what we have is a filtered quasi-isomorphism between GCc ( D1 ) and GC
c ( D2 ).
This, as we know, induces a filtered isomorphism in homology.

ii) Split cobordism. This cobordism (right in Figure 8.4) represents a band move when L2
has one more component than L1 . Take D1 with a 2 × 2 square with two X-markings,
one at the top-left and one at the bottom-right; then we claim that D2 is obtained from
D1 by deleting this two X-markings and putting two new ones: at the top-right and
the bottom-left, as shown in Figure 8.5. In order to construct the complex GCc ( D2 ) we

L1 L2 L1 L2

Figure 8.4: Merge and split cobordisms.

CEU eTD Collection

need to create one more special O-marking on the new component of L2 . To avoid this
problem we first consider the identity map in the filtered GC
f theory

f ( D1 ) −→ GC
Id : GC f ( D2 ) ,

which clearly is a chain map since now every O-marking is special; moreover, it in-
duces an isomorphism in homology, that preserves the Maslov grading, and a direct
computation gives that it is filtered of degree one.
Now we use Equation (8.5) and we get an isomorphism

ΦSplit : HF
\ L( L1 ) ⊗ W −→ HF
\ L( L2 )

that is a degree one filtered map which still preserves the Maslov grading. The W

116
 X X
X X

Figure 8.5: Band move in a grid diagram.

factor appears because in Equation (8.5) we take into account the size of Di and the
number of components of Li ; while the first quantity is the same for both diagrams
the link L2 has one more component than L1 .

iii) Merge cobordism. This cobordism (left in Figure 8.4) represents a band move when L1
has one more component than L2 . We have an isomorphism

ΦMerge : HF
\ L( L1 ) −→ HF
\ L( L2 ) ⊗ W .

The map is obtained in the same way as ΦSplit in the previous case, but with the dif-
ference that now it is filtered of degree zero.

L1 L2

Figure 8.6: Torus cobordism.

CEU eTD Collection

Sometimes we are more interested in when a split and a merge cobordism appear together,
the second just after the first, in the shape of what we call a torus cobordism (Figure 8.6).
We have the following proposition.

Proposition 8.3.1 Let Σ be a torus cobordism between two links L1 and L2 . Then Σ induces a
Maslov grading preserving isomorphism between HF
\ L( L1 ) and HF
\ L( L2 ), which is filtered of
degree one.

Proof. We can choose D1 in a way that the split and the merge band moves can be per-
formed on two disjoint bands. Then we apply twice the move shown in Figure 8.5 and we
take as map the identity. In this case the identity is a chain map because L2 has the same
number of components of L1 ; this means that the special O-markings in D1 and D2 are the

117
 L2

L1

Figure 8.7: Birth cobordism.

same and then the two differentials coincide. In this way we obtain an isomorphism in
homology with Maslov grading shift and filtered degree equal to the sum of the ones in ii)
and iii).

iv) Birth cobordism. A cobordism (Figure 8.7) representing a birth move.

Since our cobordisms have boundary in both L1 and L2 , we can always assume that a birth
move is followed (possibly after some Reidemeister moves) by a merge move. Thus it is
enough to define a map for the composition of these three cobordisms and this is what we
do in the following proposition.
Proposition 8.3.2 Let Σ be a cobordism between two links L1 and L2 like the one in Figure
8.8. Then Σ induces an isomorphism ΦBirth between HF \L( L1 ) and HF
\ L( L2 ) that preserves
the Maslov grading and it is filtered of degree zero.

L2

L1

Figure 8.8: A more useful birth cobordism.

CEU eTD Collection

Proof. The first step is to construct a map s1 associated to the grid move shown in Figure
8.9; note that we add a normal O-marking, since later we merge the new unknot compo-
nent with an already exisiting one. Let us denote with D10 the stabilized diagram and with
c = α ∩ β the point in the picture; then we have the inclusion i : S( D1 ) → S( D10 ) that sends
a grid state x in D1 to the grid state in D10 constructed from x by adding the point c. Then
c ( D 0 ) is defined by the following formula:
c ( D1 ) → GC
s1 : GC 1

∑0 ∑
n (H) n (H)
s1 ( x ) = V1 1 · ... · Vmm y for any x ∈ S( D1 )
y∈S( D1 ) H ∈SL(i ( x ),y,c)
H ∩sO=∅

where SL( x, z, p) is the set of all the snail-like domains, the exact definition can be found

118
 c X
α
D1
D1

Figure 8.9: Birth move in a grid diagram.

in [67], centered at p joining x to z, illustrated in Figure 8.10; ni ( H ) is the number of times

H passes through Oi and m is the number of the normal O-markings of D1 . In [58] is
proved that s1 is a filtered quasi-isomorphism which induces a filtered isomorphism be-
tween HF \ d O ( L1 t ); where the special O-markings on L1 t
L( L1 ) → GH coincide
with the ones on L1 (the new unknotted component has a normal O-marking). Moreover,
in [82] Sarkar showed that the map s1 is filtered of degree zero.
At this point we compose s1 with the map s2 given by the Reidemeister moves, which is
a filtered quasi-isomorphism by Theorem 8.2.3, and finally with s3 , the identity associated
to the band move of Figure 8.5. The map s3 induces an isomorphism GH d O ( L1 t ) →

Figure 8.10: Some of the snail-like domains SL( x, z, p): the coordinates of x and z are
represented by the white and black circles.

HF
\ L( L2 ) that clearly preserves the Maslov grading and again we easy compute that it is
filtered of degree zero.
Hence, the composition of these three maps induces the isomorphism in the claim.

v) Death cobordism. This cobordism (Figure 8.11) represents a death move. Since this move
CEU eTD Collection

can also be seen as a birth move between L2∗ and L1∗ , we take the dual map of

ΦBirth : HF
\ L( L2∗ ) −→ HF
\ L( L1∗ )
which exists from Proposition 8.3.2; then ΦBirth
∗ = ΦDeath , by Proposition 8.2.5, is a
map between HF L( L1 ) and HF L( L2 ). Furthermore, it is still an isomorphism that is
\ \
filtered of degree zero and preserves the Maslov grading.

The results for birth and death cobordisms in this section immediately give the following
corollary.
Corollary 8.3.3 Suppose there is a birth or a death cobordism as in Figures 8.8 and 8.11 between
two links L1 and L2 . Then we have that HF
\ L( L1 ) ∼
= HF
\ L( L2 ).

119
 L1

L2

Figure 8.11: Death cobordism.

8.3.2 Strong concordance invariance

We remark that a strong cobordism is a cobordism Σ, between two links with the same
number of components, such that every connected component of Σ is a knot cobordism be-
tween a component of the first link and one of the second link. Moreover, if the connected
components of Σ are all annuli then we call Σ a strong concordance, see Section 1.3. We
want to prove the following theorem.
Theorem 8.3.4 The function T is a strong concordance invariant. In other words, if L1 and L2 are
strongly concordant then TL1 (d, s) = TL2 (d, s) for every d, s ∈ Z.
We start by observing that Proposition 8.3.1 leads to the following corollary.
Corollary 8.3.5 Suppose there is a strong cobordism Σ between L1 and L2 such that Σ is the
composition of g(Σ) torus cobordisms, not necessarily all of them belonging to the same component
of Σ. Then Σ induces an isomorphism between HF \ L( L1 ) and HF
\ L( L2 ), which is filtered of
degree g(Σ) and preserves the Maslov grading.
Now, if we have an isomorphism F : HF \ L( L1 ) → HF
\ L( L2 ) that preserves the
Maslov
 grading and it is filtered of degree t, which means that there are inclusions
F F HF Ld ( L1 ) ⊂ F s+t HF
s \ \ Ld ( L2 ) for every d, s ∈ Z, then τ ( L2 ) 6 τ ( L1 ) + t. Hence,
we can prove the following theorem that immediately implies the invariance statement.
Theorem 8.3.6 Suppose that Σ is a strong cobordism between two links L1 and L2 . Then
τ ( L1 ) − τ ( L2 ) 6 g ( Σ ) .

Furthermore, if L1 and L2 are strongly concordant then HF

\ L( L1 ) ∼
= HF
\ L( L2 ).
Proof. From [67] we can suppose that, in Σ, 0-handles come before 1-handles while 2-
CEU eTD Collection

handles come later; moreover, we can say that Σ is the composition of birth, torus and
death cobordisms (and obviously some identity cobordisms). Each of these induces an
isomorphism in homology that also respects the Maslov grading.
For the first part, we only need to check what is the filtered degree of the isomorphism
between HF\ L( L1 ) and HF
\ L( L2 ), obtained by the composition of all the induced maps on
each piece of Σ. Birth, death and identity are filtered of degree zero, while, from Corollary
8.3.5, the torus cobordisms are filtered of degree g(Σ). Then we obtain
τ ( L2 ) 6 τ ( L1 ) + g ( Σ ) .
For the other inequality we consider the same cobordism, but this time from L2 to L1 .
Now, for the second part, we observe that now there are no torus cobordisms and then
the claim follows from Corollary 8.3.3.

