B. Ozbagci and A. I. Stipsicz - Surgery On Contact 3-Manifolds and Stein Surfaces

Surgery
on
contact
3-manifolds
and
Stein surfaces
B. Ozbagci and A. I. Stipsicz
Surgery
on
contact
3-manifolds
and
Stein surfaces
B. Ozbagci
A. I. Stipsicz
J
ANOS BOLYAI MATHEMATICAL SOCIETY

Budapest, Fo u. 68., H1027, Hungary
c _ BOLYAI J
ANOS MATEMATIKAI T
ARSULAT
Budapest, Hungary, 2004
ISBN: ??? ???? ???
Published by
J
ANOS BOLYAI MATHEMATICAL SOCIETY

Budapest, Fo u. 68.,
H1027, Hungary
Assistant editor: ???
Printed in Hungary
Budapest
Contents
Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1. Why symplectic and contact? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2. Results concerning Stein surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3. Some contact results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2. Topological surgeries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1. Surgeries and handlebodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2. Dehn surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3. Kirby calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3. Symplectic 4-manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1. Generalities about symplectic manifolds . . . . . . . . . . . . . . . . . . . . . 49
3.2. Mosers method and neighborhood theorems . . . . . . . . . . . . . . . . . 55
3.3. Appendix: The complex classication scheme for symplectic
4-manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4. Contact 3-manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.1. Generalities on contact 3-manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2. Legendrian knots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3. Tight versus overtwisted structures . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5. Convex surfaces in contact 3-manifolds . . . . . . . . . . . . . . . . . . . . 85
5.1. Convex surfaces and dividing sets . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.2. Contact structures and Heegaard decompositions . . . . . . . . . . . . 96
6. Spin
c
structures on 3- and 4-manifolds . . . . . . . . . . . . . . . . . . . . . . 99
6.1. Generalities on spin and spin
c
structures . . . . . . . . . . . . . . . . . . . . 99
4 Contents
6.2. Spin
c
structures and oriented 2-plane elds . . . . . . . . . . . . . . . . . . 102
6.3. Spin
c
structures and almost-complex structures . . . . . . . . . . . . . . 105
7. Symplectic surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1. Symplectic cut-and-paste . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2. Weinstein handles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3. Another handle attachment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8. Stein manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.1. Recollections and denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
8.2. Handle attachment to Stein manifolds . . . . . . . . . . . . . . . . . . . . . . . 125
8.3. Stein neighborhoods of surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
9. Open books and contact structures . . . . . . . . . . . . . . . . . . . . . . . . 131
9.1. Open book decompositions of 3-manifolds . . . . . . . . . . . . . . . . . . . 131
9.2. Compatible contact structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
9.3. Branched covers and contact structures . . . . . . . . . . . . . . . . . . . . . . 150
10. Lefschetz fibrations on 4-manifolds . . . . . . . . . . . . . . . . . . . . . . . . 155
10.1. Lefschetz pencils and brations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
10.2. Lefschetz brations on Stein domains . . . . . . . . . . . . . . . . . . . . . . . 162
10.3. Some applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
11. Contact Dehn surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
11.1. Contact structures on S
1
D
2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
11.2. Contact Dehn surgery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
11.3. Invariants of contact structures given by surgery diagrams . . 191
12. Fillings of contact 3-manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
12.1. Fillings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
12.2. Nonllable contact 3-manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
12.3. Topology of Stein llings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
13. Appendix: SeibergWitten invariants . . . . . . . . . . . . . . . . . . . . . . 223
13.1. SeibergWitten invariants of closed 4-manifolds . . . . . . . . . . . . 223
13.2. SeibergWitten invariants of 4-manifolds with contact boun-
dary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
13.3. The adjunction inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
14. Appendix: Heegaard Floer theory . . . . . . . . . . . . . . . . . . . . . . . . . . 235
14.1. Topological preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Contents 5
14.2. Heegaard Floer theory for 3- and 4-manifolds . . . . . . . . . . . . . . . 239
14.3. Surgery triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
14.4. Contact OzsvathSzabo invariants . . . . . . . . . . . . . . . . . . . . . . . . . . 249
15. Appendix: Mapping class groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
15.1. Short introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
15.2. Mapping class groups and geometric structures . . . . . . . . . . . . . 264
15.3. Some proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Preface
The groundbreaking results of the near past Donaldsons result on Lef-
schetz pencils on symplectic manifolds and Girouxs correspondence be-
tween contact structures and open book decompositions brought a top-
ological avor to global symplectic and contact geometry. This topological
aspect is strengthened by the existing results of Weinstein and Eliashberg
(and Gompf in dimension 4) on handle attachment in the symplectic and
Stein category, and by Girouxs theory of convex surfaces, enabling us to
perform surgeries on contact 3-manifolds. The main objective of these notes
is to provide a self-contained introduction to the theory of surgeries one can
perform on contact 3-manifolds and Stein surfaces. We will adopt a very
topological point of view based on handlebody theory, in particular, on
Kirby calculus for 3- and 4-dimensional manifolds.
Surgery is a constructive method by its very nature. Applying it in an
intricate way one can see what can be done. These results are nicely com-
plemented by the results relying on gauge theory a theory designed to
prove that certain things cannot be done. We will freely apply recent results
of gauge theory without a detailed introduction to these topics; we will be
content with a short introduction to some forms of SeibergWitten theory
and some discussions regarding Heegaard Floer theory in two Appendices.
As work of Taubes in the closed, and KronheimerMrowka in the manifold-
with-boundary case shows, the analytic approach towards symplectic and
contact topology can be very fruitfully capitalized when coupled with some
form of SeibergWitten theory. On the other hand, Lefschetz pencils on
symplectic, and open book decompositions on contact manifolds are well-
suited for the newly invented contact OzsvathSzabo invariants. Under
some fortunate circumstances these dual viewpoints provide interesting re-
sults in the subject. As a preview, Chapter 1 is devoted to the description
of problems where the above discussed techniques can be applied.
For setting up the topological background of surgeries on contact 3-
manifolds and Stein surfaces we will rst examine the smooth surgery con-
struction, with a special emphasis on 2-handle attachments to 4-manifolds
8 Preface
and Dehn surgeries on 3-manifolds. This is done in Chapter 2. Then we turn
to the symplectic cut-and-paste operation, which enables us to glue sym-
plectic 4-manifolds along contact type boundaries. To put this operation
in the right perspective, in Chapters 3 and 4 we rst briey review some
parts of symplectic and contact topology in dimensions 4 and 3, respec-
tively. We pay special attention to convex surfaces in contact 3-manifolds
(Chapter 5), with an eye on its later applicability in contact surgery. Be-
fore giving the general scheme of symplectic surgery in Chapter 7, we make
a little digression and discuss spin
c
structures from a point of view suit-
able for our later purposes. As a special case of the general gluing scheme,
we will meet Weinsteins construction for attaching symplectic 2-handles to
-convex boundaries along Legendrian knots. After having these prepara-
tions, we can turn to the discussion of the famous result of Eliashberg that
shows how to attach a Stein 2-handle to the pseudoconvex boundary of a
Stein domain along a Legendrian knot. For the convenience of the more
topologically minded reader, in Chapter 8 a short recollection of rudiments
of the theory of Stein manifolds is included. Once the gluing construction
given, we can turn to its applications, including the search for Lefschetz -
bration structures on Stein domains, embeddability of Stein domains into
closed surfaces with extra (symplectic or complex) structures, or the study
of Stein llings of contact 3-manifolds (Chapters 10 and 12). In the contact
setting, the most important technique for being able to do surgery is the
convex surface theory developed by Giroux. After recalling relevant parts
of this beautiful theory, and proving the neighborhood theorems we need in
this subject, in Chapter 11 we will be able to do contact surgeries. With
this construction at our disposal, now we can seek for applications: we will
be able to draw explicit diagrams of many contact 3-manifolds, show ways
to distinguish them and to determine the homotopy type of contact struc-
tures given by various constructions. These results together with various
versions of gauge theories, including SeibergWitten theory and Heegaard
Floer theory provide ways to examine tightness and llability properties
of numerous contact structures, which are given in Chapter 12. To make
the presentation more complete, we include Chapter 9 on open book de-
compositions and their relation to contact structures. The appearance of
mapping class groups in these theories, together with some nice applications
allows us to conclude the discussion with a short recollection of denitions
and results in that eld.
To guide the interested reader, we close this preface by listing some
monographs discussing topics we only outline here. Handlebody theory and
Preface 9
Kirby calculus, which is only sketched in Chapter 2, is discussed more thor-
oughly in [66]. A more complete introduction to symplectic geometry and
topology is provided by [111]. For additional reading on contact topology,
the reader is advised to turn to [1, 2, 57]. SeibergWitten theory is covered
by many volumes, including for example [119, 126, 149].
These notes are based on two lecture series given by the second author at
the Banach Center (Warsaw, Poland) and at the University of Lille (France).
He wants to thank these institutions for their hospitality. The nal form of
the notes were shaped while the authors visited KIAS (Seoul, Korea); they
wish to thank KIAS for its hospitality. The authors would like to thank
Selman Akbulut, John Etnyre, Sergey Finashin, David Gay, Paolo Lisca,
Gordana Matic and R obert Sz oke for many enlightening conversations.
Special thanks go to Hansjorg Geiges for suggesting numerous corrections
and improvements of an earlier version of the text. The second author
also wants to express his thanks to his family without their support this
volume would not have come into existence. The rst author acknowledges
support from the Turkish Academy of Sciences and from Ko c University.
The second author acknowledges partial support by OTKA T034885 and
T037735.
Istanbul and Budapest, 2004. Burak Ozbagci and Andr as Stipsicz
1. Introduction
1.1. Why symplectic and contact?
The intense interest of 4-manifold topologists in symplectic geometry and
topology might have the following explanation. The success of the classi-
cation of higher ( 5) dimensional manifolds relies heavily on the famous
h-cobordism theorem, in which the Whitney trick plays a fundamen-
tal role. The Whitney trick asserts that (under favorable conditions) the
algebraic and geometric intersection numbers of two submanifolds can be
made equal by isotoping one of them. In other words, by isotopy we can
get rid of excess intersections, which are present in the geometric picture
but are invisible for algebra. After eliminating these intersections algebra
will govern geometry, and the smooth classication problem of manifolds
can be translated into some (nontrivial) algebraic questions.
Remark 1.1.1. The proof of the Whitney trick involves a map of a 2-
dimensional disk into the manifold at hand. If we can achieve that this
map is an embedding (with the appropriate normal bundle), we get a local
model showing us the required isotopy. Once the dimension is high enough
(at least 5), any map from the disk admits a perturbation such that the
result is an embedding. In dimension four, however, the disk might have
self-intersections, and we cannot get rid of those by simple dimension count.
The key step of Freedmans topological classication theorem in dimen-
sion four is to show that the Whitney trick does extend to dimension four
provided we allow topological isotopies. In fact, as Donaldsons theorems on
the failure of the smooth h-cobordism theorem in dimension 4 show, in some
examples the excess intersections persist if we allow smooth maps only.
It is a standard fact that in a complex manifold complex submanifolds
intersect (locally) positively, therefore no excess intersection points appear,
12 1. Introduction
hence the above principles apply. Actually, if a 4-manifold X is only almost-
complex (and this structure is much easier to nd, since its existence de-
pends only on the homotopy type of X) then almost-complex submanifolds
still intersect positively with the usual restrictions of not sharing com-
mon components, see [109]. The problem with almost-complex manifolds is
that although the existence of the structure is guaranteed by some simple
properties of the cohomology ring H
(X
4
; Z), it is very hard to show that
almost-complex submanifolds exist (in general they do not), i.e., that the
above principle ever comes into force. Now if the almost-complex manifold
(X, J) also carries a compatible symplectic structure , then accord-
ing to fundamental results of Taubes smooth properties of X already
guarantee the existence of almost-complex (also known as J-holomorphic)
submanifolds. Since this argument only provides a few almost-complex rep-
resentatives, we cannot expect a complete solution for the classication
problem. The spectacular results built on Taubes work nevertheless show
the above described principle in action. For this reason we chose to study
symplectic 4-manifolds (and their topological counterparts, Lefschetz bra-
tions) in more detail. According to Donaldsons result, symplectic manifolds
always decompose along a circle bundle into a union of a disk bundle and
another piece which can be endowed with a Stein structure. Conversely, any
Stein surface embeds into some closed symplectic 4-manifold. The analogy
becomes even deeper if we study the topological counterparts of symplec-
tic and Stein manifolds: these are Lefschetz brations with closed or with
bounded bers. Therefore it appears natural to study topological properties
of symplectic and Stein manifolds together.
When trying to perform surgeries in the symplectic or Stein category,
we have to pay special attention to the structures induced on their 3-
dimensional boundaries this is how contact structures come into play.
The topological counterpart of contact structures (which are open book
decompositions on the 3-manifolds) ts perfectly into this picture since
open book decompositions can be interpreted as boundaries of Lefschetz
brations. (In the general case we allow achiral Lefschetz brations as
well.) The fascinating, and still not completely well-understood interplay of
the above notions provides the leading theme of these notes.
Topological questions regarding symplectic 4-manifolds and Lefschetz
brations are fairly well-treated in the literature ([66, 111]); in these in-
stances we merely restrict ourselves to quoting the necessary results. For
contact surgeries and open book decompositions the available sources are
less complete, so in these cases a more thorough treatment of the relevant
1.2. Results concerning Stein surfaces 13
material is given. In the following we address the problem of understanding
topological properties of Stein surfaces and contact 3-manifolds. In order
to attack such a problem we need two major tools, which provide existence
and nonexistence results. Complex geometry (e.g., complex surfaces, Mil-
nor bers, links of singularities) provides a rich source of examples, giving
the needed existence results. A more systematic way of studying the ex-
istence problem is provided by the theory of handlebodies initiated by
Smale, Milnor and Kirby, and extended to the symplectic and Stein cate-
gory by Weinstein, Eliashberg and Gompf. On the boundary, the handle
attachment translates into contact surgery, showing existence of a variety
of contact structures. By suitably generalizing the attachment scheme de-
scribed by Weinstein (and incorporating achiral Lefschetz brations into
the theory), in fact all contact 3-manifolds can be treated in this way. On
the other hand, gauge theory (more specically, SeibergWitten theory and
OzsvathSzabo invariants) can be used to prove that manifolds or dieo-
morphisms with certain properties do not exist. Therefore SeibergWitten
and OzsvathSzabo invariants and SeibergWitten moduli spaces provide
(in favorable cases) the needed nonexistence results. We can, for example,
show that certain 4-manifolds do not carry any Stein structure, or specic
contact 3-manifolds cannot be given as boundary of any Stein surface.
1.2. Results concerning Stein surfaces
Before turning to the detailed discussion of various surgery constructions,
we give a sample of results we would like to present in these notes. As it
turns out, the existence of a Stein structure on a 4-manifold X considerably
constrains its dierential topology. The most apparent constraint can be
summarized by the adjunction inequality given in Theorem 1.2.1. Closely
related formulae appear in many other branches in 3- and 4-dimensional
topology, and these type of results always play a central role in the theory at
hand. (See Section 4.3 for the contact version of the adjunction formula.)
Theorem 1.2.1 ([6]). If X is a 4-dimensional Stein manifold and X
is a closed, connected, oriented, embedded surface of genus g in it, then
[]
2
+[
c
1
(X), []
_
[ 2g 2
unless is a sphere with [] = 0 in H
2
(X; Z).
14 1. Introduction
Remark 1.2.2. Note that if C is a (smooth) connected complex curve in
a complex surface X then the Whitney product formula for Chern classes
implies 2g(C) 2 = [C]
2
c
1
(X), [C]
_
; this equation is frequently called
the adjunction equality. Its generalization for closed complex surfaces and
smooth submanifolds X was proved rst by Kronheimer and Mrowka
(in the case []
2
0) and in general by Ozsvath and Szab o [133]. For
more about the adjunction inequality see Section 13.3, where we will indi-
cate how such formulae for a closed symplectic 4-manifold X follow from
SeibergWitten theory, and describe the derivation of the above formula
(for Stein surfaces) from the closed case. Notice that, for example, the in-
equality shows that a Stein surface cannot contain a homologically essential,
smoothly embedded sphere S with [S]
2
1.
Below we give some surprising corollaries of the above adjunction inequality;
we hope that this demonstrates the power and diversity of the theorem.
Simple nondieomorphic 4-manifolds
The rst application gives an example of homeomorphic but nondieomor-
phic 4-manifolds.
Corollary 1.2.3 (Akbulut, [4]). The 4-manifolds X
1
, X
2
dened by the
knots K
1
, K
2
of Figure 1.1 are homeomorphic but nondieomorphic.
1 1
4
K K
1 2
Figure 1.1. Homeomorphic but nondieomorphic 4-manifolds
The meaning of such Kirby diagrams will be explained in Chapter 2; for a
more thorough treatment see [66]. Here we just note that the knots (together
with the numbers) indicate how to glue a 4-dimensional 2-handle D
2
D
2
along D
2
D
2
to D
4
in order to get X
1
and X
2
, respectively.
Proof (sketch). Using some simple operations on the diagrams one can
show that the 3-manifolds X
1
and X
2
are dieomorphic (see [66, Fig-
ure 11.4]). Now the signatures (X
1
) and (X
2
) are both equal to 1, the
Euler characteristics are both equal to 2, and since X
1
= X
2
is an in-
tegral homology sphere, the extension of Freedmans famous theorem (see
[51]) implies that X
1
and X
2
are homeomorphic. Next we show that X
1
carries a Stein structure. This follows from the theory of gluing symplectic
handles (developed by Weinstein and Eliashberg), once we realize that K
1
can be represented by a Legendrian knot with ThurstonBennequin num-
ber equal to 0. (For this theory, the denitions of the above notions and
Figure 1.2. Legendrian representative of the knot K1
constructions will be discussed in later chapters.) For such a Legendrian rep-
resentative see Figure 1.2. Therefore, in order to distinguish X
1
from X
2
it
is enough to prove that X
2
does not admit any Stein structure. This state-
ment follows from the observation that the generator of H
2
(X
2
; Z) (which
has self-intersection 1) can be represented by a sphere. Such a represen-
tative can be easily found once we get a disk D
2
D
4
with D
2
= K
2

glue the core of the 2-handle to this disk. The existence of such a disk is
shown by the movie of Figure 1.3. These pictures show how the disk in-
tersects the spheres with radius r < 1 in the 4-disk D
4
as r grows from 0 to
1. These intersections start with two circles (which are boundaries of two
disks) which get tangled as time passes (Figures 1.3 (1)(2)), and then
a ribbon is added to connect the disks, resulting an embedded disk in D
4
16 1. Introduction
with boundary given by Figure 1.3(3). (Of course, in this process the value
4
(1)
(2)
(3)
Figure 1.3. The movie showing the disk
4 plays no special role; it becomes important when proving the dieomor-
phism X
1

= X
2
.) Now the application of the adjunction inequality with
g = 0 and []
2
= 1 would give
1 +[
c
1
(X
2
, J), []
_
[ 2,
a contradiction for any Stein structure J on X
2
. Therefore X
2
cannot carry
any Stein structure, implying that X
1
and X
2
are nondieomorphic.
Remark 1.2.4. The above example was found by Akbulut [4], using dier-
ent methods in the proof of nondieomorphism. This version of the proof
is due to Akbulut and Matveyev [6].
Existence of Stein neighborhoods
Theorem 1.2.5. Let S CP
2
be a smoothly embedded sphere in CP
2
which is nontrivial in homology. Then there is no open set U containing a
neighborhood of S which admits a Stein structure.
Proof. The adjunction inequality of Theorem 1.2.1 implies that a homo-
logically nontrivial sphere in a Stein surface has self-intersection 2. For
[S] ,= 0 in H
2
(CP
2
; Z) we have that [S]
2
> 0, providing the result.
Remark 1.2.6. The same argument works for any smooth 2-dimensional
submanifold in a complex surface X with 2g() 2 < []
2
. Surprisingly
enough, if the inequality []
2
+[
c
1
(X), []
_
[ 2g() 2 does hold (notice
that because of the absolute value this is, in fact, the union of two inequal-
ities) then there is a Stein neighborhood U CP
2
of , see [49]. For the
outline of this latter argument see also Section 8.3.
This application of the adjunction inequality leads to the solution of a
seemingly unrelated problem in complex analysis.
Corollary 1.2.7 (Nemirovski, [125]). Suppose that S CP
2
is a smoothly
embedded sphere in CP
2
which is nontrivial in homology. If f is a holo-
morphic function on some neighborhood of S then f is constant.
Proof. Let us x a neighborhood U and a holomorphic function f on it.
Consider the envelope of holomorphy of U, i.e., the maximal domain

V
containing U such that every holomorphic function on U extends holomor-
phically to

V . Denote this envelope of holomorphy by

U. According to a
result of Fujita, in our case

U is either CP
2
or it is Stein. Since S U

U,
Theorem 1.2.5 shows that

U cannot be Stein. Therefore

U = CP
2
, hence
all holomorphic functions on

U (and so on U) are constant.
Remark 1.2.8. Notice that if S is a complex submanifold generating
H
2
(CP
2
; Z) then the statement is obvious: if f is a holomorphic function on
U then by restricting it to U (CP
2
CP
1
) C
2
and applying a theorem
of Hartogs we get an extension of f to CP
2
, implying that it is constant.
The question answered by the above theorem was raised by Vitushkin, and
similar results (for higher genus and immersed surfaces) are still in the focus
of current research, see [49, 125].
18 1. Introduction
The four-ball genus of knots in S
3
Let K S
3
be a given knot. The genus g(K) is dened as
min
_
g() [ S
3
is a Seifert surface for K
_
.
For example, it is fairly easy to see that g(K) = 0 holds if and only if K is
the unknot. The four-ball genus (or slice genus) g
(K) can be dened as

min
_
g(F) [ F D
4
, F = K
_
,
where F denotes a smoothly embedded connected surface in D
4
transverse
to D
4
. Obviously g
(K) g(K), and as the proof of Corollary 1.2.3

showed, g
(K) can be equal to 0 for a nontrivial knot K, e.g. for K

2
.
(Knots with vanishing four-ball genus g
are called smoothly slice.) The

adjunction inequality provides a nontrivial lower bound for g
(K) in the
following way. Approximate K with a Legendrian knot L and glue a 2-
handle to D
4
along K with surgery coecient one less than the contact
framing of L. The resulting 4-manifold X will contain a surface

F with
g(

F) = g
(K), obtained by gluing the four-ball genus minimizing surface to

the core of the 2-handle. Since X carries a Stein structure, and [

F]
2
and
c
1
(X), [

F]
_
admit expressions purely in terms of data of the Legendrian
knot L as [

F]
2
= tb(L) 1 and
c
1
(X), [

F]
_
= rot(L), the adjunction
inequality gives a lower bound for g
(K):
tb(L) +[ rot(L)[ 2g
(K) 1.
For example:
Corollary 1.2.9. The trefoil knot is not smoothly slice.
Proof. The right-handed trefoil admits a Legendrian presentation with
tb(L) = 1 and rot(L) = 0 (see Figure 1.4), hence the adjunction inequality
translates as 0 2g
(K) 2, implying 1 g
(K). It is not hard to nd

a genus-1 Seifert surface for K, therefore we see that the four-ball genus of
the trefoil knot is 1.
The unknotting number (or gordian number) u(K) of a knot K is dened
as the minimal number of crossing changes in any projection which untie
the knot.
Exercise 1.2.10. Show that u(K) g
(K) for any knot K S

3
.
Figure 1.4. A right-handed Legendrian trefoil knot
Notice that the inequality u(K) g
(K) is not an equality in general;

take for example the knot K
2
of Figure 1.1, which has g
(K
2
) = 0 but
u(K
2
) > 0 since it is not the unknot. Heegaard Floer theory provides knot
invariants which can be fruitfully used to get new constraints on the 4-ball
genus of a knot, see [129, 142].
Topological characterization of Stein domains
According to a recent result of Loi and Piergallini [104], Stein domains admit
a nice topological description in terms of Lefschetz brations.
Theorem 1.2.11 (LoiPiergallini, [104]). If S is a complex 2-dimensional
Stein domain then it admits a Lefschetz bration structure over D
2
.
This result similarly to Donaldsons result on existence of Lefschetz
pencils on closed symplectic 4-manifolds brings a topological avour into
the study of Stein domain. The original proof of Theorem 1.2.11 relies on
an approach of presenting the 4-manifolds at hand as branched covers of
D
4
along fairly complicated branch sets. A conceptually simpler proof of
the same statement was given by Akbulut and the rst author [7], making
use of the handle decomposition of a Stein domain and relating it to handle
decompositions of 4-manifolds admitting Lefschetz brations. The detailed
description of this second approach will be given in Chapter 10.
20 1. Introduction
3-manifolds which are not Stein boundaries
Our nal example in this section shows that the boundary of a Stein domain
cannot be arbitrary.
Theorem 1.2.12 (Lisca, [94]). Let E denote the boundary of the (+E
8
)-
plumbing W (as shown by the plumbing diagram of Figure 1.5). There is
2 2 2 2 2 2 2
2
Figure 1.5. The (+E8)-plumbing
no Stein domain S with S = E.
Proof. Using standard pull-apart arguments in SeibergWitten theory (see
Chapter 13) it can be shown that if X = X
1

E
X
2
and X is symplectic
then b
+
2
(X
1
) = 0 or b
+
2
(X
2
) = 0. (This argument uses the fact that
E admits a positive scalar curvature metric, since it is dieomorphic to
the Poincare homology sphere, with its standard orientation reversed, cf.
Proposition 13.1.7(5.).) Now if S is a Stein domain with S = E then S can
be embedded into a closed symplectic 4-manifold X with b
+
2
(XS) > 0. In
conclusion, from the above principle we get b
+
2
(S) = 0. Therefore the closed
4-manifold Z = S
E
(W) is negative denite. Since the intersection form
of W (which is the famous negative denite E
8
-form) does not embed
into any diagonal intersection form, the intersection form Q
Z
cannot be
diagonalized over the integers. This last consequence, however, contradicts
Donaldsons famous result about diagonalizability of denite intersection
forms of smooth 4-manifolds, showing that S cannot exist.
Remark 1.2.13. Analogous statements have been proved for the bound-
aries of the (+E
7
)- and (+E
6
)-plumbings [96]. Results of this type will
be discussed in Chapter 12 in more detail. We just note here that by ap-
plying SeibergWitten invariants of manifolds with contact type boundary
(see Section 13.2) it can be shown that these 3-manifolds (and many more
of similar type) admit no symplectically llable contact structures [94, 101].
1.3. Some contact results 21
1.3. Some contact results
As we will see, contact structures on 3-manifolds fall into two very dierent
classes. Overtwisted structures were classied by Eliashberg, and the clas-
sication scheme depends only on homotopic properties of the underlying
3-manifold. On the other hand, tight structures are expected to contain
more geometric information about the manifold. The contact counterpart
of the adjunction formula (frequently called the Bennequin inequality) char-
acterizes tight structures. This inequality reads as follows: Suppose that
is an embedded surface-with-boundary in the contact 3-manifold (Y, ) with
= L a Legendrian curve. (For the denitions of the notions used here,
see Chapter 4.) Let tb
(L) Z denote the framing induced by the contact

structure on L with respect to the framing denes on L and rot
(L) the
relative Euler number of [ with trivialized along by the tangents
of L. Now
Theorem 1.3.1 (Eliashberg, [26]). The inequality
tb
(L) +
rot
(L)
()
is satised for all L and if and only if the contact 3-manifold (Y, ) is
tight.
Contact structures and open books
Just like Donaldsons theory of symplectic Lefschetz pencils gives a topolog-
ical characterization of symplectic 4-manifolds, recent work of Giroux gives
a characterization of contact 3-manifolds in terms of open books. Giroux
proved that there is a one-to-one correspondence between open books and
contact structures on 3-manifolds up to some natural equivalence relations.
More precisely, for a given closed 3-manifold Y the following holds:
Theorem 1.3.2 (Giroux, [63]). (a) For a given open book decomposition
of Y there is a compatible contact structure on Y . Contact structures
compatible with a xed open book decomposition are isotopic.
(b) For a contact structure on Y there is a compatible open book de-
composition of Y . Two open book decompositions compatible with a xed
contact structure admit common positive stabilization.
22 1. Introduction
The reinterpretation of contact structures provided by this theorem enables
us to treat them as topological objects. The nicest manifestation of this
principle is probably the denition and application of contact Ozsvath
Szab o invariants discussed in Chapter 14. It is still an open (and very
intriguing) question how the monodromy of the open book decomposition
encodes tightness/llability properties of the corresponding compatible con-
tact structure (cf. Chapter 9). As an example of results in this direction,
we have the following theorem of Giroux:
Theorem 1.3.3. An open book decomposition gives rise to a Stein llable
contact structure if and only if it admits a positive stabilization for which
the monodromy decomposes as a product of right-handed Dehn twists.
Nonllable contact 3-manifolds
Suppose that (Y, ) is the boundary of a compact symplectic 4-manifold
(X, ) in the sense that X = Y as oriented manifolds and [
,= 0. In this
case we say that (Y, ) is (weakly) symplectically llable (or just llable),
and (X, ) is called a (weak) symplectic lling of (Y, ). The Bennequin
inequality in (Y, ) now follows from the adjunction inequality for (X, ),
i.e., llable structures are always tight. The converse of this implication,
however, does not hold: a contact manifold can be tight without being
the appropriate boundary of any symplectic 4-manifold. The rst such
structures were found by Etnyre and Honda [44]; we will give a variety of
such contact 3-manifolds, cf. also [100, 101]. Next we give a sample of these
results. Contact surgery provides a simple way for constructing contact 3-
manifolds. Because of its topological character, the surgery diagram can be
used very fruitfully to apply Heegaard Floer theory leading to the following
Theorem 1.3.4 ([101]). The contact 3-manifold (Y, ) given by the surgery
diagram of Figure 1.6 is tight but not llable.
Proof (sketch). Nonllability of this contact structure follows from the
fact that the underlying 3-manifold Y is dieomorphic to the boundary of
the (+E
7
)-plumbing (see Exercise 2.3.5(e) together with Remark 1.2.13).
Tightness follows from the fact that the contact OzsvathSzabo invariant
c(Y, ) is nonzero, although for overtwisted structures this invariant van-
ishes. (For an outline of the denition of c(Y, )

HF(Y ) see Chapter 14,
for the computation in the above case, see Section 12.2.)
1.3. Some contact results 23
+1
Figure 1.6. Surgery diagram of a tight nonllable contact 3-manifold
This simple construction leads us to a plethora of similar examples
see Section 12.2. By a variant of these ideas we get
Theorem 1.3.5 ([100]). For any n N there is a closed 3-manifold with
at least n distinct tight, nonllable contact structures.
These examples will be given by Figure 12.6. Once again, tightness will
be proved by (partially) determining the contact OzsvathSzabo invariants,
while we will show that the structures are nonllable through determining
homotopic properties of the contact structures via analyzing the diagram
and then apply a version of SeibergWitten theory. This last step is a
straightforward generalization of the proof of Theorem 1.3.4.
Topology of Stein llings
Another leading theme we will focus on in the study of a contact 3-manifold
is trying to determine all its Stein llings. As we will see, for some simple 3-
manifolds this problem can be solved, a prototype result (due to Eliashberg)
gives the following
Theorem 1.3.6 (Eliashberg). If W is a Stein lling of the standard contact
3-sphere (S
3
,
st
) then W is dieomorphic to the 4-dimensional disk D
4
.
Proof (sketch). Gauge theory as applied in the proof of Theorem 1.2.12
implies that b
+
2
(W) = b
2
(W) = 0. The surgered manifold Z = W
(CP
2
D
4
) is a symplectic 4-manifold containing a symplectic sphere of self-
intersection (+1), hence Z is symplectomorphic to CP
2
. Since a symplectic
sphere representing the generator of H
2
(CP
2
; Z) is isotopic to the complex
24 1. Introduction
line CP
1
CP
2
, we get that W is dieomorphic to D
4
. For more details of
this argument see Section 12.3.
Similar strong classication results of Stein llings have been obtained
for a variety of 3-manifolds (see [96, 110] or Section 12.3), but the general
description of all Stein llings of a contact 3-manifold is still missing. Here
we restrict ourselves to two statements along these lines:
Theorem 1.3.7 ([132]). There are contact 3-manifolds with innitely many
nondieomorphic Stein llings.
The proof of this theorem (see in Section 12.3) makes use of the connection
between Stein structures and Lefschetz brations. Using symplectic cut-
and-paste technique and applying SeibergWitten theory we will get some
restrictions on the topology of a Stein lling of a xed contact 3-manifold,
for example
Theorem 1.3.8 ([159]). For a given contact 3-manifold (Y, ) there exists
a constant K
(Y,)
such that if W is a Stein lling of (Y, ) then 3(W) +
2(W) K
(Y,)
. In other words, the number c(W) = 3(W) + 2(W) for
a Stein lling W of (Y, ) which resembles the c
2
1
-invariant of a closed
complex surface is bounded from below.
A little elaboration of the above result together with some specic cases
gives evidence for the following
Conjecture 1.3.9. For any contact 3-manifold (Y, ) there is a constant
K such that if W is a Stein lling of (Y, ) then for its Euler characteristic
(W) the inequality (W) K holds.
2. Topological surgeries
After the short Prelude given in the introductory chapter we begin our dis-
cussion by reviewing the smooth constructions behind contact and Stein
surgeries. We assume that the reader is familiar with basics in dieren-
tial topology as given, for example, in [72]. Standard facts regarding sin-
gular homology and cohomology theory will also be used without further
explanation. The manifolds appearing in these notes are all assumed to be
smooth (i.e., C
-) manifolds, possibly with nonempty boundary. The gen-

eral discussion of handlebodies will be followed by a short overview of Dehn
surgeries in dimension three, and an outline of Kirby calculus concludes the
chapter. For more details about the ideas and constructions sketched here,
see [66].
2.1. Surgeries and handlebodies
The main construction behind all surgeries can be summarized by the follow-
ing fairly simple scheme: Suppose that X
1
, X
2
are given n-dimensional man-
ifolds with boundaries and Z
i
X
i
are (n 1)-dimensional submanifolds
(with possibly nonempty boundary). For a dieomorphism f : Z
1
Z
2
we
can glue the two manifolds X
1
and X
2
together along Z
i
via f, and get a
new n-manifold X = X
1
#
f(Z
1
)=Z
2
X
2
(with possibly nonempty boundary).
In the following we will always assume that X
i
and Z
i
are compact (and
then so is X), and that the X
i
are oriented. Note that an orientation of X
i
induces one for X
i
and so orients Z
i
as well. In order to have a canonical
orientation for X, we assume that f reverses orientation.
Remark 2.1.1. In order to give a manifold structure to X we have to
round o the corners created by gluing along Z
i
(which might have (n2)-
dimensional boundaries). This process is fairly straightforward in dimen-
26 2. Topological surgeries
sion 2: we replace an angular corner by a region below a hyperbola, and by
multiplying this picture with the extra dimensions, the same can be carried
out in arbitrary dimensions, see [66].
The reason why the above construction works is that the boundaries X
i
and so Z
i
admit canonical neighborhoods (by the collar neighborhood
theorem), hence once the map f is xed, neighborhoods of Z
i
can be
identied and so the smooth structures can be patched together. The same
scheme will work for other structures (like symplectic, contact, and so on)
once the right assumptions ensuring canonical neighborhoods have been
made. The drawback of this general construction is that usually it is quite
hard to describe and identify f, although as we will see in many cases
the particular choice of the identication is crucial. Here are a few simple
examples of this operation:
Examples 2.1.2. (a) Suppose that X
1
, X
2
are compact manifolds with
boundaries X
1
, X
2
orientation reversing dieomorphic via a smooth map
f : X
1
X
2
. Then X = X
1
f
X
2
is a closed manifold.
(b) Suppose that X
i
are closed n-manifolds. Consider X
i
int D
n
and
glue them with an orientation reversing map f : S
n1
= (X
1
int D
n
)
S
n1
= (X
2
int D
n
) which extends to the disk D
n
. (This latter require-
ment species f up to isotopy.) The resulting manifold X = X
1
#X
2
is
called the connected sum of the two manifolds X
1
and X
2
. For X
i
con-
nected, the result can be proved to be independent of the choice of the
disks.
(c) Another special case of this general construction is when Z
i
X
i
(i = 1, 2) are both dieomorphic to the (n 1)-dimensional disk D
n1
.
The resulting manifold X is usually denoted by X
1
X
2
and is called the
boundary connected sum of X
1
and X
2
along Z
1
and Z
2
. As in the previous
case, the result can be proved to be independent of the choices provided the
boundaries X
1
and X
2
are connected.
(d) Suppose that X = X
1
X
X
2
and Y = Y
1
Y
Y
2
are closed manifolds
(X
i
X and Y
i
Y , i = 1, 2, are compact codimension-0 submanifolds
with boundaries and disjoint interiors). If there is an orientation preserving
dieomorphism f : X
1
Y
1
then we can use it to glue X
1
and Y
2
together
along their boundaries to get Z = X
1
Y
2
(and similarly V = Y
1
X
2
), see
Figure 2.1. The choice of f is usually crucial in this construction.
Exercises 2.1.3. (a) Let X
1
= X
2
= [0, 1] [0, 1] and Z
1
= Z
2
=
0 [0, 1] 1 [0, 1] [0, 1] [0, 1]. Find f such that the resulting
2.1. Surgeries and handlebodies 27
X
X
Y
Y
1
2
1
2
X
Y
Figure 2.1. Flipping X2 with Y2

manifold X
f
is dieomorphic to S
1
[0, 1] and g such that the result X
g
is
a Mobius band.
(b) Show that both S
3
and S
1
S
2
can be built by gluing two solid tori
S
1
D
2
together (using dierent identications of the boundary tori).
(c) More generally, show that every closed, oriented 3-manifold Y can be
given as Y = H
1
H
2
where H
1

= H
2
are solid genus-g three-dimensional
handlebodies with H
1

= H
2
=
g
, where
g
stands for the genus-g
surface. (Hint: Use a Morse function.) Such a decomposition is usually
called a Heegaard decomposition of Y .
(d) Verify that (X
1
X
2
) = X
1
#X
2
.
Notice that (S
k
D
nk
) = (D
k+1
S
nk1
) = S
k
S
nk1
, hence
if S
k
X
n
is a submanifold with trivial normal bundle S
k
= S
k
D
nk
then cutting out S
k
D
nk
and gluing back in D
k+1
S
nk1
we get
a new manifold. Once again, the chosen identications do matter. A
trivialization of S
k
in X is called a framing of the submanifold. By xing
a framing we get an embedding : S
k
D
nk
X. (If k 3 then
the parametrization of S
k
X is unique, otherwise we think of S
k
X
as given by the image of a map.) Now we can use [
(S
k
D
nk
)
to glue
D
k+1
S
nk1
back in; the new manifold is the result of the surgery of X
along S
k
(with the given framing). Notice that the connected sum operation
of Example 2.1.2(b) is just a special case of this surgery scheme: Embed
the disconnected manifold S
0
D
n
= 1 D
n
into X
1
X
2
in such a
way that 1 D
n
X
1
and 1 D
n
X
2
and then do surgery on it,
i.e., replace S
0
D
n
with the cylinder D
1
S
n1
. The eect of the above
surgery construction is the same as the following: Consider the (n + 1)-
manifold X [0, 1] and attach an (n + 1)-dimensional (k + 1)-handle (or
handle of index (k + 1)) D
k+1
D
nk
along the part D
k+1
D
nk
of
its boundary to (S
k
) X 1 with the specied framing the image
(S
k
) and the framing completely determine the gluing map f. During this
construction the tubular neighborhood (S
k
)D
nk
sinks in the interior of
the cobordism and D
k+1
S
nk1
appears on the surface. More generally,
if X is an (n + 1)-manifold with boundary X and : S
k
D
nk
X
is a given embedding, i.e., a framed sphere (S
k
) is given, then we can
glue the (n + 1)-dimensional (k + 1)-handle D
k+1
D
nk
to X using
and get a new manifold. The repeated application of the above process
(starting with a given closed n-manifold M and considering M [0, 1]) is
called a (relative) handlebody built on M. Notice that M might be the
empty manifold, in which case we get a handlebody. (In that case, to start
the process, we rst glue a 0-handle D
0
D
n+1
along D
0
D
n+1
= to
the empty manifold.) It can be shown that any compact smooth manifold
admits a handlebody decomposition, i.e., is dieomorphic to a handlebody.
(A relatively elementary proof of this statement can be found in [114, 115],
where Morse theory is applied.)
It is not hard to enumerate the possible framings an embedded sphere
can have: x a framing and try to relate all the others to this xed one.
Notice rst that framings need to be specied only up to homotopy. By
assuming linearity on the bers (which can be achieved by an isotopy),
any other framing denes a linear map at every point of the sphere (which
linear map matches up the chosen bases in the ber of the normal bundle),
so at the end we get a map S
k
GL
nk
(R). Since homotopy does not
change the framing, and GL
nk
(R) retracts to O(n k), we conclude that
the dierent framings of the k-dimensional sphere S
k
in an n-manifold
are parametrized by
k
_
O(n k)
_
. In particular, this shows that once
n 2 the framing is unique if k = n 1 or n. For k = 0 there are
two possible framings, corresponding to the two components of O(n). One
gives rise to an orientable, while the other to a nonorientable manifold.
Since we restrict our attention to the study of orientable manifolds, we get
uniqueness of framings even for k = 0. Notice that if we are dealing with
oriented 3- and 4-manifolds (so n = 2 or 3), then there is only one more
case to consider, namely when n = 3 and k = 1, i.e., when we glue 4-
dimensional 2-handles to a 3-dimensional boundary. In this case we frame
embedded circles in 3-manifolds, and the set of framings is parameterized by
1
_
O(2)
_

=
1
(S
1
)

= Z. In this special case the normal D
2
-bundle can be
regarded as a complex line bundle, hence it can be trivialized by a nowhere
vanishing section. In conclusion, a framing of a knot K in a 3-manifold can
2.1. Surgeries and handlebodies 29
be most conveniently symbolized by an appropriate push-o of K. In order
to set up an actual isomorphism between the set of framings and Z, we need
to choose a preferred framing rst (which we will call the 0-framing). This
choice, however, is canonical only in some special cases: for example, in S
3
or if the knot is null-homologous in the 3-manifold Y . Another instance
of the existence of a canonical framing is provided by the situation when
the knot is in Legendrian position in a contact 3-manifold, or if the knot
is naturally contained in a surface (which induces a natural framing by
pushing the knot o of itself inside the surface) such a surface can be
provided by a ber of a bration, or a page of an open book decomposition,
for example.
Exercises 2.1.4. (a) For K S
3
x a Seifert surface and consider the 0-
framing to be the push-o of the knot along the Seifert surface. Show that
this framing called the Seifert framing does not depend on the chosen
Seifert surface, and that the isomorphism between the space of framings
and Z is given by the linking number of K and the push-o of it along the
framing. (For the linking number to make sense, x an orientation on K
and orient any push-o accordingly.)
(b) Generalize the uniqueness of the Seifert framing for any null-homologous
knot in an arbitrary 3-manifold Y . (Hint: Argue that if two dierent
framings come from Seifert surfaces then their dierence vanishes in the
rst homology of the knot complement, contradicting the fact that the knot
is null-homologous.)
(c) Verify that the push-o K
of a null-homologous knot K Y denes

the Seifert framing if and only if the homology class [K
] of K
vanishes in
H
1
(Y K; Z).
Remark 2.1.5. When drawing the projection of a knot K R
3
in R
2
(with the usual genericity assumptions and the conventions of over- and
under-crossings), there is one more natural framing we can consider: it is
the blackboard framing bb(K) induced by the particular projection. We get
the blackboard framing bb(K) by pushing K o along a vector eld parallel
to the plane of the projection, see Figure 2.2. Notice that this framing
heavily depends on the chosen projection. The conversion between the
Seifert framing and the blackboard framing is given by the linking number
of K and its parallel push-o.
Exercise 2.1.6. For a given knot K S
3
compute the blackboard framing
bb(K) of one of its projections with respect to the Seifert framing of the knot.
Figure 2.2. The blackboard framing
Conclude that w(K) = bb(K), where w(K) is the writhe of the projection.
(The writhe is dened as the signed sum of crossings in the projection. For
this to make sense, we need to x an orientation on K, but w(K) can be
proved to be independent of this choice of orientation. For the sign of a
crossing see Figure 2.3.)
_
+
Figure 2.3. Positive and negative crossings
As a consequence of our framing computation above, we get that in
order to build 3-dimensional orientable handlebodies one only needs to keep
track of the attaching spheres. For 1-handles these are essentially unique
since S
0
= 1 1 embeds into a connected manifold uniquely up to
isotopy. For 2-handles we get embedded circles in a genus-g surface or
equivalently, in the plane with 2g holes, which are glued together in pairs.
For convenience, we identify S
2
with R
2
and draw the diagrams in
its nite part R
2
. These pictures are called Heegaard diagrams. (For
2.2. Dehn surgery 31
an example see Figure 2.4. Here the shaded disks are identied with each
other. In case of more pairs of shaded disks we connect the pairs to be
identied with dotted lines. The curve of Figure 2.4 might seem to have
many components, but after identifying the shaded disks it becomes a
connected 1-manifold.) Finally, a 3-handle can be attached uniquely to
Figure 2.4. A Heegaard diagram
a 3-manifold with boundary dieomorphic to S
2
. In dimension four the
2-handle attachment is somewhat more complicated; we will return to the
detailed discussion of this question in Section 2.3.
2.2. Dehn surgery
There is one more purely 3-dimensional construction we would like to
discuss, frequently called Dehn or rational surgery. The basic idea is again
pretty simple: consider a 3-manifold Y (for simplicity we assume that it is
closed), and x a knot K Y in it. By deleting a tubular neighborhood
K (
= S
1
D
2
) of K and regluing it via a dieomorphism f : K
(Y K) we get a new 3-manifold. Obviously, the resulting manifold will
depend on the chosen gluing map f. Notice that we reglue S
1
D
2
along
a (2-dimensional) torus, and the self-dieomorphisms f : T
2
T
2
are well-
understood: up to isotopy such an f is determined by the induced map
f
: H
1
(T
2
; Z) H
1
(T
2
; Z)

= Z
2
, i.e., (after xing a basis of H
1
(T
2
; Z))
by a 2 2 integer matrix. Since f (and so f
) is invertible, the matrix

is of determinant 1; the fact that f reverses orientation implies that
det f
= 1. Consequently, after xing a basis of H

1
(T
2
; Z), four integers
specify f. Dierent matrices might yield dieomorphic 3-manifolds, for
example, if for the maps f
1
, f
2
: (S
1
D
2
) (Y int K) the composition
f
1
2
f
1
: (S
1
D
2
) (S
1
D
2
) extends to a dieomorphism of S
1
D
2
then the surgered manifolds using f
1
or f
2
will be dieomorphic. Actually,
two of the four numbers already determine the surgery; we show this fact
from a slightly dierent point of view. Notice that S
1
D
2
can be thought
of as the union of a 3-dimensional 2-handle and a 3-handle, and remember
that the gluing of 3-handles is unique, while for 2-handles one only needs
to specify the gluing circle, which is an embedded simple closed curve in
(Y int K). (Recall that in dimension three there is no framing issue.)
Consequently the Dehn surgery is determined by a simple closed curve in
(Y int K), which can be given (up to isotopy) by xing its homology
class a H
1
_
(Y int K); Z
_

= Z
2
. This is the curve which bounds the
disk pt. D
2
in the surgered manifold.
Denition 2.2.1. For a xed closed 3-manifold Y , knot K Y and
primitive element a H
1
_
(Y int K); Z
_
the manifold (Y int K)
f
(S
1
D
2
) will be denoted by Y
a
(K), where f : (S
1
D
2
) (Y int K)
is specied by f
_
pt. D
2
= a H
1
_
(Y int K); Z
_
. The resulting
manifold Y
a
(K) is called the Dehn surgery of Y along K with slope a. Notice
that since
_
pt.D
2
is nondivisible and f
is invertible, the chosen class

a should be a primitive class. It can be proved that the simple closed curve
representing such a homology class a is unique up to isotopy.
A choice of a basis of H
1
_
(Y int K); Z
_

= Z
2
converts a into a pair
of relatively prime integers. A canonical choice for one basis element is pro-
vided by the meridian H
1
_
(Y int K); Z
_
of the knot K i.e., a
nontrivial primitive element which vanishes under the embedding of K
into K. By xing an orientation on K, the element is uniquely deter-
mined by the requirement that it links K with multiplicity +1 (otherwise
is determined only up to sign). Informally, is the homology class of the
circle which is the boundary of a small normal disk to K (i.e., a small disk
intersecting K transversely in a unique point), see Figure 2.5(a). The choice
of a longitude (another basis element in H
1
_
(Y int K); Z
_
) is, how-
ever, not canonical. For an example see Figure 2.5(b). The longitude can be
xed without further choices only if K admits a canonical framing. In fact,
xing a longitude or a framing is equivalent, since the normal R
2
-bundle
(regarded as a C-bundle) is trivialized by a nonvanishing section, i.e., by a
longitude. Therefore, if K Y is null-homologous (for example, Y = S
3
)
or if K is a Legendrian knot in a contact manifold, then using the canonical
longitude the homology class a can be converted into a pair of relatively
prime integers (p, q) by setting a = p+q, but in general the integers will

K
(b) (a)
K
Figure 2.5. (a) the meridian and (b) an example of a longitude for the knot K
depend on the choice of . An orientation of K xes an orientation for both
and , and by reversing the orientation on K these elements switch signs.
Therefore, although (p, q) depends on the chosen orientation, their ratio
p
q
does not. Notice that =
1
0
is also allowed. Therefore after xing a lon-
gitude if needed, the rational number
p
q
Q encodes all the gluing
information we need. For this reason Dehn surgery is also called rational
surgery. Notice nally that by denition the dierence of two framings
1
and
2
is some integral multiple of the class .
Remark 2.2.2. The two integers p and q can be easily recovered from the
matrix f
, since it maps the meridian of (S

1
D
2
) into p + q. This
shows that after the appropriate trivializations
f
=
_
p p
q q
_
with pq
q = 1.
Lemma 2.2.3. Fix K Y and a framing for K. If
p
q
Z (i.e., q = 1)
then Y
a
(K) = Y
p
q
(K) can be given by an ordinary surgery, i.e., by a (4-
dimensional) 2-handle attachment. If
p
q
= (i.e., q = 0) then Y
a
(K) = Y
for any knot K.
Proof. The coecient
p
q
being an integer means that the curve representing
a is simply a push-o of the knot K, therefore it determines a framing on it.
This shows that integral Dehn surgery has the same eect as 4-dimensional
handle attachment. If q = 0 then a = , so we simply glue back the 2-
handle of S
1
D
2
in the way it was before the surgery.
Remark 2.2.4. Notice that the fact that
p
q
is an integer is independent
of the choice of , since is canonical, and just species a parallel
circle to K. Similarly,
p
q
= is independent of the choice of since in
this case a can be represented by a meridian. Alternatively, observe that
for a = p
1
+ q
1
1
= p
2
+ q
2
2
we have that q
1
= q
2
and (provided
1

2
= k) that p
2
= p
1
+ q
1
k. This argument shows again that the
value of [q[ is independent of the chosen longitude.
The following fundamental theorem asserts representability of 3-mani-
folds by Dehn surgeries:
Theorem 2.2.5 (Lickorish and Wallace, [93]). Every closed, oriented 3-
manifold can be given as Dehn surgery on a link in S
3
. In fact, all rational
numbers used in the surgery presentation can be assumed to be integers.
Proof. Since the third cobordism group
3
vanishes, for a 3-manifold Y
there is an oriented 4-manifold W such that W = Y . Surgering out the
1- and 3-handles we get a presentation of Y as the boundary of D
4
some
2-handles, hence as an integral surgery on a link.
For a particular example of 3-manifolds we consider lens spaces. For this
matter, for coprime integers p > q 1 take the group
G
p,q
=
__
z 0
0 z
q
_

z C, z
p
= 1
_
U(2)
and denote the factor S
3
/G
p,q
by L(p, q). (By viewing S
3
C
2
as vectors of
unit length, the action of U(2) on S
3
is obvious.) As Exercise 2.2.6 shows,
L(p, q) is the result of
p
q
-surgery on the unknot. In fact, lens spaces are
exactly those 3-manifolds which admit Heegaard decompositions along the
2-dimensional torus T
2
.
Exercise 2.2.6. Show that the lens space L(p, q) is dieomorphic to the
result of a
p
q
-surgery on the unknot in S
3
. (Hint: See [148, Section 9B].)
There are certain operations we can use to manipulate our surgery dia-
grams without changing the resulting 3-manifold. In the following exercise
we list those moves which will be useful in our subsequent discussion.
Exercises 2.2.7. (a) Verify the slam-dunk operation, i.e., that the two
surgeries given by Figure 2.6 give dieomorphic 3-manifolds. Here it is
assumed that n Z and r Q . (Hint: Perform surgery on K
2
rst
and isotope K
1
into the glued-up solid torus T. Since rst we performed an
integral surgery, K
1
will be isotopic to the core of T, hence when performing
K
r
2
n
1
K
K
n
r 1
2
Figure 2.6. The slam-dunk operation
the second surgery we cut T out again and reglue it. Therefore it can be
done by one surgery; the coecient can be computed by rst assuming n = 0
and then adding n extra twists. For more details see [66, pp. 163164].)
(b) Turn a rational surgery in S
3
with coecient r into a sequence of
integral surgeries. (Hint: Use the continued fraction expansion of r and
apply (a) above. For the convention regarding continued fraction expansions
see Section 11.1.)
(c) Using the above result transform the Dehn surgery diagram of a lens
space into an integral surgery on a linear chain of unknots. Using this
diagram verify that L(p, q) = L(p, q
) if qq
1 (mod p).
Warning 2.2.8. Notice that the slam-dunk operation can be performed
only for n Z. Take, for example the 3-manifold given by the diagram of
Figure 2.7. Applying a slam-dunk on the
1
2
-framed circle we get L(2, 1) =
RP
3
. But if we perform the illegal slam-dunk on the (4)-framed circle, we
get S
3
as a result.
Exercises 2.2.9. (a) Verify the Rolfsen twist operation, i.e., that for n Z
the two surgeries given by Figure 2.8 give dieomorphic 3-manifolds. Here
the framing of K is r =
p
q
on the left and (
1
r
+n)
1
=
p
q+np
on the right; the
box with an n inside means n full twists (right-handed for n > 0 and left-
handed for n < 0) and the surgery coecient on a component of the link
RP
3
S
3
1
2
1
4
4
2
Figure 2.7. Warning with slam-dunks
intersecting the spanning disk of K changes from r
i
to r
i
+n
_
k(K, K
i
)
_
2
.
The term k(K, K
i
) denotes the linking number of the two knots K and K
i
.
(Hint: See [66, page 162].)
q+np
p
n
K
q
p
Figure 2.8. Rolfsen twist
(b) Verify that L(p, q) is dieomorphic to L(p, q + np) for any integer n.
(Hint: Introduce an -framed normal circle to the
p
q
-framed unknot,
perform Rolfsen twists and delete any -framed surgery curve.)
(c) Show that surgery on the disjoint union of two framed links yields the
connected sum of the two corresponding 3-manifolds. (We say that two
links are disjoint if they can be separated by a plane.)
(d) Verify that adding a disjoint unknot with surgery coecient (1) or
does not change the 3-manifold. (Hint: Show that (1)-surgery along the
unknot provides S
3
.)
(e) Describe a diagram for Y (the 3-manifold Y with opposite orientation)
in terms of a diagram for Y . (Hint: Take the mirror image of the link
presenting Y and multiply the framings by (1).)
In fact, there is a complete set of moves which determines when the
resulting 3-manifolds are dieomorphic:
Theorem 2.2.10 (Kirby, [80]). Two links L, L
with rational coecients in

S
3
determine dieomorphic 3-manifolds through Dehn surgery if and only if
L can be transformed into L
by a nite sequence of Rolfsen twists, isotopies

and inserting and deleting components with coecient .
We have to note here that in particular cases it might be quite dicult to
nd the actual nite sequence of moves transforming one surgery picture of
a given 3-manifold into another. Before turning to the 4-dimensional case,
we show a way to read o the rst homology of the 3-manifold at hand
from its rational surgery diagram. For this matter, suppose that Y is given
by rational surgery on the n-component link L = (K
1
, . . . , K
n
) S
3
with
surgery coecients
p
i
q
i
with respect to the meridians
i
and longitudes
i
,
where these latter provide the Seifert framings for K
i
in S
3
. It is not hard
to see that H
1
(S
3
n
i=1
int K
i
; Z) = Z
n
, freely generated by the homology
classes of the meridians: simply use the long exact homology sequence of
the pair (S
3
, S
3
n
i=1
int K
i
). Next, as the surgery procedure dictates, we
add a 3-dimensional 2- and a 3-handle to every T
2
-boundary component of
S
3
n
i=1
int K
i
. Notice that if
i
is a Seifert surface for K
i
(containing the
longitude
i
) then it provides the relation
i
=
j=i
k(K
i
, K
j
)
j
, where
k(K
i
, K
j
) stands for the linking number of the two knots in S
3
. Now
each 2-handle provides a relation among the
i
s: by denition p
i
i
+ q
i
i
becomes zero after the surgery (since this is the curve which bounds the
core of the new 2-handle). Therefore we conclude
Theorem 2.2.11. If Y is given by Dehn surgery along (K
1
, . . . , K
n
) S
3
with surgery coecients
p
i
q
i
(i = 1, . . . , n) then H
1
(Y ; Z) can be presented
by the meridians
i
(i = 1, . . . , n) as generators and the expressions
p
i
i
+q
i
j=i
k(K
i
, K
j
)
j
= 0
as relators.
Corollary 2.2.12. If Y is given by
p
q
-surgery along a knot K S
3
then
H
1
(Y ; Z)
= Z
p
. (Here Z
0
is interpreted as Z.)
We say that a 3-manifold Y is an integral homology sphere if H
(Y ; Z) =
H
(S
3
; Z); equivalently if H
1
(Y ; Z) = 0. The 3-manifold Y is a rational ho-
mology sphere if H
(Y ; Q) = H
(S
3
; Q); in other words, if
H
1
(Y ; Z)
< .
Alternatively, Y is a rational homology sphere if and only if its rst Betti
number b
1
(Y ) is zero.
Exercises 2.2.13. (a) Show that the 3-manifold S
3
r
(K) we get by r-surgery
on K S
3
is an integral homology sphere if and only if r =
1
k
for some
k Z.
(b) Suppose now that Y is given by (n
1
, . . . , n
k
)-surgery on the link L =
(K
1
, . . . , K
k
) S
3
(n
i
Z). Verify that Y is an integral homology sphere if
and only if the determinant of the linking matrix of L is 1. Show that Y
is a rational homology sphere if this determinant is nonzero. The diagonal
entries of the linking matrix are given by the surgery coecients. (Hint:
Use the long exact sequence for the pair of the 4-manifold X given by the
4-dimensional 2-handle attachment along L and the 3-manifold Y = X.)
2.3. Kirby calculus
Suppose that X
n
is a given smooth n-dimensional manifold. By choosing
an appropriate Morse function on X we see that it admits a handlebody
decomposition and we can always assume that our handlebody is built
by attaching handles in the order with increasing index to the 0-handle
D
n
. In this section we will focus on the n = 4 case. If X
4
is closed then
(according to a result of Laudenbach and Poenaru [91]) the gluing of the
union of 3- and 4-handles (which union is dieomorphic to
k
S
1
D
3
for
some k) is unique. Therefore, in order to present closed 4-manifolds, we may
restrict our attention to the discussion of 4-dimensional 2-handlebodies, i.e.,
2.3. Kirby calculus 39
handlebodies involving handles with index 2. In addition, a Stein surface
always admits a handle decomposition involving 0-, 1- and 2-handles only,
hence the study of 2-handlebodies is sucient for the purposes of these
notes. The attaching of a 1-handle (at least if we assume orientability,
which we always do) is unique up to isotopy. There are two common ways
of picturing the attachment of a 1-handle to the boundary S
3
of the unique
0-handle D
4
. (For convenience we identify S
3
with R
3
and use only
its nite part R
3
). We can draw a pair of D
3
s in R
3
, indicating where the
feet of the 1-handle are attached, or alternatively we can draw an unknot
with a dot on it, symbolizing that we consider the 4-manifold D
4
a
neighborhood of a spanning disk for the above unknot in D
4
, i.e., a dotted
circle refers to the compact 4-manifold
_
D
2
p
_
D
2
, where p
denotes a small tubular neighborhood of a point p int D
2
. Obviously, in
both ways we get a 4-manifold dieomorphic to S
1
D
3
. We will follow
the latter convention, therefore the subhandlebody X
1
=union of 1-handles
=
k
(S
1
D
3
) will be symbolized by a k-component unlink in S
3
with a
dot on every component: the unknots in S
3
simply denote the boundaries
of the disks D
2
p
i
(i = 1, . . . , k) deleted from D
4
. The 2-handles are
attached along a framed link in (
k
S
1
D
3
). By the above convention this
link can be regarded as lying in S
3
, therefore (using the Seifert framings)
the surgery coecients can be naturally converted into integers. A 2-handle
passes through a 1-handle exactly when its attaching circle links with the
dotted circle of the 1-handle. Such a link presentation of the 4-dimensional
2-handlebody is called a Kirby diagram.
Remark 2.3.1. One can easily convert a handle picture using the rst
convention into the dotted circle notation. To do this, rst isotope all
attaching circles away from the region between the two feet D
3
R
3
of
the 1-handle. Then delete the embedded 3-balls, connect the attaching
circles of the 2-handles and link them with a dotted circle. An example for
this procedure is given by Figure 2.9. The rst convention (which uses the
attaching balls of the 1-handle) is probably conceptually clearer, but when
manipulating the diagram of an explicitly given 4-manifold, the dotted circle
notation introduced by Akbulut in [3] is much more convenient.
Exercises 2.3.2. (a) Verify that Figure 2.10 gives a diagram for D
2
T
2
.
Visualize the bration on the diagram.
(b) Show the equivalence of Figure 2.10 with the diagram of Figure 2.11.
A given 4-manifold might admit many dierent Kirby diagrams. Since
any two Morse functions can be joined by a path of functions, by analyzing
(1) (2)
(3)
Figure 2.9. Converting 1-handle into dotted circle
0
Figure 2.10. D
2
T
2
0
Figure 2.11. An alternative diagram for D
2
T
2
the changes during such a path one can prove that two diagrams repre-
sent the same manifold if and only if they can be connected by repeated
applications of the following moves:
isotopies of the link in S
3
,
handle slides and
adding/deleting cancelling 1/2- and 2/3-handle pairs. (A pair of
handles is cancelling if their union amounts to a connected sum with
D
4
.)
In the diagram we visualize a 2-handle slide corresponding to circles K
1
, K
2
by connect summing K
1
to a push-o of K
2
corresponding to its framing
along an arbitrary band. The new surgery coecient K
1
becomes the sum of
the old coecients of the two knots twice their linking number the sign
depends on whether the connecting band respects or disrespects a chosen
orientation on K
1
and K
2
. One can slide 1-handles over each other as 0-
framed 2-handles, and a 2-handle slides over a 1-handle by treating the latter
as a 0-framed 2-handle. When sliding a 1-handle over an other 1-handle we
must be careful with the choice of the band, since the resulting dotted circles
should still form an unlink. A 1-handle/2-handle pair cancels if the 2-handle
intersects the spanning disk of the 1-handle in a single point; in this case
rst we slide o all the 2-handles geometrically linking the dotted circle in
question (using, for example, the cancelling 2-handle) and then erase the
1/2-handle pair from the picture. The process can of course be reversed
by introducing a pair of knots geometrically linking once (and one is the
unknot); then by putting a dot on the unknot and an arbitrary surgery
coecient to the other knot the 4-manifold remains the same. Finally, a
2-handle can be cancelled against a 3-handle if (possibly after handleslides)
it can be represented by a 0-framed unknot disjoint from the rest of the
picture. Notice that on an unknot (disjoint from the other dotted circles)
we can have surgery coecient 0 or a dot such a change corresponds to
surgery along the sphere given by the 0-framed unknot.
The operations listed above obviously do not change the boundary X
of a 4-manifold X given by a diagram. Changing a dot to 0-framing (or vice
versa) changes the 4-manifold but leaves the boundary intact. Besides these
moves, we can also insert or delete a (1)-framed unknot disjoint form the
rest of the picture which corresponds to adding and removing a copy of
CP
2
or CP
2
without changing the boundary of the 4-manifold. For more
details about these operations see [66]. The art of manipulating diagrams
using the above rules and understanding the structure of smooth 3- and 4-
manifolds in this way is frequently called Kirby calculus. Here we restricted
ourselves to outline the very basics of this theory, and highlighted only the
aspects which are important in our contact geometric studies. For a more
complete treatment of Kirby calculus the reader is advised to turn to [66].
Example 2.3.3. The sequence of moves given by [66, Figure 11.14] provides
a proof for the fact that the 4-manifolds X
1
, X
2
of Corollary 1.2.3 have
dieomorphic boundaries.
Suppose that X admits a handlebody decomposition with a single 0-
handle and some 1- and 2-handles. The homology groups H
i
(X; Z) and
H
i
(X; Z) can be easily read o from a diagram corresponding to such
a handle decomposition; this method will be discussed in the following.
Consider the Abelian groups C
1
and C
2
freely generated by [K
1
], . . . , [K
t
]
and [K
1
], . . . , [K
n
], corresponding to the t dotted circles and the n attaching
circles of the 2-handles respectively, and dene the map : C
2
C
1
by
[K]
t
i=1
k(K, K
i
)[K
i
]
on the generators and extend linearly. As for CW-complexes, we get
C
1
/ im

= H
1
(X; Z) and ker

= H
2
(X; Z). This latter identity follows
from C
3
= 0, which is the consequence of the absence of 3-handles. Now
the universal coecient theorem and Poincare duality allows us to com-
pute all homologies and cohomologies of X. In fact, the ring structure of
H
(X, X; Z) can also be read o from the picture. We restrict ourselves

to the case when there are no 1-handles in the decomposition: if the homo-
logy classes
1
, . . . ,
n
H
2
(X; Z) are represented by Seifert surfaces of K
i
together with the cores of the 2-handles (i = 1, . . . , n) then we can easily
see that PD(
i
) PD(
j
) = k(K
i
, K
j
); as before, PD(
i
) PD(
i
) = n
i
,
the framing of K
i
. In other words, in the basis PD(
1
), . . . , PD(
n
) of
H
2
(X, X; Z)

= H
2
(X; Z) the intersection form of X is represented by the
linking matrix of the link K
i
n
i=1
. Here PD denotes the Poincare duality
isomorphism between H
2
(X; Z) and H
2
(X, X; Z). By considering surfaces
F
i
in D
4
with F
i
= K
i
and gluing the core disks to them we might nd
lower genus representatives of the homology class
i
H
2
(X; Z). The geo-
metric intersections of these F
i
s are, however, harder to visualize.
Next we discuss the computation of H
i
(X; Z). First perform surg-
eries along the 1-handles, i.e., replace the dots on the dotted circles by 0.
This transforms X into a simply connected 4-manifold Z but leaves X
unchanged. Notice that Z is the union of a 0-handle and m(= t + n) 2-
handles which are attached (after renaming) along the knots K
1
, . . . , K
m
with framings n
1
, . . . , n
m
. According to the above said, H
2
(Z; Z) is freely
generated by the closed surfaces
i
we get by gluing an orientable Seifert
surface of the knot K
i
and the core of the 2-handle together. After xing
an orientation on K
i
these surfaces are canonically oriented: x the ori-
entation making K
i
the oriented boundary of the Seifert surface. Let D
i
denote a small meridional disk to K
i
. It is fairly straightforward to see that
H
2
(Z, Z; Z) is generated by the relative homologies represented by [D
i
]
(i = 1, . . . , m). Here we choose an orientation on these disks in such a way
that K
i
intersects D
i
positively when we use the orientation on K
i
xed
above. The long exact homology sequence of the pair (Z, Z) reduces to
0 H
2
(Z; Z) H
2
(Z; Z)

1
H
2
(Z, Z; Z)

2
H
1
(Z; Z) 0
(since H
3
(Z, Z; Z)

= H
1
(Z; Z) = 0 and H
1
(Z; Z) = 0 by the simple
connectivity of Z). As Theorem 2.2.11 shows, the map
1
is given by
1
_
[
i
]
_
= n
i
[D
i
] +
j=i
k(K
i
, K
j
)[D
j
],
while
2
is simply
2
_
[D
i
]
_
= [D
i
] =
i
, where
i
denotes the homology
class of the linking normal circle of the knot K
i
oriented in such a way that
their linking number is (+1). The exact sequence (with the maps described
above) provides an explicit presentation for both H
1
(Z; Z) = H
1
(X; Z)

=
H
2
(X; Z) and H
2
(Z; Z) = H
2
(X; Z)
= H
1
(X; Z).
With introducing more notation, in fact we can picture cobordisms
involving only 1- and 2-handles. To this end, consider the cobordism W
from Y
1
to Y
2
. First present Y
1
as X
4
for some 4-manifold X and draw
a diagram for X. Next, add the knots corresponding to the handles of W,
and distinguish the two sets of curves by putting the framings of the link
producing X into brackets; for a simple example see Figure 2.12. There
< 2 >
1
Figure 2.12. A relative Kirby diagram of a cobordism from RP
3
to S
3
is one rule we have to obey with handleslides and cancellations in such a
cobordism: handles in X cannot be slid over handles in the cobordism W
and handles in X cannot be cancelled against handles in W. On the other
hand, we can obviously slide handles of W over handles of X.
It is only a little more complicated to investigate homologies in cobor-
disms. Suppose that W is a given cobordism from Y
1
to Y
2
. Fix a 4-manifold
X with X = Y
1
, and suppose that it is given by attaching 2-handles to D
4
along a framed link L. For the sake of simplicity, suppose furthermore that
W is given by a single 2-handle attachment to Y
1
. Denote the 4-manifold
X W by X
.
Exercises 2.3.4. (a) Determine the homology class in H
2
(X
; Z) generating
H
2
(W, W; Z). (Hint: Consider a primitive homology class H
2
(X
; Z)
such that Q
X
(, ) = 0 for all H
2
(X; Z) H
2
(X
; Z).)
(b) Determine the self-intersection Q
W
(, ) of this generator.
(c) Find a surface in W representing the above H
2
(W; Z). (Hint:
Use the above computation to represent H
2
(X
; Z) with a surface. By
adding extra handles make sure that the surface is disjoint from the cores
of all the 2-handles dening X. Now show that the surface is in W.) Notice
that dierent presentations of Y
1
as X
4
might provide dierent estimates
on the genus of a surface representing .
(d) Go through the above computations for the cobordism provided by
Figure 2.12. Find a torus of square (2) in this cobordism.
(e) Let K S
3
be a given knot with 4-ball genus g
s
(K). Perform n-surgery
along K and denote the resulting 3-manifold S
3
n
(K) by Y . Let K
be a
meridian to K and dene the cobordism W by attaching a 2-handle along
K
with surgery coecient k. Let H