120
 8.3.3 A lower bound for the slice genus
Suppose Σ is a cobordism (not necessarily strong) between two links L1 and L2 . Denote
with Σ1 , ..., Σ J the connected components of Σ. For i = 1, 2 we define the integers lik (Σ) as
the number of components of Li that belong to Σk minus 1; in particular lik (Σ) > 0 for any
J
k, i. Finally, we say that li (Σ) = ∑ lik (Σ) = ni − J, where ni is the number of component
k =1
of Li . For example, if Σ is the cobordism in Figure 8.12 then J = 2, ordering Σ1 and Σ2
from above to bottom, we have that l1 (Σ) = 3 and l2 (Σ) = 4; while l11 (Σ) = 2, l12 (Σ) = 1,
l21 (Σ) = 3 and l22 (Σ) = 1. We have the following lemma.

Lemma 8.3.7 If Σ has no 0,2-handles then, up to rearranging 1-handles, we can suppose that Σ
is like in Figure 8.12: there are l1 (Σ) merge cobordisms between (0, t1 ), l2 (Σ) split cobordisms
between (t2 , 1) and g(Σ) torus cobordisms between (t1 , t2 ). We have no other 1-handles except for
the ones we considered before.

L2

L1

0 t1 t2 1

Figure 8.12: An example of a cobordism between two links after rearranging handles.

Proof. We consider a connected component Σk , which is a cobordism between L1k and L2k ,
and we fix a Morse function f : S3 × [0, 1] → [0, 1]. After some Reidemeister moves, from
CEU eTD Collection

[67] we know that we can assume that all the band moves are performed on disjoint bands,
in particular we can apply them in every possible order.
Since Σk has boundary in both L1k and L2k by the definition of cobordism given at begin-
ning of Section 1.3, if we take an ordering for the components of L1k , we can find a merge
cobordism joining the first and the second component of L1k at some point t in (0, 1); we
assume that the associated band move is the first we apply on L1k . Now we just do the same
thing on the other components, but taking the new component instead of the first two. In
this way we have that there is a t1 ∈ (0, 1) such that Σk ∩ f −1 [0, t1 ] is composed by l1k (Σ)
merge cobordisms and Σk ∩ f −1 (t1 ) is a knot.
In the same way we find that, for a certain t2 ∈ (0, 1), the cobordism Σk ∩ f −1 [t2 , 1] is
composed by l2k (Σ) split cobordisms and Σk ∩ f −1 (t2 ) is a knot.

121
 At this point Σk ∩ f −1 [t1 , t2 ] is a knot cobordism of genus g(Σk ) and from [67], see also
[51], we can rearrange the saddles to obtain a composition of g(Σk ) torus cobordisms like
in Figure 8.12.
To see that there are no other 1-handles left it is enough to compute the Euler charac-
teristic of Σk :

2 − 2g(Σk ) − l1k (Σ) − 1 − l2k (Σ) − 1 = χ(Σk ) = −|1-handles| .

This means that the number of 1-handles in Σk is precisely 2g(Σk ) + l1k (Σ) + l2k (Σ).

From Figure 8.12 we realize that merge and split cobordisms can appear alone in Σ
and not always in pair like in strong cobordisms. In case ii) and iii) of Subsection 8.3.1
we see that they do not induce isomorphisms in homology, but we find maps ΦSplit and
ΦMerge that are indeed isomorphisms if restricted to HF
\ L0 ( L1 ) → HF
\ L0 ( L2 ); moreover,
the filtered degree is one for split cobordisms and zero for merge cobordisms. Since we
are looking for information on τ, this is enough for our goal and then we can prove the
following inequality.

Proposition 8.3.8 Suppose Σ is a cobordism between two links L1 and L2 . Then

τ ( L1 ) − τ ( L2 ) 6 g(Σ) + max {l1 (Σ), l2 (Σ)} .

Proof. From Corollary 8.3.3 and Theorem 8.3.6, we can suppose that there are no 0 and no
2-handles in Σ. We can also assume that Σ is like in Lemma 8.3.7.
All of these cobordisms induce isomorphisms of the homology in Maslov grading zero.
The number of torus cobordisms is g(Σ) while the number of split cobordisms (that are not
part of torus cobordisms) is l2 (Σ). This means that

τ ( L 2 ) 6 τ ( L 1 ) + g ( Σ ) + l2 ( Σ ) .

Now we do the same, but considering the cobordism going from L2 to L1 , as we did in the
proof of Theorem 8.3.6. We obtain that

τ ( L 1 ) 6 τ ( L 2 ) + g ( Σ ) + l1 ( Σ ) .

Putting the two inequalities together proves the relation in the statement of the theorem.
CEU eTD Collection

If L is an n-component link, from Proposition 8.3.8 we have immediately that

|τ ( L)| + 1 − n 6 g4 ( L) (8.9)

which, as we already said, is a lower bound for the slice genus of our link. Indeed, we can
say more by using Equation (8.6) and observing that g4 ( L∗ ) = g4 ( L):

max |τ ( L)|, |τ ∗ ( L)| + 1 − n 6 g4 ( L) .



122
 8.4 The HF L− version of the homology
8.4.1 A different point of view
The collapsed filtered homology cHF L− can also be obtained from a grid diagram D
as the free F[V1 , ..., Vgrd( D)−n , V ]-module generated by the set of grid states S( D ). This ring
has one more variable V, compared to the ring we considered for GC c ( D ), associated to the
−
special O-markings. The differential ∂ is defined as following:

∑ ∑◦
O1 (r ) O (r )
∂− x = V1 · ... · Vm m · V O(r) y for any x ∈ S( D )
y∈S( D ) r ∈Rect ( x,y)

where m = grd( D ) − n and O(r ) is the number of special O-markings in r.

It is clear from the definition that
cGC − ( D ), ∂−
 

c ( D ), b
GC ∂ = .
V=0
The collapsed filtered link Floer homology cHF L− ( L) can be seen as the homology of our
new complex, in the same manner as in Subsection 8.1.4; moreover, now each level
F s cHF L− ( L) has also a structure of an F[U ]-module given by U [ p] = [Vi p] = [V p] for
every i = 1, ..., m and [ p] ∈ cHF L− ( L); see [67]. The groups F s cHF L−
d ( L ) are still finite
dimensional over F and so we can define the function N as
F s cHF L−
d ( L)
NL (d, s) = dimF .
F cHF L−
s − 1
d ( L)

We expect the function N to be a strong concordance invariant, possibly better than T.

We can compute the function N of the unknot:
(
1 if (d, s) = (2t, t), t 6 0
N (d, s) =
0 otherwise

and we know that

n −1
cHF L− ( L) ∼
=F[U ] cHF L− ( n)
∼
= F [U ] F [ U ] 2
as an F[U ]-module, where n is the number of component of L.
Since dimF cHF L0− ( L) is still equal to one, we can define an invariant ν exactly like
CEU eTD Collection

we did in Subsection 8.2.1 for τ. A version of the ν-invariant has been introduced first by
Rasmussen in [78] and he proved that it is a concordance invariant for knots. In [45] Hom
and Wu found knots whose ν-invariant gives better lower bound for the slice genus than
τ.
Since H∗,∗ gr(cGC − ( D ), ∂− ) is isomorphic to the homology cGH − ( L) ∼= cHFL− ( L)


−
of [67, 72] and NL (0, ν( L)) = 1 for every diagram D of L, we have that cHFL0,ν ( L)
( L) is
non-trivial. Hence, if the homology group cHFL0− ( L) is non-zero only for one Alexander
grading s, we can argue that ν( L) = s. This method can be used to compute the ν-invariant
of some links.
We can also define
the (uncollapsed) filtered link Floer homology as the homology of the
complex GC − ( D ), ∂− , where GC − ( D ) is the free F[V1 , ..., Vgrd ( D) ]-module over the grid
states of D; there are no special O-markings this time.

123
 We see immediately that for every n-component link L is

HF L− ( L) ∼
= F[U1 , ..., Un ]

with generator in Maslov grading zero, but the filtration will of course depend on L.

8.4.2 Filtered link homology and the τ-set

In this subsection we prove the following theorem.

Theorem 8.4.1 For an n-component link the τ-set, defined in [67] as -1 times the Alexander grad-
ings of a homogeneous, free generating set of the torsion-free quotient of cHFL− ( L) as an F[U ]-
module, coincides with the 2n−1 (with multiplicity) values of s where the function T is supported.

Proof. We use the complex cGC − to show that, for every n-component link L, an integer
s gives Td,s ( L) 6= 0 for some d if and only if s belongs to the τ-set of L. To do this, given
C a freely and finitely generated F[U ]-complex, we define two set of integers: τ (C) and
t(C). First we call Bτ (C) a homogeneous, free generating set of the torsion-free quotient
of H∗,∗ (gr(C)) as an F[U ]-module; then τ (C) is the set of s ∈ Z such that there is a [ p] ∈
Bτ (C) with bigrading (d, −s) for some d ∈ Z. Similarly, t(C) is the set of the integers s
such that the inclusion
C C
   
s −1 s
is : F H∗ ,−→ F H∗
U=0 U=0

is not surjective. Note that the set Bτ (C) is not unique, but τ (C) and t(C) are well-defined.
We say that s ∈ τ (C) has multiplicity k if there are k distinct elements in Bτ (C) with
bigrading (∗, −s), while a number u ∈ t(C) has multiplicity k if Coker iu has dimension k
as an F-vector space.
Clearly, if D is a grid diagram of L, t cGC − ( D ) is the set of the values of s where

the function TL is supported; moreover, we already remarked that H∗,∗ (gr (cGC − ( D )))
is isomorphic to cHFL− ( L) and then τ cGC − ( D ) is the τ-set of L. Hence our goal is to

prove that t cGC − ( D ) = τ cGC − ( D ) .