2
(W, W; Z) denote a generator.
Compute the self-intersection of and give an estimate for the genus of a
surface representing it in W.
We conclude this section with a few examples and exercises. The 3-
manifold Y given by Figure 2.13 is called a Seifert bered 3-manifold with
.........
.........
.........
.........
n
g
r
k
r
1
1 1
r
1
2
Figure 2.13. A Seifert bered 3-manifold
Seifert invariants (g, n; r
1
, . . . , r
k
) (g, n N, r
i
Q). Notice that the dotted
circles form an unlink in the diagram. If r
i
1 then we say that this set
of invariants is in standard form. Note that by applying Rolfsen twists any
such diagram can be transformed into standard form. When g = n = 0,
the 3-manifold with Seifert invariants (g, n; r
1
, . . . , r
k
) is usually denoted
by M(r
1
, . . . , r
k
). Notice that according to this convention the surgery
coecients are negative reciprocals of the given data.
Exercises 2.3.5. (a) Determine the intersection matrix of the 4-manifold
X given by Figure 2.14. (This manifold is frequently called the Gompf
nucleus.) What is H
1
(X; Z)?
n
0
Figure 2.14. Kirby diagram for the nucleus Nn
(b) Verify that Figure 2.15 gives a 4-manifold X dieomorphic to the disk
n
Figure 2.15. Disk bundle over a genus-3 surface with Euler number n
bundle : D
3,n

3
over the genus-3 surface
3
with Euler number n.
Draw the diagram of D
g,n
for an arbitrary positive integer g and n Z.
Compute the intersection form, signature and Euler characteristic for D
g,n
.
(c) By inverse slam-dunks nd a 4-manifold X such that X = M(g, n; r
1
,
. . . , r
k
). (Hint: Use the continued fraction expansions of r
i
Q, cf. Exer-
cise 2.2.7(b).)
(d) Verify that the boundary of the (+E
7
)-plumbing (a truncation of the
long leg of the diagram of Figure 1.5) is dieomorphic to the 3-manifold
we get by doing (+2)-surgery on the right-handed trefoil knot see Fig-
ure 2.16. (Hint: Adapt [66, Figure 12.9] to the present problem.)
Figure 2.16. Right-handed trefoil knot
(e) Prove that (+5)-surgery on the right-handed trefoil knot is a lens space.
(Hint: Use the exercise above and truncate the long leg of the (+E
7
)-
plumbing.)
(f ) Show that r-surgery on the right-handed trefoil knot gives the Seifert
bered manifold M(
1
2
,
1
3
,
1
r6
). Use this fact to reprove (c) above.
Determine the Seifert invariants of the result of (+6)- and (+7)-surgeries
on the trefoil knot.
(g) Generalize the above result to a (2, 2n +1) torus knot T
(2,2n+1)
. (Hint:
S
3
r
(T
(2,2n+1)
) is dieomorphic to M(
1
2
,
n
2n+1
,
1
r4n2
), cf. [102].)
Another family of 3manifolds is provided by the Brieskorn spheres
(p, q, r) (p, q, r N). Such a 3manifold can be most conveniently dened
as the oriented boundary of the compactied Milnor ber V (p, q, r), where
V (p, q, r) =
_
(x, y, z) C
3
[ x
p
+y
q
+z
r
= , [x[
p
+[y[
q
+[z[
r
1
_
for 0 < small. In other words, (p, q, r) can be identied with the link of
the isolated singularity x
p
+y
q
+z
r
= 0. By perturbing the equation we
rather consider the smoothing of this singularity the introduction of the
perturbing term leaves the topology of the link (p, q, r) unchanged.
It can be shown that the smooth Milnor ber V (p, q, r) admits a plumb-
ing description, and the 3manifolds (p, q, r) are Seifert bered manifolds.
The computation of the Seifert invariants from the triple (p, q, r) N
3
can
be rather involved.
In order to x our convention, we remark here that we orient (2, 3, 5)
(i.e., the Poincare homology sphere) as the boundary of the negative de-
nite E
8
-plumbing, which is the same as (1)-surgery on the left-handed tre-
foil knot. Consequently, (+1)-surgery on the right-handed trefoil provides
(2, 3, 5), which is the boundary of the positive denite E
8
-plumbing. This
orientation convention is consistent with complex geometry the Poincare
sphere with its natural orientation is the oriented boundary of the compact-
ied Milnor ber V (2, 3, 5), where we equip this latter 4-manifold with the
orientation naturally induced by its complex structure.
Performing (+1)-surgery on the left-handed trefoil knot we get (2, 3, 7)
and (1)-surgery on the right-handed trefoil gives (2, 3, 7).
Examples 2.3.6. (a) As it follows from the above discussion, (2, 3, 5) =
M(
1
2
,
1
3
,
1
5
) and in a similar vein (2, 3, 4) = M(
1
2
,
1
3
,
1
4
) and
(2, 3, 7) = M(
1
2
,
1
3
,
1
7
).
(b) In general, however, the transition from (p, q, r) to M(r
1
, r
2
, r
3
) is less
simple, for example (2, 3, 6n 1) = M(
1
2
,
1
3
,
n
6n1
).
3. Symplectic 4-manifolds
In this section we recall some general facts about symplectic manifolds.
Then we give a short discussion of Mosers method, which is applied in the
proof of numerous fundamental statements discussed in the text. The chap-
ter concludes with a short review on what is known about the classication
of symplectic 4-manifolds. For a more detailed treatment of symplectic geo-
metry and topology the reader is advised to turn to [111]; here we restrict
our attention mostly to the 4-dimensional case.
3.1. Generalities about symplectic manifolds
Denition 3.1.1. A 2-form on the smooth n-manifold X is a symplectic
form if is closed (i.e., d = 0) and nondegenerate (i.e., for any nonzero
tangent vector v there is w with (v, w) ,= 0). The pair (X, ) is called a
symplectic manifold.
Since any antisymmetric form is degenerate on an odd dimensional vector
space, a symplectic manifold is necessarily even dimensional.
Examples 3.1.2. (a) For R
2n
with coordinates (x
1
, y
1
, . . . , x
n
, y
n
) the 2-
form
st
=
n
i=1
dx
i
dy
i
is symplectic, called the standard symplectic
structure on R
2n
.
(b) The above form is invariant under translations, hence denes a sym-
plectic form on the 2n-torus T
2n
= R
2n
/Z
2n
.
(c) Let g denote the FubiniStudy metric on the complex projective space
CP
n
. Then
FS
(u, v) = g(iu, v) is a symplectic form on CP
2
.
50 3. Symplectic 4-manifolds
(d) If (X
i
,
i
) are symplectic (i = 1, 2) then their product X
1
X
2
with
any of the pulled back forms
2
is a symplectic manifold. (The
map
i
: X
1
X
2
X
i
denotes the projection to the i
th
factor.)
(e) A volume form on an oriented surface is a symplectic form.
Exercises 3.1.3. (a) Show that the nondegeneracy of is equivalent to
the nonvanishing of
n
= . . . (n times). Notice that in this way
provides an orientation for X; for oriented manifolds we require the two
orientations to agree.
(b) Show that the sphere S
n
admits a symplectic structure only if n = 2.
(c) Prove that S
1
S
3
does not carry any symplectic structure. Show
the same for CP
2
. Here CP
2
denotes the complex projective plane with its
natural (complex) orientation reversed. (Hint: Note that d = 0 implies
that [] represents a cohomology class in H
2
(X; R), and []
n
> 0 follows
from nondegeneracy and the compatibility with the given orientation. Use
compactness of the above manifolds.)
(d) Prove that a smooth projective variety (i.e., a complex manifold with
a holomorphic embedding into some complex projective space) admits a
symplectic structure.
(e) Verify that for any smooth manifold V the cotangent bundle T
V with
the 2-form d is a symplectic manifold, where the Liouville 1-form is
dened as
p
(v) = p(
v) for p T
V , v T
p
(T
V ) and : T
V V .
It turns out that symplectic manifolds are close to complex manifolds
in the sense that their tangent bundles can be equipped with complex
structures. For this to make sense we need a denition:
Denition 3.1.4. A linear map J : TX TX is an almost-complex struc-
ture if J
2
= id
TX
. An almost-complex structure is said to be compatible
with a given symplectic structure if (Ju, Jv) = (u, v) and for u ,= 0 we
have (u, Ju) > 0, that is g(u, v) = (u, Jv) is a Riemannian metric. If
and J are compatible then (X, , J, g) is called an almost-K ahler manifold.
For any symplectic structure there exists a compatible almost-complex
structure J, moreover the space of such Js is contractible. (This state-
ment can be proved berwise.) In conclusion, the tangent bundle TX of a
symplectic manifold (X, ) carries a complex structure. In fact, all compat-
ible almost-complex structures are homotopic to one another, therefore the
Chern classes c
i
(X, ) H
2i
(X; Z) are well-dened.
3.1. Generalities about symplectic manifolds 51
Denition 3.1.5. A submanifold of an almost-Kahler manifold
(X, , J, g) is symplectic if the restriction [
T
is a symplectic form on
. The submanifold is J-holomorphic (or pseudo-holomorphic) if T is J-
invariant, that is, v T TX implies Jv T. The submanifold L X
is Lagrangian if [
L
= 0. Finally, L X is totally real if T
l
L JT
l
L = 0
for all l L.
Example 3.1.6. Recall that for a symplectic manifold (X, ) the product
X X with () is a symplectic manifold. It is not hard to see that
the submanifolds X pt. and pt. X are symplectic submanifolds of
_
XX, ()
_
while the diagonal
_
(x, x) X [ x X
_
is Lagrangian.
Exercises 3.1.7. (a) Show that (X, ) is symplectic if and only if
[
T
is nondegenerate.
(b) Suppose that and J are compatible. Show that a J-holomorphic
submanifold is symplectic. Find a counterexample for the converse.
(c) Show that a submanifold is symplectic if and only if there is a
compatible J for which it is J-holomorphic.
(d) Show that if L is Lagrangian and J is an -compatible almost-complex
structure then L is totally real. (Hint: Show that if V is a complex subspace
of T
x
X then [
V
,= 0.)
Following the holomorphic analogy, J-holomorphic curves in an almost-
complex 4-manifold intersect positively, more precisely:
Theorem 3.1.8 ([109]). Suppose that the surfaces
1
and
2
are J-
holomorphic submanifolds of the almost-complex 4-manifold X. If
1
and
2
do not share a component then [
1
] [
2
] 0, with equality if and only
if the submanifolds are disjoint.
One of the most important formulae in the study of symplectic manifolds
is the following adjunction equality, which is just a simple manifestation of
the Whitney product formula for characteristic classes:
Theorem 3.1.9. If
2
X
4
is a symplectic submanifold then () =
[]
2
c
1
(X), []
_
.
Proof. Notice that in order for c
1
(X) to make sense we need to x an -
compatible almost-complex structure. If is symplectic, one can choose
a compatible J such that becomes a J-holomorphic curve. Then the
splitting TX[
= T (as complex bundles) gives
c
1
(X), []
_
=
c
1
(T), []
_
+
c
1
(), []
_
.
The identity
c
1
(T), []
_
=
e(T), []
_
= ()
for the Euler characteristic is fairly straightforward, while
c
1
(), []
_
= []
2
needs only a little argument realizing that a push-o of gives rise to a
section of .
One of the goals of symplectic topology is to understand which man-
ifolds admit symplectic structures and if they do, how many inequivalent
structures do they carry. Using Gromovs h-principle it can be shown that
every open 2n-manifold admits a symplectic structure (see [33, 56]); the
question is more subtle for closed manifolds. In order to understand top-
ological properties of symplectic 4-manifolds, rst we have to understand
obstructions to the existence of symplectic structure and describe construc-
tions of symplectic manifolds. According to the following proposition, the
existence of an almost-complex structure depends only on the homotopy
type of a 4-manifold. Recall that the existence of a symplectic structure
implies the existence of an almost-complex structure.
Proposition 3.1.10 (Wu, [175]). A closed, oriented 4-manifold X carries
an almost-complex structure if and only if there is a class h H
2
(X; Z)
such that h w
2
(X) (mod 2) and h
2
= 3(X) + 2(X). In particular,
a simply connected, closed 4-manifold X is almost-complex if and only if
b
+
2
(X) is odd.
Remark 3.1.11. If X is not closed then the 4-dimensional cohomology
class h
2
H
4
(X; Z) might have noncompact support, hence it might not
be integrable on X. Therefore
h
2
, [X]
_
might not be dened. If X is
compact with nonempty boundary X and the restriction of h to X is
torsion then h
2
can be dened as a rational number as follows: the multiple
nh will vanish on the boundary X, hence the square (nh)
2
has compact
support, so the expression h
2
=
1
n
2
(nh)
2
Q is a well-dened quantity. The
dierence h
2
3(X) 2(X), however, is not necessarily zero anymore for
an almost-complex structure. It provides an invariant of the oriented 2-
plane eld induced by the complex tangencies on X; for more about this
topic see Section 6.2. We note here that since the congruence h w
2
(X)
(mod 2) always admits a solution (which implies, in particular, the existence
of a spin
c
structure on X), every nonclosed 4-manifold carries an almost-
complex structure.
3.1. Generalities about symplectic manifolds 53
A ner obstruction to the existence of is given by the following theorem
of Taubes. (For the brief denition of the SeibergWitten function SW
X
and more on Taubes work see Chapter 13.)
Theorem 3.1.12. If X carries a symplectic structure then for the
SeibergWitten invariant SW
X
we have that
SW
X
_
c
1
(X, )
_
= 1.
Moreover, if K H
2
(X; Z) and SW
X
(K) ,= 0 then
K []
c
1
(X, ) []
,
with equality if and only if K = c
1
(X, ).
It can be shown, for example, that SW
3CP
2 0, hence although 3CP
2
ad-
mits an almost-complex structure, it cannot be equipped with a symplectic
structure.
Above we saw obstructions to the existence of symplectic structures,
in the following we will describe some constructions to produce symplectic
manifolds. As we already mentioned, all K ahler surfaces are symplectic.
One of the most eective ways of constructing symplectic manifolds is
the symplectic normal connected sum operation, which we will describe
in Section 7.1. Another source of examples comes from surface bundles over
surfaces, since we have
Theorem 3.1.13 (Thurston, [166]). If the 4-manifold X admits a bration
X
g
such that the ber has genus dierent from 1 then X admits a
symplectic structure.
Remark 3.1.14. If the ber is a torus, similar result cannot be expected,
since S
1
S
3
admits a torus bration over the sphere: multiply the Hopf
bration S
3
S
2
by a circle. On the other hand, as a theorem of Geiges
[55] shows, torus bundles over tori are all symplectic. A generalization of
Theorem 3.1.13 to more general Lefschetz brations will be discussed in
Section 10.1.
These constructions give partial results regarding the existence of a sym-
plectic structure on a given smooth 4-manifold. Such studies are usually
called geographic questions of symplectic 4-manifolds. For overviews of
various aspects of such geographic questions see [14, 144, 152, 156]. The
next problem is: how many symplectic structures can a 4-manifold carry.
Such investigations are usually called botany. To make the question pre-
cise, we have to clarify what do we mean by distinct symplectic structures.
Denition 3.1.15. Let X be a given 4-manifold and
0
,
1
two symplectic
forms on it. The forms
0
and
1
are said to be deformation equivalent if
there is a smooth path of symplectic forms interpolating between them. The
form
0
is the pullback of
1
if there is a dieomorphism f : X X such
that f
1
=
0
. Finally,
0
and
1
are equivalent if they lie in the same
equivalence class under the equivalence relation generated by the above two
relations.
Theorem 3.1.16 ([112, 150, 170]). For any n N there is a simply
connected 4-manifold X
n
which carries at least n inequivalent symplectic
structures.
The construction of the manifolds and the symplectic structures uses
Gompfs symplectic normal connected sum operation (given in Theorem
7.1.10). By computing c
1
of the various symplectic structures it is easy to
show that they are deformation inequivalent (since the c
1
s are distinct).
By proving that the dierent c
1
s lie in dierent orbits of Diff
+
(X) it fol-
lows that the symplectic structures are inequivalent. In this last step either
the divisibilities of the integral cohomology classes show the nonexistence
of certain dieomorphisms [150], or the SeibergWitten equations pose re-
strictions on the action of Diff
+
(X) on H
2
(X; Z), see [112, 170].
The spectacular success of the results of Taubes on SeibergWitten
invariants of symplectic 4-manifolds indicates that appropriate extensions
of these techniques might lead to new results for a much broader class of
4-manifolds. Such a potential extension was initiated by Taubes [165] by
considering singular symplectic forms, that is, closed 2-forms nondegenerate
only away from a subset of the given closed 4-manifold X.
Exercise 3.1.17. Suppose that for a given 4-manifold X the condition
b
+
2
(X) 1 holds. Show that there exists a closed 2-form which is
nondegenerate away from the closed 1-manifold Z = x X [
x
= 0 of
its zeros. (Hint: Fix a metric on X and consider harmonic representatives
of a second cohomology class of positive square. Choose generic metric.)
The analysis for setting up a correspondence between J-holomorphic curves
in XZ (with appropriate boundary conditions) and SeibergWitten solu-
tions on X is much more complicated than in the symplectic case, and it is
in the focus of current research. A fairly explicit way of nding a singular
symplectic form on a closed 4-manifold X with b
+
2
(X) 1 is given by Gay
and Kirby [54]. This procedure makes use of symplectic surgery in the spirit
it is discussed in later chapters.
3.2. Mosers method and neighborhood theorems 55
3.2. Mosers method and neighborhood theorems
In this section we shortly outline the circle of ideas usually referred to as
Mosers method. Using this method we can prove that symplectic manifolds
have no local invariants. This last statement can be interpreted in two
ways: (i) a symplectic manifold is locally standard, or (ii) a small deforma-
tion of the symplectic structure produces symplectomorphic manifold. The
rst statement actually generalizes to neighborhoods of special submani-
folds, while the second holds for deformations keeping the cohomology class
dened by the symplectic form xed (see Mosers Stability Theorem 3.2.1).
The main idea can be easily summarized: Suppose that X is compact
and
t

2
(X) is a family of symplectic forms with exact derivative:
d
dt
t
= d
t
. We claim that in this case there is a family
t
Diff(X)
of dieomorphisms such that
t
=
0
. The dieomorphisms
t
can be
constructed via the ow of the family of vector elds X
t
they induce by
d
dt
t
= X
t

t
,
0
= id .
The key point is that if
t
=
0
then for X
t
we have
0 =
d
dt
t
=
t
_
d
dt
t
+
Xt
d
t
+d
Xt
t
_
=
t
d(
t
+
Xt
t
),
since d
t
= 0 and
d
dt
t
= d
t
. Therefore a vector eld X
t
satisfying
t
=
Xt
t
will be appropriate for our purposes. This equation is, indeed, easy to
solve for X
t
since
t
is nondegenerate. Then solving
d
dt
t
= X
t

t
for
t
we get the family
t
with the desired property. (This last step can
be achieved without any problem provided the manifold X is compact; the
general case needs some more care.) Applying this principle, one can deduce
the following (see [111]):
Theorem 3.2.1 (Mosers Stability Theorem). Suppose that
t
_
t [0, 1]
_
is a family of symplectic forms on the closed manifold X and [
t
] = [
0
].
Then there is an isotopy
t
such that
0
= id
X
and
t
=
0
.
Proof. In order to apply the above principle we need to show the existence
of a family
t
with
d
dt
t
= d
t
. Since [
t
] is constant, we obviously get
that
d
dt
t
is exact; applying Hodge theory, for example, a smooth family of
appropriate
t
can be chosen.
More interestingly, Mosers method shows that symplectic manifolds are
locally the same. This principle rests on the following result:
Theorem 3.2.2. Suppose that X is a smooth manifold with Y X a
compact submanifold and
1
,
2

2
(X) two closed 2-forms which are
equal and nondegenerate on T
y
X for all y Y . Then there are open
neighborhoods N
1
, N
2
of Y in X and a dieomorphism : N
1
N
2
such
that [
Y
= id and
2
=
1
.
Proof (sketch). By applying Mosers argument described above, the the-
orem reduces to nding a 1-form
1
(N
1
) with d =
2

1
and
[
T
Y
X
= 0. In fact, with such the family
t
= (1 t)
1
+t
2
=
1
+td
will be symplectic on a neighborhood of Y . This follows from the fact that
nondegeneracy is an open condition, while d
t
= (1 t)d
1
+ td
2
= 0.
Therefore the argument of Moser provides an appropriate vector eld X
t
which vanishes along Y . By possibly shrinking N
1
, this implies the exis-
tence of and N
2
with the properties given in the theorem. For the explicit
construction of see [111, page 95].
Applying this theorem for Y = pt. we get Darbouxs theorem:
Theorem 3.2.3 (Darboux). For a point x X in the symplectic manifold
(X, ) there is a chart U X containing x such that
_
U, [
U
_
is symplecto-
morphic to some open set V R
2n
equipped with the standard symplectic
form
st
[
V
.
Remark 3.2.4. One can dene symplectic manifolds by requiring that ev-
ery point admits a neighborhood symplectomorphic to some open set in
(R
2n
,
st
) and the transition functions between such charts respect the sym-
plectic structures on the charts. This approach turns out to be equivalent
to Denition 3.1.1.
In fact, symplectic structures are standard not only around points, but also
near symplectic and Lagrangian submanifolds. In the following we formulate
these theorems only for 4-dimensional symplectic manifolds.
Theorem 3.2.5 (Symplectic neighborhood theorem, Weinstein [172]). Sup-
pose that (X
i
,
i
) is a symplectic 4-manifold with 2-dimensional closed sym-
plectic submanifolds
i
X
i
for i = 1, 2. Suppose furthermore that there is
an isomorphism F : (
1
) (
2
) of the normal bundles (
i
)
i
cov-
ering a symplectomorphism f :
_
1
,
1
[
1
_

_
2
,
2
[
2
_
. Then f extends
to a symplectomorphism on some tubular neighborhoods of the surfaces
i
.
3.2. Mosers method and neighborhood theorems 57
Proof (sketch). The symplectomorphism f guarantees that
1
and f
2
coincide on T
1
TX
1
. By choosing appropriate neighborhoods we can
assume that the two structures are equal in the normal direction as well
using the isomorphism of the normal bundles. Then an application of
Theorem 3.2.2 yields the result.
With a small modication of this argument we get:
Theorem 3.2.6 (Lagrangian neighborhood theorem, Weinstein [172]). Let
(X, ) be a symplectic 4-manifold and L X a compact Lagrangian sub-
manifold. Then there is a neighborhood V X of L in X and a neigh-
borhood U T
L of the zero-section in the cotangent bundle of L and a

dieomorphism : U V such that
= d and [
L
= id, where is
the Liouville form on T
L.
Recall that the Liouville form on T
L is dened by
p
(v) = p(
v) for
: T
L L and v T
p
(T
L). For more details of the proofs see [111]

or McDus lectures in [34]. Notice that for symplectic submanifolds we
need the existence of a dieomorphism F : (
1
) (
2
) and a symplecto-
morphism f :
_
1
,
1
[
1
_

_
2
,
2
[
2
_
; the topology around a symplectic
submanifold is not unique and symplectic structures might be dierent
for dieomorphic s. (For example, the volume
_
n/2
is an invariant.)
The isomorphism type of the normal bundle () of X
4
is determined
by the self-intersection number []
2
Z. After possibly rescaling
2
on X
2
there exists a symplectomorphism f :
_
1
,
1
[
1
_

_
2
,
2
[
2
_
once
1
and
2
are dieomorphic, that is, the genera g(
1
) and g(
2
) are equal.
In conclusion, the assumptions of Theorem 3.2.5 can be checked from the
topology of the situation.
In the Lagrangian case, on the other hand, [
L
= 0 holds, so the topology
of L already determines its neighborhood, as the following exercise shows.
Exercise 3.2.7. Show that for L
2
X
4
Lagrangian we have [L]
2
= (L).
(Hint: Fix a compatible almost-complex structure J and show that JT
p
L
is the orthogonal complement of T
p
L with respect to the metric g
J
induced
by and J.) In conclusion, two orientable Lagrangian 2-manifolds in a
symplectic 4-manifold admit symplectomorphic neighborhoods if and only
if the genera of the surfaces are equal.
3.3. Appendix: The complex classification scheme for
symplectic 4-manifolds
In the following we give a short overview about the present status of the
smooth classication of closed symplectic 4-manifolds. The classication
scheme tries to imitate the classication results obtained for compact com-
plex surfaces (for a detailed description of the latter see [10]), hence rst
we introduce the notion of minimality and Kodaira dimension of symplec-
tic 4-manifolds. Since this discussion falls outside the main theme of this
volume, we mainly give the statements without proofs.
Denition 3.3.1. A symplectic 4-manifold X is minimal if it does not
contain a symplectic submanifold S X dieomorphic to the 2-sphere S
2
with [S] [S] = 1.
Remark 3.3.2. A detailed analysis of the SeibergWitten invariants shows
that minimality is equivalent to requiring that X does not contain any
smoothly embedded 2-sphere S with [S] [S] = 1.
A symplectic 4-manifold can always be blown up in a point by imitating
the corresponding complex operation (for an extended discussion see [111,
page 233]); i.e., if X admits a symplectic structure then so does its blow-
up X
= X#CP
2
. In this latter symplectic 4-manifold the generator of
the H
2
(CP
2
; Z)-factor can be represented by a symplectic sphere of square
(1). Using the symplectic normal connected sum operation (for the de-
tailed description see Section 7.1), we can prove the converse: if S X
is a symplectic sphere with square (1) then X is the blow-up of another
symplectic manifold. This implies
Lemma 3.3.3. A symplectic 4-manifold X can be written as Y #nCP
2
where Y is a minimal symplectic manifold. Y is called a minimal model
of X.
Proof. If S X is a symplectic sphere of self-intersection (1) then
X = X
1
#CP
2
since S is dieomorphic to CP
2
int D
4
. Taking the
symplectic sum of X and CP
2
along S and CP
1
CP
2
(as it is discussed in
Theorem 7.1.10) we nd that X
1
is symplectic. Repeating the above process
completes the proof. Notice that each step reduces b
2
(X) by 1, hence this
procedure will terminate after nitely many steps.
3.3. Appendix: The complex classication scheme for symplectic 4-manifolds 59
Remark 3.3.4. The minimal model is not necessarily unique; for example
CP
2
#2CP
2
= S
2
S
2
#CP
2
admits dierent minimal models (CP
2
and
S
2
S
2
) according to the order of blow-downs. For a related discussion see
Remark 3.3.9.
Let us assume that (X, ) is a minimal symplectic 4-manifold. Following
the complex analogy, its Kodaira dimension is dened as follows: Fix an
-compatible almost-complex structure J and consider c
1
(X, ) = c
1
(X, J).
Denition 3.3.5.
If c
1
(X, )[] > 0 or c
2
1
(X, ) < 0 then the Kodaira dimension (X)
of X is .
In case c
1
(X, )[] = 0 we say that X is of Kodaira dimension 0.
For c
1
(X, )[] < 0 and c
2
1
(X, ) = 0 we dene (X) = 1.
Finally, if c
1
(X, )[] < 0 and c
2
1
(X, ) > 0 then (X) = 2.
If X is nonminimal then (X) is dened as the Kodaira dimension of
its minimal model.
Theorem 3.3.6 (Liu, [103]). If (X, ) is a minimal symplectic 4-manifold
with c
2
1
(X, ) < 0 then X is a ruled surface, that is, an S
2
-bundle over a
Riemann surface.
It follows that (X) is dened for any minimal symplectic 4-manifold X,
and it is well-dened, since by Theorem 3.3.6 the above cases are mutually
disjoint. In principle (X) might depend on the minimal model chosen,
since X
min
might not be unique.
Proposition 3.3.7. If the symplectic 4manifold has a minimal model X
min
with (X
min
) 0 then this minimal model is unique up to dieomorphism.
Therefore the Kodaira dimension (X) of any symplectic 4-manifold is well-
dened.
Note that the quantity c
2
1
(X, ) is equal to 3(X) + 2(X), hence depends
only on the topology of X. As a consequence, it can be shown that (X)
depends only on the oriented dieomorphism type of X.
As it turns out, we have a fairly good understanding of the topology of
symplectic 4-manifolds with = :
Theorem 3.3.8 (Liu, [103]). If X is minimal and (X) = then X is
dieomorphic either to CP
2
or to a ruled surface.
Remark 3.3.9. All these manifolds carry complex, in fact, K ahler struc-
tures. According to the classication of complex surfaces, these are all the
K ahler surfaces with (complex) Kodaira dimension [10]. Suppose that
we blow up a ber of a ruled surface X
g
. When constructing the
minimal model of this symplectic 4-manifold, we can choose which (1)-
sphere to blow down: the exceptional sphere of the blow-up or the proper
transform of the ber. It is not very hard to see that the result of one
blow-down is spin, while the other is not. Therefore the minimal model of
the blown-up 4-manifold is not unique. It can be shown that further blow-
ups of these symplectic 4-manifolds are the only ones admitting nonunique
minimal models.
The next theorem follows from Taubes correspondence between Seiberg
Witten and GromovWitten invariants (see Chapter 13 for the statement).
Theorem 3.3.10 ([111]). If (X) = 0 and b
+
2
(X) > 1 then c
1
(X, ) = 0.
In the case b
+
2
(X) = 1, the assumption (X) = 0 implies 2c
1
(X, ) = 0.
Examples of such manifolds are provided by the K3-surface, T
2
-bundles
over T
2
(which are all symplectic by the quoted result of Geiges) and the
Enriques surface. This latter manifold is the quotient of a K3-surface by
an appropriate free Z
2
-action, therefore its fundamental group is Z
2
and
the rst Chern class is a nonzero torsion element of order two. (For a con-
struction see [66].) Note that according to Theorem 3.1.12 the assumption
c
1
(X, ) = 0 implies that for a manifold X with b
+
2
(X) > 1 there is a unique
basic class (i.e., K H
2
(X; Z) with SW
X
(K) ,= 0), and this unique class is
equal to 0. According to a result of Morgan and Szab o, a simply connected
4-manifold X with SW
X
(0) odd is homeomorphic to the K3-surface, hence
it can be proved that
Theorem 3.3.11 (MorganSzab o, [120]). If (X, ) is a simply connected
symplectic 4-manifold with (X) = 0 then X is homeomorphic to the K3-
surface.
Remark 3.3.12. Complex surfaces with Kodaira dimension 0 are classied:
besides torus bundles over the torus or the sphere there are the K3-surfaces
and Enrique surfaces. For more detail see [10, pp. 188189]. It seems
reasonable to expect that all symplectic 4-manifolds with = 0 admit a
genus-1 Lefschetz bration, hence these are essentially torus bundles, the
K3-surface and the Enriques surface.
Much less is known about symplectic manifolds with = 1 in general.
For example,
3.3. Appendix: The complex classication scheme for symplectic 4-manifolds 61
Theorem 3.3.13 (Gompf, [64]). If G is a nitely presented group then
there is a symplectic 4-manifold (X, ) with (X) = 1 and
1
(X)
= G.
In the simply connected case, however, the homeomorphism type of (mini-
mal) symplectic 4-manifolds with (X) = 1 is understood:
Theorem 3.3.14. If X is a minimal simply connected symplectic 4-mani-
fold with (X) = 1 then X is homeomorphic to an elliptic surface.
All complex surfaces of Kodaira dimension 1 are elliptic surfaces and can be
constructed from E(1) = CP
2
#9CP
2
and torus bundles using ber sum and
an additional operation called logarithmic transformation. For additional
discussion on the topology of elliptic surfaces see [66].
If the symplectic 4-manifold (X, ) has (X) = 2 then we say that it is
of general type. Such examples are provided by complex surfaces of general
type (since these complex manifolds are all algebraic, hence K ahler). Recall
that for a complex surface the condition (X) = 2 implies c
2
1
(X) 3c
2
(X),
the famous BogomolovMiyaokaYau inequality. (For a related discussion
see [153].) This inequality, or some similar relation between c
2
1
and c
2
is
conjectured to hold for symplectic 4-manifolds of general type. We do not
know too much about the topology of symplectic 4-manifolds of general
type. The following conjecture (usually attributed to Gompf) would provide
a strong topological restriction:
Conjecture 3.3.15. If a symplectic 4-manifold (X, ) satises (X) 0
then for its Euler characteristic (X) 0 holds.
Notice that ruled surfaces might have negative Euler characteristic (depend-
ing on the genus of the base), but 0 excludes them in the conjecture.
The following lemma provides a tool for studying (X) of symplectic 4-
manifolds with b
+
2
(X) = 1:
Lemma 3.3.16. If (X, ) is a symplectic 4-manifold and b
+
2
(X) = 1 then
either (X) = or b
1
(X) 0, 2.
Proof. By the existence of an almost-complex structure we get that 1
b
1
+ b
+
2
= 2 b
1
is even, therefore b
1
(X) is even. Now by (X) 0 and
Theorem 3.3.8 we have that 0 c
2
1
(X
min
, ) = 3(X
min
) + 2(X
min
)
= 3
_
b
+
2
(X
min
) b
2
(X
min
)
_
+ 2
_
2 2b
1
(X
min
) +b
+
2
(X
min
) +b
2
(X
min
)
_
= 9 b
2
(X
min
) 4b
1
(X
min
),
showing that 4b
1
(X
min
) = 4b
1
(X) 9, which yields the result.
Conjecture 3.3.17. There is no symplectic 4-manifold X with b
+
2
(X) = 1,
b
1
(X) = 2 and b
2
(X) = 0. (Notice that such a 4-manifold would provide a
counterexample to Conjecture 3.3.15.)
4. Contact 3-manifolds
This chapter is devoted to the recollection of basic facts about contact
manifolds. As before, we start with the general case, but very quickly
specialize to 3-manifolds. To understand the topology of contact 3-mani-
folds we consider submanifolds and the contact structures near them. The
contact version of Darbouxs theorem says that every point in a contact
3-manifold has a neighborhood which is standard regardless of the contact
structure. Then we consider knots which are always tangent or always
transverse to the contact planes and examine their classical invariants. It
turns out that the contact structures near these types of knots are essentially
unique. For an arbitrary surface embedded in a contact 3-manifold we look
at the characteristic foliation induced by the contact structure to extract
information. It is typical to move a surface by a small isotopy to modify its
characteristic foliation to get a generic picture and/or to eliminate certain
type of singularities. As it turns out, the characteristic foliation determines
the contact structure near the surface. A more complete treatment of the
ideas and theorems collected here can be found in e.g. [1, 39, 56, 57].
4.1. Generalities on contact 3-manifolds
Denition 4.1.1. Suppose that Y is a given (2n+1)-dimensional manifold.
A 1-form
1
(M) is a contact form if (d)
n
is nowhere zero. The
2n-dimensional distribution TM is a contact structure if locally it can
be dened by a contact 1-form as = ker .
Example 4.1.2. The standard contact structure
st
on R
2n+1
can be given
in the coordinates (x
1
, y
1
, . . . , x
n
, y
n
, z) as ker (dz +
n
i=1
x
i
dy
i
). The com-
plex tangents to S
2n1
C
n
also form a contact structure.
64 4. Contact 3-manifolds
According to a classical result of Frobenius, the plane eld = ker
is integrable if and only if d = 0. Integrability is equivalent to be-
ing closed under Lie bracket, hence = ker is integrable if and only if
_
[v
1
, v
2
]
_
= 0 whenever (v
i
) = 0 (i = 1, 2). Recall that d
_
[v
1
, v
2
]
_
=
L
v
2
(v
1
) L
v
1
(v
2
) +
_
[v
1
, v
2
]
_
, hence if is integrable then d vanishes
on = ker . So is maximally nonintegrable if d is nondegenerate on
, i.e., (d)
n
,= 0. Therefore the contact condition can be interpreted as
maximally nonintegrable. In other words, is a contact form if d is a
symplectic form on the hyperplane distribution ker . As we will see, Dar-
bouxs theorem generalizes to the contact setting (Theorem 4.1.13), hence in
conjunction with Remark 3.2.4 it can be shown that contact structures can
be given by patching open subsets of R
2n+1
together with transition func-
tions respecting
st
. The existence of contact structures on open manifolds
(similarly to the symplectic case) follows from an appropriate h-principle
see [33, 56], for example. The question becomes more subtle on closed (odd
dimensional) manifolds.
From now on we will assume that Y is a 3-manifold, that is, in the above
denition n = 1. In the following we describe a few examples of contact
structures on R
3
, T
3
and S
3
just to illustrate how contact structures may
look like in 3-manifolds.
Exercise 4.1.3. Verify that
1
= dz + xdy and
2
= dz y dx dene
contact structures on R
3
. Visualize the contact planes
i
= ker
i
, for
i = 1, 2. (Hint: For the latter see Figure 4.1.)
Examples 4.1.4. (a) The form
3
= dz + r
2
d (with polar coordinates
(r, ) on the (x, y)-plane) gives a contact structure
3
= ker
3
on R
3
.
To check this, realize that in (x, y, z)-coordinates the form
3
is equal to
dz +xdy y dx, and so
3
d
3
= 2 dxdy dz = 2r dr d dz. One can
easily see that the contact planes are spanned by
r
, r
2
z

. These
planes are horizontal (i.e., parallel to the xy-plane) along the z-axis and
as we move out along any ray perpendicular to the z-axis the planes will
twist in a clockwise manner. The contact structure
3
is obviously invariant
under translation in the z-direction and under rotation in the (x, y)-plane.
Notice that the planes will not twist too quickly as the twisting angle is
an increasing function of r which monotone converges to

2
as r tends to .
(b) Similarly, the form = cos r dz + r sin r d is a contact 1-form on R
3
.
To check this we calculate
d = sinr dr dz + (sin r +r cos r) dr d,
4.1. Generalities on contact 3-manifolds 65
x
y
z
Figure 4.1. The contact planes of the standard contact structure on R
3
d = r sin
2
r d dr dz + (sin r cos r +r cos
2
r) dz dr d
=
_
1 +
sin r cos r
r
_
r dr d dz,
and it is easy to see that 1 +
sin r cos r
r
> 0 for r > 0. Again, as in the
previous example, ker admits the same symmetries as
3
and the contact
planes (spanned by
r
, r tan r

z

) are horizontal along the z-axis

and they will twist in a clockwise manner as we move out along any ray
perpendicular to the z-axis. This time, however, the contact planes will
make innitely many full twists as r goes to . More generally, the 1-form
n
= cos f
n
(r) dz +sin f
n
(r) d, where f
n
(r) is a strictly monotone function
equal to r
2
near r = 0 and asymptotic to n +

2
(as r ) provides a
contact form on R
3
for every nonnegative integer n. For n = 0 this form
gives the standard contact structure on R
3
.
(c) Let us identify the 3-torus T
3
with R
3
/Z
3
. For any positive integer
n the 1-form sin(2nx) dy + cos(2nx) dz dened on R
3
induces a contact
structure
n
on T
3
.
(d) Consider the smooth map f : R
4
R dened by f(x
1
, y
1
, x
2
, y
2
) =
x
2
1
+ y
2
1
+ x
2
2
+ y
2
2
. Let p denote the point (x
1
, y
1
, x
2
, y
2
). It is clear that
S
3
= f
1
(1) and T
p
S
3
= ker df
p
= ker(2x
1
dx
1
+2y
1
dy
1
+2x
2
dx
2
+2y
2
dy
2
).
By identifying R
4
with C
2
we can dene a complex structure J on each
tangent space of R
4
by J

x
i
=

y
i
and J

y
i
=

x
i
for i = 1, 2. Let be
the plane eld of complex tangencies of J along S
3
, i.e.,
p
= T
p
S
3
J(T
p
S
3
).
We claim that is a contact structure on S
3
. To show this we will nd a
contact 1-form on S
3
such that = ker . Consider the 1-form df J
on R
4
. By evaluating on the basis vectors

x
1
,

y
1
,

x
2
,

y
2
it is easy to
check that df
p
J = 2x
1
dy
1
2y
1
dx
1
+ 2x
2
dy
2
2y
2
dx
2
. Moreover we
have J(T
p
S
3
) = ker(df J) since J
2
= id. Let =
1
2
(df J)
S
3
. It is
clear that = ker . It is straightforward to check that d is nonzero
on S
3
. We will check this only at a point p = (x
1
, y
1
, x
2
, y
2
) on S
3
where
x
1
,= 0, y
1
,= 0, y
2
,= 0. A basis for the tangent space T
p
S
3
can be chosen as
_

x
1
x
1
y
1
y
1
,

x
2
x
2
y
2
y
2
,

x
1
x
1
y
2
y
2
_
.
Now it is easy to see that d > 0 on this basis. Hence we conclude
that = (x
1
dy
1
y
1
dx
1
+x
2
dy
2
y
2
dx
2
)
S
3
is a contact form. We dene
= ker as the standard contact structure on S
3
and denote it by
st
.
(e) For a more subtle source of examples consider a complex manifold (X, J)
with a function : X R such that the symmetric 2-form g
(u, v) =
dJ
d(u, Jv) is a Riemannian metric. (Here J denotes multiplication by

i on TX and J
is the induced map on T
X.) Then the 1-form
given by
(v) = dJ
d(
g
, v) denes a contact form on
1
(a) for a regular
value a of . We will return to these examples in Chapter 8.
Exercise 4.1.5. Show that the 1-form
n
dened in Example 4.1.4(b) is a
contact form for every nonnegative integer n.
Denition 4.1.6. Two contact 3-manifolds (Y, ) and (Y
) are called con-

tactomorphic if there is a dieomorphism f : Y Y
such that f
() =
.
If = ker and
= ker
, this is equivalent to the existence of a nowhere

zero function g: Y R such that f
= g. Two contact structures and
on a manifold Y are said to be isotopic if there is a contactomorphism

h: (Y, ) (Y,
) which is isotopic to the identity.

In fact, two contact structures on a closed manifold are isotopic if and only
if they are homotopic through contact structures by Theorem 4.1.16 below.
Notice also that there exist contact structures which are contactomorphic
but not isotopic, see Exercise 11.3.12(c).
Exercises 4.1.7. (a) Prove that
1
= dz +xdy,
2
= dz y dx and
3
=
dz +xdy y dx = dz +r
2
d dene contactomorphic contact structures on
R
3
. (Hint: For identifying the contact structures given by
1
and
3
, use the
dieomorphism (x, y, z) = (x,
y
2
, z+
xy
2
) or (x, y, z) = (
x+y
2
,
yx
2
, z+
xy
2
).)
(b) Let p be any point in S
3
. Show that (S
3
p,
st
[
S
3
{p}
) is contacto-
morphic to (R
3
,
st
). (Hint: See [57] for a complete solution.)
In case can be dened by a global 1-form, we say that the contact
structure is coorientable. This implies that is orientable as a 2-plane
eld on Y . In fact, the 2-plane eld underlying the contact structure is
orientable if and only if can be dened by a global 1-form. Given an
oriented 3-manifold Y , we say that = ker is a positive contact structure
on Y if the orientation of Y coincides with the orientation given by d.
Notice that the orientation induced on Y by d is independent of
the contact form dening the contact structure . In the following we
will always assume that the contact structures at hand are positive and
cooriented. We orient the normal direction to the contact planes by , or
equivalently the contact planes are oriented by d. If we choose as
our contact form then the normal orientation and hence the orientation of
the contact planes will be reversed but d() will induce the same
orientation on the 3-manifold.
Denition 4.1.8. Suppose that the contact structure is given as ker
for the contact 1-form
1
(Y ). The vector eld R
on Y satisfying
d(R
, .) = 0 and (R
) = 1 is called the Reeb vector eld of . In other

words, at each point p Y , the Reeb vector eld points in the direction
where the skew-symmetric 2-form d
p
(of rank 2) degenerates in the tangent
space T
p
Y and it is uniquely determined by the normalization condition
(R
) = 1. Notice that R
is transverse to the contact planes and depends

on the contact 1-form, not just on the contact structure. In general the
Reeb vector eld R
f
will be very dierent from R
for a nowhere vanishing

function f : Y R.
Exercises 4.1.9. (a) Suppose that a 1-form on Y such that (R
) = 0.
Show that there is a unique vector eld X with X(p)
p
on Y such that
X
d = . (Hint: Choose X =
(e
2
)e
1
(e
1
)e
2
d(e
1
,e
2
)
with respect to some (local)
frame e
1
, e
2
of .)
(b) Find the Reeb vector elds for the contact forms dened in Exam-
ples 4.1.4(a) and (b). (Hint: The answers are R
3
=

z
, and
R
= (r sin r +cosr +r cot r cos r)

1

+
+ (1 +r cot r)
_
r sin r + (1 +r cot r) cos r
_
1

r
.)
Next we turn to the study of submanifolds in contact 3-manifolds.
Denition 4.1.10. Suppose that (Y, ) is a given contact 3-manifold. A
knot K Y isLegendrian if the tangent vectors TK satisfy TK , i.e.,
(TK) = 0 for the contact 1-form dening . The knot K is transverse
if TK is transverse to along the knot K, i.e., if (TK) ,= 0. The contact
framing of a Legendrian knot is dened by the orthogonal of along K.
(In other words, push K o in the normal direction to .) Equivalently,
we can take the framing obtained by pushing K o in the direction of a
nonzero vector eld transverse to K which stays inside the contact planes.
This framing is the ThurstonBennequin framing of the Legendrian knot.
Remark 4.1.11. If K is null-homologous in (Y, ) then it admits a natural
0-framing provided by any embedded surface Y with = K, cf.
Exercise 2.1.4(b). In this case the ThurstonBennequin framing can be
converted into an integer tb(K) Z: measure the ThurstonBennequin
framing with respect to the Seifert framing, i.e., the natural 0-framing.
Notice that the 0-framing does not depend on the chosen surface , therefore
the resulting number tb(K) will be independent of . Also notice that by
the coorientation of a transverse knot T (Y, ) comes with a natural
orientation: choose the nonzero tangent vector v to be positive if (v) > 0
for the contact 1-form (Y ) providing the given coorientation for .
Similarly to the symplectic case, contact structures are the same locally
in either sense. As before, these theorems rest on the following result.
Theorem 4.1.12. Let Y be a given 3-manifold with N Y a com-
pact subset. Consider contact structures
0
,
1
on Y which coincide as
cooriented contact structures on N, i.e.,
0
[
N
=
1
[
N
as oriented 2-plane
elds. Then there exists a neighborhood U of N and a contactomorphism
:
_
U,
0
[
U
_

_
U,
1
[
U
_
which is isotopic to id
U
rel N.
Proof (sketch). The proof makes use of Mosers method again (see Sec-
tion 3.2). Assume that
0
,
1
are contact forms giving rise to the contact
structures
0
,
1
. Since
0
[
N
=
1
[
N
, there exists a function f : N R
+
such that
1
= f
0
on the compact set N. Then consider the family
t
= (1 t)
0
+ t
1
of 1-forms on Y . To see that
t
is a contact form on
N for every t [0, 1], we calculate that
t
d
t
=
_
(1 t)
2
+ 2f(1 t)t
0
d
0
+t
2
1
d
1
> 0.
In addition, one can show that there is a neighborhood U of N such that
t
is a contact form on U for every t. (Notice that
t
is not necessarily contact
on the entire Y .) We would like to represent the map in the theorem
as the time1 map of a family of dieomorphisms
t
with (
t
)
0
=
t
and
0
= id. Equivalently, we start with the equation
t
=
t
0
and
dierentiate it with respect to t, yielding
d
dt
(
t
) =
t
_
d
dt
t
+ L
Xt
t
_
=
d
dt
0
=
d
dt
t
.
(The rst equality is an exercise in dierential forms and its proof can be
found, for example, in [12].) As before, X
t
denotes the family of vector
elds induced by
t
, i.e.,
d
dt
t
= X
t

t
. Taking
t
=
d
dt
(log
t
)
1
t
,
the above equation gets the form
t
_
d
dt
t
+d
_
t
(X
t
)
_
+
Xt
d
t
_
=
t
(
t
t
),
which is solved by X
t
ker
t
provided
(4.1.1)
d
dt
t
+
Xt
d
t
=
t
t
.
Now consider the Reeb vector eld R
t
of the contact 1-form
t
and plug
it into the above equation to get
d
dt
t
(R
t
) =
t
.
This denes the function
t
, hence Equation (4.1.1) above can be uniquely
solved for X
t
, since d
t
is nondegenerate on ker
t
. (Recall that this non-
degenracy is equivalent for
t
being a contact 1-form.) Now X
t
integrates
to the desired ow
t
.
Applying this principle for N = pt. we get
Theorem 4.1.13 (Darbouxs theorem for contact structures). For every
y Y in the contact 3-manifold (Y, ) there is a neighborhood U Y such
that
_
U, [
U
_
is contactomorphic to
_
V,
st
[
V
_
for some open set V R
3
.
As in the symplectic case, similar argument extends to special sub-
manifolds. For stating the relevant results, consider the contact structure
1
= ker
_
cos(2) dx sin(2)dy
_
and
2
= ker(cos rd + r sin r d) on
S
1
R
2
. (Here is the coordinate in the S
1
-direction, while (x, y) are
Cartesian and (r, ) are polar coordinates on R
2
.)
Exercise 4.1.14. Show that S
1
0 is Legendrian for
1
and transverse
for
2
.
By taking N = S
1
, the neighborhood theorems 3.2.5 and 3.2.6 now translate
to
Theorem 4.1.15 (Contact neighborhood theorems). If K (Y, ) is a
Legendrian knot then there are neighborhoods U
1
Y of K and U
2

S
1
R
2
of S
1
0 such that
_
U
1
, [
U
1
_
and
_
U
2
,
1
[
U
2
_
are contactomorphic
via a contactomorphism mapping K to S
1
0. If K is transverse,
then some neighborhood of it is contactomorphic to some neighborhood
of S
1
0 in (S
1
R
2
,
2
) again K is mapped to S
1
0.
For a detailed proof of the Legendrian case which will be more important
from our present point of view see [57] or [1, Section 2.2]. The proof
relies on Theorem 4.1.12 after nding a map f taking K to S
1
0 in
such a way that f
maps [
K
to
i
[
S
1
{0}
. Notice that since both the
1-manifold K and its normal bundle N(K) is topologically unique, no
topological assumption is needed for the neighborhood theorems to hold.
Another important corollary of the principle of Theorem 4.1.12 is Grays
stability theorem, the contact version of Mosers stability Theorem 3.2.1
the other manifestation of the principle that contact structures admit no
local invariants.
Theorem 4.1.16 (Gray, [71]). If
t
(t [0, 1]) is a smooth family of
contact forms on a closed 3-manifold Y then there is an isotopy
t
of Y
such that
0
= id and
t
=
t
0
for some smooth family of smooth
functions
t
: Y R
+
. In particular, (
t
)
0
=
t
for
t
= ker
t
.
Remark 4.1.17. Notice that the theorem deals with contact structures as
opposed to contact forms; in general one cannot nd
t
satisfying
t
=
0
.
There is an intimate relationship between symplectic and contact man-
ifolds, which will not be discussed in full detail here. (See Denition 12.1.1
as an example of this relationship.) In our cut-and-paste construction we
will frequently refer to a symplectic manifold Symp(Y, ) associated to a
contact manifold (Y, ) called the symplectization of (Y, ). In order to
dene Symp(Y, ), choose a contact form for and consider
Symp(Y, ) = v T
m
Y [ v = t
m
for some t > 0.
It is easy to see that Symp(Y, ) is dieomorphic to Y (0, ) and that
for any other contact 1-form with ker = the 1-form or is
simply a section of Symp(Y, ). (This trivially follows from the fact that
if ker = ker then = f for some f C
(Y ) with f > 0 or f < 0.)

By taking = d([
Symp(Y,)
) we get a closed 2-form on Symp(Y, ) here
stands for the Liouville 1-form on T
Y dened as
p
(v) = p
_
(v)
_
for
p T
Y, v T
p
(T
Y ) and projection : T
Y Y . We claim that is a
symplectic form, that is, ,= 0.
Exercise 4.1.18. By considering the contact 1-form as a map : Y
T
Y show that
= . Using the same simple idea verify that on (Y )

the forms
and coincide.
Therefore t
= [
Symp(Y,)
, hence = d(t
) = dt
+ t
d. Now
since d d = 0, we get that = 2t
_
dt
( d)
_
,= 0; showing
that denes a symplectic structure on Symp(Y, ).
Remark 4.1.19. An alternative way to describe the symplectic 2-form
on Symp(Y, ) is to take the 1-form = t on Y (0, ) and dene as
d = t d +dt ; the result is clearly the same. Notice that the resulting
symplectic form is exact.
Exercise 4.1.20. Show that L (Y, ) is Legendrian if and only if LR
Symp(Y, ) is Lagrangian.
4.2. Legendrian knots
In order to have a better understanding of the topological constructions we
will introduce in Chapter 8, we discuss a way to visualize Legendrian knots in
the standard contact S
3
(or, equivalently, in R
3
) equipped with the standard
contact structure
st
= ker(dz+xdy). See [40] for more on Legendrian knots.
Consider a Legendrian knot L (R
3
,
st
) and take its front projection, i.e.,
its projection to the yz-plane. Notice that the projection has no vertical
tangencies (since
dz
dy
= x ,= ), and for the same reason at a crossing
the strand with smaller slope is in front. A straightforward computation
(see [57]) shows that L can be C
2
-approximated by a Legendrian knot
for which the projection has only transverse double points and (2, 3)-cusp
singularities (see Figure 4.2). Vice versa, a knot projection with these
(a) (b)
Figure 4.2. Cusp singularity of the projection
properties (that is, cusps instead of vertical tangencies and no crossings
depicted by Figure 4.3(a)) gives rise to a unique Legendrian knot in (R
3
,
st
)
dene x from the projection as x =
dz
dy
. Since any projection can be
isotoped to satisfy the above properties, we can easily show that every knot
can be isotoped to Legendrian position. (This knot is, however, far from
being unique up to Legendrian isotopy.)
Lemma 4.2.1. Any knot K S
3
can be isotoped to a Legendrian knot.
Proof. Consider a generic projection of K R
3
S
3
onto the yz-plane.
Isotope the knot near the nitely many points where
dz
dy
= by adding
cusps. At each crossing make sure that the strand with more negative slope
crosses in front by adding zig-zags if necessary (see Figure 4.3). The Legen-
drian knot can be recovered from a projection with these properties.
4.2. Legendrian knots 73
(a) (b)
Figure 4.3. Introducing new zig-zags at an illegal crossing
Remark 4.2.2. In fact, any knot K in a contact 3-manifold can be C
0
-
approximated by a Legendrian knot; for the proof of this statement see [57],
for example.
The contact framing tb(L) of a knot L can be computed as follows.
(Recall that we measure the contact framing with respect to the Seifert
framing in S
3
.) Dene w(L) (the writhe of L) as the sum of signs of the
double points (see Figure 4.4) for this to make sense we need to x an
orientation on the knot, but the answer will be independent of this choice,
cf. also Exercise 2.1.6.
Lemma 4.2.3. If c(L) is the number of cusps, then the ThurstonBennequin
framing tb(L) given by the contact structure is equal to w(L)
1
2
c(L) with
respect to the framing given by a Seifert surface.
_
+
Figure 4.4. Positive and negative crossings
Proof. The equality tb(L) = w(L)
1
2
c(L) can be seen by noting that

z
is transverse to = ker(dz + xdy) hence tb(L) is just the linking number
k(L, L
) where L
is a small vertical push-o of L. Now Figure 4.5 shows

that the canonical framing diers from the blackboard framing by a left
half-twist for each cusp, and this veries our formula for tb(L). Recall from
Exercise 2.1.4 that the blackboard framing of a knot diers from the framing
given by the Seifert surface by the writhe of the knot projection at hand.
Notice that c(L) is always even, since any cusp pointing right is followed
by one pointing left and vice versa. Consequently c(L) = 2c
r
(L) = 2c
l
(L)
where c
r
(L) and c
l
(L) denote the number of right and left cusps, resp.
Another invariant, the rotation number rot(L) can be dened by trivializing
st
along L and then taking the winding number of TL. For this invariant
to make sense we need to orient L, and the result will change sign when
reversing orientation. Since H
2
(S
3
; Z) = 0, this number will be independent
of the chosen trivialization.
Lemma 4.2.4. For the rotation number we have rot(L) =
1
2
_
c
d
(L)c
u
(L)
_
where c
d
(L) (and c
u
(L)) denotes the number of down (and up) cusps in the
projection.
Proof. Notice that the vector eld

x
gives rise to a trivialization of
st
,
hence the rotation number can be computed as the winding number with
respect to this vector eld. In conclusion, we have to count how many
times the tangent of L passes

x
as we traverse L. Dene l
(resp. r
)
as the number of left (resp. right) cusps where the knot L is oriented
upward/downward. Then the above principle shows that rot(L) = l
r
+
.
Doing the same count with

x
we get that rot(L) = r
l
+
, and taking
the average of the two expressions gives the result.
4.2. Legendrian knots 75
contact
blackboard
Figure 4.5. Contact and blackboard framings
Exercise 4.2.5. Show that for a Legendrian knot L (S
3
,
st
) the sum
tb(L) + rot(L) Z is always odd.
The ThurstonBennequin invariant and the rotation number admit nat-
ural generalization to any (homologically trivial) Legendrian knot L in any
contact 3-manifold (Y, ): Suppose that for an embedded orientable (com-
pact) surface Y we have = L. Then the contact framing can be
measured with respect to the framing on L induced by the resulting
number tb
(L) is the ThurstonBennequin invariant of L with respect to

. (As we already pointed out, this quantity is independent of .) By
considering the SO(2)-bundle [
with the trivialization along given by

the tangents of L (after xing an orientation on it), we get a relative Euler
number e
_
[
_
Z, which is called the rotation number of L with respect
to . Equivalently, since is trivial over we can x a trivialization which
indeed induces a trivialization of over = L. Also x a vector eld
v of tangents to L inducing the given orientation of L. Then the winding
number of v along L with respect to the xed trivialization of on L is the
rotation number of L. The rotation number of L depends on the orienta-
tion of L and will change sign when the orientation of L is reversed. It also
might depend on the chosen surface Y .
Exercise 4.2.6. Find a contact structure (Y, ), a Legendrian knot L
(Y, ) and surfaces
1
,
2
such that rot
1
(L) ,= rot
2
(L). (Hint: Start with
a contact structure and closed surface such that
e(), []
_
,= 0 and
nd L on separating it.)
Recall that any knot K (Y, ) can be C
0
-approximated by a Leg-
endrian knot. It has been extensively studied recently to what extent a
Legendrian knot is determined by the knot type in R
3
and the two classi-
cal invariants (the ThurstonBennequin number and the rotation number).
It has been proved [31] that if the Legendrian knot L is smoothly isotopic to
the unknot then the above classical invariants determine L (S
3
,
st
) up to
Legendrian isotopy. In particular, up to Legendrian isotopy there is a unique
knot L which is smoothly the unknot and has tb(L) = 1, rot(L) = 0. This
L is usually called the Legendrian unknot see Figure 4.2(a). Similar re-
sults have been achieved for torus knots and gure eight knots [41, 47]. The
answer to the above question in general is negative, though: according to
results of Chekanov [15, 37] there are Legendrian knots which have the same
classical invariants but are not Legendrian isotopic. For further reading on
this topic see [40].
4.3. Tight versus overtwisted structures
Special to dimension three, contact structures fall into two distinct cate-
gories.
Denition 4.3.1. (a) An embedded disk D (Y, ) is an overtwisted
disk in the contact 3-manifold (Y, ) if D = L is a Legendrian knot with
tb
D
(L) = 0, i.e., if the contact framing of L coincides with the framing
given by the disk D.
(b) The contact manifold (Y, ) is overtwisted if it contains an overtwisted
disk; (Y, ) is called tight otherwise. The contact structure on Y is
universally tight if its pull-back to the universal cover of Y is tight. If
4.3. Tight versus overtwisted structures 77
becomes overtwisted when pulled back to some nite cover of Y then it is
called virtually overtwisted.
Remark 4.3.2. A contact structure covered by a tight contact structure is
tight.
Exercises 4.3.3. (a) Show that the contact form in Example 4.1.4(b)
denes an overtwisted contact structure. (Hint: The disk
_
z = (
2
r
2
),
r
_
is an overtwisted disk for suciently small [[.)
(b) More generally, show that
n
in Example 4.1.4(b) is overtwisted for
n 1. (Hint: For n 1 consider the overtwisted disk z = 0, r r
0
for
f
n
(r
0
) = .)
(c) Prove the assertion in Remark 4.3.2.
According to a fundamental result of Eliashberg [24], overtwisted contact
structures on closed 3-manifolds can be classied using homotopy theory,
since
Theorem 4.3.4 (Eliashberg, [24]). Two overtwisted contact structures are
isotopic if and only if they are homotopic as oriented 2-plane elds. More-
over, every homotopy class of oriented 2-plane elds contains an overtwisted
contact structure.
In summary, the classication of overtwisted contact structures reduces to a
homotopy theoretic problem which is not very hard to solve. We will return
to the discussion of the homotopy classication of oriented 2-plane elds in
Section 6.2. In fact, using contact surgery we will verify the second assertion
of the theorem, usually attributed to Lutz and Martinet. Notice that so far
we do not have any example of tight contact structures. In general it is very
hard to show that there is no overtwisted disk present in a given contact
3-manifold. This fact gives particular interest to the following result.
Theorem 4.3.5 (The Bennequin inequality, [11]). If L is a Legendrian
knot in (R
3
,
st
) or in (S
3
,
st
) and Y is a Seifert surface for L then
tb(L) +
rot(L)
().
Since an overtwisted disk D has (D) = 1 and tb
D
(L) = 0, this theorem
implies
Corollary 4.3.6. The standard contact structures (S
3
,
st
) and (R
3
,
st
)
are tight.
As we will see later, the examples given in Example 4.1.4(e) are all tight.
Theorem 4.3.5 admits a natural generalization.
Theorem 4.3.7 (Eliashberg). The contact 3-manifold (Y, ) is tight if and
only if for all Y with = L Legendrian we have tb
(L)+
rot
(L)

().
This inequality resembles to the adjunction inequality we saw in Theo-
rem 1.2.1, so informally tight contact structures are those which obey the
appropriate adjunction inequality. Later we will see that the analogy be-
tween the adjunction inequality of Theorem 1.2.1 and the above inequality
is even deeper. Another inequality of the same spirit states that
Theorem 4.3.8 (Eliashberg, [26]). If e() denotes the Euler class of a tight
contact structure then
e(), []
_
() for any closed embedded

surface ,= S
2
and
e(), [S
2
]
_
= 0.
Notice that since []
2
= 0 in H
2
(Y ; Z), this formula can again be regarded
as an analogue of the adjunction inequality for 4-manifolds. Once again,
this inequality fails to hold for overtwisted structures, in general. In order
to sketch the proofs of these inequalities, we need a tool for studying contact
structures near surfaces. Notice that by the nonintegrability of the plane
eld , a surface generically intersects the plane eld (through the tangent
planes T) in lines.
Denition 4.3.9. Fix a contact structure on Y . For a surface Y we
can consider T, and for generic this intersection is a line eld except
at nitely many points (where is tangent to , hence T = = T).
Integrating T we get a foliation of with singularities at the tangencies,
called the characteristic foliation T
of in (Y, ).
Examples 4.3.10. (a) Consider the unit sphere S in the contact manifold
(R
3
,
3
) where
3
= ker(dz +r
2
d) as in Example 4.1.4(a). Since the contact
planes are horizontal along the z-axis, they are tangent to S at the points
(0, 0, 1), and hence the characteristic foliation on S has singularities at
these points. By visualizing the contact planes as they slowly twist while
moving out along any ray perpendicular to the z-axis one can see that
(0, 0, 1) are the only singular points and each leaf of the characteristic
foliation will spiral around the sphere connecting the two singular points
as shown in Figure 4.6.
(b) Consider the disk D of radius in the (r, )-plane in (R
3
, ) as in Exam-
ple 4.1.4(b). Recall that the contact planes are spanned by
r
, r tan r

z

Figure 4.6. Characteristic foliation on S
2
(R
3
, st)
. So it is clear that the center of the disk and each point on the boun-
dary of D (where r = ) is a singular point. Each leaf of the characteristic
foliation is a line segment connecting the center of the disk to a boundary
point. This gives an example of a nongeneric characteristic foliation on a
surface. Notice that D is an overtwisted disk since tb
D
(D) = 0. Now
imagine that we slightly push up (or push down) the interior of D with-
out moving its boundary to obtain a new disk D
. Notice that the planes

tangent to D
along its boundary are no longer horizontal. It is clear that

the boundary of D
becomes a closed leaf of the characteristic foliation with

only one singularity in the center of D
, see Figure 4.7.