Consider the complex

cGC − ( D )
C= ,
V1 = ... = Vm = U
where m = grd( D ) − n and U is the variable associated to the special O-markings. Then
CEU eTD Collection

we define C as the complex C J1 − n − m, −mK.

We introduce a new complex C 0 = C ⊗F[U ] F[U, U −1 ] and we define a Z ⊕ Z-filtration
on C 0 in the following way: F x,∗ C 0 = U − x C for every x ∈ Z, F ∗,y C 0 = F y C 0 = p ∈


C 0 | A( p) 6 y for every y ∈ Z and F x,y C 0 = F x,∗ C 0 ∩ F ∗,y C 0 = U −x F y−x C for every

x, y ∈ Z. The first step is to prove that τ (C) = t(C).
We have that
C ∼ f F 0,∗ C 0
= GC ( D ) ∼= −1,∗ 0 ;
U=0 F C
moreover, for every integer s we claim that

f (D) ∼ F 0,s C 0
F s GC = −1,s 0 .
F C

124
 Since, from [67], it is F x,y C 0 ∼
= F y,x C 0 for every x, y ∈ Z, then we have that
C ∼ sf F s,0 C 0
Fs = F GC ( D )J1 − n − m, −mK ∼
= s,−1 0 . (8.10)
U=0 F C
Using this identification we obtain that t(C) coincides with set of s ∈ Z such that the map
 s−1,0 0   s,0 0 
F C F C
H∗ ,−→ H∗ (8.11)
F s−1,−1 C 0 F s,−1 C 0
is not surjective.
M F 0,t C 0
Now we consider the complex gr(C), which is equal to . We have that the
t ∈Z
F 0,t−1 C 0
map
F 0,t C 0 F −t,0 C 0
Ut : −→
F 0,t−1 C 0 F −t,−1 C 0
F − t,0 C 0 F ∗,0 C 0
is an isomorphism and −t,−1 0 is a subspace of ∗,−1 0 for every integer t. From a
F C F C
gr(C)
result in [67] the latter filtered complex is isomorphic to , but it is also isomorphic to
U=1
f ( D )J1 − n − m, −mK for Equation (8.10). In this way we can define a surjective map
GC
gr(C)
Ψ : gr(C) −→
U=1
and it is easy to see that Ψ (Bτ (C)) is still a homogeneous, free generating set of the homol-
ogy; furthermore, if [ p] is a torsion element in H∗,∗ (gr(C)) then [Ψ( p)] = [0]. This means
that τ (C) is the set of −t ∈ Z such that the map
 −t−1,0 0   −t,0 0 
F C F C
H∗ ,−→ H∗ (8.12)
F −t−1,−1 C 0 F −t,−1 C 0
is not surjective.
If we change −t with s in Equation (8.12) then we immediately see that it coincides
with Equation (8.11) and so τ (C) = t(C). Moreover, we can say that an integer in τ (C) has
multiplicity k if and only if it has multiplicity k in t(C). Finally, since the map U t drops the
Maslov grading by 2t, we have that if there is a [ p] ∈ Bτ (C) with bigrading (d, s) then there
is a generator of GH(
g D )J1 − n − m, −mK with grading and minimal level (d − 2s, −s).
The second and final step is to show that the previous claim implies t cGC − ( D ) =

τ cGC − ( D ) . From Lemma 8.2.4 we have that the filtered quasi-isomorphisms C ∼

=
CEU eTD Collection

−
cGC ( D ) ⊗ W ⊗ m ∼
and C = cGC ( D ) ⊗ (W ) , where W is the two dimensional F-vector
− ∗ ⊗ m

space with generators in grading and
minimal level (d, s) = (0, 0) and (d, s) = (−1, −1).
Thus t cGC − ( D ) and τ cGC − ( D ) are completely determined by τ (C) and t(C), so this

means that they coincide and the proof is complete. Obviously, the conclusions about
multiplicities and Maslov shifts are still true.

From [67] we know that there is only one element [ p] ∈ Bτ (cGC − ( D )) with bigrading
(−2τ1 , −τ1 ) and only another one [q] with bigrading (−2τ2 + 1 − n, −τ2 ). Then the proof
of Theorem 8.4.1 implies that there are two non-zero elements in HF \ L( L) in grading and
minimal level (0, τ1 ) and (1 − n, τ2 ). Since, from the definition of τ and τ ∗ , we also know
that there are only two generators of HF \ L( L) in Maslov grading zero and 1 − n; we have
the following corollary.

125
 Corollary 8.4.2 Take a grid diagram D of an n-component link L and consider a set
Bτ (cGC − ( D )). If [ p], [q] ∈ Bτ (cGC − ( D )) are such that [ p] is in bigrading (−2τ1 , −τ1 ) and
[q] is in bigrading (−2τ2 + 1 − n, −τ2 ) then τ ( L) = τ1 and τ ∗ ( L) = τ2 .

8.5 Applications
8.5.1 Computation for some specific links
In general it is hard to say when a sum of grid states is a generator of the homology,
but the following lemma provides an example where we have useful information.
Lemma 8.5.1 Suppose L is an n-component link with grid diagram D and x ∈ S( D ) as in Figure
8.13. Then [ x ] is always the generator of HF
\ L0 ( L) and cHF L0− ( L). Furthermore, τ ( L) =
ν ( L ) = A ( x ).

Figure 8.13: We denote with x the grid state in the picture.

∂x = ∂− x = 0 and that for every other grid state y of D it

Proof. We show that M( x ) = 0, b
is M (y) 6 0.
i) The fact that M ( x ) = 0 is trivial.
ii) For every y ∈ S( D ) there are always 2 rectangles in Rect◦ ( x, y) and they contain no O,
so they cancel when we compute the differential.
iii) We prove by induction on grd( D ) that M (y) 6 0 for every y ∈ S( D ).
If grd( D ) = 1 then x is the only grid state.
If grd( D ) = 2 then there are only x and y and it is M (y) = −1.
CEU eTD Collection

Suppose the claim is true for the diagrams with dimension equal or smaller than α and
let grd( D ) = α + 1. We denote with I ( D ) ⊂ S( D ) the subset of grid states that contain
the point (0, α) as in Figure 8.14. Every y ∈ I ( D ) is the extension of a grid state y0 of
the diagram D 0 obtained by removing the first column and the last row from D. By
the inductive hypothesis we have M(y) = −1 + M(y0 ) 6 −1.
Now it easy to see that every other z ∈ S( D ) is obtained by a rectangle move from a
y ∈ I ( D ). Then, if r is the rectangle, we have

M(z) − M(y) = 1 − 2 · r ∩ O + 2 · Int(r ) ∩ y ,

but Int(r ) ∩ y 6 min{π1 (r ), π2 (r )} = r ∩ O where πi (r ) is the lenght of the edges

of r. Hence M (z) 6 0.

126
 (0, α)

Figure 8.14: The state y ∈ I ( D ) is marked with the black circles.

In Lemma 8.5.1 we used that the grid diagram D has all the O-markings aligned on a
diagonal. It is easy to see that if a link admits such a diagram then it is positive. On the
other hand, it seems difficult for the converse to be true.

8.5.2 Torus links

We compute the τ-invariant of every torus link. Consider the grid diagram Dq,p in Fig-
ure 8.15, representing the torus link Tq,p with q 6 p and all the components oriented in the
same direction. From Lemma 8.5.1 we know that [ x ] is the only generator of HF\ L0 ( Tq,p ).

X
X p
X
X
X
X q
X

Figure 8.15: x is the grid state in the picture.

If we denote with n the number of components of Tq,p then a simple computation gives

1 1
CEU eTD Collection

A( x ) = M ( x ) − MX ( x ) − grd( Dq,p ) + n = (− MX ( x ) − p − q + n) =
2" # 2
q −1
1 1
= 2 ∑ i + q ( p − q + 1) − p − q + n = [ q ( q − 1) + q ( p − q + 1) − p − q + n ] =
2 i =1
2
1 ( p − 1)(q − 1) − 1 + n
= ( pq − p − q + n) = .
2 2
Now we use Lemma 8.5.1 again and obtain that

( p − 1)(q − 1) − 1 + n
τ ( Tq,p ) = .
2

127
 Using the lower bound of Equation (8.9) gives a different way to compute the slice genus
of a torus link with respect to what we did in [8]:

( p − 1)(q − 1) 1 − n ( p − 1)(q − 1) + 1 − n
g4 ( Tq,p ) > − +1−n = .
2 2 2
Since the Seifert algorithm applied to the standard diagram of Tq,p gives the opposite in-
equality, we conclude that

( p − 1)(q − 1) + 1 − n
g4 ( Tq,p ) = for any q 6 p .
2

8.5.3 Applications to Legendrian invariants

We equip S3 with its unique tight contact structure ξ st , whose definition can be found
in Section 3.4. We remark that, if D = D1 ∪ ... ∪ Dn is a front projection of the Legendrian
link L in the standard contact 3-sphere, the Thurston-Bennequin and rotation number of
L are given by
n n
tb(L) = ∑ tbi (L) and rot(L) = ∑ roti (L)
i =1 i =1

where
1
tbi (L) = wr(Di ) + lk (Di , D \ Di ) − |cusps in Di |
2
and  
1
roti (L) = |down-ward cusps in Di | − |up-ward cusps in Di | .
2
More information can be found in Section 3.2.

Proposition 8.5.2 Consider a Legendrian n-component link L of link type L in S3 equipped with
the standard contact structure. Then the following inequality holds:

tb(L) + | rot(L)| 6 2τ ( L) − n . (8.13)

Proof. If L is a Legendrian link then, from [67], we know that L can be represented by a
grid diagram D of the link L∗ (the mirror of L). This diagram D is such that

tbi (L) − roti (L) + 1 tbi (L) + roti (L) + 1

= Ai ( x + ) = Ai ( x − )
CEU eTD Collection

2 2
tb(L) − rot(L) + 1 = M( x + ) tb(L) + rot(L) + 1 = M ( x − ) ,
where x ± are the grid states in D obtained by taking a point in the northeast (southwest for
x − ) corner of every square decorated with an X ∈ X. Moreover, Ai is defined as follows:

grd( D )i − 1
 
1
A i ( x ) = J x − ( X + O ) , ( Xi − Oi ) − for any x ∈ S( D )
2 2

with Oi ⊂ O and Xi ⊂ X the markings on the i-th component of L∗ and grd( D )i the
number of elements in Oi .
In [67] is proved that x ± represent non-torsion elements in the homology group
cHFL− ( L∗ ); in fact these classes are the Legendrian grid invariants λ± (L). Since A( x ) =

128
 n
∑ Ai (x) for every grid state x, we have that M (x± ) = 2A (x± ) + 1 − n. Therefore, Corol-
i =1
lary 8.4.2 implies that A ( x ± ) 6 −τ ∗ ( L∗ ). Combining the latter claim with Corollary 8.2.6,
that gives τ ∗ ( L∗ ) = −τ ( L), we have
tb(L) ∓ rot(L) + n
= A( x± ) 6 τ ( L)
2
that is precisely Equation (8.13).

From Equation (8.13), together with Equation (8.9), we obtain the following lower
bound for the slice genus:

tb(L) + | rot(L)| 6 2g4 ( L) + n − 2 6 2u( L) − n , (8.14)

where u( L) is the unlinking number of L; see [55].

This bound is sharp for positive torus links, but here we show that there are other links
for which this happens.
In Figure 8.16 we have a front projection D of a Legendrian two component link L.
n . A simple computation gives tb(L) = 6 and rot(L) = 0,
The link type of L is the link L919
n ) > 3. Since it is easy to see that the link repre-
therefore Equation (8.14) says that g4 ( L919
sented by D can be unlinked by changing the four crossings highlighted in the picture, we
n ) 6 u ( L9n ) − 1 6 3 and then we conclude that g ( L9n ) = 3.
have g4 ( L919 19 4 19

n .
Figure 8.16: A diagram of the link L919

From Equation (8.13) we also have the following upper bound for the maximal
Thurston-Bennequin number.
CEU eTD Collection

Proposition 8.5.3 For every n-component link L we have

TB( L) 6 2τ ( L) − n .

Furthermore, if L is a quasi-alternating link then we have that

TB( L) 6 −1 − σ( L) .

Although this bound is much less powerful than the Kauffmann or the HOMFLY poly-
nomial, we can still get some interesting conclusions.
Consider a Legendrian link L such that each component Li is algebraically unlinked.
Then tbi (L) = tb(Li ) and so tb(L) is precisely the sum of the Thurston-Bennequin num-
bers of its components. For example this happens for the Borromean link B, whose com-
ponents Bi are three (algebraically unlinked) unknots. It was shown in [61] that there is no

129
 −2k +2k

Figure 8.17: A diagram of Lk . For k = 0 we have the link L940

a .

Legendrian representation of B, where the Thurston-Bennequin number of each compo-

nent is -1; in fact we have TB( B) = −4, while TB( ) = −1. In particular, this means that
the difference between TB( B) and the sum of TB( Bi ) is -1.
We prove Proposition 8.5.4, where we give a family of 2-components links Lk such that
the components of Lk are two unknots with lk( L1k , L2k ) = 0 and the difference between
TB( Lk ) and the sum of TB( Lik ) is actually arbitrarily small, improving the latter result for
B. The links Lk are shown in Figure 8.17.

Proposition 8.5.4 The links Lk in Figure 8.17 are a family of two component links, whose compo-
nents Lik are unknots with linking number zero, such that TB( Lk ) is arbitrarily small.

Proof. Since for every k > 0 the link Lk is non-split alternating, we can easily compute
the signature that is equal to 3 + 2k. Now we apply Proposition 8.5.3 and we obtain that
TB( Lk ) 6 −4 − 2k.

As a final obervation, we note that the same argument used by Baldwin, Vela-Vick and
Vértesi in [3] gives that the invariant L( L) defined in Chapter 6, for Legendrian links in
the standard contact 3-sphere, coincides with the grid invariant λ+ ( L) of [67].
CEU eTD Collection

130
 Chapter 9

Quasi-positive links in the 3-sphere

9.1 Maximal self-linking number and the slice genus

A quasi-positive link is any link which can be realized as the closure of a d-braid of the
form
b
∏ wi σj wi−1 ,
i
i =1

where σj for j = 1, ..., d − 1 are the generators of the d-braids group that we defined in
Subsection 3.4.2. Thus quasi-positive links are closures of braids consisting of arbitrary
conjugates of positive generators.
It is worth noting that quasi-positive links are equivalent to another, more geometric
class of links: the transverse C-links. These links arise as the transverse intersection of the
3-sphere S3 ⊂ C2 , with a complex curve. Transverse C-links include links of isolated curve
singularities, but are in fact a much larger class. The fact that quasi-positive links can be
realized as transverse C-links is due to Rudolph [80], while the fact that every transverse
C-link is quasi-positive is due to Boileau and Orekov [6].
Given a quasi-positive braid B = (w1 σj1 w1−1 ) · ... · (wb σjb wb−1 ), we can associate to B a
surface Σ B as follows. Let us consider the braid B0 , obtained by removing the σji ’s from

B B0
CEU eTD Collection

ΣB

r1 r2

Figure 9.1: The disk Σ B constructed from the quasi-positive braid B, representing the un-
knot. The segments r1 and r2 are ribbon intersections.

131
 the presentation of B. Then B0 is the boundary of d disks with some ribbon interesctions
between themselves. If we push these disks in the 4-ball D4 then the intersections disap-
pear and we obtain a surface which is properly embedded in D4 . At this point, we add b
positive bands in correspondence of the σji ’s; the result is an oriented and compact surface
that is not embedded in S3 , but it is properly embedded in D4 , whose boundary is clearly
the closure of the braid B. See Figure 9.1 for an example.
A connected transverse C-link is a link which is the closure of a quasi-positive braid B
such that Σ B is connected. Then we have the following proposition.

Proposition 9.1.1 Every connected transverse C-link is non-split quasi-positive.

Proof. Suppose that B is a quasi-positive braid, presented as before, such that Σ B is con-
nected and its closure L is a split link. Then we can write L = L1 t L2 and Li is the closure
of the subbraid Bi for i = 1, 2.
Since lk( L1 , L2 ) = 0, we have that the arcs corresponding to the σji ’s both belong to
B1 or B2 . Therefore, there is no positive band connecting a disk, which appears in the
construction of Σ B1 , with the ones in Σ B2 . This implies that Σ Bi is a connected component
of Σ B for i = 1, 2 and this is a contradiction.

Later, in Corollary 9.2.4 we show that this inclusion is strict.

The Thurston-Bennequin inequality stated in Section 3.6 can be improved using the
invariant τ ( L), defined from the filtered link Floer homology group HF
\ L( L) in Chapter
8. More specifically, in Subsection 8.5.3 we prove Equation (8.13):

tb(L) + | rot(L)| 6 2τ ( L) − n ,

where L is an n-component Legendrian link in the tight 3-sphere. From this result we
obtain the following corollary.

Corollary 9.1.2 Suppose that L is an n-component Legendrian link in (S3 , ξ st ) with smooth link
type L. Then we have that

sl(TL± ) 6 tb(L) + | rot(L)| 6 2τ ( L) − n 6 −χ(Σ) , (9.1)

where Σ is an oriented, compact surface, properly embedded in D4 , such that ∂Σ = L.

Proof. The first two inequalities follow from Proposition 3.3.2 and Equation (8.13). The last
one is a consequence of Proposition 8.3.8.
CEU eTD Collection

It is interesting to observe that Equation (9.1) is a version of the adjunction inequality,

see [52, 56].
We recall that, from the Thurston-Bennequin inequality, a transverse link in (S3 , ξ st )
has bounded self-linking number. This means that we can denote the maximal self-linking
number of a link L with SL( L).

Theorem 9.1.3 Suppose that L is an n-component quasi-positive link in S3 . Then we have that

b−d+n
SL( L) = b − d and τ ( L) = ,
2

where L is the closure of a quasi-positive d-braid B = (w1 σj1 w1−1 ) · ... · (wb σjb wb−1 ).