Figure 4.7. The overtwisted disk, before and after
Denition 4.3.11. Consider the eigenvalues
1
,
2
of the linearization of
the ow at a generic isolated singular point p. We dene the index of p to
be equal to +1 if
1
2
> 0 and 1 if
1
2
< 0. A generic isolated singular
point of index +1 (resp. 1) is called an elliptic (resp. hyperbolic) singular
point. We depict a generic elliptic and a hyperbolic point in Figure 4.8.
(a) (b)
Figure 4.8. Isolated (a) elliptic and (b) hyperbolic singular points
By a vague analogy we can think of elliptic points as maxima and minima
of a Morse function on a surface, while hyperbolic points correspond to
saddle points. This analogy gets even deeper when recognizing that for a
generic ow hyperbolic points cannot be connected by a leaf similar to
the saddle points of a MorseSmale function. In addition, we can assign a
sign to each (isolated) singular point p of the characteristic foliation: The
singularity is positive (resp. negative) if the orientation of
p
agrees (resp.
disagrees) with the orientation of T
p
. Notice that this makes sense once
and are both oriented. In Example 4.3.10(a) both singular points are
elliptic with opposite signs. See Section 8.3 for similar notions in dimension
four. The characteristic foliation T
can be oriented as follows: If p is a

nonsingular point of a leaf L of T
, then we choose v T
p
L so that (v, n)
is an oriented basis for T
p
, where n T
p
is an oriented normal vector to
p
. With this choice of orientation a positive elliptic point becomes a source
and a negative elliptic point becomes a sink.
In order to understand the topology of contact 3-manifolds we need to
have a good grasp on how to cut and paste contact structures along surfaces.
It turns out that the characteristic foliation determines the contact structure
near the surface. The following result can be obtained as an application of
Theorem 4.1.12; for a proof see for example [57].
Theorem 4.3.12. If
i
(Y
i
,
i
) (i = 1, 2) embedded surfaces are dieo-
morphic through a dieomorphism f :
1

2
which preserves the char-
acteristic foliations then f extends to a contactomorphism on some neigh-
borhood of
1
.
Using the concept and count of positive and negative elliptic/hyperbolic
points we can outline proofs of Theorems 4.3.8 and 4.3.7.
Proof of Theorem 4.3.8 (sketch). Suppose that is a closed, embedded,
connected, oriented surface in a contact 3-manifold. We assume that the
characteristic foliation T
is generic, i.e., the singular points are isolated and

no two hyperbolic points are connected by a leaf. We can express
e(), []
_
and () =
e(T), []
_
in terms of the number of various types of singular
points of T
. Let e
and h
denote the number of elliptic/hyperbolic

points of T
. Fix a vector eld w which directs T
. Now it easily follows

from the PoincareHopf theorem that
() = (e
+
+e
) (h
+
+h
),
since each elliptic (resp. hyperbolic) point is a zero for w of index +1 (resp.
1). To calculate
e(), []
_
we need to count the oriented intersection
number of a generic section of the bundle [
with the zero section by

considering them as embedded oriented surfaces in the total space of the
bundle [
. We choose the section of [
given by w which also gives a

section of the tangent bundle T of . Notice that to calculate () =
e(T), []
_
we count the oriented intersection number of the zero section
of the tangent bundle T with a generic section (e.g., given by w ). The
count of oriented intersection number of sections to calculate
e(), []
_
will dier from the calculation of () exactly at those intersection points
where the orientations of the contact planes disagree with the orientations
of the tangent planes. So we need a sign reversal in the count exactly at
the negative singular points of T
to derive the formula
e(), []
_
= (e
+
e
) (h
+
h
).
By adding the above equations we get
e(), []
_
+() = 2(e
+
h
+
).
It is a theorem of Giroux (called the Elimination lemma) that if an elliptic
and hyperbolic point of the same sign are connected by a leaf of the char-
acteristic foliation on a surface then there is an isotopy of the surface such
that both singular points disappear. (For the corresponding phenomenon in
dimension four see Section 8.3.) Conversely we can always create a pair of
elliptic and hyperbolic points of the same sign on a given leaf. Therefore we
can assume that there is no closed leaf in T
. We will call the new surface

we obtain after such isotopies again, and clearly
e(), []
_
and () will
not change under these isotopies. Notice that until now we have not used
the tightness of the contact structure. Suppose that p is a positive elliptic
point on the surface . Now let O
p
be the union of all leaves limiting to p
and let D
p
be the closure of it. Suppose that D
p
is an embedded disk so
that D
p
= D
p
O
p
. Then all the singular points of T
on D
p
other than
p will be on D
p
. Since T
is oriented, there is no positive elliptic point on

D
p
and no two elliptic points can be adjacent. This is because a positive
elliptic point is a source and a negative elliptic point is a sink, so a leaf
connecting two elliptic points is directed form the positive to the negative.
Therefore the arcs on D
p
between elliptic points are divided by hyperbolic
points and, by the assumption we made at the beginning of the proof about
T
, no two hyperbolic points are adjacent. Suppose that there is no posi-

tive hyperbolic point on D
p
. Then we can eliminate all the singular points
on D
p
using the Elimination lemma and thus D
p
becomes an overtwisted
disk which cannot exist in a tight contact manifold. Hence there has to be
a positive hyperbolic point q on D
p
. But then we can eliminate the posi-
tive elliptic point p using this positive hyperbolic point q. The dicult part
of the proof is to show that we can eliminate a positive elliptic point even
if D
p
is not embedded. For details of this part of the proof the reader is
advised to turn to [26, 39]. By completing this last step we conclude that
e
+
= 0 can be assumed, trivially implying
e(), []
_
().
Moreover by subtracting the above equations and eliminating the negative
elliptic points we prove that
e(), []
_
(). In conclusion we get
the inequality
e(), []
_
().
Denition 4.3.13. Let be an arbitrary transverse knot in a contact 3-
manifold bounded by a Seifert surface . We dene the self-linking number
sl
() of as the linking number of and
, where
is a push-o obtained
by a nonzero vector eld in the contact planes. That is, sl
() is the oriented
intersection number of
with . If R
3
or S
3
then sl
() can be shown
to be independent of ; in this case we drop from the notation.
Given a Legendrian knot L, we can construct two copies of L by pushing
L in opposite directions in a suciently small annulus neighborhood of
L to obtain positive and negative transverse push-os L
of L. If L
(R
3
,
st
) then it is easy to obtain the front projections of the transverse
push-os L
from the front projection of a Legendrian knot L: For L

+
just smooth out the upward cusps and replace downward cusps by negative
kinks. See Figure 4.9. (For details regarding projections of transverse knots
Figure 4.9. From Legendrian to transverse knot
see [40, 57].) By using these projections and the fact that the self-linking
number of a transverse knot in (R
3
,
st
) is equal to its writhe in its front
projection, we get that for a Legendrian knot L (R
3
,
st
)
sl(L
+
) = w(L) c
d
(L) = w(L)
1
2
_
c
d
(L) +c
u
(L)
_

1
2
_
c
d
(L) c
u
(L)
_
= tb(L) rot(L).
This equation holds for a null-homologous transverse knot in an arbitrary
contact 3-manifold, leading us to the equation
sl
(L
) = tb(L) rot
(L).
Proof of Theorem 4.3.7. Notice that one direction of this equivalence
is clear: an overtwisted disk provides a surface violating the inequality.
To prove Eliashbergs theorem x with = L a Legendrian knot and
consider the positive and negative transverse push-os L
of the Legendrian
knot L. We can interpret the self-linking number of a transverse knot
as a relative Euler number and by the use of the method of the proof of
Theorem 4.3.8 we derive the equation
sl
() = (e
+
h
+
) + (e
).
In conclusion, we get a relation between sl
() and the number of dierent

types of singular points of the characteristic foliation on the surface
bounded by . Combining this result with
() = (e
+
+e
) (h
+
+h
)
we get sl
(L
) + () = 2(e
). By using the Elimination lemma

and the tightness of the contact structure as in the proof of Theorem 4.3.8
we can assume that e
= 0 and thus sl
(L
) (), clearly implying

Eliashbergs inequality. (See [39] and [40] for further details.)
We close this section by remarking that Legendrian knots in over-
twisted contact structures might have arbitrarily high ThurstonBennequin
invariants. More precisely, if L (Y, ) is homologically trivial and
_
Y L, [
Y L
_
is overtwisted then for every n there is L
smoothly iso-
topic to L such that tb
(L
) = n. It turns out that by taking enough copies

of the boundary of the overtwisted disk and connect sum them we get an un-
knot with the desired property and then the general statement easily follows
by an additional connect sum. For Legendrian knots with tight complement
the situation is more complicated.
5. Convex surfaces in contact 3-manifolds
When trying to do surgery on contact 3-manifolds we need to understand
contact structures in neighborhoods of embedded surfaces. As we already
pointed out in Chapter 4, for a given surface (Y, ) the characteristic
foliation T
determines the contact structure near . But it is not easy

to describe or relate characteristic foliations. It turns out that the same
information can be captured by certain congurations of curves on the
surface at hand once the surface is in a special position with respect to the
contact structure. This theory has been developed and fruitfully applied
by Giroux and Honda in various circumstances in 3-dimensional contact
geometry. For the sake of completeness, in this Chapter we recall the
fundamental denitions and results regarding convex surfaces and dividing
sets. These statements will be used in our study of contact Dehn surgery
in Chapter 11. For a more detailed introduction to the subject see [43, 76].
5.1. Convex surfaces and dividing sets
Denition 5.1.1. A vector eld v on a contact manifold (Y, ) is called
contact if its ow
t
preserves the contact planes, i.e., (
t
)
= .
The following lemma gives a convenient characterization of contact vector
elds on contact 3-manifolds.
Lemma 5.1.2. Let (Y, ) be a contact 3-manifold, where = ker for some
contact 1-form . A vector eld v on Y is contact if and only if L
v
= f
for some smooth function f : Y R.
Exercises 5.1.3. (a) Show that for each smooth function H: Y R there
is a unique vector eld V
H
such that v
H
= HR
+ V
H
is a contact
86 5. Convex surfaces in contact 3-manifolds
vector eld, where R
denotes the Reeb vector eld for . (Hint: V

H
is
the unique vector eld in satisfying
V
H
d = dH(R
) dH, cf. also

Exercise 4.1.9(a).)
(b) Verify that for each contact vector eld v there is a unique smooth
function H: Y R such that v = v
H
.
Inspired by the higher dimensional analogue we make the following
Denition 5.1.4. A smooth surface (Y, ) is convex if there is a contact
vector eld v transverse to . If ,= then we also require that is
Legendrian.
Remark 5.1.5. Notice that the direction of the contact vector eld v in
the denition is irrelevant, therefore there is no distinguished side of . In
that respect the term convex is unfortunate, since there is no concavity
present.
It can be shown that (Y, ) is convex if and only if it has a neighborhood
N = = I such that [
N
is invariant in the I-direction. Consequently,
in the neighborhood = I of the convex surface the contact 1-form
can be written as f dt +, where f is a function, is a 1-form on and
t denotes the I-coordinate.
Proposition 5.1.6 (Giroux, [61]). Any closed surface admits a C
-small
perturbation which puts it into convex position.
Remark 5.1.7. In [76] it was shown that this is also true for a surface with
boundary as long as the surface has Legendrian boundary and the twisting
of the contact planes with respect to the surface is not positive.
We should point out that even though every surface in a contact 3-manifold
can be perturbed into a convex surface it is the existence of non-convex
surfaces which makes the theory interesting. In Example 5.1.11 we will
describe a non-convex surface.
Denition 5.1.8. Suppose that (Y, ) is a convex surface with the
contact vector eld v. Dene =
_
x [ v(x)
x
_
as the dividing
set of .
As the following proposition shows, the dividing set (generically) is a multi-
curve, i.e., a properly embedded smooth 1-manifold, possibly disconnected
and possibly with boundary. We will often refer to this set of nite union
of disjoint simple closed curves and properly embedded arcs on as the
dividing curves.
5.1. Convex surfaces and dividing sets 87
Proposition 5.1.9 (Giroux, [61]). The dividing set is a 1-dimensional
submanifold of the surface transverse to the characteristic foliation T
and =
+

where the ow of a vector eld w which directs T
expands (contracts) a volume form on

+
(on
, resp.) and w points

outward from
+
along =
+
.
Proof. We choose coordinates x and t in the I-direction for the I-
invariant neighborhood = I of the convex surface . In these
coordinates the 1-form dening can be expressed as = f dt +. The
vertical vector eld v =

t
is a contact vector eld for since

t
is clearly
transverse to and L
t
= 0. Then for a point x we have
x
(
t
) = 0 if
and only if f(x) = 0, and therefore = f
1
(0). Now the contact condition
0 < d = ( +f dt) (d +df dt) = df dt +f dt d
= dt ( df +fd)
implies that df+fd > 0. (Notice that d = 0 on I.) In particular,
f(x) = 0 implies that df ,= 0 and hence df ,= 0. Consequently, is a
submanifold of , transversely cut out by f. Let u be a vector tangent to .
Then df(u) = 0 and thus df ,= 0 implies that (u) ,= 0. That is, u is not
in TT
= ker and it follows that is transverse to T
. Let w be a vector
eld which directs T
. The vector eld w can be dened by the equation
w
= [
= for a volume form on . Notice that w vanishes exactly

at the zeros of . Moreover if we take a dierent volume form on we get
a positive multiple of w directing T
. We dene the region

+
(resp.
)
as the set of points on where the normal orientation of agrees (resp.,
disagrees) with the orientation of the contact vector eld v. Equivalently,
+
(resp.
) is the subsurface where f > 0 (resp. f < 0). To see this

rst notice that f changes sign at : Consider the oriented basis (w, u)
of at a point x . Then ( df)(w, u) > 0 implies that df(w) < 0.
Now it is easy to calculate the spanning vectors for the contact planes and
we can see that the normal orientation of the planes agree with v =

t
if
and only if f > 0. Furthermore the vector eld w points outward from the
boundary of
+
. To see that the ow of w expands on
+
we observe
that L
w
f
= d
w
f
+
w
f
d = d
w
f
= d(
f
) =
1
f
2
(fd + df) > 0.
The choice of a contact vector eld is not unique; nevertheless we have
Proposition 5.1.10 (Giroux, [61]). The isotopy class of the dividing curves
is independent of the choice of the contact vector eld.
The following example of a non-convex torus is given in [41], cf. also [43].
Example 5.1.11. Consider the contact structure on Y = R
2
S
1
induced by
the contact structure
3
= ker(dz+r
2
d) on R
3
(with cylindrical coordinates
(r, , z)) through the identication z z+1. Let k be a positive real number.
We will show that the torus T = T
k
= (r, , z) Y [ r = k is not convex.
Recall that the contact planes of the given contact structure
3
are spanned
by
r
, r
2
z

(see Example 4.1.4(a)). Therefore at any point p on T

the intersection of the tangent plane to T and the contact plane
3
is given
by the line generated by the vector k
2
z

. Here we can view this line

in the (, z)-plane when we consider T as obtained by the identications
z z +1 and +2. Thus we conclude that the characteristic foliation
on T is linear as shown in Figure 5.1(a). Suppose that T is convex. Then
the contact 1-form on Y can be written as f dt + in a vertically invariant
neighborhood of T as explained above, where f is a smooth function and
is a 1-form on T. The form is given by (dz + r
2
d)[
T
= dz + k
2
d
and hence d = 0. On the other hand the contact condition implies that
df +fd > 0 as shown in the proof of Proposition 5.1.9. It follows that
df(w) < 0 for some vector eld w directing the characteristic foliation on T
which is a contradiction since the function f on T has to be periodic in
and z and thus f can not be decreasing along a linear foliation on T.
Exercises 5.1.12. (a) Perturb the torus T = T
k
in the example above into
a convex torus in (Y,
3
). (Hint: First consider the two disjoint annuli in the
complement of two orbits of the characteristic foliation on T. Then push
slightly one of the annuli (xing its boundary) towards the z-axis while
pushing the other one slightly in the opposite direction to get a smooth
embedded torus. Show that the dividing curves look like the dashed lines
in Figure 5.1(b).)
(b) Show that the unit sphere S
2
in (R
3
,
3
) is convex. Determine the
dividing set on S
2
. (Hint: Try the vector eld v = z

z
+
r
2
r
.)
Denition 5.1.13. Let L be a Legendrian curve on a convex surface in
a contact 3-manifold (Y, ). Then tw(L, ) denotes the twisting number of
the contact planes along L measured with respect to the surface framing
on L. Notice that tw(L, ) gives tb(L) if is a Seifert surface for L.
Exercise 5.1.14. Suppose that L is a Legendrian curve on a convex surface
which is transverse to the dividing set . Show that tw(L, ) =
1
2
(L).
(Hint: Fix a contact vector eld v for the convex surface . The twisting
z z
(a)
(b)

Figure 5.1. (a) Linear foliation on the non-convex torus T and (b) the dividing set
(dashed lines) on its convex perturbation
of with respect to is the twisting of relative to v. Observe that each
point in L contributes
1
2
to tw(L, ).)
Proposition 5.1.15. Suppose that is a closed convex surface in a contact
manifold (Y, ). Then
e(), []
_
= (
+
) (
).
Proof. In Theorem 4.3.8 we showed that
e(), []
_
= (e
+
h
+
)(e
).
It follows by denitions that the positive (resp. negative) singular points
will be in
+
(resp.
). Then using the PoincareHopf theorem for

a vector eld on a manifold which is transverse to the boundary we get
(
+
) = e
+
h
+
and (
) = e
.
If T is any singular foliation on the surface then a multicurve on
is said to divide T if the pair (T, ) satises the properties proved in
Proposition 5.1.9, where T
is replaced by T. The power of studying the

dividing set comes from the fact that (rather than the full characteristic
foliation) already determines the contact structure near :
Theorem 5.1.16 (Girouxs exibility, [61]). If T is another singular fo-
liation on divided by then there is an isotopy
s
: N =
_
s [0, 1]
_
,
0
= id
and
s
[ = id
such that
s
() is convex for all s
and T
1
()
=
1
(T).
Therefore, by xing , any foliation divided by can be thought of as
the characteristic foliation; in conclusion determines the germ of the
contact structure along . The next lemma shows a connection between
convex surfaces and Legendrian knots on them. First we need the following
denition.
Denition 5.1.17. A properly embedded 1-submanifold C of a convex
surface is nonisolating if C is transverse to and the closure of every
component of ( C) intersects .
Lemma 5.1.18 (Legendrian Realization Principle, [79, 76]). If C is non-
isolating on a convex surface then C can be made Legendrian, i.e., there
exists an isotopy
s
: N =
_
s [0, 1]
_
,
0
= id
such that
s
()
is convex for all s [0, 1],
1
(
) =
1
()
, and
1
(C) is Legendrian.
Remark 5.1.19. The nonisolating condition guarantees that C can be
extended to a singular foliation divided by . Then by Theorem 5.1.16
we can realize this foliation on as the characteristic foliation and hence
C becomes Legendrian after an isotopy of (, C) xing .
The set of dividing curves can in principle be very complicated.
A constraint on is posed by the following result of Giroux:
Theorem 5.1.20 (Girouxs criterion). Suppose that (Y, ) is a convex
surface (possibly with Legendrian boundary) and ,= S
2
. Then has
a tight neighborhood if and only if contains no homotopically trivial
component. If = S
2
then is tight if and only if consists of a single
component.
Proof (sketch). We give a proof for the only if direction. Suppose that
contains at least two components one of which is homotopically trivial.
Let denote the homotopically trivial curve which bounds a disk D. Let
be a curve parallel to such that
= . Then
is nonisolating
on and hence by the Legendrian realization principle we can make
Legendrian (so that it stays disjoint from ). This implies that the surface
framing of
agrees with its contact framing by Exercise 5.1.14. Thus the

disk bounded by
on is an overtwisted disk by denition. Now suppose

that has only one component which is homotopically trivial. Take a
homotopically essential non-separating simple closed curve
and
use the folding method of Honda [76] to introduce a pair of dividing curves
parallel to . Then repeat the previous argument to nd an overtwisted
disk.
Exercises 5.1.21. (a) Use Theorem 4.3.8 to show that if the dividing set
on a closed convex surface ,= S
2
in a contact 3-manifold (Y, ) consists
of only one homotopically trivial curve then (Y, ) is overtwisted. (Hint:
Observe that
e(), []
_
= (
+
) (
) = 2g.)
(b) Use Girouxs criterion to prove Theorem 4.3.8. (Hint: Put the given
in a tight contact 3-manifold (Y, ) into convex position. It is clear that
() = (
+
) +(
). Compare this fact with the equation
e(), []
_
=
(
+
) (
) of Proposition 5.1.15 and observe that (
) 0 when
,= S
2
. If = S
2
then
is the disjoint union of two disks. See [43]

for further details.)
In the following we focus on the special case of = T
2
.
Exercise 5.1.22. Suppose that a convex torus T
2
has a tight neighborhood
(e.g., it is embedded in a tight contact 3-manifold). Then show that the
dividing set on T
2
consists of 2n parallel circles (n 1).
By xing an identication of T
2
with R
2
/Z
2
, the slope of these parallel
curves is called the slope of the torus at hand. Of course, there is no
canonical choice of identication of T
2
with R
2
/Z
2
in general. In particular
cases, however, there are natural directions to choose for example if
T
2
= (S
1
D
2
) or T
2
= K is the boundary of the neighborhood of a
Legendrian knot then the meridian provides an obvious direction.
Example 5.1.23. Consider N = R
2
(R/Z) R
2
S
1
with the 1-form
= cos(2z) dx sin(2z) dy
in the coordinates (x, y) for R
2
and z for R/Z. First we check that is a
contact form on N: since
d = 2 sin(2z) dx dz + 2 cos(2z) dy dz,
we have
d = 2 cos
2
(2z) dx dy dz + 2 sin
2
(2z) dx dy dz
= 2 dx dy dz.
The contact form on N induces a contact form on the solid torus N
d
=
_
(x, y, z) [ x
2
+ y
2
d
2
_
for d > 0. We claim that N
d
is a convex torus.
To this end, consider the vector eld v = x

x
+ y

y
. It is clear that v is
transverse to N
d
. To show that v is a contact vector eld, we check that
L
v
= (see Lemma 5.1.2). For the given we calculate
v
=
_
x

x
+y

y
_
=
x

x
+
y

y
= xcos(2z) y sin(2z),
d
v
= cos(2z) dx 2xsin(2z) dz sin(2z) dy 2y cos(2z) dz,
and
v
d =
x

x
d +
y

y
d = 2xsin(2z) dz + 2y cos(2z) dz.
Then it follows by Cartans formula L
v
=
v
d +d
v
that L
v
= . The
central circle
C =
_
(x, y, z) N
d
[ x = y = 0
_
acquires a canonical contact framing. This framing can be given by the
longitude that is obtained by pushing C along a vector eld which is
transverse to C and stays inside the contact planes = ker . If we
choose the vector eld that is orthogonal to C, it is easy to calculate that
=
_
d sin(2z), d cos(2z), z
_
. Since N
d
is a convex torus in a contact
3-manifold, there are dividing curves on N
d
induced by the contact vector
eld v. By denition, the dividing curves consist of the points x N
d
such
that v(x) (x), i.e, when
x
_
v(x)
_
= 0. The solution of the equation
_
cos(2z) dx sin(2z) dy
_
_
x

x
+y

y
_
= xcos(2z) y sin(2z) = 0
or equivalently the equation
z =
1
2
tan
1
_
x
y
_
can be given by the set
=
_
d sin(2z), d cos(2z), z
_
,
which consists of two parallel copies of the longitude . Consequently, with
the trivialization of N
d
by and the meridian the slope of the dividing
curves comes out to be equal to . Here will correspond to the x- and
to the y-axis; hence the slope being
p
q
means that is parallel to the curve
p +q. In fact, we can visualize the contact planes as follows: The planes
are horizontal at z = 0 and start twisting as z is increasing and they become
horizontal again when z = 1. So the characteristic foliation consists of two
singular lines of slope = and parallel nonsingular leaves of slope ,= .
(Notice that this characteristic foliation is not generic.)
Remark 5.1.24. In general, on S
1
D
2
only the meridian is canonical,
hence the slope of (S
1
D
2
) is well-dened only up to an action of SL
2
(Z)
leaving xed, i.e., of the action
_
1 m
0 1
_
the Dehn twists changing
the framing. It is not hard to see that using this equivalence any nonzero
slope can be transformed into the form
p
q
with (p, q) = 1 and p > q 0;
moreover this form is unique: just notice that under the action of the above
matrix the slopes
p
q
and
p
pm+q
are equivalent.
Exercise 5.1.25. Find slopes equivalent to
2
3
and 1.
For topologically simple 3-manifolds the dividing curves may determine the
entire contact structure. The following is a fundamental result which is
essential for the classication of tight contact structures.
Theorem 5.1.26 (Eliashberg). Assume that there exists a contact struc-
ture on a neighborhood of D
3
which makes D
3
convex with connected
dividing set. Then there exists a unique extension of to a tight contact
structure on the 3-disk D
3
up to an isotopy which xes the boundary.
Exercise 5.1.27. Using Theorem 5.1.26 show that the 3-sphere S
3
admits
(up to isotopy) a unique tight contact structure.
The exercise above can be solved by a simple-minded approach to nd
an upper bound on the number of tight contact structures on a given 3-
manifold. In order to calculate an upper bound we cut the 3-manifold along
convex surfaces until we end up with a disjoint union of 3-disks. At each
step we keep track of all possible congurations of dividing curves on these
surfaces along which we cut our 3-manifold. We will apply this strategy
below to nd an upper bound for the number of tight contact structures
on the solid torus for the case when the boundary slope of the dividing
curves is equal to
1
n
. We will rst state a basic lemma called the edge
rounding which is frequently used to transfer dividing sets between two
convex surfaces meeting along a Legendrian curve.
Exercise 5.1.28. Let
i
be a convex surface with dividing set
i
for i = 1, 2.
Assume that
2
is a Legendrian curve in
1
. Let A =
1

2
and
B =
2
2
. Then between two adjacent points of A there is a point in B
and between two adjacent points of B there is a point in A. (Hint: Consider
the unique geometric model of contact structures in a neighborhood of the
Legendrian curve.)
Lemma 5.1.29 (Edge rounding, [76]). Let
i
be a convex surface with
the dividing set
i
for i = 1, 2 and assume that
1
=
2
is Legendrian.
Then using the standard local model around
1
we can glue
1
to
2
by
rounding the edge
1
=
2
to get a smooth surface so that the dividing
curves
i
connect up as shown in Figure 5.1.29 to form a dividing set on
.
2
Figure 5.2. Connecting up the dividing curves while rounding an edge
Theorem 5.1.30. Suppose that
1
and
2
are two tight contact structures
on S
1
D
2
with two parallel dividing curves on the convex boundary
(S
1
D
2
) having slope equal to
1
n
for some n Z. Then
1
and
2
are isotopic.
Proof (sketch). Notice rst that
1
n
and
1
m
are equivalent boundary slopes
for any m, n Z and 1 =
1
1
also represents this class. Hence it suces
to classify the tight contact structures for any one of these slopes. It is
clear that a meridian on the convex surface (S
1
D
2
) is nonisolating
and therefore we can isotope this meridian into Legendrian position by the
Legendrian Realization Principle. Notice that the twisting tw(D, D) of the
contact planes along D with respect to a spanning disk D of the meridian is
negative. Thus D can be isotoped to a convex disk by Remark 5.1.7. Then
tightness of the contact structures at hand implies by Girouxs criterion
(Theorem 5.1.20) that the dividing set
D
on the disk D contains no closed
components, hence
D
is a single arc connecting two points a
1
and a
2
on
D. Let b
1
, b
2
D denote the points of the intersection of D with the
dividing set on the convex boundary of the solid torus. Now we have a
convex torus intersecting a convex disk along a Legendrian curve and we
know the dividing sets on these surfaces. Hence by Exercise 5.1.28, b
1
is
positioned between a
1
and a
2
while b
2
is positioned between a
2
and a
1
on the
(oriented) circle D. Next we cut S
1
D
2
along D and smooth the corners
by rounding the edges using Lemma 5.1.29. Notice that when we remove
a neighborhood D of D from S
1
D
2
we get a 3-disk D
3
such that the
dividing set on its boundary is connected. Now Eliashbergs Theorem 5.1.26
concludes the proof: near the boundary and near the spanning disk D the
contact structures
1
and
2
are isotopic (shown by the dividing curves),
and the complement of D in S
1
D
2
is D
3
with connected dividing set
on its boundary. Therefore Theorem 5.1.26 extends the above isotopy to
S
1
D
2
, nishing the proof.
The case of general boundary slope follows by the same line of argument:
By considering the disk D, however, there are more possible congurations
for the dividing curves on it, since the dividing curves on (S
1
D
2
) will
intersect
_
pt.D
2
_
in more points: if the slope is r =
p
q
then D inter-
sects the dividing set of (S
1
D
2
) in 2p points. Every conguration gives
a potential tight contact structure, and so this argument gives a (poten-
tially weak) upper bound for the number of tight structures. In fact, many
of the dierent congurations correspond to isotopic tight structures. In
order to get the classication, Honda followed a slightly dierent path, and
manipulated the set of dividing curves on the boundary slope by applying
bypasses. For details see [76].
Remark 5.1.31. Notice that we assumed that the boundary slope is dif-
ferent from zero. The reason is that there is no tight contact structure on
S
1
D
2
with boundary slope zero: in this case
_
pt. D
2
_
is disjoint
from the dividing curves of the boundary, therefore pt.D
2
(after having
been isotoped to have Legendrian boundary) provides an overtwisted disk.
5.2. Contact structures and Heegaard decompositions
In this section we review a construction of Torisu [168] associating a unique
contact structure to an open book decomposition of a 3-manifold. Torisus
result is based on the work of Giroux on convex contact structures. We
follow an alternative line of proof which is based on the discussion in Sec-
tion 5.1. It turns out that Torisus contact structure is compatible with the
given open book decomposition in the sense of Giroux. (See Chapter 9 for
relevant denitions regarding open book decompositions and their relation
to contact structures.) Suppose that (L, ) is a given open book decom-
position on a closed 3-manifold Y . (Here L Y is a bered link, while
: Y L S
1
denotes the bration of the open book decomposition.)
Then by presenting the circle S
1
as the union of two closed (connected) arcs
S
1
= A
1
A
2
intersecting each other in two points, the open book decom-
position (L, ) naturally induces a Heegaard decomposition Y = U
1

U
2
of the 3-manifold Y : one only needs to verify the simple observation that
U
i
=
1
(A
i
) L are solid handlebodies. The surface along which these
handlebodies are glued is simply the union of two pages
1
(A
1
A
2
) to-
gether with the binding. This is illustrated in Figure 5.3.
Figure 5.3. The handlebody Ui

5.2. Contact structures and Heegaard decompositions 97
Theorem 5.2.1 (Torisu, [168]). Suppose that
1
,
2
are contact structures
on Y satisfying:
(i)
i
[
U
j
(i = 1, 2; j = 1, 2) are tight, and
(ii) is convex in (Y,
i
) and L is the dividing set for both contact structures.
Then
1
and
2
are isotopic. In addition, the set of such contact structures
is nonempty.
Proof. Suppose that a page F of the given open book is a genus-g surface
with r boundary components. Then U
1
= is a closed surface of genus
h = 2g + r 1. First we would like to argue that there is at most one
tight contact structure on the handlebody U
1
such that L is the dividing
set on . Since U
1
is a genus-h handlebody, it is clear that we can nd h
homologically linearly independent curves
1
,
2
, . . . ,
h
on which bound
h disjoint disks in U
1
so that when we cut along these disks we get the 3-disk
D
3
. The key point of our construction is that we can choose
1
,
2
, . . . ,
h
in such a way that each
k
intersects the dividing set L twice for
k = 1, 2, . . . , h. We depicted a choice of such curves for r = 3, g = 2 in
Figure 5.4. The disk D
k
spanned by
k
can be visualized as the disk which
is swept out in U
1
(see Figure 5.3) by swinging the left-half of the curve
k
until it coincides with its right-half.

L
L
L
6
5
4
3
1
2
Figure 5.4. The -curves.
Now we proceed exactly as in the proof of Theorem 5.1.30. First we put
the curves
1
,
2
, . . . ,
h
into Legendrian position and make the spanning
disks D
1
, D
2
, . . . , D
h
convex. The diving set on each D
k
will be an arc
connecting two points on the boundary, for k = 1, 2, . . . , h. Then we cut
along these disks and round the edges to get a connected dividing set on
the remaining D
3
and use Eliashbergs theorem to show the uniqueness of a
tight contact structure on U
1
with the assumed boundary condition. Clearly
we can prove the same result for the handlebody U
2
. To nish this part of
the proof of the theorem we need to show the existence of a tight contact
structure
1
on U
1
(and
2
on U
2
) which has L as its dividing set on . The
idea is to embed U
1
into an open book whose compatible contact structure
(see Chapter 9) is Stein llable, and hence tight. Such an embedding of
the genus-g handlebody into a Stein llable contact structure will be shown
in Exercise 11.3.5(c). Suppose now that
j
is a tight contact structure on
U
j
whose dividing set is equal to the binding L U
j
for j = 1, 2. Let
: U
1
U
2
be the dieomorphism dening the Heegaard decomposition
Y = U
1
U
2
. The tight contact structure
j
on U
j
induces a foliation T
j
on . Now (T
1
) (as well as T
2
) is a singular foliation on U
2
divided by L
(since is the identity on L). Then by Girouxs exibility Theorem 5.1.16
we can isotope in Y so that (T
1
) and T
2
agree by this isotopy and hence
we can glue the tight contact structure
1
and
2
to get a contact structure
on Y . The uniqueness of such a contact structure on Y follows from the
uniqueness of
1
and
2
. Notice that the tightness of
1
and
2
does not
imply tightness for the glued up contact structure .
Example 5.2.2. Consider the open book decomposition of S
3
induced by
the positive (resp. negative) Hopf link. The associated contact structure
is tight (resp. overtwisted).
Let (Y
i
,
i
) be a contact 3-manifold for i = 1, 2. To dene the contact
connected sum (Y = Y
1
#Y
2
, =
1
#
2
) just delete a Darboux ball D
i
from Y
i
(i = 1, 2) and glue Y
1
int D
1
to Y
2
int D
2
by a dieomorphism
f : (Y
1
int D
1
) (Y
2
int D
2
) which takes the dividing set
1
on
(Y
1
int D
1
) to the dividing set
2
on (Y
2
int D
2
). This operation is
well dened by Eliashbergs Theorem 5.1.26.
Remark 5.2.3. Torisu [168] also proves that the contact structure asso-
ciated to a plumbing of two bered links (cf. Chapter 9) is the contact
connected sum of the corresponding contact structures. (See [68] for an
alternative proof.)
Exercise 5.2.4. Show that the contact structure associated to the open
book of S
3
induced by a (p, q)-torus knot is tight. (Hint: See Example 9.1.4.)
6. Spin
c
structures on 3- and 4-manifolds
Spin
c
structures turn out to be very useful tools in understanding homotopic
properties of contact structures. In addition, gauge theoretic invariants
such as SeibergWitten and OzsvathSzabo invariants are dened for
spin
c
3- and 4-manifolds. This chapter is devoted to the review of spin
c
structures with a special emphasis on the 3- and 4-dimensional case.
Throughout this chapter we will assume that the reader is familiar with
the basics of the theory of characteristic classes. (For an excellent reference
see [116].) For a more complete treatment of spin
c
structures the reader is
advised to turn to [113].
c
structures
We begin our discussion by recalling the related and much more standard
subject of spin structures.
Spin structures
By denition the n-dimensional (n 3) spin group Spin(n) is the universal
(double) cover of SO(n). In other words, Spin(n) is a simply connected
Lie group with a map : Spin(n) SO(n) which is the principal Z
2
-
bundle of the unique nontrivial real line bundle on SO(n) (n 3). (Recall
that
1
_
SO(n)
_
= H
1
_
SO(n); Z
_
= Z
2
.) Let X be a given oriented
Riemannian n-manifold and let p: P
SO(n)
X denote the principal SO(n)-
bundle of orthonormal frames in TX. A spin structure on X is a principal
Spin(n)-bundle s: P
Spin(n)
X with a map : P
Spin(n)
P
SO(n)
such
that p = s and berwise is just the double cover : Spin(n) SO(n).
100 6. Spin
c
In other words, the associated principal SO(n)-bundle P
Spin(n)
SO(n) is
isomorphic to P
SO(n)
. Two spin structures P
1
and P
2
are equivalent if there
is a bundle isomorphism : P
1
P
2
such that s
1
= s
2
where s
i
: P
i
X
are the bundle maps of the principal Spin(n)-bundles for i = 1, 2. The set
of equivalence classes of spin structures on X will be denoted by Spin(X).
Remark 6.1.1. More generally, for any principal SO(n)-bundle E X a
similar denition provides spin structures on E.
Theorem 6.1.2. An oriented Riemannian n-manifold X admits a spin
structure if and only if w
2
(X) = 0. In that case the number of inequivalent
spin structures is equal to
H
1
(X; Z
2
)
. In fact, H
1
(X; Z
2
) admits a free
and transitive action on Spin(X).
In a similar fashion, it can be shown that a principal SO(n)-bundle E X
admits a spin structure if and only if w
2
(E) = 0, and the number of
inequivalent spin structures is again equal to
H
1
(X; Z
2
)
.
Spin
c
structures
The group Spin
c
(n) is dened as S
1
Spin(n)/Z
2
where Z
2
=
_
(1, 1)
S
1
Spin
c
(n)
_
, and 1 is dened as ker Spin(n). It follows
that Spin
c
(n) admits an S
1
-bration over SO(n) (n 3); this map
: Spin
c
(n) SO(n) can be characterized as the principal S
1
-bundle of
the unique nontrivial complex line bundle on SO(n) (n 3). Notice that
H
2
(SO(n); Z) = Z
2
for n 3. Again, a spin
c
structure on an n-dimensional
manifold X is a principal Spin
c
(n)-bundle s: P
Spin
c
(n)
X with a map
: P
Spin
c
(n)
P
SO(n)
such that s = p and berwise is just ; equiva-
lently, in bundle theoretic terms P
Spin
c
(n)
SO(n)
= P
SO(n)
. As in the spin
case, spin
c
structures can be dened for any principal SO(n)-bundle. The
map : Spin
c
(n) S
1
we get by the formula
_
(z, A)
_
z
2
enables us to
associate a line bundle the determinant line bundle L = P
Spin
c
(n)
C
to a given spin
c
structure P
Spin
c
(n)
. In an equivalent way, a spin
c
structure
on X can be regarded as an element u H
2
(P
SO(n)
; Z) whose restriction to
every ber of P
SO(n)
X is the unique nontrivial element of H
2
(SO(n); Z):
by considering the S
1
-bundle s
u
: L P
SO(n)
corresponding to u, the com-
position s
u
p provides a principal Spin
c
(n)-bundle structure on L and
hence a spin
c
structure in the above sense. Similarly to the spin case, we
say that spin
c
structures P
1
, P
2
X are equivalent if there is a bundle
c
structures 101
isomorphism h: P
1
P
2
satisfying s
1
= s
2
h, where s
i
: P
i
X are the
bundle projections. The set of equivalence classes of spin
c
structures on a
xed manifold X will be denoted by Spin
c
(X). As the above reformulation
shows, we can regard Spin
c
(X) as subset of H
2
(P
SO(n)
; Z).
Theorem 6.1.3. Let P
Spin
c
(n)
be a given spin
c
structure with determinant
line bundle det P
Spin
c
(n)
= L. Then c
1
(L) w
2
(X) (mod 2). In addition,
if c H
2
(X; Z) satises c w
2
(X) (mod 2) then there is a spin
c
structure
with determinant line bundle L satisfying c
1
(L) = c.
Proof (sketch). The natural map : Spin
c
(n) SO(2) SO(n) can
be shown to be the unique double cover of SO(2) SO(n) which extends
to a double cover of SO(n + 2), hence TX admits a spin
c
structure if and
only if there is a line bundle L such that TX L admits a spin structure.
This latter is equivalent to w
2
(TX L) = w
2
(TX) + c
1
(L)
2
= 0, proving
the claim.
The group H
2
(X; Z) admits a free and transitive action on Spin
c
(X) (if
the latter is nonempty) as follows: for s Spin
c
(X) H
2
(P
SO(n)
; Z) and
a H
2
(X; Z) the action of a on s is given by
s s +p
(a)
where p: P
SO(n)
X is the bundle map of the frame bundle. The natural
group homomorphism Spin(n) Spin
c
(n) shows that a spin structure
induces a spin
c
structure. It follows that such an induced spin
c
structure
has trivial determinant line bundle. Conversely, if det (P
Spin
c
(n)
) is trivial
for a spin
c
structure then it can be induced by a spin structure, since the
triviality of the determinant line bundle shows that the cocycle structure of
P
Spin
c
(n)
can be homotoped into the kernel ker = Spin(n).
The collar neighborhood theorem for the embedding X X provides
a splitting of TX[
= T(X)R near the boundary, implying in particular,

that a spin (or spin
c
) structure on X naturally induces a similar structure
on X. (As always, R denotes the trivial real line bundle.) After having
dispensed with the above general discussion, in the rest of this chapter we
focus on the 3- and 4-dimensional case and relate spin
c
structures to other
geometric objects on such manifolds.
102 6. Spin
c
6.2. Spin
c
structures and oriented 2-plane fields
We start with the 3-dimensional case. It is fairly easy to see that Spin(3) =
SU(2)