132
 Proof. As we saw in Subsection 3.4.2, the braid B determines a transverse link T in (S3 , ξ st )
and
sl(T ) = wr( B) − d = b − d .
Then from Theorem 6.6.5 and Equation (9.1) we have that
sl(T ) 6 SL( L) 6 2τ ( L) − n 6 −χ(Σ B ) = b − d
and these inequalities are all equalities.
When we consider connected surfaces, the inequality in Equation (9.1) becomes
sl(T ) 6 2τ ( L) − n 6 2g4 ( L) + n − 2 , (9.2)
for every n-component transverse link T in (S3 , ξ st ), with link type L, and where we recall
that g4 ( L) is the minimal genus of a compact, oriented surface Σ properly embedded in D4
and such that ∂Σ = T , see Section 1.3. Then we have the following proposition.
Proposition 9.1.4 Suppose that L is an n-component link, embedded in S3 , which is a connected
transverse C-link. Then we have that τ ( L) = g4 ( L) + n − 1.
Proof. The claim follows from the same argument in the proof of Theorem 9.1.3 and Equa-
tion (9.2).
An implication of this proposition is the additivity of the slice genus under connected
sums.
Corollary 9.1.5 The slice genus g4 is additive under connected sums if we restrict to the family of
connected transverse C-links.
Proof. If L1 and L2 are two connected transverse C-links then it is easy to see that every
connected sum L = L1 # L2 has the same property. In fact, take two quasi-positive braids,
representing L1 and L2 ; then L is obtained by putting the second below the first one and
adding a positive band between the components that we want to sum, say the i-th and the
j-th ones with i < j. This move can be seen as the composition of B with wσj−1 w−1 for
some w and then the resulting braid is still quasi-positive.
Therefore, Proposition 9.1.4 gives that τ ( Li ) = g4 ( Li ) + ni − 1, where ni are the num-
ber of components of Li for i = 1, 2. Moreover, what we said before also implies that
τ ( L) = g4 ( L) + (n1 + n2 − 1) − 1. At this point, we use that the invariant τ is additive
under connected sums of links, see Equation (8.7) in Subsection 8.2.3, and then the proof
is complete.
In Theorem 9.1.3 we showed a way to compute the τ-invariant for quasi-positive links,
CEU eTD Collection

using braids. It is known that a similar formula holds for the Rasmussen link invariant s,
see [76, 83]. Then we can prove Proposition 9.1.6, which says that the two invariants are
related to each other in this case.
Proposition 9.1.6 Take an n-component quasi-positive link L in the 3-sphere. Then the invariant
τ ( L) from link Floer homology and the invariant s( L) from Khovanov homology satisfy the relation
s( L) = 2τ ( L) + 1 − n .
Proof. The invariant s( L) can be computed from a quasi-positive braid B, presented as
usual in this section, in the following way:
s( L) = b − d + 1 .
This is proved in [76, 83]. Then the statement follows immediately from Theorem 9.1.3.

133
 9.2 Strongly quasi-positive links
We apply the results obtained in Chapter 8 and in the previous subsection to recompute
the invariant tau for strongly quasi-positive links in S3 . A link L ,→ S3 is strongly quasi-
positive if it is the boundary of a quasi-positive surface F. Such surfaces are constructed

Figure 9.2: The quasi-positive surface determined by d = 4 and ~b = (σ13 , σ24 , σ13 , σ24 ).

in the following way: take d disjoint parallel, embedded disks, all of them oriented in the
same way, and attach b negative bands on them, each one between a pair of distinct disks.
This procedure is shown in Figure 9.2. The negative bands cannot be knotted with each
other; this means that, up to isotopy, F only depends on the number d and the ordered
b-tuple ~b = (σi1 j1 , ..., σib jb ), where σij denotes that a negative band is put between the i-th
and the j-th disk with i < j.
Denote with σ1 , ..., σd−1 the generators of the d-braids group. Then it is easy to see that
the boundary of the quasi-positive surface (1, σij ) is isotopic to the closure of the d-braid
given by
(σi · · · σj−2 )σj−1 (σi · · · σj−2 )−1 if j > i + 2
(
;
σi if j = i + 1
see [44] for more details. This immediately implies that strongly quasi-positive links are
quasi-positive.
Suppose S3 is equipped with its unique tight contact structure ξ st and L ,→ (S3 , ξ st )
CEU eTD Collection

is an n-component transverse link with smooth link type L. Then we have the following
corollary.

Corollary 9.2.1 Suppose that T is an n-component transverse link in (S3 , ξ st ) with smooth link
type L. Then we have that

sl(T ) 6 2τ ( L) − n 6 k Lk T − o ( L) ,

where o ( L) is the number of disjoint unknots in L.

Proof. The first inequality follows immediately from Equation (9.2). For the second one we
[ ∗,∗ ( L) is non-zero in
apply Theorem 2.1.2; in fact, it is known, see [67], that the group HFL
bigrading (0, τ ( L)).

134
 We recall that k Lk T is the Thurston norm of L, see Chapter 2. Then we can now prove
the following theorem.
Theorem 9.2.2 Consider a strongly quasi-positive link L with n components in S3 . Then the
Thurston-Bennequin inequality is sharp for L, in the sense that there exists a transverse represen-
tative T of L such that
sl(T ) = 2τ ( L) − n = k Lk T − o ( L) = −χ( F ) ,
where o ( L) is the number of disjoint unknots in L and F is a quasi-positive surface for L. In
i
i

j j
Figure 9.3: Transverse realization of L1 .

particular
k LkT − o ( L) + n b−d+n
τ ( L) = = ,
2 2
where F consists of d disks and b negative bands.
Proof. From Lemma 2.2.5 and Corollary 9.2.1 it is enough to show that, given a quasi-
positive surface F whose boundary is L, we can find a transverse link T , with smooth link
type L, such that sl(T ) = b − d.
Let us start with the d parallel disks in F and see what happens when we attach the
first negative band. The surface we obtain is the quasi-positive surface determined by d
and ~b = (σij ), where i < j correspond to the disks on which we glue the negative band,
and we call L1 its boundary.
At this point, we choose a transverse representative of L1 as shown in Figure 9.3. Then
T is defined by iterating this procedure with all the b bands and, since we computed its
self-linking number in the previous section, we obtain that
sl(T ) = b − d .
This proves the claim.
CEU eTD Collection

Since we remarked in the proof of Corollary 9.2.1 that HFL [ 0,τ ( L) ( L) 6= {0}, Theorems
2.1.2 and 9.2.2 tell us that for strongly quasi-positive links
n o
τ ( L) = max s ∈ Z | HFL
[ ∗,s ( L) 6= {0} .

Furthermore, applying Lemma 2.2.5 to Theorem 9.2.2, we can say more when the link L
also bounds a connected quasi-positive surface, that we call quasi-positive Seifert surface for
L.
Corollary 9.2.3 If L is a link as in Theorem 9.2.2, which also admits a quasi-positive Seifert surface
F 0 , then we have that
τ ( L ) = g3 ( L ) + n − 1 = g ( F 0 ) + n − 1 ,
where g3 ( L) is the Seifert genus of L.

135
 This corollary implies that all the quasi-positive Seifert surfaces of a given link have
the same genus. Moreover, we observe that not every strongly quasi-positive link admits
a quasi-positive Seifert surface. In fact, the 4-component link L in Figure 9.2 is such that
τ ( L) = 2 from Theorem 9.2.2, but L has Seifert genus zero: to see this it is enough to add a
tube between disks 1 and 2 in Figure 9.2. Then we have that

2 = τ ( L ) 6 = g3 ( L ) + n − 1 = 3

and so this would contradict Corollary 9.2.3. This argument also implies the following
corollary.

Corollary 9.2.4 The link L in Figure 9.2 is non-split quasi-positive, but it is not a connected
C-transverse link.

Proof. By construction L is non-split strongly quasi-positive and then it is quasi-positive. If

we suppose that L is a connected transverse C-link then we can use Proposition 9.1.4 and
we obtain that
g4 ( L ) = τ ( L ) + 1 − n = − 1
which is clearly impossible.

9.3 The case of positive links

We recall that, from [81], positive links are always strongly quasi-positive. Moreover,
if a link is positive non-split then it also admits a quasi-positive Seifert surface. This means
that the results in Section 9.2 also hold in this case. Moreover, we state a proposition from
[49], which tells us that the maximal Thurston-Bennequin number TB of a positive link can
be determined easily.

Proposition 9.3.1 (Kálmán) Suppose that L is an n-component link in S3 with a positive diagram
D. Then we have that
TB( L) = c( D ) − k ( D ) ,
where c( D ) and k ( D ) are the numbers of crossings in D and of circles in the oriented resolution of
D.

We want to prove the following corollary.

CEU eTD Collection

Corollary 9.3.2 Suppose that L is a link as in Proposition 9.3.1 with split components L1 , ..., Lr .
Then we have that
TB( L) = 2τ ( L) − n = 2g3 ( L) + n − 2r .

When a link has a diagram D which is positive, the oriented resolution defined in Sub-
section 1.2.2 happens to possess some peculiar properties. Let us start with the following
lemma.

Lemma 9.3.3 Suppose that D is a positive diagram for a link L and consider D 0 ⊂ D a subdiagram,
which represents a split component L0 of L.
Then the surface F 0 , obtained from the oriented resolution of D 0 , is contained as subset in F, the
one obtained in the same way from D.

136
 Proof. The surface F 0 is completely determined by K 0 , the collection of circles in the ori-
ented resolution of D 0 , and the crossings in D 0 . Denote with K the circles in the oriented
resolution of D. If C ∈ K 0 then C ∈ K; this is because lk( L0 , L \ L0 ) = 0, since L0 is a split
component of L. Therefore, there is no crossing between D 0 and D \ D 0 , because D is posi-
tive, and then neither between K 0 and K \ K 0 . Moreover, the crossings in D 0 also appear in
D because D 0 is a subdiagram of D.
It follows that F 0 ⊂ F and F \ F 0 is precisely the surface obtained from the oriented
resolution of D \ D 0 . This appears clearly from Figure 1.7.