= S
3
and Spin
c
(3) = U(2). Notice also that from the theory of
characteristic classes it follows that for a 3-manifold X we have w
2
(X) =
w
2
1
(X), therefore any oriented 3-manifold admits a spin (and so spin
c
)
structure. A spin
c
structure on an oriented 3-dimensional Euclidean vector
space V can be given by specifying a complex hermitian plane W with a
map : V Hom
C
(W, W) satisfying (v)
(v) = [v[
2
id
W
. Globally, a
spin
c
structure on an oriented Riemannian 3-manifold is simply a continuous
family of spin
c
structures on the tangent spaces, i.e., a pair (W, ) where
W Y is a hermitian C
2
-bundle (a U(2)-bundle) on the 3-manifold Y and
: T
C
Y Hom
C
(W, W) is a bundle homomorphism satisfying (v)
(v) =
[v[
2
id
W
. The equivalence with the denition given in Section 6.1 is clear:
P
Spin
c
(3)
= P
U(2)
corresponds to the principal U(2)-bundle of W while
: T
C
Y Hom
C
(W, W) and the map : P
Spin
c
(3)
P
SO(3)
determine each
other. Next we discuss a more geometric presentation of spin
c
structures
on 3-manifolds; this presentation will be more suitable for our purposes in
our subsequent discussions. Let (Y ) denote the space of oriented 2-plane
elds on Y , while V ect(Y ) stands for the set of vector elds of length 1.
Notice that by considering the oriented unit normal of an oriented 2-plane
eld we get a bijection (Y ) V ect(Y ).
Denition 6.2.1. Two nowhere vanishing vector elds v
1
and v
2
are said
to be homologous if v
1
is homotopic to v
2
outside a disk D
3
Y (through
nowhere vanishing vector elds).
This equivalence relation together with the identication given above
induces an equivalence relation on (Y ) and hence on
0
_
(Y )
_
. For the
proof of the following statement see [169].
Proposition 6.2.2 (Turaev, [169]). Spin
c
(Y ) can be identied with the
set of equivalence classes of elements of
0
_
(Y )
_
under the equivalence
relation given by homology.
Exercises 6.2.3. (a) Show that an oriented 2-plane eld reduces the struc-
ture group of TY Y from SO(3) to U(1), and this latter group admits a
natural lift to U(2).
6.2. Spin
c
structures and oriented 2-plane elds 103
(b) Verify that a C
2
-bundle W Y admits a nowhere vanishing section.
Show that a spin
c
structure (W, ) induces a nowhere vanishing section of
TY Y .
(c) Using the solutions of the above exercises prove Proposition 6.2.2.
In fact, the above correspondence can be rened as follows:
Lemma 6.2.4 (KronheimerMrowka, [86]). Let us x a closed, oriented 3-
manifold Y . There is a one-to-one correspondence between the space (Y )
of oriented 2-plane elds on Y and isomorphism classes of pairs (t, ) where
t Spin
c
(Y ) and (W) is of unit length.
An oriented 2-plane eld, or more specically a contact structure
naturally induces a spin
c
structure which will be denoted by t
. Let
p:
0
_
(Y )
_
Spin
c
(Y ) denote the map associating t
to . In the rest of
this section we briey recall the classication of oriented 2-plane elds (up
to homotopy) on Y , cf. also Chapter 11 of [66]. Trivializing TY and consid-
ering the oriented normal of an oriented 2-plane eld, a map Y S
2
can be
associated to (Y ). In particular, on Y = S
3
the oriented 2-plane elds
are in one-to-one correspondence with elements of [S
3
, S
2
] =
3
(S
2
)

= Z.
Using the Pontrjagin-Thom construction, the space [Y, S
2
] can be identied
with the framed cobordism classes of framed 1-manifolds in Y . Homotopies
outside of a disk (i.e., spin
c
structures) can be parameterized by the 1-
manifolds in Y up to cobordism, i.e., with H
1
(Y ; Z)

= H
2
(Y ; Z). A ber
p
1
(t) for a spin
c
structure t admits an [S
3
, S
2
]
= Z-action: for any n we can

twist the given framing of the framed link corresponding to the oriented
2-plane eld by n. Viewing this action from another point of view, oriented
2-plane elds (or, equivalently, the orthogonal vector elds corresponding
to them) inducing a xed spin
c
structure t Spin
c
(Y ) can be assumed to
be identical outside of a disk D
3
Y . Then Z acts on p
1
(t) by connect
summing a given (Y, v) (where [v] p
1
(t)) with the elements of
_
(S
3
, w) [w is a nowhere zero vector eld on S
3
_
.
By pulling back the generator of H
2
(S
2
; Z) with the map f
: Y S
2
associated to (Y ) we get a second cohomology class
H
2
(Y ; Z).
This class will depend on the chosen trivialization of TY , but for
1
,
2

(Y ) the dierence
2
is independent of this choice, since it can be
identied with the obstruction of f
1
being homotopic to f
2
on Y D
3
.
This observation again shows the existence of a natural free and transitive
H
2
(Y ; Z)-action on Spin
c
(Y ). It is not hard to see that c
1
() H
2
(Y ; Z)
104 6. Spin
c
(where we regard as an oriented R
2
-, hence a complex line bundle) is equal
to 2
: by denition
is the pull-back of [S
2
] = PD[point] H
2
(S
2
; Z)
while is the pull-back of the tangent bundle TS
2
, hence
c
1
() = f
_
c
1
(TS
2
)
_
= f
(PD[two points]).
Consequently, if H
2
(Y ; Z)

= H
1
(Y ; Z) has no 2-torsion, then c
1
() deter-
mines the spin
c
structure t
induced by . Notice that c

1
() = c
1
(t
) for
the induced spin
c
structure t
, since a second cohomology class uniquely ex-

tends from Y D
3
to Y on a 3-dimensional manifold. Recall that Z admits
a transitive action on p
1
(t) for any spin
c
structure t. In the statement
below, Z
0
is understood to be equal to Z.
Proposition 6.2.5 (Gompf, [64]). Let t Spin
c
(Y ) be a given spin
c
structure. The ber p
1
(t) can be identied with Z
d(t)
where d(t) denotes
the divisibility of c
1
(t), and is zero if c
1
(t
) is a torsion class.
In conclusion, the homotopy type of a 2-plane eld is uniquely specied
by the induced spin
c
structure t
and the framing of the corresponding 1-

manifold in Y . This latter invariant is an element of Z
d(t)
in general, and it
is hard to work with, except in the case of torsion rst Chern class c
1
(t
).
In this case the set of framings (an ane set for Z) can be lifted to a subset
of Q as follows: Suppose that (Y ) has torsion rst Chern class c
1
()
and suppose furthermore that (X, J) is an almost-complex 4-manifold with
X = Y and is homotopic (as an oriented 2-plane eld) to the complex
tangencies along X, i.e., to TY JTY .
Lemma 6.2.6 (Gompf, [64]). The expression
d
3
() =
1
4
_
c
2
1
(X, J) 3(X) 2(X)
_
Q
denes an invariant of .
The proof is a standard exercise relying on the following
Theorem 6.2.7 (Hirzebruch signature theorem for 4-manifolds). For a
closed almost-complex 4-manifold (X, J) we have c
2
1
(X, J) = 3(X) +
2(X).
Exercises 6.2.8. (a) Verify that if c
1
() is torsion then c
2
1
(X, J) Q is
well-dened. (Hint: Cf. Remark 3.1.11.)
(b) Show that
_
d
3
() [ (S
3
)
_
Z +
1
2
. (Hint: Use the fact that for
a unimodular form Q and characteristic vector c we have Q(c, c) (Q)
(mod 8).) In fact, the above two sets are equal.
6.3. Spin
c
structures and almost-complex structures 105
It is not hard to see (cf. [64]) that for any oriented 2-plane eld (Y )
there is an almost-complex 4-manifold (X, J) such that X = Y and is
homotopic to the oriented 2-plane eld of complex tangencies along X.
The rational number d
3
() dened for those (Y ) which have torsion
c
1
() is called the 3-dimensional invariant of the oriented 2-plane eld .
Proposition 6.2.5 now specializes to
Theorem 6.2.9 (Gompf, [64]). Suppose that the oriented 2-plane elds
1
,
2
induce the same spin
c
structure t and that c
1
(t) is torsion. Then
[
1
] = [
2
] if and only if d
3
(
1
) = d
3
(
2
).
Later we will show explicit computations for d
3
() of some contact struc-
tures .
6.3. Spin
c
structures and almost-complex structures
Next we consider the geometric interpretation of spin
c
structures on 4-
manifolds. Recall that Spin(4) = SU(2) SU(2) and so
Spin
c
(4) = S
1
SU(2) SU(2)/ (1, id, id);
alternatively, Spin
c
(4) =
_
(A, B) U(2) U(2) [ det A = det B
_
. The
isomorphism between the above groups is given by the map
(A, B)
_
, A
_
1
0
0
1
_
, B
_
1
0
0
1
__
with
2
= det A = det B. Spin and spin
c
structures in dimension 4 can also
be dened as follows. First let V be a 4-dimensional oriented Euclidean
vector space. A spin structure on V is a pair (V
+
, V
) of 1-dimensional
quaternionic vector spaces with hermitian metrics together with an isomor-
phism : V Hom
H
(V
+
, V
) compatible with the metrics. (Note that

the group of symmetries of V is SO(4), while for the spin structure (V
, )
the group of symmetries is Spin(4).) Globally, for an oriented, Rieman-
nian 4-manifold X a spin structure is a triple (S
+
, S
, ) where S
X
are quaternionic line bundles with hermitian metrics (i.e., SU(2)-bundles)
and : TX Hom
H
(S
+
, S
) is a bundle isomorphism compatible with the

chosen metrics. Using the cocycle structures of the bundles S
X it is
fairly easy to see that this denition coincides with the general one given in
Section 6.1.
106 6. Spin
c
A spin
c
structure on a vector space V is a pair (V
+
, V

) of com-
plex planes with hermitian metrics such that det
C
V
+
= det
C
V
to-
gether with an isomorphism : V C Hom
C
(V
+
, V

) which satises
(v)
(v) = [v[
2
id
V
+. (It is not hard to verify that the symmetry group
of (V

, ) is isomorphic to Spin
c
(4).) Once again, by globalizing the above
construction, we can dene a spin
c
structure on X by a triple (W
+
, W
, )
where W
X are hermitian C
2
-bundles with det W
+
= det W
and
: TX C Hom
C
(W
+
, W
) is a bundle isomorphism which satises

(v)
(v) = [v[
2
id
W
+. The proof of equivalence is again an easy exercise.
As a simple homological argument shows (see e.g. [66]) the set of spin
c
structures on an oriented 4-manifold X is always nonempty, and hence it
is (noncanonically) isomorphic to H
2
(X; Z). An almost-complex structure
J naturally induces a spin
c
structure s
J
: the almost-complex structure re-
duces the structure group of TX from SO(4) to U(2), and the map
A
__
det A 0
0 1
_
, A
_
U(2) U(2)
provides the desired lift from U(2) to Spin
c
(4). Alternatively, since J gives
rise to the bundles
p,q
J
(X), we can take W
+
to be equal to
0,0
J
(X)
0,2
J
(X)
and W
=
0,1
J
(X) together with dened as
(x)(, ) =
2((x +iJx)
_
(x +iJx)
_
).
If J is dened away from nitely many points on X, we still get an induced
spin
c
structure s
J
Spin
c
(X) since the above construction provides a spin
c
structure on X x
1
, . . . , x
n
where J is dened and (since both S
3
and
D
4
admit unique spin
c
structures) it extends uniquely to X. It can be
shown that J
1
, J
2
induces the same spin
c
structure if J
1
is homotopic to J
2
outside of a 1-dimensional submanifold (containing all points where J
i
are
undened). Hence
Proposition 6.3.1. The set of spin
c
structures Spin
c
(X) on X can be
identied with
J almost-complex structure on X x
1
, . . . , x
n
for some x
1
, . . . , x
n
X/ ,
where is the equivalence relation described above.
6.3. Spin
c
If X is a compact 4-manifold with J as above, then the complex tangencies
along X provide an oriented 2-plane eld, giving a geometric interpretation
of the restriction map Spin
c
(X) Spin
c
(X) of spin
c
structures.
Exercises 6.3.2. (a) Show that s = (W
, ) Spin
c
(X) is induced by an
almost-complex structure if and only if c
2
(W
+
) = 0.
(b) Verify that for a closed 4-manifold X the identity
c
2
(W
+
) =
1
4
_
c
2
1
(W
+
) 3(X) 2(X)
_
holds.
If the second cohomology group contains torsion elements (for example, if Y
is a rational homology 3-sphere) it is quite complicated to work with spin
c
structures directly. In such cases c
1
might not determine the spin
c
structure,
and we cannot work with torsion second cohomology classes through their
values on embedded surfaces. The underlying smooth 3-manifold can always
be presented as the oriented boundary of a smooth 4-manifold built from a
0-handle and some 2-handles only. Studying spin
c
structures on Y through
their extensions to simply connected 4-manifolds (i.e., to manifolds where we
do not have torsion (co)homologies) turns out to be very useful in numerous
situations.
Exercise 6.3.3. Show that if X is simply connected then the restriction map
Spin
c
(X) Spin
c
(X) is onto. (Hint: Apply the long exact cohomology
sequence of the pair (X, X) and use the fact that H
1
(X; Z) = 0.)
Therefore, instead of studying t Spin
c
(X) we can focus on some s
Spin
c
(X) with s[
X
= t. Since
1
(X) = 1, the spin
c
structure s Spin
c
(X)
is uniquely determined by c
1
(s), and this class is specied by its values on
the second homologies of X. So suppose that X is a compact 4-manifold
with boundary, given by a Kirby diagram involving a unique 0-handle and
t 2-handles, attached along the knots K
i
(i = 1, . . . , t) with framings n
i
(i = 1, . . . , t). The corresponding basis of H
2
(X; Z) is denoted by
1
, . . . ,
t
.
Suppose furthermore that J is an almost-complex structure on X with rst
Chern class c
1
(X, J) satisfying
c
1
(X, J),
i
_
= m
i
for i = 1, . . . , t. Recall that c
1
(X, J) is a characteristic cohomology element,
that is,
c
1
(X, J),
i
_
Q
X
(
i
,
i
) (mod 2).
108 6. Spin
c
Denote the induced oriented 2-plane eld of complex tangencies on Y = X
by . Using the notation of Section 2.3 (see text preceding Exercises 2.3.4)
we get that the Poincare dual of c
1
(X, J) is equal to
t
i=1
m
i
[D
i
]. This
element maps to
t
i=1
m
i
i
H
1
(Y ; Z). From the relations among the
homology classes
i
we get a presentation of H
1
(Y ; Z), and we can easily
identify c
1
() and decide whether it is torsion or not. If c
1
() is torsion then
for some n N the class nc
1
() = 0, implying that nPD
_
c
1
(X, J)
_
maps
to zero under the map
H
2
(X, Y ; Z)

2
H
1
(Y ; Z),
hence it is in the image of
1
: H
2
(X; Z) H
2
(X, Y ; Z). Since
1
is
explicitly described in Section 2.3, it is a simple matter of solving a linear
system of equations to nd c H
2
(X; Z) with the property that
1
(c) =
nPD
_
c
1
(X, J)
_
. The linking matrix of the Kirby diagram dening X
enables us to determine cc = Q
X
(c, c), leading to a computation of c
2
1
(X, J)
since this latter term is equal to
1
n
2
c c
1
n
Z. Having this quantity at hand
now it is an easy exercise to determine d
3
() since the linking matrix of the
Kirby diagram provides (X) and (X). Of course, in general it is rather
hard to nd an appropriate (X, J) for a given (Y, ). As we will see, for a
contact 3-manifold (Y, ) given by a contact surgery diagram such (X, J)
can be described quite easily.
The above discussion naturally extends to cobordisms as well. Suppose
that (Y, t) is a given spin
c
3-manifold and the cobordism is dened by
attaching a (4-dimensional) 2-handle along K Y . Fix a 4-manifold X
with X = Y which admits a handle decomposition with 0- and 2-handles
only, and let s Spin
c
(X) be chosen in such a way that s[
X
= t. Note
that since X is simply connected, the spin
c
structure s is determined by
the values of c
1
(s) on a generating system of the second homology group
H
2
(X; Z). A spin
c
structure s
1
on X W extending s Spin
c
(X), that
is, a spin
c
cobordism (W, s
1
) from (Y, t) can be specied now by the value
m of c
1
(s
1
) on the 2-homology dened by the 2-handle giving rise to W.
Since H
2
(W, W; Z)

= Zg) with g a generator, the value
c
1
(s
1
), g
_
= n
species the extension s
1
of t. The computation of the self-intersection of g
and so of c
2
1
(s
1
) follows the same line as it is discussed in Section 2.3.
Exercises 6.3.4. (a) Suppose that [K] = 0 in H
1
(Y ; Z). Determine the
number of possible extensions of a given spin
c
structure t to the cobor-
dism given by the handle attachment along the knot K Y with surgery
coecient being equal to 0 (with respect to the Seifert framing).
6.3. Spin
c
(b) Suppose that (Y, t) is a spin 3-manifold. Show that s
1
Spin
c
(W) is a
spin extension if and only if, with the above notations,
c
1
(s
1
), g
_
= 0.
(c) Find an example of a spin
c
cobordism (W, s) such that
c
1
(s
1
), g
_
= 0
but s
1
Spin
c
(W) is not spin. (Hint: Start with a nonspin structure
t Spin
c
(Y ) and extend it.)
(d) Consider the cobordism W given by Figure 2.12. Let the spin
c
structure
s Spin
c
(W) satisfy
c
1
(s), g
_
= n. Determine c
2
1
(s) and
1
4
_
c
2
1
(s)
3(W) 2(W)
_
Q.
7. Symplectic surgery
After these preparatory chapters now we are ready to describe the surgery
scheme in the symplectic category. First we will deal with the general cut-
and-paste operation and then examine the handle attachment procedure in
detail. The chapter concludes with the description of a version of surgery
which will be useful in the contact setting, see Chapter 11.
7.1. Symplectic cut-and-paste
Denition 7.1.1. A vector eld v on a symplectic manifold (X, ) is a
symplectic dilation or Liouville vector eld if L
v
= . Notice that since
d = 0 we have that L
v
= d
v
+
v
d = d
v
, therefore the above
equation translates to d
v
= . A codimension-1 submanifold Y (X, )
is of contact type if there is a vector eld v dened on some neighborhood
Y of Y which is a symplectic dilation and is transverse to Y .
Remark 7.1.2. Notice the similarity with the denition of convex surfaces
in contact manifolds. The important dierence is that now a symplectic
dilation v has a direction: v is not a dilation anymore, since L
v
= .
This orientation property is also reected in the following denition:
Denition 7.1.3. A codimension-0 submanifold U X in (X, ) is -
convex (-concave) if U is of contact type and the vector eld v points out
of (into) U.
Let L
Y
= TY
=
_
v TX [ (v, x) = 0 for all x TY
_
. Since is
antisymmetric, we have that TY
TY ; here is taken with respect of .

From the nondegeneracy of it follows that L
Y
is a line eld on Y . Consider
the special case when Y is given as H
1
(a) for a function H: X R and
112 7. Symplectic surgery
regular value a R. Then L
Y
is equal to the vector eld v
H
, where v
H
is
specied by the equation dH =
v
H
that is, the vector eld v
H
is in the
line eld L
Y
. Recall that since is nondegenerate, the formula dH =
v
H
uniquely determines v
H
. Therefore the fact that v
H
L
Y
easily follows
from the fact that for any u TY we have (u, v
H
) = dH(u) = 0 since H
does not change in the Y direction.
Theorem 7.1.4 (Weinstein, [173]). The codimension-1 submanifold i : Y
X is of contact type if and only if there is a 1-form on Y such that d = i
and is nonzero on L
Y
.
Proof. We prove the theorem in the special case when Y = H
1
(a) for
some function H: X R and regular value a. Suppose that Y is of contact
type with vector eld v. Consider
=
v
= (., v) and take = i
.
Now = L
v
= d
v
= d
implies i
= d. In order to evaluate on
L
Y
notice that (v
H
) = (
v
)(v
H
) = (v
H
, v) = dH(v) ,= 0 since v is
transverse to Y = H
1
(a). Since v
H
spans L
Y
, the second property follows
for . For the converse direction extend the given to
dened on the
neighborhood Y in such a way that d
= . (This can be done since Y

retracts to Y .) If v is dened by the equation
v
=
we get that L
v
=
d
v
= d
= , moreover dH(v) = (v
H
, v) =
v
(v
H
) =
(v
H
) ,= 0
implies that v is transverse to the level set Y = H
1
(a).
Remark 7.1.5. The assumption that Y = H
1
(a) for some function H
is not very restrictive. For a codimension-1 submanifold Y X we can
always nd H such that Y H
1
(a) for some regular value a and this
description is enough for our purposes. Equality cannot always be achieved,
since the complement of Y might be connected, preventing the existence of
an appropriate H this is the case, for example, if X = T
2
and Y is a
homologically essential circle on it.
Proposition 7.1.6. If Y (X, ) is of contact type then the 1-form
provided by Theorem 7.1.4 is a contact form on Y .
Proof. We need to examine d on ker . Notice that d = i
and
ker
= TY/L
Y
, on which i
is obviously a symplectic form, proving that

is a contact form. (For ker
= TY/L
Y
notice that ker L
Y
= 0, hence
the map sending u ker TY to [u] TY/L
Y
is injective, and so an
isomorphism by dimension reasons.) Notice that the contact structure on
Y induced by is cooriented by .
7.1. Symplectic cut-and-paste 113
Remark 7.1.7. Alternatively, using =
v
we can compute d =
v
d(
v
) =
v
L
v
=
v
=
1
2
v
( ), so d is nowhere
zero on a hypersurface Y transverse to v, therefore = ker is a contact
structure on Y (after using the appropriate orientation).
Informally, the Liouville vector eld v in the denition of a hypersurface of
contact type helps us to determine the symplectic structure near Y . This
means that if we know on Y (through, for example, the induced contact
form ) then we know near Y . To make this picture more rigorous, we
prove the following
Proposition 7.1.8. Suppose that Y (X, ) is a hypersurface of contact
type (with vector eld v). Then Y admits a neighborhood Y symplecto-
morphic to a neighborhood
_
(Y )
_
of (Y ) Symp(Y, ), where =
v
and = ker
_
[
Y
_
.
Proof. Let us denote the symplectic form d(t) on Symp(Y, ) by
.
According to the Tubular Neighborhood Theorem there are neighborhoods
Y X and
_
(Y )
_
Symp(Y, ) which are dieomorphic through
a dieomorphism sending the ow of v to the ow of

t
. Notice that
[
(Y )
= d = [
Y
; furthermore we can arrange
[
T(Symp)|(Y )
= [
TX|
Y
;
so
= also holds in the normal direction. Using Mosers method now

the dieomorphism can be isotoped to a symplectomorphism.
Now we are in the position to prove the theorem which allows us to
perform symplectic cut-and-paste.
Theorem 7.1.9 ([38]). Suppose that U
i
X
i
are codimension-0 submani-
folds and
i
symplectic forms on X
i
(i = 1, 2) such that U
i
are
i
-convex and
the boundaries Y
i
= U
i
with the induced contact structures are contacto-
morphic. Then the surgered manifold (X
1
U
1
) U
2
admits a symplectic
structure.
Proof. Consider the contact forms
i
=
v
i
i
(i = 1, 2) and take the
symplectization Symp(Y, ) with contact structure = ker
1
induced by
1
. Now
1
(Y ) = 1 Y Symp(Y, ) = (0, ) Y . For the con-
tactomorphism : (Y, ker
1
) (Y, ker
2
) there is a function f : Y R
such that
2
= f
1
; the graph of f : Y R in Symp(Y, ) will be de-
noted by
2
(Y ) Symp(Y, = ker
1
). Fix neighborhoods N
i
X
i
and
N
i
Symp(Y, ) of Y and
i
(Y ), respectively, which are pairwise sym-
plectomorphic (such neighborhoods are provided by Proposition 7.1.8). By
rescaling
2
we can achieve that f < 1, hence N
1
and N
2
can be chosen to
be disjoint. Considering V Symp(Y, ) bounded by N
1
and N
2
we can
form [X
1
(U
1
N
1
)] V (U
2
N
2
). (Notice that topologically V is triv-
ial, it serves as an interpolation between the symplectic structures on the
two pieces.) By applying the above symplectomorphisms on the overlapping
regions we can glue the symplectic forms together, producing a symplectic
structure on the smooth 4-manifold (X U
1
) U
2
.
As a special case of the above construction we outline a proof of a
theorem of Gompf:
Theorem 7.1.10 (Gompf, [64]). Suppose that for i = 1, 2 the closed sym-
plectic 4-manifolds (X
i
,
i
) contain closed symplectic 2-dimensional sub-
manifolds
i
X
i
satisfying g(
1
) = g(
2
) and [
1
]
2
+ [
2
]
2
= 0. Fix
an identication f :
1

2
and consider an orientation reversing lift
F :
1

2
of f. Then the normal connected sum X
1
#
F
X
2
=
(X
1
1
)
F
(X
2
2
) admits a symplectic structure.
Proof (sketch). We assume rst that [
1
]
2
< 0. In that case
1
admits
an
1
-convex, and
2
(with [
2
]
2
> 0) an
2
-concave neighborhood as
their local models show. Then all we need to do is to show that the contact
structures on the boundaries are contactomorphic. Let
1
be the contact
form on
1
and
2
the pull-back of the contact form of
2
by F. Let
t
= t
1
+(1 t)
2
_
t [0, 1]
_
be a path connecting them. Since d
1
and
d
2
are both positive multiples of
1
[
1
for :
1

1
, we conclude
that the
t
are all contact forms on
1
: notice that d
t
= t d
1
+ (1
t) d
2
is also a positive multiple of
1
, and ker
t
is always transverse
to the bers of
1

1
, hence d
t
is nondegenerate on ker
t
. This
shows that
1
and
2
are isotopic, therefore Grays Theorem 4.1.16 shows
that they are contactomorphic, hence the previous construction proves the
theorem. If [
1
]
2
= [
2
]
2
= 0 then just use a function which berwise turns
the punctured unit disk in R
2
symplectically inside out.
Remark 7.1.11. The original proof of Gompf for the above theorem rests
on the symplectic neighborhood theorem, for details see [64]. Notice that
this construction applied for CP
1
CP
2
as it is described in Lemma 3.3.3
veries the existence of a minimal model of a symplectic 4-manifold.
In general it is quite a delicate question whether X of a symplectic mani-
fold X is of contact type or not (i.e., whether an appropriate vector eld
exists). In addition, to apply the above gluing procedure we have to relate
7.2. Weinstein handles 115
induced contact structures on hypersurfaces of contact type. In some spe-
cial cases the vector eld comes with the construction (for example, for a
Stein manifold), and contactomorphism can be proved by relying on some
form of classication results of contact structures on X.
7.2. Weinstein handles
In the following we work out a special case (rst described by Weinstein) of
the above gluing procedure when we glue a 4-dimensional 2-handle to a
symplectic 4-manifold (X, ) with -convex boundary. For a more general
discussion of gluing handles see [173]. Let us take the standard 2-handle H
as the closure of the component of
R
4
__
x
2
1
+x
2
2
1
2
(y
2
1
+y
2
2
) = 1
_
_
x
2
1
+x
2
2

6
(y
2
1
+y
2
2
) =

2
_
_
which contains the origin, see the shaded region in Figure 7.1, cf. [38]. It
inherits the symplectic structure
0
= dx
1
dy
1
+dx
2
dy
2
from the standard
structure on R
4
. Consider the vector eld
v = 2x
1
x
1
y
1
y
1
+ 2x
2
x
2
y
2
y
2
.
Exercises 7.2.1. (a) Show that v is equal to f for the function f =
x
2
1
+x
2
2
1
2
y
2
1

1
2
y
2
2
with respect to the standard Euclidean metric.
(b) Check that =
v
(
0
) = 2x
1
dy
1
+y
1
dx
1
+ 2x
2
dy
2
+y
2
dx
2
.
(c) Show that L
v
0
=
0
and that v is transverse to the boundary of
the standard 2-handle. (Hint: Use the fact that L
v
0
= d
v
0
. For
transversality compute df(v) and show that it is equal to 4x
2
1
+4x
2
2
+y
2
1
+y
2
2
.)
(d) Show that the attaching circle S = x
1
= x
2
= 0, y
2
1
+y
2
2
= 2 H is
Legendrian with respect to the contact structure = ker generated by v
on the boundary of the 2-handle. Similarly, the belt circle B = y
1
= y
2
=
0, x
2
1
+ x
2
2
=

2
H is Legendrian. Notice that in this part of H the
vector eld points out of H. The orientation of H near S given by v is
opposite to the orientation H inherits from H, while the two orientations
coincide near B.
Now applying the previous construction we get
, y y
1 2
, x x
1 2
B B
S
S
H
Figure 7.1. The standard 4-dimensional 2-handle H
Theorem 7.2.2. Suppose that (X, ) is a symplectic 4-manifold with -
convex boundary X and L X is a Legendrian curve (with respect to
the induced contact structure). Then a 2-handle H can be attached to X
along L in such a way that extends to X H as
and (X H) is
-convex.
Proof. According to the Legendrian neighborhood theorem, L X and
S H admit contactomorphic neighborhoods. Choose in the denition
of H so small that the attaching region of H becomes a subset of this
neighborhood. The contactomorphism between the neighborhoods of L
X and S H will provide a suitable gluing map. Notice that since this
map is dictated by Theorem 4.1.15, the framing of the handle attachment
is also given. The last statement follows from patching the vector elds
together.
Remark 7.2.3. The same gluing scheme has been developed for any 2n-
dimensional index k-handle (with k n) in [173]. For example, the 4-
manifold #
m
S
1
D
3
admits a symplectic structure with -convex boundary
7.2. Weinstein handles 117
just repeat the handle attachment for m 1-handles starting with the
standard 4-disk
_
D
4
,
st
[
D
4
_
and vector eld v = x

x
+y

y
+z

z
+t

t
.
In order to have a complete picture about the topology of X H we
need to identify the framing of the 2-handle H we have to use in the above
construction. Recall that L admits a canonical framing (as a Legendrian
knot), hence we need to understand the framing of the gluing relative to
this canonical one. We think of a framing as a vector eld in the tangent
space along the knot transverse to its tangent vector eld. Therefore we
need to identify two vector elds along L: v
1
is the vector eld in which is
transverse to the tangent of L (providing the contact framing), while v
2
is
the image of the direction we push-o the attaching circle when measuring
framings. Notice that for this computation we can work in the standard
handle H: the vector eld v
2
is dened as an image of a vector eld
in H, while v
1
is the image of the corresponding vector eld along the
Legendrian knot S H, since the gluing map is a contactomorphism.
For this computation, x a parametrization of the attaching circle S =
x
1
= x
2
= 0, y
2
1
+ y
2
2
= 2 as (0, 0,
2 cos t,
2 sin t). Then the unit

tangent vectors along S are given by s
(t) = (0, 0, sin t, cos t). Restricting

the contact form to the tangent to S we get
2 cos t dx
1
+
2 sint dx
2
,
therefore the contact framing (i.e., the vector eld along S(t) which is
orthogonal to s
(t) and is in the kernel of ) can be represented by the

vector eld V (t) = (sin t, cos t, 0, 0). Now the framing of the gluing is
measured by pushing o S in the (x
1
, x
2
)-direction in the handle. The
corresponding vector eld can be chosen, for example as (1, 0, 0, 0). Since
the two unit length vector elds intersect each other once, the dierence
of the two framings can be clearly represented by a meridian of the knot.
Taking the orientations into account, we see that the contact framing makes
one positive full turn around the origin, therefore the framing we get by
pushing the knot slightly in the (x
1
, x
2
)-direction is (1) when compared
to the contact framing. In conclusion
Theorem 7.2.4 (Weinstein). Suppose that (X, ), X and L are as in
Proposition 7.2.2. If we attach a (4-dimensional) 2-handle H with framing
1 with respect to its canonical contact framing to X along L then
extends to X H as in Proposition 7.2.2.
Corollary 7.2.5. Let L (S
3
,
st
) be a given Legendrian link. Then L
equips the 4-manifold X dened by handle attachment along the smooth
link underlying L (with framings tb(L
i
) 1) with a symplectic structure
such that X is -convex.
Proof. Apply the above theorem for D
4
with the symplectic structure
it inherits from (R
4
,
0
) and for the outward pointing radial vector eld
v = x

x
+y

y
+z

z
+t

t
.
Exercises 7.2.6. (a) Let the 3-manifold Y be given as 0-surgery on the
right-handed trefoil knot. Present Y as the -convex boundary of a sym-
plectic 4-manifold. (Hint: Take the Legendrian knot of Figure 1.4 and
compare framings.)
(b) Find a symplectic 4-manifold with -convex boundary dieomorphic to
the 3-manifold given by the surgery diagram of Figure 7.2. (Hint: Convert
the 0-framed circle into a 1-handle, see Figure 7.3.)
0
4
Figure 7.2. Stein llable 3-manifold
0
4
4 4
Figure 7.3. Convert appropriate 2-handle into 1-handle
In the construction above we always assumed that X has -convex boun-
dary, that is, a symplectic dilating vector eld transverse to the boundary
exists. In the gluing construction discussed above, however, we only need
the existence of this vectoreld near the Legendrian knot L X. As it
truns out, the necessary vector eld exists near the given knot under much
weaker assumptions: it exists if (X, ) is only a weak lling of the contact
3-manifold (Y = X, ), see Section 12.1.
7.3. Another handle attachment 119
7.3. Another handle attachment
A similar scheme would work by gluing the handle to X along the Leg-
endrian knot B of Figure 7.1, i.e., along the belt circle of H. This time,
however, the symplectic structures of X and H do not match up, since the
vector eld on the handle points in the wrong direction. Therefore the re-
sulting 4-manifold carries no natural symplectic structure. Notice also that
the framing coecient of this latter operation is +1 with respect to the
contact framing of the knot in X. This operation will have interesting in-
terpretation in the realm of contact surgery, see Chapter 11. Viewing this
latter construction from another point of view, we see that a symplectic
2-handle H can be glued along B (with framing (+1) relative to the con-
tact framing) to a symplectic 4-manifold X along a Legendrian knot lying
in an -concave part of X. In this case the symplectic structure will ex-
tend to the handle attachment. Return now to the picture when gluing the
handle along -convex boundary with framing (+1) (relative to the contact
framing).
Lemma 7.3.1. Take a Legendrian curve L (X, ) and push it o along
its contact framing to get another Legendrian knot L
. If we attach a 2-
handle along L with framing (1) (with respect to the contact framing) and
another 2-handle along L
with framing (+1) (again, measured with respect

to the contact framing) then the resulting 4-manifold will have boundary
dieomorphic to X.
Proof. This fact is quite obvious from the smooth point of view, since by
sliding L
over L we will get a 0-framed circle which is just the boundary of

a small normal disk to L. Surgering out the corresponding sphere of self-
intersection 0, we end up with L passing through a 1-handle, i.e., we get
a cancelling pair of handles. This means that we can erase them without
changing the 4-manifold. Since the surgery along the sphere does not change
the boundary, we conclude that the boundary after the handle attachments
is dieomorphic to X.
As we will see, the new 3-manifold we get after the handle attachment
with framing (+1) carries a natural contact structure. In Section 11.2 we
will sharpen the above lemma to prove contactomorphism for the resulting
structure after a (1)- and a (+1)-surgery on L and its contact push o
L
. In another context we will see that although the symplectic structure

does not extend through the handle when glued along the belt circle B, an
appropriate almost-complex structure does extend to H pt., providing,
for example, a spin
c
structure on the manifold X H. We will discuss this
aspect of the gluing theorem in the next section.
8. Stein manifolds
In this chapter we interpret the Weinstein handle attachment in the Stein
category, leading us to Eliashbergs celebrated theorem. To put this result
in the right perspective, we rst recall rudiments of Stein manifold theory.
The chapter concludes with a discussion about surfaces in Stein manifolds.
For a more detailed treatment of this topic the reader is advised to turn to
[70].
8.1. Recollections and definitions
Let X be a complex manifold. The holomorphic convex hull of K X is
K =
_
x X
f(x)
sup
yK
f(y)
for all f holomorphic on X