In order to prove Proposition 9.3.1 we need another lemma.

Lemma 9.3.4 Suppose that D is a positive diagram for a link L = L1 t ... t Lr , where { Li }ri=1 are
the split components of L.
Then the surface F, obtained from the oriented resolution of D, can be isotoped into F1 t ... t Fr ;
where each Fi is a Seifert surface for Li . Furthermore, Fi is obtained from the oriented resolution of
Di ⊂ D, the subdiagram representing Li , for every i = 1, ..., r.

Proof. Let us work on the case when L is non-split first. The surface F is compact and
oriented, but now it is also connected; in fact, otherwise we could write D = D1 ∪ D2
where there is no crossing between D1 and D2 . This implies that, if D ⊂ {z = 0} in R3 , we
could push D2 up in the plane {z = 1} and then move it away from D1 . This determines
an isotopy of S3 and then L would be split, which is a contradiction.
Now, for the general case, we observe that, from Lemma 9.3.3 and the non-split case we
did before, the surface F possesses as connected components the surfaces { Fi }ri=1 each one
being a Seifert surface for Li ; moreover, the Fi ’s are obtained from the oriented resolution
of the subdiagrams Di ⊂ D that represent Li for i = 1, ..., r.
We want to isotope F in a way that it becomes the disjoint union of F1 , ..., Fr . To do
this we consider the planar subspaces D and {Di }ri=1 of R2 , associated to D and { Di }ri=1
respectively as in Subsection 1.2.2, and we call Fi an innermost component of F if only one
component of R2 \ Di intersects F. Hence, we reason in the same way as before and we
push Fi slightly up from the plane and move it away from F \ Fi . We repeat this procedure
for all the components of F and at the end the surface will be as we wanted.

We are now ready to prove Corollary 9.3.2.

Proof of Proposition 9.3.1 and Corollary 9.3.2. We suppose first that L is a non-split link. Then
from Corollaries 9.1.2 and 9.2.1 we know that each Legendrian representative of L satisfies
CEU eTD Collection

the following inequality:

tb(L) 6 TB( L) 6 2τ ( L) − n = 2g3 ( L) + n − 2 6 −χ( F ) = c( D ) − k ( D ) ; (9.3)

where F is the surface obtained from the oriented resolution of D, which is a Seifert surface
for Lemma 9.3.4. The first equality follows from the fact that a non-split positive link
admits a quasi-positive Seifert surface and Corollary 9.2.3.
We define the Legendrian link L in the following way. Let us consider the planar
subspace D of R2 , associated to D, and the set K consisting of the circles in the oriented
resolution. We isotope the circles in K, starting from the innermost ones, until they have
only two vertical tangency points, one on the left and the other one on the right of the
circle. At this point, we isotope D in a way that the lines between the circles in K appear

137
 D

Figure 9.4: Legendrianization of an oriented resolution.

vertical in the plane, see Figure 9.4; this is done by moving the innermost ones first as
before, more details can be found in [49].
Then we just change the vertical tangencies into cusps and the vertical lines into cross-
ings; we obtain the front projection of a Legendrian link because D is positive and then
all the lines correspond to negative bands. An easy computation gives that tb(L) =
c( D ) − k ( D ) and rot(L) = 0, which means that all the inequalities in Equation (9.3) are
equalities.
Suppose now that L = L1 t ... t Lr . Take the Legendrian links Li , representing Li
for i = 1, ..., r, defined following the same procedure as before and denote with L the
Legendrian link L1 t ... t Lr . Then we have that
r r
tb(L) = ∑ tb(Li ) = ∑ c( Di ) − k( Di ) = c( D) − k( D) , (9.4)
i =1 i =1

where Di ⊂ D is the subdiagram representing Li . Here the final equality follows from
Lemma 9.3.4. The Thurston-Bennequin inequality in Corollary 9.2.3 again gives
r
tb(L) 6 TB( L) 6 2τ ( L) − n = k Lk T − o ( L) 6 ∑ (2g3 ( Li ) + ni − 1) 6
i =1
r
(9.5)
6 − ∑ χ( Fi ) = c( D ) − k( D ) ,
CEU eTD Collection

i =1

where Fi are the Seifert surfaces obtained from the oriented resolution of Di for i = 1, ..., r,
the number o ( L) tells us how many disjoint unknots there are in L and ni is the number
of components of each Li . We also used Proposition 2.2.2. Clearly, Equation (9.4) says that
all the inequalities in Equation (9.5) are equalities and then the claim follows from the fact
that
r r
∑ (2g3 ( Li ) + ni − 1) = 2 ∑ g3 ( Li ) + n − r = 2g3 ( L) + n − r ,
i =1 i =1

where the final equality holds because the Seifert genus is additive under disjoint unions.

138
 The invariant τ ( L) of a positive link can be determined from the oriented resolution of
a positive diagram D for L. In fact, Proposition 9.3.1 and its proof immediately imply the
following corollary.

Corollary 9.3.5 Suppose L is a positive n-component link in S3 with r split components. Then we
have that
r
c( D ) − k( D ) + n
τ ( L) = ∑ g( Fi ) + n − r = ,
i =1
2
where { Fi }ri=1 are the connected components of the surface obtained from the oriented resolution of
a positive diagram D for L, while c( D ) and k ( D ) are as in Proposition 9.3.1.

In particular, all the surfaces obtained from the oriented resolution of a positive dia-
gram for a given link have the same genus. Furthermore, Corollary 9.3.5 also gives that a
positive link admits a quasi-positive Seifert surface if and only if it is non-split.
CEU eTD Collection

139
CEU eTD Collection
 Bibliography

[1] M. F. Atiyah and I. G. Macdonald, Introduction to commutative algebra, Addison-Wesley

Publishing Co., Reading, Mass.-London-Don Mills, Ont. 1969, pp. ix+128.

[2] K. Baker and J. Etnyre, Rational linking and contact geometry, Progr. Math., 296,
Birkhäuser/Springer, New York, 2012.

[3] J. Baldwin, D. S. Vela-Vick and V. Vértesi, On the equivalence of Legendrian and transverse
invariants in knot Floer homology, Geom. Topol., 17 (2013), no. 2, pp. 925-974.

[4] D. Bennequin, Entrelacements et équations de Pfaff, Astérisque, 107-108 (1983), pp. 87-161.

[5] J. Birman, Braids, Links, and Mapping Class Groups, Annals of Math. Studies 82, Prince-
ton University Press, Princeton 1975.

[6] M. Boileau and S. Orevkov, Quasipositivite dune courbe analytique dans une boule pseudo-
convexe, C. R. Acad. Sci. Paris., 332 2001, pp. 825-830.

[7] D. Calegari, Foliations and the geometry of 3-manifolds, Oxford Mathematical Mono-
graphs, Oxford University Press, Oxford, 2007.

[8] A. Cavallo, On the slice genus and some concordance invariants of links, J. Knot Theory
Ramifications, 24 (2015), no. 4, 1550021.

[9] A. Cavallo, The concordance invariant tau in link grid homology, to appear on Alg. Geom.
Topol., arXiv:1512.08778.

[10] A. Cavallo, An invariant of Legendrian and transverse links from open book decompositions
of contact 3-manifolds, arXiv:1711.09508.
CEU eTD Collection

[11] A. Cavallo, On loose Legendrian knots in rational homology spheres, Topology Appl., 235
(2018), pp. 339-345.

[12] A. Cavallo, On Bennequin type inequalities for links in tight contact 3-manifolds,
arXiv:1801.00614.

[13] V. Colin, Chirurgies d’indice un et isotopies de sphères dans les variétés de contact tendues,
C. R. Acad. Sci. Paris Sér. I Math., 324 (1997), pp. 659-663.

[14] O. Dasbach and B. Mangum, On McMullen’s and other inequalities for the Thurston norm
of link complements, Algebr. Geom. Topol., 1 (2001), pp. 321-347.

[15] F. Ding and H. Geiges, A Legendrian surgery presentation of contact 3-manifolds, Math.
Proc. Cambridge Philos. Soc., 136 (2004), no. 3, pp. 583-598.

141
 [16] F. Ding and H. Geiges, A unique decomposition theorem for tight contact 3-manifolds, En-
seign. Math. (2), 53 (2007), no. 3-4, pp. 333-345.

[17] K. Dymara, Legendrian knots in overtwisted contact structures on S3 , Ann. Global Anal.
Geom., 19 (2001), no. 3, pp. 293-305.

[18] Y. Eliashberg, Classification of overtwisted contact structures on 3-manifolds, Invent.

Math., 98 (1989), no. 3, pp. 623-637.

[19] Y. Eliashberg, Topological characterization of Stein manifolds of dimension > 2, Int. J. of

Math., 1 (1990), pp. 29-46.

[20] Y. Eliashberg, On symplectic manifolds with some contact properties, J. Diff. Geom., 33
(1991), pp. 233-238.

[21] Y. Eliashberg, Contact 3-manifolds twenty years since J. Martinet’s work, Ann. Inst.
Fourier (Grenoble), 42 (1992), no. 1-2, pp. 165-192.

[22] Y. Eliashberg, Classification of contact structures on R3 , Internat. Math. Res. Notices.,

(1993), no. 3, pp. 87-91.

[23] Y. Eliashberg, Legendrian and transversal knots in tight contact 3-manifolds, Topological
methods in modern mathematics, pp. 171-193, Publish or Perish, Houston, TX, 1993.

[24] Y. Eliashberg and M. Fraser, Classification of topologically trivial Legendrian knots, CRM
Proc. Lecture Notes, 15, Amer. Math. Soc., Providence, RI, 1998.

[25] Y. Eliashberg and M. Gromov, Convex symplectic manifolds, Proc. Sympos. Pure Math.,
52, Part 2, Amer. Math. Soc., Providence, RI, 1991.

[26] J. Etnyre, Legendrian and transversal knots. Handbook of knot theory, Elsevier B. V., Ams-
terdam, 2005, pp. 105-185.

[27] J. Etnyre, Lectures on open book decompositions and contact structures, Clay Math., Proc.
(5), Amer. Math. Soc., Providence, RI, 2006, pp. 103-141.

[28] J. Etnyre, On contact surgery, Proc. Amer. Math. Soc., 136 (2008), no. 9, pp. 3355-3362.

[29] J. Etnyre, On knots in overtwisted contact structures, Quantum Topol., 4 (2013), no. 3, pp.
229-264.
CEU eTD Collection

[30] J. Etnyre and K. Honda, Knots and contact geometry I: torus knots and the figure eight,
Journal of Sympl. Geom., 1 (2001), no. 1, pp. 63-120.

[31] J. Etnyre and K. Honda, On connected sums and Legendrian knots, Adv. Math., 179
(2003), no. 1, pp. 59-74.

[32] A. Floer, A relative Morse index for the symplectic action, Comm. Pure Appl. Math., 41
(1988), no. 4, pp. 393-407.

[33] A. Floer, The unregularized gradient flow of the symplectic action, Comm. Pure Appl.
Math., 41 1988,no. 6, pp. 775-813.

[34] R. Fox and J. Milnor, Singularities of 2-spheres in 4-space and cobordism of knots, Osaka J.
Math., 3 (1966), pp. 257-267.

142
 [35] H. Geiges, Contact geometry, Handbook of differential geometry. Vol. II, pp. 315-382,
Elsevier/North-Holland, Amsterdam, 2006.

[36] P. Ghiggini, Strongly fillable contact 3-manifolds without Stein fillings, Geom. Topol., 9
(2005), pp. 1677-1687.

[37] E. Giroux, Convexité en topologie de contact, Comment. Math. Helv., 66 (1991), no. 4, pp.
637-677.

[38] E. Giroux, Structures de contact en dimension trois et bifurcations des feuilletages de surfaces,
Invent. Math., 141 (2000), no. 3, pp. 615-689.

[39] E. Giroux, Géométrie de contact: de la dimension trois vers les dimensions supérieures, Pro-
ceedings of the International Congress of Mathematicians, Vol. II, Higher Ed. Press,
Beijing, 2002, pp. 405-414.

[40] R. Gompf, Handlebody construction of Stein surfaces, Ann. of Math. (2), 148 (1998), no. 2,
pp. 619-693.

[41] R. Gompf and A. Stipsicz, 4-manifolds and Kirby calculus, Graduate Studies in Mathe-
matics, 20. American Mathematical Society, Providence, RI, 1999, pp. xvi+558.

[42] M. Gromov, Pseudoholomorphic curves in symplectic manifolds, Invent. Math., 82 (1985),

no. 2, pp. 307-347.

[43] A. Hatcher, Algebraic topology, Cambridge University Press, Cambridge, 2002.

[44] M. Hedden, Notions of positivity and the Ozsváth-Szabó concordance invariant, J. Knot
Theory Ramifications, 19 (2010), no. 5, pp. 617-629.

[45] J. Hom and Z. Wu, Four-ball genus bounds and a refinement of the Ozsváth-Szabó τ-
invariant, J. Symplectic Geom., 14 (2016), no. 1, pp. 305-323.

[46] K. Honda, On the classification of tight contact structures I, Geom. Topol., 4 (2000), pp.
309-368.

[47] K. Honda, W. Kazez and G. Matić, On the contact class in Heegaard Floer homology, J.
Differential Geom., 83 (2009), no. 2, pp. 289-311.

[48] H. Hopf, Über die Abbildungen der dreidimensionalen Sphäre auf die Kugelfläche, Math.
CEU eTD Collection

Ann., 104 (1931), pp. 637-665.

[49] T. Kálmán, Maximal Thurston-Bennequin number of +adequate links, Proc. Amer. Math.
Soc., 136 (2008), no. 8, pp. 2969-2977.

[50] Y. Kanda, The classification of tight contact structures on the 3-torus, Comm. in Anal. and
Geom., 5 (1997), pp. 413-438.

[51] A. Kawauchi, T. Shibuya and S. Suzuki, Descriptions on surfaces in four-space. I. Normal

forms, Math. Sem. Notes Kobe Univ., 10 (1982), no. 1, pp. 75-125.

[52] P. Kronheimer and T. Mrowka, The genus of embedded surfaces in the projective plane,
Math. Research Letters, 1 (1994), pp. 797-808.

143
 [53] J. Levine, Invariants of knot cobordism, Invent. Math., 8 (1969), pp. 98-110.

[54] J. Levine, Knot cobordism groups in codimension two, Comment. Math. Helv., 44 (1969),
pp. 229-244.

[55] W. B. R. Lickorish, An introduction to knot theory, Graduate Texts in Mathematics, 175.

Springer-Verlag, New York, 1997.

[56] P. Lisca and G. Matić, Stein 4-manifolds with boundary and contact structures, Topology
Appl., 88 (1998), no. 1-2, pp. 55-66.

[57] P. Lisca, P. Ozsváth, A. Stipsicz and Z. Szabó, Heegaard Floer invariants of Legendrian
knots in contact three-manifolds, J. Eur. Math. Soc., 11 (2009), no. 6, pp. 1307-1363.

[58] C. Manolescu, P. Ozsváth, Z. Szabó and D. Thurston, On combinatorial link Floer homol-
ogy, Geom. Topol., 11 (2007), pp. 2339-2412.

[59] J. Milnor, Lectures on the h-cobordism theorem, Princeton University Press, Princeton,
NJ, 1965.

[60] J. Milnor and J. Stasheff, Characteristic classes, Annals of Mathematics Studies, No. 76.
Princeton University Press, Princeton, NJ; University of Tokyo Press, Tokyo, 1974, pp.
vii+331.

[61] K. Mohnke, Legendrian links of topological unknots, AMS, volume 279 of Contemporary
Mathematics, 2001, pp. 209-211.

[62] Y. Ni, Link Floer homology detects the Thurston norm, Geom. Topol., 13 (2009), no. 5, pp.
2991-3019.

[63] H. Ohta and K. Ono, Simple singularities and topology of symplectically filling 4-manifold,
Comment. Math. Helv., 74 (1999), pp. 575-590.

[64] S. Orevkov and V. Shevchishin, Markov Theorem for Transverse Links, J. Knot Theory
Ramifications, 12 (2003), no. 7, pp. 905-913.

[65] B. Ozbagci and A. Stipsicz, Surgery on contact 3-manifolds and Stein surfaces, Bolyai
Society Mathematical Studies, 13. Springer-Verlag, Berlin; János Bolyai Mathematical
Society, Budapest, 2004. pp. 281.
CEU eTD Collection

[66] P. Ozsváth and A. Stipsicz, Contact surgeries and the transverse invariant in knot Floer
homology, J. Inst. Math. Jussieu, 9 (2010), no. 3, pp. 601-632.

[67] P. Ozsváth, A. Stipsicz and Z. Szabó, Grid homology for knots and links, AMS, volume
208 of Mathematical Surveys and Monographs, 2015.

[68] P. Ozsváth and Z. Szabó, Holomorphic disks and topological invariants of closed three-
manifolds, Ann. of Math. (2), 159 (2004), no. 3, pp. 1159-1245.

[69] P. Ozsváth and Z. Szabó, Holomorphic disks and knot invariants, Adv. Math., 186 (2004),
no. 1, pp. 58-116.

[70] P. Ozsváth and Z. Szabó, Heegaard Floer homology and contact structures, Duke Math. J.,
129 (2005), no. 1, pp. 39-61.

144
 [71] P. Ozsváth and Z. Szabó, On the Heegaard Floer homology of branched double-covers, Adv.
Math., 194 (2005), no. 1, pp. 1-33.

[72] P. Ozsváth and Z. Szabó, Holomorphic disks, link invariants and the multi-variable Alexan-
der polynomial, Algebr. Geom. Topol., 8 (2008), no. 2, pp. 615-692.

[73] P. Ozsváth and Z. Szabó, Link Floer homology and the Thurston norm, J. Amer. Math.
Soc., 21 (2008), no. 3, pp. 671-709.

[74] J. Pardon, The link concordance invariant from Lee homology, Algebr. Geom. Topol., 12
(2012), no. 2, pp. 1081-1098.

[75] T. Perutz, Hamiltonian handleslides for Heegaard Floer homology, In Proceedings of

Gökova Geometry-Topology Conference 2007, pp. 15-35. Gökova Geometry/Topology
Conference (GGT), Gk̈ova, 2008.

[76] O. Plamenevskaya, Transverse knots and Khovanov homology, Math. Res. Lett., 13 (2006),
no. 4, pp. 571-586.

[77] O. Plamenevskaya, A combinatorial description of the Heegaard Floer contact invariant,

Algebr. Geom. Topol., 7 (2007), pp. 1201-1209.

[78] J. Rasmussen, Floer homology and knot complements, Thesis (Ph.D.)-Harvard University,
2003, pp. 126.