_
.
The manifold X is holomorphically convex if for all K X compact the
holomorphic convex hull

K is compact. This property is equivalent to the
requirement that for any innite discrete set D X there is a holomorphic
function f : X C which is unbounded on D. Yet another equivalent
property for a domain C
n
is the existence of a holomorphic function
f : C which cannot be extended holomorphically to any larger domain.
The traditional denition of Stein manifolds requires holomorph convexity
(which, as we remarked above, resembles to being a domain of holomorphy)
and the existence of many holomorphic functions. More precisely:
Denition 8.1.1. A complex manifold X is a Stein manifold if it is holo-
morphically convex, for each x ,= y X there is a holomorphic function
f : X C such that f(x) ,= f(y) and for every x X there are holomor-
phic functions f
1
, . . . , f
n
and a neighborhood U of x such that z
i
= f
i
[
U
(i = 1, . . . , n) give local coordinates on U.
122 8. Stein manifolds
Exercise 8.1.2. Show that if X is Stein then it is noncompact.
For example, C
n
and every closed analytic submanifold of C
n
is Stein. The
next theorem asserts that the converse also holds:
Theorem 8.1.3 (Narasimhan, Bishop, Remmert, EliashbergGromov).
Let q be an integer greater than or equal to
3n
2
+1. Then an n-dimensional
Stein manifold biholomorphically and properly embeds into C
q
.
Therefore the following denition (in the dimension of our interest) is equiv-
alent to the one given above:
Denition 8.1.4. The 2-dimensional complex manifold S is a Stein surface
if it admits a proper biholomorphic embedding S C
4
. That is, S is a
smooth ane 2-dimensional complex analytic submanifold in C
4
.
Another, technically more involved equivalent way of dening Stein man-
ifolds is to require the vanishing of the sheaf cohomology groups H
q
(X, o)
for q 1 and any coherent sheaf o (see [70], for example). For our purposes
yet another, more topological denition will be the most suitable. First a
few related denitions are in place:
Denition 8.1.5. A smooth function : X R on a complex manifold
X is (strictly) plurisubharmonic if is (strictly) subharmonic on every
holomorphic curve C X. Recall that is subharmonic if for r small
enough (x
0
)
1
2r
_
B(x
0
,r)
(x) dx; or alternatively 0 for the Laplace
operator . A function : X R is an exhausting function if
_
x X [
(x) < c
_
is relatively compact in X for all c R. Recall that a map
: X Y is proper if the inverse image of a compact set is compact.
(Hence a proper function : X [0, ) is exhausting.)
For example, the function z [z[
2
=
n
i=1
z
i
z
i
is a plurisubharmonic
exhausting function on C
n
. Moreover, if f is holomorphic (and not identi-
cally 0 on any component) then log [f[ is plurisubharmonic; e.g. log [z[
2
is
plurisubharmonic.
Let now Y X be a codimension-1 submanifold of X. The complex
tangencies along Y form a complex hyperplane distribution in TY , which
can be (at least locally) given as ker for some 1-form . The Levi form
L
Y
(x, y) is dened as L
Y
(x, y) = d(x, Jy) (where J is multiplication by
1). Taking the orientation of Y into account, L

Y
is dened up to
multiplication by a positive function. If Y is given as
1
(a) for some
8.1. Recollections and denitions 123
smooth function : X R (which can be assumed, at least, locally) then
its Levi form can be given as L
Y
(x, y) =
n
i,j=1
z
i
z
j
x
i
y
j
.
Denition 8.1.6. The hypersurface Y X is strictly pseudoconvex (or
J-convex) if its Levi form is positive denite.
In this case the complex hyperplane distribution can be proved to give a
contact structure, since the denition requires d to be nondegenerate on
ker .
Remark 8.1.7. It can be proved that if Y is strictly pseudoconvex then it
cannot be touched by a holomorphic curve from inside. More precisely,
if Y =
1
(0) is pseudoconvex and Y is oriented as the boundary of
X =
1
_
(, 0]
_
then any holomorphic curve C X with boundary
is transverse to Y , in particular int C Y = . This property explains
convexity in the denition.
A function : X R turns out to be strictly plurisubharmonic if the
associated Levi form (

2
z
i
z
j
) is positive denite. More precisely, if
1
(a)
is J-convex for all a (oriented as the boundary of the sublevel set a)
then there is a dieomorphism h: R R such that the function = h is
plurisubharmonic. A more invariant reformulation of the above fact can be
given as follows. Suppose that a smooth function : X R is given on the
complex manifold X. Consider the associated 2-form
= dJ
d. (Here
J
: T
X T
X is the dual of J. The operator J
d is frequently denoted
by d
C
, hence
= dd
C
.) This 2-form gives rise to a symmetric tensor
g
(x, y) =
(x, Jy).
Proposition 8.1.8. The smooth function : X R on the complex
manifold X is strictly plurisubharmonic if and only if g
is a Riemannian
metric. In particular, this property implies that the exact 2-form
is
nondegenerate, hence is an exact symplectic form, while g
is a K ahler
metric on X.
Note that if X is complex 2- (hence real 4-) dimensional then the denite-
ness of L
Y
on the (real) 2-dimensional distribution on a 3-manifold Y X
is in fact equivalent to requiring that the distribution is a contact structure.
Therefore in this dimension : X R is strictly plurisubharmonic if and
only if the complex tangencies provide a contact structure on the smooth
points of the level sets
1
(a). The gradient vector eld (with respect to
the Riemannian metric g
) is a symplectic dilation since L
= d(
)
and (
)(v) =
(, v) = g
(, Jv) = d(Jv) = J
d(v), hence
d(
) = dJ
d =
. Recall that on C
n
there are many plurisubhar-
monic functions, and so Stein manifolds (being complex submanifolds of C
n
for some n) admit many plurisubharmonic functions as well. In fact, the
converse of this statement also holds:
Theorem 8.1.9 (Grauert, [69]). If a complex manifold X admits a proper
plurisubharmonic function : X [0, ) then X is Stein.
For X
n
C
m
the distance function f(z) = [z p[
2
from a generic point
p C
m
denes a proper Morse function f : X
n
[0, ) with critical points
of index at most n [114]. In conclusion
Theorem 8.1.10 (Grauert, [69]). The complex surface S is Stein if and
only if it admits a proper Morse function f : S [0, ) such that away from
the critical points f
1
(t) is a contact 3-manifold (with complex tangent lines
as ) for all t.
Remarks 8.1.11. (a) It is not very hard to see that a plurisubharmonic
function satises the maximum principle, i.e., on a connected compact com-
plex space it is constant. In other words, if C is a holomorphic curve with
boundary then for a plurisubharmonic function the restriction [
C
has no
local maximum in the interior int C. This property is closely related to the
alternative reformulation of pseudoconvexity described in Remark 8.1.7.
(b) According to a result of Eliashberg and Gromov, two plurisubharmonic
functions and with complete gradient ows dene symplectomorphic
symplectic structures (X,
) and (X,
) on a Stein manifold X.
A compact manifold W with boundary will be called a Stein domain if there
is a Stein manifold X with plurisubharmonic function : X [0, ) such
that W =
1
_
[0, a]
_
for some regular value a. So a compact manifold
with boundary (and complex structure on its interior) is a Stein domain
if it admits a proper plurisubharmonic function which is constant on the
boundary. More generally, a cobordism W (with boundary Y
1
Y
2
) is
a Stein cobordism if W is a complex cobordism with a plurisubharmonic
function f : W R such that f
1
(t
i
) = Y
i
, t
1
< t
2
.
8.2. Handle attachment to Stein manifolds 125
8.2. Handle attachment to Stein manifolds
In this section we show a way to adapt the handle attachment scheme given
in the previous chapter to the case of Stein surfaces. Recall that in this
setting we assume that W is the level set of a plurisubharmonic function
and we consider the contact structure on W provided by the distribution of
complex tangencies. The main theorem (due to Eliashberg) is the following
Theorem 8.2.1 (Eliashberg, [25]). Suppose that W is a (complex) 2-
dimensional Stein domain and L W is a Legendrian knot. By attaching
a Weinstein handle H to W along L, the Stein structure can be extended
to W H.
Proof (sketch). The idea of the proof is the following: rst we approximate
the Legendrian knot L with a C
-close real algebraic Legendrian knot. In

this way the attaching map can be chosen to be complex analytic, providing
us a complex structure on W H. Notice that by the framing assumption
complex lines will match up under the gluing. Therefore the proof reduces
to extending the plurisubharmonic function which already exists on W.
Now suppose that we glue the 2-handle to the Stein domain
1
_
[0, c +]
_
,
hence the plurisubharmonic function is already dened on some parts
(containing the attaching circle) of the 2-handle. The extension of to
the 2-handle now proceeds by turning it into a standard model and then
extending. For details see [25]; for an explicit description of the shape of
the 2-handle see [50].
Corollary 8.2.2. A Legendrian link L (S
3
,
st
) determines a Stein mani-
fold X
L
. Topologically X
L
is given by 2-handle attachments along the link
L with framings tb(L
i
) 1 on the individual components.
Remarks 8.2.3. (a) Similar (simpler) result holds for attachment of 1-
handles: after attaching a 1-handle to a Stein domain the Stein structure
always extends.
(b) The product Y I of a contact 3-manifold (Y, ) and I = [0, 1] can be
equipped with a Stein structure. In addition, if (Y, ) is overtwisted then the
framing condition yields no restriction in the gluing. Therefore a cobordism
(involving only 2-handles) on an overtwisted 3-manifold always admits Stein
structure. (For the denition of overtwisted structures see Section 4.)
(c) The result generalizes to arbitrary dimension n > 2, with the simpli-
cation of dropping the framing condition.
The theory of Stein manifolds from the point of view of handle calculus
was carefully developed by Gompf in [65], see also Chapter 11 of [66]. Here
we highlight only one result of [65], which will be important in our later
considerations: the identication of the rst Chern class of a Stein structure
given by handle attachments. Suppose that the Stein surface (X, J) is given
by attaching Weinstein 2-handles to (D
4
, J
st
) along the Legendrian link
L = (L
1
, . . . , L
t
) (S
3
,
st
), and suppose that
L
i
X denotes the surface
corresponding to the knot L
i
(see Section 2.3). The value of c
1
(X, J) of the
resulting complex structure on the homology dened by the knot is given
by the following
Proposition 8.2.4 (Gompf, [65]). The rst Chern class of the resulting
complex structure is given by
c
1
(X, J), [
L
i
]
_
= rot(L
i
).
Proof. By denition, the value of the rst Chern class c
1
(X, J) is equal
to the obstruction of extending a complex trivialization of TD
4
to the
complex 4-manifold X we get after the handle attachment. To determine
this obstruction, we x trivializations on D
4
and on the handle and compute
the obstruction for splicing them together. Along the boundary S
3
we
can take the vector eld

x
(spanning the standard contact structure
st
,
regarded as ker(dz +xdy) on the nite part of S
3
), and an inward pointing
normal v these two vector elds span TD
4
over C , and extend them
over the 4-disk. In the handle consider the tangent vector eld and the
outward pointing normal w along the attaching circle. These two vector
elds extend to a complex trivialization of the tangent bundle of the handle
where the spanning disk of the attaching circle is viewed as part of iR
2
C
2
.
Now under the handle attachment we map w to v and into . The
obstruction for extending the trivialization given on D
4
is now simply the
rotation number of with respect to the chosen vector eld

x
, which is by
denition the rotation number of the Legendrian knot K (S
3
,
st
).
Remark 8.2.5. Similar statement holds when we glue Stein 1-handles rst
to D
4
and then add the 2-handles; for details see [65].
Exercises 8.2.6. (a) Equip the handlebodies given by Figure 2.14 with
Stein structures.
(b) Equip RP
3
with a contact structure.
(c) Find contact structures on lens spaces.
8.3. Stein neighborhoods of surfaces 127
8.3. Stein neighborhoods of surfaces
In this section we describe a way to nd Stein neighborhoods of certain
embedded surfaces in complex 4-manifolds. Let us x a complex manifold
(X
4
, J) and an embedded oriented surface
2
X
4
. As always, J : TX
TX denotes the almost-complex structure induced by the complex structure
of X; all dimensions are understood to be real. The theory presented
below resembles to the discussion about Bennequins inequality given in
Section 4.3.
Denition 8.3.1. A point p is a complex point if T
p
= JT
p
, i.e.,
the tangent space of at p is a complex line. The noncomplex points of the
embedding are called real points. For a generic embedding complex points
are isolated.
Saying the above condition in another way, p is an isolated complex
point if there are complex coordinates (z
1
, z
2
) in X such that locally can
be given by
_
z
2
= f(z
1
)
_
with p = (0, 0) and
f
z
(0, 0) = 0. This can be
seen by noting that the vectors
X(z
1
) = (1,
f
x
(z
1
)) and Y (z
1
) = (1,
f
y
(z
1
))
(with z
1
= x + iy) form a real basis of T
(z
1
,f(z
1
))
and a point is complex
if and only if these vectors are complex scalar multiples of each other, i.e.,
the determinant
_
1
f
x
(z
1
)
1
f
y
(z
1
)
_
= 2i
f
z
(z
1
) = 0.
Yet another way to see the picture is the following: If Gr
2
(4) denotes
the Grassmannian of oriented 2-planes in R
4
then consider the associated
bundle Gr
2
(X) = P
X

Gl
2
(R)
Gr
2
(4) for the principal frame bundle P
X

X. By taking the tangent planes of we get a lift of the embedding
X to F : Gr
2
(X). The complex tangent lines dene a subset
CGr(X) Gr
2
(X) and for the projection : Gr
2
X we have that
_
CGr(X)F()
_
is precisely the set of complex tangencies. The fact
that complex tangencies are isolated for a generic embedding now follows
from a general transversality result of Thom.
Denition 8.3.2. The index I
p
Z of an isolated complex point p is
dened as the winding number of
f
z
around a small circle around p.
To have a well-dened notion, one has to check that this quantity is inde-
pendent of f, see [49]. In fact, by choosing appropriate coordinates f can be
written either as z
2
= z
k
1
z
1
with k 0 or as z
2
= z
k+1
with k < 0, and so
I
p
= k. The index of the embedding X is dened as I() =
p
I
p
.
Denition 8.3.3. The complex point p is elliptic if I
p
= 1 and
hyperbolic if I
p
= 1.
In the Grassmannian picture the index I
p
can be interpreted as an
intersection multiplicity: if (P) = p for P CGr(X) F() Gr
2
(X)
then I
p
is the multiplicity of the intersection of CGr(X) with F() at
P. For a generic embedding X all complex points are isolated and
either elliptic or hyperbolic. Now the orientation of makes us able to
assign a sign to each complex point p: this sign is positive if the complex
orientation of T
p
coincides with the orientation of it inherited from and
it is negative otherwise. Said another way, CGr(X) falls into the disjoint
union CGr (X)
+
CGr (X)
according to whether the orientation of the 2-

plane is the complex one or its opposite. The complex point p is positive
if p = (P) for P CGr (X)
+
F() and negative if P CGr (X)
F().
We dene I
+
() (I
()) as the sum of I

p
for all positive (resp. negative)
complex points of X. Obviously I
+
() + I
() = I(). It turns out

that I
() are topological invariants, more precisely

Theorem 8.3.4 (Lai, [90]). With the above conventions I() = I
+
() +
I
() = () + []
2
and I
+
() I
() =
c
1
(X), []
_
. In conclusion
I
() =
1
2
(() + []
2
c
1
(X), []
_
).
Proof (sketch). By choosing a metric on X x a projection of TX[
to
the normal bundle . For proving the rst identity, choose a vector eld
v in the tangent bundle T. Applying J to it and projecting the result to
we get a section of the normal bundle . The projection provides
a zero of at p if and only if Jv
p
is in the tangent plane T
p
, which
happens if and only if either p is a complex point or v
p
= 0. Checking the
signs of the zeros the rst formula follows.
For the second formula consider the section s = (v
1
, v
2
)
1
v
1

C
v
2
of
the complex line bundle
2
C
(TX)
, where is a symplectic form on and

v
1
, v
2
is a local frame for T. Now the zeros of s correspond to those
points of where v
1
and v
2
are not independent over C, i.e., in the complex
points of . A careful checking of the signs expresses the (signed) sum of
zeros of s as I
+
() I
(). On the other hand the sum of zeros of s is

equal to c
1
(
C
(TX)
_
[]
_
=
c
1
(X), []
_
. The expressions for I
() now
8.3. Stein neighborhoods of surfaces 129
follow by adding and subtracting the above formulae.
Suppose now that either [] ,= 0 or is not a sphere.
Theorem 8.3.5. If there is an open subset U X such that U and U
admits a Stein structure then I
() 0.
Proof. By the adjunction inequality of Theorem 1.2.1 we have
[]
2
+[
c
1
(U), []
_
[ ()
once U is Stein (and is not a null-homologous sphere). Since c
1
(U) =
c
1
(X), the formulae for I
() imply the result.

According to a beautiful result of Forstneric, the converse of the above
theorem also holds:
Theorem 8.3.6 (Forstneric, [49]). If I
() 0 for a generic embedding

X then there is a Stein domain U X containing an isotopic copy
of .
Proof (sketch). The proof of this theorem involves two major steps. First
we use a cancellation theorem due to Eliashberg and Kharlamov: if p, q
are complex points with equal sign and I
p
+I
q
= 0 then can be isotoped
to cancel p and q without changing the other complex points or introducing
new complex tangencies. Therefore the assumption guarantees that we can
isotope to have only hyperbolic complex point. Then a local construc-
tion near real and hyperbolic points together with a patching argument
provides a neighborhood U of with a plurisubharmonic function show-
ing its Stein property: In a complex chart around a complex hyperbolic
point p
j
take the nonnegative function
j
(z
1
, z
2
) = [z
2
z
2
1
[. Now dene
0
on a tubular neighborhood of as h(v) where we implicitly identied the
tubular neighborhood with the normal bundle of and h is the quadratic
form of a Riemannian metric on the normal bundle . For the complex
points p
1
, . . . , p
m
choose
j
smooth cut-o functions (j = 1, . . . , m) sup-
ported by the (disjoint) complex coordinate neighborhoods which are con-
stant 1 near p
j
. The function =
m
j=1
j
+ (1
j
)
0
can be shown to
be plurisubharmonic, proving the fact that admits a neighborhood with
Stein structure.
In fact, this argument provides a Stein neighborhood basis for , that
is, Stein neighborhoods U
(0,1)
with U
=
<
U
and U
=
>
U
and all U
retract to . It is known that the presence of an elliptic point on

obstructs the existence of such a basis: the existing holomorphic Bishop
disks [27] around the elliptic point (which cannot t into all U
) would
provide holomorphic extensions of functions. Notice the similarity between
these ideas and the ones involved in the proof of Bennequins inequality
(Section 4.3).
9. Open books and contact structures
Recently Giroux [63] proved a central result about the topology of contact
3-manifolds. He showed that there is a one-to-one correspondence between
contact structures (up to isotopy) and open book decompositions (up to
positive stabilization/destabilization) on a closed oriented 3-manifold. This
chapter is devoted to the introduction of relevant notions and also some
parts of the proof of this beautiful correspondence.
9.1. Open book decompositions of 3-manifolds
Denition 9.1.1. Suppose that for a link L in a 3-manifold Y the com-
plement Y L bers as : Y L S
1
such that the bers are interiors
of Seifert surfaces of L. Then (L, ) is an open book decomposition of Y .
Traditionally the Seifert surface F =
1
(t) is called a page, while L the
binding of the open book decomposition. The monodromy of the bration
is called the monodromy of the open book decomposition.
Any locally trivial bundle with ber F over an oriented circle is canonically
isomorphic to the bration I F/(1, x)
_
0, h(x)
_
I/I S
1
for
some self-dieomorphism h of F. In fact, the map h is determined by
the bration up to isotopy and conjugation by an orientation preserving
self-dieomorphism of F. The isotopy class represented by the map h is
called the monodromy of the bration. Conversely given a compact oriented
surface F with nonempty boundary and h
F
(the mapping class group of
F) we can form the mapping torus F(h) = I F/(1, x)
_
0, h(x)
_
. Since
h is the identity on F, the boundary F(h) of the mapping torus F(h) can
be canonically identied with r copies of T
2
= S
1
S
1
, where the rst S
1
factor is identied with I/I and the second one comes from a component
132 9. Open books and contact structures
of F. Hence by gluing in r copies of D
2
S
1
to F(h) so that D
2
is
identied with S
1
= I/I and the S
1
factor in D
2
S
1
is identied with a
boundary component of F, F(h) can be completed to a closed 3-manifold
Y equipped with an open book decomposition. In conclusion, an element
h
F
determines a 3-manifold together with an abstract open book
decomposition on it. Notice that by conjugating the monodromy h of an
open book on a 3-manifold Y by an element in
F
we get an equivalent open
book on a 3-manifold Y
which is dieomorphic to Y . In Example 9.1.4(b)

we illustrate a method to convert an abstract open book to an open book
concretely embedded into an ambient 3-manifold. See also Section 15.2.
Remark 9.1.2. We dene a bered link as a link L Y whose complement
Y L bers over S
1
, in such a way that each ber intersects a tubular
neighborhood of L in a curve isotopic to L. It is clear that the binding of
an open book decomposition is a bered link and conversely a bered link
naturally induces an open book decomposition with our denition.
Theorem 9.1.3 (Alexander, [9]). Every closed and oriented 3-manifold
admits an open book decomposition.
Proof (sketch). There are several dierent proofs of this classical theorem
of Alexander. We rst outline a proof using branched covers. Every closed
and oriented 3-manifold Y is a 3-fold branched cover Y S
3
. The proof
of this fact rests on the following. Choose a Heegaard decomposition of Y
as U
1
f
U
2
(here U
1
and U
2
are solid genus-g handlebodies and f
g
is a
mapping class) and represent U
1
, U
2
as triple branched covers of B
3
. Gluing
the two B
3
together we can realize any mapping class for the gluing of the
two handlebodies [75, 117], hence the result follows. We can assume that
the branch locus of this cover is transverse to the pages of an open book
decomposition of S
3
. Notice that S
3
admits several open book decomposi-
tions (see examples below). Thus we get an open book decomposition of Y
by lifting an open book decomposition of S
3
to the cover.
Next we describe a proof given in [148]. It is well-known that every
closed and oriented 3-manifold is obtained from S
3
by a (1)-surgery along
a link L. Moreover we can assume that there is an unknot K S
3
such
that each component L
i
of L links K exactly once. Consider the trivial open
book of S
3
whose binding is K and whose pages are the spanning disks for
K with trivial monodromy (cf. Example 9.1.4(a)). It is clear that when we
remove a neighborhood V
i
of L
i
to perform surgery along L
i
we puncture
once every page of this open book on S
3
. Observe that the boundary of
9.1. Open book decompositions of 3-manifolds 133
a puncture (which is a meridional curve to L
i
in S
3
) becomes longitudinal
by a (1)-surgery along L
i
. By performing surgery along L
i
we glue in an
annulus (bounded by L
i
and this longitudinal curve on V
i
) to a puncture
on each page of the trivial open book of S
3
to obtain a page of an open
book on Y . Notice that a page of this open book on Y is planar, i.e., it is a
disk with holes since to each puncture we glue in an annulus. The binding
is given by K L and the monodromy is given by an (appropriate) Dehn
twist along a curve parallel to each boundary component L
i
and identity
near K. (Yet another proof of this theorem using Lefschetz brations was
given by Harer, see Section 10.2.)
Examples 9.1.4. (a) Taking L to be the z-axis and considering the
half-planes with boundary L we get an open book decomposition on R
3
.
Alternatively, take : R
3
_
(0, 0, z)
_
S
1
R
2
given by
(x, y, z) =
1
_
x
2
+y
2
(x, y).
This open book decomposition extends to the one point compactication
S
3
as a genus-0 open book decomposition with binding the unknot and
monodromy equal the identity id
D
2 of the disk. The resulting open book
decomposition is called the standard open book decomposition on S
3
. This
picture is another manifestation of the fact that S
3
is the union of two solid
tori, one is the neighborhood of the binding and the other one is the union
of the pages.
(b) Let h be the right-handed Dehn twist along the middle circle S
1
1
2
in S
1
[0, 1]. Using h as monodromy, it denes a 3-manifold together with
an (abstract) open book decomposition. We denote the corresponding open
book decomposition by ob
+
. Using methods we will discuss in the next
chapter, we can see that the 3-manifold given by h is the 3-sphere S
3
: Just
consider the Lefschetz bration given by h and realize that it can be built
using a single 1-handle and a 2-handle cancelling it; see Figure 9.1 for a
Kirby diagram of this 4-manifold. In fact, the binding of the resulting open
book decomposition can be identied with the positive Hopf link and the
pages are just the obvious Seifert surfaces. To see this, slide the circles
representing
_
S
1
[0, 1]
_
over the (1)-framed 2-handle and cancel the
1-handle/2-handle pair. Then the circles a = S
1
0 and b = S
1
1
will be linked once in the new S
3
. This is depicted by the ne lines in
Figure 9.1. Taking h
1
corresponds to reversing the orientation on the
Lefschetz bration and hence on its boundary S
3
. Therefore the resulting
1
a
a
b
b
Figure 9.1. The circles a and b after handle cancellation
open book decomposition has the negative Hopf link as its binding this
open book decomposition will be denoted by ob
. An alternative way to
give ob
is by considering H
= r
1
r
2
= 0 S
3
equipped with polar
coordinates (r
1
,
1
, r
2
,
2
) coming from S
3
C
2
and
(r
1
,
1
, r
2
,
2
) =
2
. The standard open book decomposition becomes L = r
2
= 0 and
(r
1
,
1
, r
2
,
2
) =
2
in these coordinates.
(c) Let p and q be relatively prime integers such that p, q 2. It is
well-known that a (p, q) torus knot T
(p,q)
is a bered knot. This gives
an open book decomposition of S
3
, where the ber is a surface of genus
1
2
(p 1)(q 1) with one boundary component and the monodromy is a
product of (p 1)(q 1) right-handed Dehn twists along nonseparating
(i.e., homologically essential) curves.
Next we will briey describe the plumbing operation (which is a special
case of the Murasugi sum) and explain how to construct the bered surface
of a torus knot (i.e., a page of the open book of S
3
induced by the torus
knot) by plumbing positive Hopf bands (cf. [7, 74]) depicted in Figure 9.2.
Recall that the monodromy of a positive (resp. negative) Hopf link is a right-
handed (resp. left-handed) Dehn twist along the core circle of the Hopf band.
Warning 9.1.5. Our convention for monodromy diers from Harers [74].
We glue the end (1, x) to
_
0, h(x)
_
in the mapping torus to calculate the
monodromy h as opposed to gluing (0, x) to
_
1, h(x)
_
as Harer does.
Let H
+
denote the positive Hopf link and F
+
its bered surface (the positive
Hopf band) in S
3
as it is shown in Figure 9.2. Suppose that (L, F) is another
bered link with its bered surface in an arbitrary 3-manifold Y . Choose
an arc in F connecting two points on the boundary L = F. Take a
neighborhood of in F and thicken this into a 3-ball. Now apply the
(a) (b)
Figure 9.2. (a) positive and (b) negative Hopf bands
same for the curve depicted in Figure 9.2 on the Hopf band F
+
and take
a connected sum of Y with S
3
along these 3-balls such that = I and
= I are glued in a way that is identied with I and is
identied with I . In fact plumbing a Hopf band is a special case of a
more general operation called the Murasugi sum which is dened similarly
for gluing arbitrary bered links along their bered surfaces. It is proven in
[151] (see also [52]) that by plumbing two bered links we get a new bered
link whose monodromy is the product of the monodromies of these bered
links in the following sense: First extend the monodromies of each bered
surface onto the glued up surface (obtained by plumbing) by identity and
then take the product of the resulting dieomorphisms.
For example, we can plumb two positive Hopf links to get a (2, 3) torus
knot (the right-handed trefoil) with its bered surface. Simply identify a
neighborhood of the arc in one Hopf band with a neighborhood of the arc
in the other Hopf band, transversely as shown in Figure 9.3. The resulting
monodromy will be the product of two right-handed Dehn twists along the
curves also drawn in Figure 9.3(c). Note that the two curves (one of which
is drawn thicker) intersect each other only once and they stay parallel when
they go through the left twist on the surface. We can iterate this plumbing
operation to express the monodromy of a (2, q) torus knot as a product of
(q 1) right-handed Dehn twists.
(a) (b)
(c)
Figure 9.3. Plumbing two Hopf bands

Exercise 9.1.6. Describe an abstract open book corresponding to the (2, q)
torus knot.
By attaching more positive Hopf bands we can construct the bered
surface of a (p, q) torus knot for arbitrary p and q. We depicted the (3, 5)
torus knot with its bered surface in Figure 9.4. We would like to view
this gure as two rows of gates. First construct the row of gates in the
back as described above and then plumb a Hopf band in the front row and
proceed as above to obtain a second row of gates. It should be clear that we
can iterate this process to build as many rows of gates as we wish. Hence
the monodromy of the (p, q) torus knot is a product of right-handed Dehn
twists.
Figure 9.4. Monodromy of the (3, 5) torus knot
Exercise 9.1.7. Show that the curves depicted on the Seifert surface of the
(3, 5) torus knot in Figure 9.4 are homologically essential.
Denition 9.1.8. Suppose that an open book decomposition with page F
is specied by h
F
. Attach a 1-handle to the surface F connecting two
points on F to obtain a new surface F
. Let be a closed curve in F
going
over the new 1-handle exactly once. Dene a new open book decomposition
with h t

F
, where t
denotes the right-handed Dehn twist along .