[79] D. Rolfsen, Knots and links, Mathematics Lecture Series, 7. Publish or Perish, Inc.,
Houston, TX, 1990.

[80] L. Rudolph, Algebraic functions and closed braids, Topology, 22 1983, no. 2, pp. 191-202.

[81] L. Rudolph, Positive links are strongly quasipositive, Geom. Topol. Monogr., (2), Geom.
Topol. Publ., Coventry, 1999.

[82] S. Sarkar, Grid diagrams and the Ozsváth-Szabó tau-invariant, Math. Res. Lett., 18 (2011),
no. 6, pp. 1239-1257.

[83] A. Shumakovitch, Rasmussen invariant, slice-Bennequin inequality, and sliceness of knots,

J. Knot Theory Ramifications, 16 (2007), no. 10, pp. 1403-1412.

[84] J. Singer, Three-dimensional manifolds and their Heegaard diagrams, Trans. Amer. Math.
CEU eTD Collection

Soc., 35 (1933), no. 1, pp. 88-111.

[85] J. Światkowski.
˛ On the isotopy of Legendrian knots, Ann. Global Anal. Geom., 10 (1992),
no. 3, pp. 195-207.

[86] W. Thurston, A norm for the homology of 3-manifolds, Mem. Amer. Math. Soc., 59 (1986),
no. 339, pp. i-vi and 99-130.

[87] W. Thurston and H. Winkelnkemper, On the existence of contact forms, Proc. Amer.
Math. Soc., 52 (1975), pp. 345-347.

[88] V. Turaev, Torsions of 3-manifolds, Geom. Topol. Monogr. 4, Geom. Topol. Publ., Coven-
try, 2002.

145
 [89] T. Vogel, Non-loose unknots, overtwisted discs, and the contact mapping class group of S3 ,
arXiv:1612.06557.

[90] F. Warner, Foundations of differentiable manifolds and Lie groups, Graduate Texts in Math-
ematics, 94. Springer-Verlag, New York-Berlin, 1983. pp. ix+272.

[91] N. Wrinkle, The Markov theorem for transverse knots, Thesis (Ph.D.)-Columbia Univer-
sity, 2002, pp. 51.
CEU eTD Collection

146
 Index

J-holomorphic strip, 57 homology, 18, 19

moduli space, 58 maps, 19
Spinc structure, 47 Chain homotopy
relative, 49 filtered, 19
τ-set, 124 Chain homotopy equivalence
Rect, 107 filtered, 19
Rect◦ , 107 Chain map, 19
filtered, 19
Abstract open book, 53–54 associated mapping cone, 20
associated to an adapted open book de- Characteristic foliation, 42
composition, 55 Classical invariants
basepoint, 54 Legendrian, 34, 38
monodromy, 53 surgery presentation, 99–100
positive stabilization, 74 transverse, 36, 39
strip position, 54 Cobordism, 15
Alexander grading, 60, 106 birth, 118
symmetry, 63 death, 119
Alexander pair, 93 elementary, 16
Arc identity, 116
dead, 51 merge, 117
diagram, 12 split, 116
distinguished, 51 strong, 16
separating, 51 torus, 117
Braid, 38–39 Complexity, 25
Braids group, 39 Component
of a link, 9
Chain complex, 18 split, 9
associated to a grid diagram, 106–107, Compressibility term, 30
CEU eTD Collection

113 Compressible surface, 28

Alexander grading, 106 Concordance, 16
Maslov grading, 106 Connected sum
associated to a Heegaard diagram, 58– contact, 40
62, 64 diagram, 15
Alexander multi-grading, 60 Legendrian, 40
Maslov grading, 59 links, 10
differential, 18 Contact 3-manifold, 31
filtered, 18 contact cell decomposition, 51–53
associated graded object, 18 prime decomposition, 33
quasi-isomorphic, 19 Stein fillable, 33
filtered isomorphism, 20 symplectically fillable, 33

147
 Contact framing, 31, 34 invariance, 108–109, 111–112
Contact structure, 31
ξ n , 33 Handlebody, 45
ξ st , 37, 39 Heegaard diagram
overtwisted, 31 handleslide, 46
at infinity, 33 Heegaard diagram, 45–46
tight, 31 admissible, 61
at infinity, 33 basepoint, 46, 47, 50
Contact surgery, 34 compatible Morse function, 46
Contact vector field, 32 domain, 58
Contactomorphism, 31 handleslide, 48
Convex surface, 32 intersection point, 47
Cooriented 2-plane field, 31 Legendrian, 73–74
Crossing, 12 multi-pointed, 47–48, 50
associated to an abstract open book,
Darboux ball, 40 55
Differential, 18 pointed, 46
associated to a grid diagram, 107 stabilization, 46, 48
associated to a Heegaard diagram, 64 Heegaard Floer homology, 64–66
associated to a multi-pointed Heegaard connected sum, 114
diagram, 61–62 disjoint union, 114
mapping cone, 20 invariance, 66–70
Disjoint union, 9, 29 choice of the almost-complex struc-
Legendrian, 42 ture, 67–68
Dividing set, 32 handleslide, 69
isotopy, 68–69
Elementary cobordism, 16 stabilization, 70
Elementary tangle, 10 link, 57, 63
Legendrian, 40 mirror image, 113–114
trivial, 10 quasi-alternating links, 115
Equivalence unknot, 123
of abstract open books, 54 unlink, 66, 111
of filtered chain complexes, 19 Hopf invariant, 32
of links, 9 Hopf link, 17
Filtered isomorphism, 20 Invariant L, 80–81
CEU eTD Collection

Filtration, 18 connected sum, 88

level, 18 disjoint union, 91
Front projection, 37 gradings, 87–88
Function T, 109 invariance, 83–84
admissible arc slide, 82
Genus
isotopy, 81–82
Seifert, 9
positive stabilization, 82–83
slice, 16
stabilization, 91
Graded object, 18
Invariant T, 81
Grid diagram, 105–106
gradings, 88
grid state, 105
invariance, 92
stabilization with an unknot, 109–111
Invariant ν, 123
Grid homology, 107–108, 111, 123

148
 Invariant τ, 111 Link type, 34
τ ∗ , 114 simple, 103
connected sum, 114 Linking number, 10
disjoint union, 114 diagram, 14
invariance, 120 Loose link, 95
mirror image, 114
positive links, 139 Mapping cone, 20
quasi-alternating links, 115 Markings
quasi-positive links, 133 in a grid diagram, 105
slice genus, 121–122 special O-marking, 105
strongly quasi-positive links, 135 Maslov grading, 59–60, 106
Invariant cb, 84 Maslov index, 58
invariance, 85–86 Meridian, 10
Isotopy Mirror image, 11
contact, 31 diagram, 15
Legendrian, 34 Morse moves, 16
smooth, 9 Non-loose, 95
transverse, 36 Non-split, 9
Knot, 9 Null-homologous link, 9
compressible, 28 Open book decomposition, 50–51
concordance, 17 adapted to a Legendrian link, 51
Legendrian, 34 admissible arc slide, 77–78
loose, 95 associated to a contact cell decomposi-
meridian, 10 tion, 52
transverse, 36 isotopy, 74
trivial, 9 Murasugi sum, 88
Knot concordance group, 17 page, 50
positive stabilization, 74–75
Lens space, 28
L-elementary, 75
Link, 9
supporting a contact structure, 50
cobordism, 15
Order
connected sum, 10
of a link, 26
diagram, 12
Orientation
Legendrian, 34
of a link, 9
linking number, 10
Oriented resolution, 12
CEU eTD Collection

loose, 95
Overtwisted disk, 31
mirror image, 11
oriented, 9 Planar diagram, 12
positive, 136 Positive link, 136–138
quasi-alternating, 115
quasi-positive, 131 Quasi-alternating link, 115
slice, 16 Quasi-isomorphism
split, 9 filtered, 19
strongly quasi-positive, 134 Quasi-positive braid, 131
transverse, 36 Quasi-positive link, 131
a , 130
Link L940 Quasi-positive surface, 134
n
Link L919 , 129
Rasmussen s-invariant, 133

149
 Rational homology sphere, 10 maximal, 129, 136
Rationally bounded surface, 26 Torus links, 127
Reidemeister moves, 12 Transverse C-link, 131
Legendrian, 37 connected, 132
Resolution Transverse push-off, 36
of a crossing, 12
Ribbon Unknot, 9
of a Legendrian graph, 52 Legendrian, 37
Rotation number, 35 non-loose, 95

Seifert genus, 9 Weakly slice, 16

strongly quasi-positive links, 135 Writhe, 14
Seifert surface, 9
quasi-positive, 135
rational, 26
Self-linking number, 36
maximal, 132
Sign
of a crossing, 12
Slice genus, 16
additivity, 133
torus links, 128
Slice link, 16
Snail-like domain, 119
Split diagram, 13
Split link, 9, 42
Stabilization
L-elementary, 75
of a Legendrian link, 41
of an open book decomposition
positive, 74
Stein filling, 33
Strong cobordism, 16
Strongly quasi-positive link, 134–135
Strongly slice, 16
Symmetric power of a surface, 47
CEU eTD Collection

Symplectic 4-manifold, 33
System of attaching circles, 45
System of generators
adapted to a link, 51

Tensor product, 21
of K[ x ]-modules, 22
of vector spaces, 22
Thurston norm, 25–27
Thurston semi-norm, 25
Thurston-Bennequin inequality, 42–43, 132
Thurston-Bennequin number, 35

150