The resulting open book decomposition is called a positive stabilization of
the one dened by h. If we use a left-handed Dehn twist instead then we call
the result a negative stabilization. The inverse of the above process is called
positive (negative) destabilization. Notice that the resulting monodromy
depends on the chosen curve .
We can view the stabilization/destabilization as plumbing/deplumbing Hopf
bands. Since plumbing a Hopf band on the 3-manifold level is equal to the
connected sum with S
3
, by denition we do not change the topology of the
underlying 3-manifold.
There is another technique, called twisting, for constructing new open
book decompositions of 3-manifolds. Suppose that C is a curve embedded in
a page F of a given open book decomposition in S
3
. Twisting is dened as
performing a (1)-surgery along C with respect to the framing C acquires
by the page. It is easy to see that by this operation we add a Dehn twist
along C to the monodromy of the original open book decomposition. In
particular if C is unknotted in S
3
and the surgery coecient of C in S
3
turns out to be 1 (with respect to its Seifert framing) then the resulting
manifold is again S
3
and hence we get a new open book of S
3
.
Theorem 9.1.9 (Harer, [73]). Every open book decomposition in S
3
is
related to the trivial one by a sequence of plumbings and twistings.
In fact Harer conjectured that one can entirely omit the twisting operation in
the theorem above. Harers conjecture was recently veried by Giroux (and
also independently by Goodman [68]). It was showed that any two open
book decompositions of an arbitrary integral homology sphere are related
by a sequence of plumbings. (See Corollary 9.2.14.)
9.2. Compatible contact structures
Denition 9.2.1. An open book decomposition of a 3-manifold Y and
a (cooriented) contact structure on Y are called compatible if can be
represented by a contact form such that the binding is a transverse link,
d is a volume form on every page and the orientation of the transverse
binding induced by agrees with the boundary orientation of the pages.
This denition of compatibility is natural in the sense that the conditions
that > 0 on the binding and d > 0 on the pages is a strengthening of
the contact condition d > 0 in the presence of an open book on Y .
Exercises 9.2.2. (a) Show that the condition that d is a volume form
on every page is equivalent to the condition that the Reeb vector eld of
is transverse to the pages. (Hint: Recall that the Reeb vector eld R
is determined as the unique direction where d degenerates, and a volume

form is nondegenerate.)
9.2. Compatible contact structures 139
(b) Show that an open book decomposition and a contact structure are
compatible if and only if the Reeb vector eld of is transverse to the
pages (in their interiors) and tangent to the binding.
Intuitively an open book is compatible with a contact structure if we can
push the contact planes arbitrarily close to the tangents of the pages (except
on the binding) of the open book. Recall that the Reeb vector eld R
for
is transverse to the contact planes.
Next we would like to look at the simplest example of a compatible
open book decomposition and a contact structure. Recall that the standard
contact structure on R
3
can be given as the kernel of the form
1
= dz+xdy
and the pages of the standard open book decomposition on R
3
is given by
the half-planes around the z-axis (see Example 9.1.4). Notice that d
1
is
degenerate when restricted to a page, and therefore clearly is not a volume
form. In fact

z
is the Reeb vector eld for
1
and it is tangent to all
the pages of the open book decomposition. However, if we multiply
1
by
the positive function f(x, y, z) = e
x
2
+ e
y
2
we get a contact form which
represents the same contact structure and using this form, rather than the
standard one, we show that that the standard contact structure and the
standard open book decomposition on R
3
are compatible. It is clear that
the binding (the z-axis) is transverse to the contact planes and

z
orients
the binding. Let = (e
x
2
+e
y
2
)(dz +xdy). Then
d = 2xe
x
2
dx dz 2ye
y
2
dy dz +
_
(1 2x
2
)e
x
2
+e
y
2
_
dx dy
is a volume form on the pages since we can easily check that

z
is not
the direction that d degenerates so the Reeb vector eld of cannot be
tangent to the pages. So we checked two conditions in the denition but we
still have to verify the condition about the orientations. We want to show
that the orientation on the binding induced from a page (which is oriented
by d) agrees with

z
. This can be easily checked by evaluating d on the
basis
z
, u for any vector u in the xy-plane.
Exercise 9.2.3. Consider the contact form = (x
2
+y
2
+1)(dz +xdy) on
R
3
. Show that d is a volume from on the pages of the standard open book
decomposition but the orientation it induces on the binding is

z
.
Example 9.2.4. Next we give an example of a compatible contact struc-
ture and an open book decomposition on a closed manifold. Consider the
standard tight contact structure
st
on
S
3
=
_
(z
1
, z
2
) C
2
[ [z
1
[
2
+[z
2
[
2
= 1
_
.
The contact structure
st
can be given by the kernel of the 1-form
= r
2
1
d
1
+r
2
2
d
2
,
where z
j
= r
j
e
i
j
, for j = 1, 2. The simplest open book decomposition of S
3
is given by the binding L = r
2
= 0 and the bration (r
1
,
1
, r
2
,
2
) =
2
.
Then, for a xed
2
,
1
(
2
) is the interior of a disk bounded by the binding
L. This is the trivial open book decomposition of S
3
where the binding is
an unknot, the pages are disks and the monodromy is the identity. We can
see that the trivial open book decomposition of S
3
is compatible with
st
as follows: The tangent to the binding L is given by

1
and the contact
form is d
1
when restricted to r
2
= 0. Therefore the binding is transverse
to the contact structure
st
. The contact form restricted to a page is r
2
1
d
1
and thus d(r
2
1
d
1
) = 2r
1
dr
1
d
1
is a volume form. This also shows that
the orientation induced on the binding as the boundary of a page coincides
with the orientation induced by the contact form .
The roots of Girouxs result invoked at the beginning of this chapter go
back to the following classical result of Thurston and Winkelnkemper:
Theorem 9.2.5 (ThurstonWinkelnkemper, [167]). Every open book de-
composition of a closed and oriented 3-manifold Y admits a compatible
contact structure.
Proof. We describe the construction of Thurston and Winkelnkemper fol-
lowing the expositions given in [1, 122]. Recall that if an open book decom-
position of Y is given then Y is dieomorphic to
F(h)
_
(F D
2
),
where F is an oriented surface with boundary, h is a self-dieomorphism of F
preserving F pointwise and F(h) =
_
F [0, 1]
_
/((x, 1)
_
h(x), 0
_
) is the
relative mapping torus of the element h
F
. To simplify notation, below
we assume that the boundary F has only one component. To nd a contact
form on Y , we nd a contact form on F [0, 1] rst, which descends to the
quotient F(h) and then extend it over the solid torus S
1
D
2
FD
2
. Let
(t, ) be coordinates for a collar neighborhood C of F such that t (
1
2
, 1]
and F = t = 1. We claim that the set o of 1-forms satisfying
(1) d is a volume form on F, and
(2) = td near F,
is nonempty and convex. For proving the claim, choose a volume form
on F with
_
F
= 1 and [
C
= dt d.
Let
1
be any 1-form on F which equals td near F. Then by Stokes
Theorem we obtain
_
F
( d
1
) =
_
F

_
F
d
1
= 1
_
F
1
= 1
_
F
d = 0.
Hence the closed 2-form d
1
represents the trivial class in cohomology
and vanishes near F. By deRhams theorem there is a 1-form on F with
d = d
1
and vanishes near F. Dene
2
=
1
+ . Then d
2
= is a volume
form on F and
2
= td near F, showing that o , = . Let
1
and
2
be
two 1-forms in o. Then
d
_
1
+ (1 )
2
_
= d
1
+ (1 )d
2
> 0
on F and
1
+ (1 )
2
= td
near F, which shows the convexity of the set o.
Let be any 1-form in o. Then h
also belongs to the set o: dh
=
h
d is a volume form on F and h
= = td near F. By convexity, the

1-form

(x,)
=
x
+ (1 )(h
)
x
is in o for each and descends to the quotient F(h) where x is in the ber
and is in the base circle. Thus d induces a volume form when restricted
to a page of our open book decomposition. Notice that when we glue the
two ends of F I, the forms and h
match up on that ber. Moreover,

since h and hence h
is the identity near F, we have

(x,)
= td for
all (x, ) =
_
(t, ),
_
near F(h) = F S
1
. Let d be a volume form on
S
1
. We claim that
1
= +
d
is a contact form on F(h) for some suciently large constant > 0, where
denotes the projection of F(h) onto the circle S
1
= [0, 1]/ . To prove
the claim we x a point (x, ) F(h) and choose an oriented basis u, v, w
of T
(x,)
F(h) such that d
(x,)
(u, v) > 0 and
(u) =
(v) = 0. This means

that the vectors u and v are tangent to the ber and w is transverse to the
bration. Thus we get
(
1
d
1
)
(x,)
(u, v, w)
= ( d )
(x,)
(u, v, w) +
(x,)
_
d(
w)d
(x,)
(u, v)
_
.
Hence we conclude that
(
1
d
1
)
(x,)
(u, v, w) > 0
for
(x,)
suciently large since d(
w)d
(x,)
(u, v) is positive by the choice
of the oriented basis (u, v, w). By compactness of F(h), there is a suciently
large > 0 such that
1
d
1
> 0 on F(h).
Now we would like to identify a collar neighborhood of F(h) with a
tubular neighborhood of the binding. Let D(r) denote a disk of radius r.
We use polar coordinates (r, ) for D(1.5) (near the binding) and identify
coordinates as (, r, ) (, 2 t, ) where 1 r 1.5. Then the 1-form
1
dened above is given by
1
= (2 r)d +d
on F (D(1.5) D(1)) since = td near the boundary and
d is
identied with d. We have to extend this form smoothly onto F D(1.5).
Note that the form (2 r)d + d is a positive contact form away from
r = 0 but it does not extend across r = 0. Consider the 1-form
2
= (2 r
2
)d +r
2
d
instead, which is a contact form near r = 0 since
2
d
2
= 4rd dr d.
Now we claim that we can connect
1
and
2
by contact 1-forms, i.e., we
can nd smooth functions
f
1
, f
2
: [0, 1.5] R
so that the 1-form = f
1
(r)d +f
2
(r)d is a contact form on F D(1.5)
where equals
2
near r = 0 and equals
1
for 1 r 1.5. Note that the
necessary and sucient condition for to be a positive contact form is that
d > 0 which is equivalent to the condition
f
1
f
2
f
2
f
1
> 0
(away from r = 0) as the following simple calculation shows:
d = (f
1
d +f
2
d) (f
1
dr d +f
2
dr d)
= (f
1
f
2
f
2
f
1
) d dr d.
To guarantee the condition f
1
f
2
f
2
f
1
> 0 we choose smooth functions
f
1
(r) and f
2
(r) such that
f
1
(r) =
_
2 r
2
if 0 r 0.5
2 r if 1 r 1.5
f
2
(r) =
_
r
2
if 0 r 0.5
if 1 r 1.5
satisfying f
1
(r) < 0 (0.5 r 1), and f
2
(r) > 0 (0.5 r 1). It
is clear that we can nd such smooth functions. We claim now that the
smooth 1-form is compatible with the given open book decomposition.
Note that by construction d = d
1
= d is a volume form on the bers
of F(h). On the other hand d = f
1
dr d + f
2
dr d is a volume form
on = constant, 0 r 1.5 in F D(1.5). Note that the orientation
dt d on the collar of a ber F in F(h) and the orientation dr d on the
surface = constant, 0 < r 1.5 in the solid torus F D(1.5) match
up (via the identication t = 2 r ) to orient the pages of the given open
book decomposition. So we conclude that d induces a volume form when
restricted to a page of our open book decomposition. Moreover, the tangent
to the binding is given by

and it is clearly transverse to (2r

2
)d+r
2
d.
Finally notice that the orientation induced on the binding by the volume
form d agrees with

.
Exercise 9.2.6. We proved above that the 1-form
1
= +
d is a
contact 1-form for suciently large . Show that the contact planes ker
1
approach the tangents of the pages in the open book decomposition as
.
Theorem 9.2.5 was substantially rened by Giroux [63]. He proved
Proposition 9.2.7. Any two contact structures compatible with a given
open book decomposition are isotopic.
Proof. Suppose that
0
and
1
are two contact structures compatible with
a given open book decomposition of a closed oriented 3-manifold Y . Then
there are contact forms
0
and
1
such that
i
= ker
i
, where d
i
is
a positive volume form on the pages and
i
is transverse to the binding
L, for i = 0, 1. Choose coordinates (, r, ) near L in which the binding
and the pages are given by r = 0 and = const, respectively. Let
= f(r)d, where f(r) is a nondecreasing function which equals 0 for
small r and which equals 1 for r r
0
. Extend to Y as
d (which agrees
with d in a tubular neighborhood of the binding), where : Y L S
1
is the bration and d is a volume form on S
1
. Notice that in this way we
dene a global 1-form on Y which vanishes near the binding. Then the
1-forms
i,t
=
i
+ t, t 0 are all contact. It is easy to see that
i,t
is a
contact form away from the binding:
i,t
d
i,t
=
i
d
i
+t
d d
i
> 0
since
i
is a contact form and d
i
is a volume form on the pages. Moreover
for t large enough, the forms
s,t
= (1 s)
0,t
+ s
1,t
(0 s 1) are
also contact. Again, when we consider
s,t
d
s,t
away from the binding,
the only terms which are not necessarily positive are s(1 s)
1
d
2
and
s(1 s)
2
d
1
. But the rest of the terms are positive and some of them
are multiplied with the parameter t. This shows that for large enough t,
s,t
is contact for all 0 s 1 and hence
0,t
and
1,t
are isotopic, which
in turn implies that
0
and
1
are isotopic.
For the converse direction, we get the following theorem.
Theorem 9.2.8. Every closed contact 3-manifold (Y, ) admits a compati-
ble open book decomposition.
Here we outline a construction of a compatible open book described in
Goodmans thesis [68] which is a slight modication of Girouxs original
construction. An alternative construction will be given in Chapter 11.
Denition 9.2.9. A contact cell decomposition of (Y, ) is a CW-decom-
position of Y such that
(i) the 1-skeleton of Y is a Legendrian graph,
(ii) each 2-cell D is convex with tw(D, D) = 1, i.e., the contact planes
twist negatively once (along D) with respect to the surface D, and
(iii) the contact structure is tight when restricted to the 3-cells.
In fact every contact 3-manifold admits a contact cell decomposition. In
order to nd such a cell decomposition rst cover (Y, ) by a nite number
of Darboux balls. Then take a cell decomposition of Y such that each 3-cell
lies in the interior of one of these Darboux balls. Isotope the 1-skeleton to
be Legendrian by the Legendrian Realization Principle (cf. Lemma 5.1.18)
and make each 2-cell convex (cf. Remark 5.1.7). If tw(D, D) < 1 then
take a renement of the cell decomposition by appropriately subdividing D.
Denition 9.2.10. Let G be the 1-skeleton of a contact cell decomposition
of (Y, ). The ribbon R of G is a (smoothly embedded) surface in Y such
that the surface R retracts onto G, T
p
R =
p
for p G and T
p
R ,=
p
for
p / G.
A ribbon R for the 1-skeleton of any contact cell decomposition of (Y, )
can be constructed such that B = R is the binding of an open book on Y
compatible with . Notice that the ribbon R is a page of this open book.
Theorem 9.2.8 can be sharpened by determining the relation between two
open book decompositions compatible with a xed contact structure, in
a similar fashion as Proposition 9.2.7 does in the converse direction. This
result says that two open book decompositions are compatible with the same
contact structure if and only if they admit common positive stabilization.
We will not deal with the proof of this statement in these notes, although
the proof involves similar ideas as described above for the converse direction.
Summarizing the above results, together with this last missing identication,
we get Girouxs theorem announced in the introductory section:
Theorem 9.2.11 (Giroux, [63]). (a) For a given open book decomposition
of Y there is a compatible contact structure on Y . Contact structures
compatible with a xed open book decomposition are isotopic.
(b) For a contact structure on Y there is a compatible open book de-
composition of Y . Two open book decompositions compatible with a xed
contact structure admit common positive stabilization.
Remark 9.2.12. When we stabilize a compatible open book ob
on (Y, )
we take a connected sum of Y with S
3
at the topological level, so that the
resulting manifold is dieomorphic to Y . In the case of positive stabilization
the resulting open book is obtained by plumbing a positive Hopf band
F
+
to a page of ob
. Recall that the contact structure on S

3
compatible
with the open book ob
+
(induced by the positive Hopf link H
+
) is the
standard tight contact structure. Hence by Remark 5.2.3, the contact
structure compatible with the resulting open book is a contact connected
sum #
st
, which is isotopic to . In conclusion, positive stabilization does
not change the (compatible) contact structure. On the other hand, negative
stabilization does change the contact structure since the contact structure
on S
3
compatible with the open book ob
(induced by the negative Hopf

link H
) is an overtwisted contact structure. Consequently, the contact

structure compatible with an open book obtained by negative stabilization
is necessarily overtwisted.
The next natural question is how to read o contact topological properties
of a given contact structure from a compatible open book decomposition.
Recall that an open book decomposition can be specied by giving a map-
ping class in the mapping class group of the page. It seems that tightness
and llability properties of the contact structure translate to factorizabil-
ity of the monodromy of some compatible open book decomposition. The
diculty in using such characterizations lies in the fact that the open book
decomposition is not uniquely dened for a contact structure, rather it is
given up to a complicated equivalence relation between various mapping
class groups. The relation between properties of mapping classes and the
corresponding contact structures is still not completely understood. By the
classication of overtwisted contact structures one can prove
Corollary 9.2.13 (Giroux). A contact 3-manifold (Y, ) is overtwisted if
and only if it admits a compatible open book decomposition which is a
negative stabilization of another open book decomposition.
Proof. We already showed in Remark 9.2.12 that the contact structure
compatible with an open book obtained by a negative stabilization is over-
twisted. It is not hard to see that negative stabilization of an open book
changes the oriented 2-plane eld induced by the compatible contact struc-
ture by adding one to its 3-dimensional invariant. This follows from the
connected sum formula for the 3-dimensional invariant (cf. Chapter 11).
The classication of overtwisted contact structures shows that any oriented
2-plane eld v
0
is homotopic to an overtwisted contact structure
0
and we
know that there is an open book ob
0
compatible with
0
. Let (Y, ) be an
overtwisted contact 3-manifold. Now take the oriented 2-plane eld v
0
for
which the contact structure compatible with the negative stabilization of
the corresponding open book ob
0
is homotopic to the oriented 2-plane eld
induced by . Consider the open book ob which is obtained by a negative
stabilization of ob
0
. By construction, the contact structure
ob
compatible
with ob is homotopic (as an oriented 2-plane eld) to . Now since is over-
twisted, then (again by the classication of overtwisted contact structures)
and
ob
are isotopic, hence and ob are compatible, proving the result.
Corollary 9.2.14 (Giroux). In an integral homology 3-sphere, any two
open book decompositions can be related by a sequence of plumbing and
deplumbing positive and negative Hopf bands.
Proof (sketch). We plumb suciently many negative Hopf bands to one
of the given open book decompositions so that the 3-dimensional invariants
of the resulting contact structures become equal. (We might have to plumb
extra negative Hopf bands to guarantee that both contact structures are
overtwisted (cf. Corollary 9.2.13)). Thus these overtwisted contact struc-
tures are homotopic and therefore isotopic. Consequently, the resulting
open book decompositions have a common positive stabilization by Theo-
rem 9.2.11. Notice that by the assumption it follows that the 3-manifold
supports a unique spin
c
structure, hence oriented 2-plane elds are classi-
ed by their 3-dimensional invariants d
3
Z up to homotopy.
Another corollary makes use of Legendrian surgery:
Corollary 9.2.15 (Giroux, Matveyev). The contact 3-manifold (Y, ) is
Stein llable if and only if Y admits an open book decomposition compatible
with whose monodromy admits a factorization into right-handed Dehn
twists only.
A proof of this theorem is given in Theorem 10.3.4. For various notions
of llability of contact structures see Chapter 12, and for factorizations of
mapping classes into Dehn twists see Chapter 15.
Next we will discuss a criterion given by Goodman ([68]) to detect over-
twistedness of a contact structure by examining the monodromy of a com-
patible open book decomposition based on a dierent point of view. Notice
that the contact structures compatible with a given open book decompo-
sition are all tight or all overtwisted by Proposition 9.2.7. Hence we will
call an open book decomposition overtwisted if a contact structure com-
patible with this open book decomposition is overtwisted. Let , F be
properly embedded oriented arcs which intersect transversely on an oriented
surface F. The algebraic intersection number i
alg
(, ) is the oriented sum
over interior intersections. The geometric intersection number i
geom
(, ) is
the unassigned count of interior intersections, minimized over all boundary
xing isotopies of and . The boundary intersection number i
(, ) is
one-half the oriented sum over intersections at the boundaries of the arcs,
after the arcs have been isotoped, xing boundary, to minimize geometric
intersection. See Figure 9.5 for sign conventions. In particular, given an
arc on a page F of an open book decomposition with monodromy h, we

+
+
Figure 9.5. Sign convention for intersection numbers on a surface
can consider i
alg
_
, h()
_
, i
geom
_
, h()
_
and i
_
, h()
_
. Here h() is
oriented by reversing the orientation on h() which is obtained by pushing
forward the orientation of by the monodromy h.
Denition 9.2.16. A properly embedded arc is sobering for a monodromy
h if
i
alg
_
, h()
_
+i
geom
_
, h()
_
+i
_
, h()
_
0,
and is not isotopic to h().
Proposition 9.2.17 (Goodman, [68]). If there is a sobering arc F for
h then the given open book decomposition is overtwisted.
In order to prove this result, Goodman constructs a surface with Legendrian
boundary which violates Eliasbergs inequality given by Theorem 4.3.7. As
an example, we consider the simplest case: the open book decomposition
ob
of S
3
induced by the negative Hopf link H
with its bered surface F
.
Exercise 9.2.18. Show that the arc across the annulus F
in Figure 9.6
is a sobering arc for the monodromy h of the open book decomposition.
Recall that h is a left-handed Dehn twist along the middle circle of the
annulus. (Hint: Observe that i
_
, h()
_
= 1, while i
alg
_
, h()
_
=
i
geom
_
, h()
_
= 0.)
In the light of Proposition 9.2.17 this implies that the induced open book
decomposition is overtwisted. The arc , however, is not a sobering arc
( ) h
Figure 9.6. A sobering arc
for H
+
. In fact, since the monodromy of the open book decomposition
ob
+
is a right-handed Dehn twist, the compatible contact structure is Stein
llable by Corollary 9.2.15 and therefore it gives the standard tight contact
structure by the classication of contact structures on S
3
.
Proposition 9.2.19 (Goodman, [68]). If an arc F satises
i
alg
_
, h()
_
+i
geom
_
, h()
_
+i
_
, h()
_
= 1,
then the open book decomposition with page F and monodromy h
n
is
overtwisted for any n > 0.
For an application of Proposition 9.2.19 consider a genus-g surface F with
only one boundary component and let be a curve parallel to the boundary.
Then the open book decomposition with page F and monodromy t
n
is
overtwisted for n > 0, where t
denotes the right-handed Dehn twist along

. To see this, we rst observe that
t
= (t
a
0
t
a
1
t
a
2g1
t
a
2g
)
4g+2
where the curves a
0
, a
1
, . . . , a
2g
are depicted in Figure 15.5. Now the arc
shown in Figure 9.7 will be a sobering arc for h = t
1
a
2g
t
1
a
2g1
t
1
a
1
t
1
a
0
, since
i
alg
_
, h()
_
+i
geom
_
, h()
_
+i
_
, h()
_
= 1.
We depict and h() in Figure 9.7. Then since t
n
= h
n(4g+2)
, the
open book decomposition with monodromy t
n
is overtwisted for n > 0

by Proposition 9.2.19. On the other hand, the open book decomposition
with page F and monodromy t
n
(n 0) is Stein llable and hence tight.

() h
2g

a
Figure 9.7. The action of the mapping class h on the arc
Corollary 9.2.20 (Goodman, [68]). An open book decomposition is over-
twisted if and only if it has a common positive stabilization with an open
book decomposition which has a sobering arc.
Proof. If the nal open book decomposition has a sobering arc then it is
overtwisted by Theorem 9.2.17, and positive stabilization/destabilization
does not change the contact structure (up to isotopy). Conversely, if the
contact structure is overtwisted then it has a positive stabilization which
is a negative stabilization of some other open book decomposition. On the
other hand, negative stabilization can be realized by plumbing a negative
Hopf band. Now the solution of Exercise 9.2.18 shows the existence of a
sobering arc, concluding the proof.
9.3. Branched covers and contact structures
Let F denote an orientable compact surface and Y a closed, orientable 3-
manifold.
Denition 9.3.1. A smooth surjective map : F D
2
is called a simple
d-fold cover if there is a nite set Q in the interior of D
2
, called the branch
set and each p D
2
has a neighborhood U over which :
1
(U) U
behaves as follows:
(1) if p / Q then [
1
(U)
is a trivial d-fold cover, and
(2) if p Q then
1
(U) has d 1 components, one of which is a disk
projecting to U as a double cover branched over p, i.e., can be modeled
by the complex map z z
2
around the origin, while the others are disks
projecting dieomorphically.
9.3. Branched covers and contact structures 151
Denition 9.3.2. A smooth map h: Y S
3
is called a simple d-fold cover
with branch set B S
3
if it is locally dieomorphic to the product of
an interval with a simple d-fold cover of a disk D
2
and the branch points
(multiplied by the interval) form the set B.
Theorem 9.3.3 ([118]). Every open book decomposition of Y with con-
nected binding is a simple 3-fold cover of S
3
branched over a closed braid.
Proof. Let F be the page of a given open book decomposition on Y with
connected binding. Choose a 3-fold simple branched cover : F D
2
.
Then we can realize the monodromy of the open book decomposition as the
lift of a braid (viewed as a dieomorphism of the disk D
2
xing D
2
and the
branch set in D
2
) under the branched covering map . Now consider the
closure of this braid (viewed as a geometric object in the usual sense) in S
3
with respect to an axis A. Note that S
3
has an open book decomposition
with disks D
t
as pages, A = D
t
as binding, and the identity map as
monodromy. This construction gives a branched covering map h: Y S
3
such that each page of the given open book decomposition is realized as
h
1
(D
t
) and the binding is simply equal to h
1
(A).
Corollary 9.3.4 (Giroux). Every closed contact 3-manifold (Y, ) is a
simple 3-fold cover of (S
3
,
st
) branched along a transverse link.
Proof. For a given contact 3-manifold (Y, ) Theorem 9.2.8 provides an open
book decomposition of Y (with connected binding) which is compatible with
the contact structure . By Theorem 9.3.3, on the other hand, this open
book decomposition is a simple 3-fold cover h: Y S
3
branched over a
closed braid B. Let be the standard contact form on S
3
. Note that h
is not a contact form on Y . We denote by C the set of points in Y where h

fails to be a local dieomorphism. Next we follow the discussion in [67] to
construct a contact form on Y using the branched covering map h. There
is a tubular neighborhood U
1
of B with coordinates (, x, y) where
g = d +xdy y dx
for some positive function g on U
1
. Extend g to a positive function on S
3
such that g = 1 outside a tube U
2
slightly larger than U
1
and then dene
= g. Using polar coordinates,
can be given by
d +r
2
d
on U
1
B. Then we deform h by an isotopy of S
3
supported in U
1
to a
nonsmooth covering map H which is a local dieomorphism on Y C. Since
H is a local dieomorphism, = H
is a contact form on Y C. Moreover,

we can nd local coordinates (
0
, r
0
,
0
) in a tubular neighborhood U of C
such that the form is given by
(1 +nr
2
0
)d
0
+r
2
0
d
0
for some integer n. We can extend this form to a smooth contact form
on Y , again denoted by . To see this we write the form in Cartesian
coordinates as
=
_
1 +n(x
2
0
+y
2
0
)
_
d
0
+x
0
dy
0
y
0
dx
0
and check that
d = 2 d
0
dx
0
dy
0
> 0.
Let A denote the axis of the closed braid B which is the branch set of
h: Y S
3
. We can assume that both the braid B and the axis A are
transverse to the standard contact structure
st
on S
3
. Consider the trivial
open book decomposition of S
3
, whose pages are the disks D
t
, the binding
is the axis A = D
t
and the monodromy is the identity map. The pages
of the open book decomposition on Y are then given by h
1
(D
t
) and the
binding is equal to h
1
(A). To show that the contact form is compatible
with the open book decomposition on Y , we need to check the conditions
given in Denition 9.2.1. It is easy to see that the binding h
1
(A) of our
open book decomposition is transverse to the contact structure on Y
since A is transverse to and the covering map is a local dieomorphism
on the points of the binding. We also need to check that d induces a
volume form on every page of the open book decomposition on Y . Since
= H
= h
away from the set C, the contact form is simply the pull
back of by the covering map h. Now d = dh
= h
d is a volume form
on h
1
(D
t
) (h
1
(D
t
) U), because d induces a volume form on a page
D
t
of the trivial open book decomposition on S
3
and the covering map h is
a local dieomorphism. Near the set C, the contact form is given by
(1 +nr
2
0
)d
0
+r
2
0
d
0
in local coordinates. Note that
d = 2r
0
dr
0
d
0
= 2 dx
0
dy
0
on a small disk obtained by xing
0
and clearly it is a volume form on this
small disk which doubly covers a disk in S
3
with a branching point at r
0
= 0.
9.3. Branched covers and contact structures 153
This proves that d is a volume form on every page. The condition about
the orientation is clear since we use the same orientation preserving map to
pull back the contact form and to construct the branched cover. Thus, by
Proposition 9.2.7 the contact structure ker is isotopic to , which nishes
the proof.
10. Lefschetz fibrations on 4-manifolds
In the light of recent results it turns out that both closed symplectic 4-man-
ifolds and Stein surfaces admit a purely topological description in terms
of Lefschetz brations and Lefschetz pencils. In this chapter we give the
necessary denitions and sketch this topological descriptions of symplectic
and Stein manifolds. In the discussion we include achiral Lefschetz bra-
tions as well; these more general objects are useful in viewing open book
decompositions as boundaries of certain achiral Lefschetz brations. The
chapter concludes with some applications of Lefschetz brations in various
low dimensional problems.
10.1. Lefschetz pencils and fibrations
Denition 10.1.1. (a) Suppose that X and are given oriented 4- and 2-
dimensional manifolds. The smooth map f : X
4
2
is an achiral Lefschetz
bration if df is onto with nitely many exceptions p
1
, . . . , p
k
= C int X
(called the set of critical points), the map f is a locally trivial surface bundle
over f(C) and around p
i
C and q
i
= f(p
i
) f(C) there are complex
charts U
i
, V
i
on which f is of the form z
2
1
+ z
2
2
. We call the bers f
1
(q
i
)
(q
i
f(C)) singular, while the other bers are regular. A bration is
relatively minimal if no ber contains a sphere with self-intersection 1,
i.e., we cannot blow down X without destroying its bration structure.
(b) An achiral Lefschetz pencil on X (with X = ) is a nonempty nite
set B X (called the base point set) together with a map f : XB CP
1
such that each point b B has a coordinate chart on which f can be given
by the projectivization C
2
0 CP
1
and around its critical points f
behaves as in (a).
156 10. Lefschetz brations on 4-manifolds
(c) An achiral Lefschetz bration/pencil is called a Lefschetz bration/pencil
if the complex charts U
i
, V
i
around the critical and base points p
i
and
q
i
= f(p
i
) appearing in the above denition respect the given orientations
of X and .
Remark 10.1.2. Notice that in Denition 10.1.1(a) the manifolds might
have boundaries. If the typical ber f
1
(t) is a closed surface then
f
1
() = X, but the denition also allows f
1
(t) to have boundary,
in which case f
1
() forms only part of X. Notice that the notion of
achiral Lefschetz brations/pencils is more general than the one without
the adjective achiral ; although the terminology might suggest the contrary.
We did not take the courage for changing this unfortunate phenomenon, we
will rather remind the reader for this subtlety of the subject.
Denition 10.1.3. Two Lefschetz brations f : X and f
: X
are equivalent if there are dieomorphisms : X X
and :
such
that f
= f.
Next we show that a Lefschetz critical point corresponds to gluing a 4-
dimensional 2-handle, and then we determine its attaching map (see also
[66]). Recall that near the critical point f(z
1
, z
2
) equals z
2
1
+z
2
2
, so a nearby
regular ber is given by z
2
1
+ z
2
2
= t, and after multiplying f by a unit
complex number we can assume t > 0. For the discussion below, assume
that the chart does respect the orientation of X around the critical point.
If we intersect the ber with R
2
C
2
, we obtain the circle x
2
1
+ x
2
2
= t
in R
2
(where z
j
= x
j
+ iy
j
and R
2
is spanned by x
1
and x
2
). This cir-
cle bounds a disk D
t
R
2
and as t 0 the disk D
t
shrinks to a point
in R
2
. By denition D
t
= F
t
R
2
is the vanishing cycle of the critical
point, and we explicitly see the singular ber F
0
being created from F
t
by
the collapse of D
t
. A regular neighborhood F
0
of the singular ber is
obtained from the neighborhood F
t
by adding a regular neighborhood of
D
t
. This latter neighborhood is clearly a 2-handle H attached to F
t
, with
core disk equal to D
t
. (In fact, a corresponding Morse function can be
given locally by g = Re f, or g(z
1
, z
2
) = y
2
1
+ y
2
2
x
2
1
x
2
2
. This Morse
function provides a handlebody decomposition of the relative handlebody
built on F
t
for a regular ber F
t
.) Suppose that F
t
contains a ber
F
s
, 0 < s < t. Then the core of the 2-handle H is D
s
and the attach-
ing circle is the vanishing cycle D
s
F
s
. We describe the framing of the
handle attachment by comparing it with the framing on D
s
F
t
deter-
mined by the surface F
s
. At a point (
s cos ,
s sin , 0, 0) D
s

R
2
R
2
iR
2
= C
2
the vector (sin , cos , 0, 0) is tangent to D
s
.
10.1. Lefschetz pencils and brations 157
Since D
s
lies in F
s
, which is holomorphic in the given local coordi-
nates, the vector eld v() = (0, 0, i sin , i cos ) on D
s
is also tan-
gent to F
s
. This can be seen explicitly by taking, for example, the curve
(
s t
2
cos ,
s t
2
sin , it sin, it cos ) on the ber and consider its
tangent at the point (
s cos ,
s sin , 0, 0). Notice that the dot product of

v() with the tangent vector of the circle is zero, therefore v() provides the
normal to D
s
inside the surface F
s
. This framing has to be compared with
the one we get by considering a parallel nearby circle to the attaching cir-
cle in the 2-handle. In the tangent space of the 2-handle the corresponding
vector eld can be chosen to be (0, 0, 0, i), for example. This choice imme-
diately shows that the two framings dier by one. By taking the orientation
into account, if we measure the surface framing with respect to the push-
o on the 2-handle, we have to compute the winding number of the curve
(sin , cos ) around the origin, and this quantity can be easily veried to
be 1. On the other hand, we would like to specify the framing coecient
of the 2-handle with respect to the surface framing, therefore the above
argument translates to 1. For the case of a critical point admitting an
orientation reversing coordinate chart the above argument passes through
with an orientation reversal at the last moment, implying that the framing
coecient in that case is +1 (with respect to the ber framing). The above
reasoning can be obviously inverted in the following sense: Suppose that
f : X is an achiral Lefschetz bration with ,= and X is a
given knot which lies in a ber. Let X
be given by attaching a 2-handle

to X along with framing 1 relative to the surface framing of . Then f
extends as an achiral Lefschetz bration to X
.
Remark 10.1.4. By adding the standard shaped 2-handle, the map will
not extend to X
as a Lefschetz bration, since we added only a small neigh-

borhood of the critical point of the new singular ber, but not the whole
ber. On the other hand, f can be extended to a manifold dieomorphic
to X
(similarly to the procedure of smoothing corners, encountered in

Remark 2.1.1), and this is the content of the above argument.
Next we determine the monodromy around a critical value. For any
bundle with ber F over an oriented circle, the monodromy is determined by
a single dieomorphism representing the image of the canonical generator
of
1
(S
1
) in
F
. The bundle is then canonically isomorphic to the bration
I F/((1, x)
_
0, (x)
_
I/I S
1
. Given a Lefschetz bration
f : X and a disk D inheriting the orientation of , we can consider
the monodromy of the bundle f[
D
provided that the oriented circle D
avoids the critical values of f. If D contains no critical values then f[
D
is
trivial, as is the monodromy (i.e., is isotopic to id
F
). If D contains a
unique critical value, however, the monodromy is nontrivial provided f[
D
is
relatively minimal. A local computation shows
Proposition 10.1.5. If f
1
(c) contains a unique critical point then the
monodromy around c is a Dehn twist along a simple closed curve. If the
orientation of the chart containing c C respects the orientation of X
then the Dehn twist is right-handed, otherwise it is left-handed. The simple
closed curve is isotopic to the vanishing cycle of the singular ber under
examination.
(For the denition of a Dehn twist see Appendix 15.) Notice the assumption
about the number of critical points in a ber. It is not hard to see that any
Lefschetz bration admits a perturbation such that f is injective on C.
Therefore by xing a natural generating system of
1
( f(C)) we get a
word in
g
: if g
i
in the generating system is dened as g
i
= [
i
] with either
i
= D
i
for disks D
i
satisfying [f(C) D
i
[ = 1 and [h
i
] (i = 1, . . . , 2g())
is a natural generating system of
1
() then the bration can be encoded
by the word t
1
. . . t
n
[
i
,
i
] where the t
i
are the monodromies around
i
and
i
(resp.
i
) are the monodromies around h
i
(resp. h
i+g()
). The word
uniquely determines the bration since a Dehn twist determines its dening
vanishing cycle up to isotopy, and from this information the bration can be
recovered by adding 2-handles along the vanishing cycles with appropriate
framings. In fact, if the ber of the bration f : X is a manifold
with r boundary components then the resulting word naturally lives in
g,r
.
We also note that by blowing up the points of the base point set B (cf.
Denition 10.1.1) we can turn a Lefschetz pencil into a Lefschetz bration.
In conclusion, Lefschetz brations can be thought of being the geometric
counterparts of certain special words in various mapping class groups. This
relation will be discussed in more details in Section 15.2.
Suppose that f : X D
2
is an achiral Lefschetz bration, such that
each singular ber contains a unique critical point. We will describe an
elementary handlebody decomposition of X using essentially the denition
of an achiral Lefschetz bration and Proposition 10.1.5. We select a regular
value q
0
of the map f in the interior of D
2
, an identication of the ber
f
1
(q
0
)

= F (a compact surface with possibly nonempty boundary), and
a collection of arcs s
i
in the interior of D
2
with each s
i
connecting q
0
to q
i
, and otherwise disjoint from each other. We also assume that the
critical values are indexed so that the arcs s
1
, . . . , s
m
appear in order as we
travel counterclockwise in a small circle about q
0
. Let V
0
, . . . , V
m
denote a
collection of small disjoint open disks with q
i
V
i
for each i, see Figure 10.1.
Since an achiral Lefschetz bration is a locally trivial F-bundle away from
0
1
3
m
s
V
2
V
3
q
2
q
m
q
4
q
V
V
1
q
s
1
2
s
4
V
V
m
3
4
s
s
q
0
Figure 10.1. Fibration over the disk
the critical points, we have f
1
(V
0
)
= D
2
F with
_
f
1
(V
0
)
_

= S
1
F.
Let (s
i
) be a regular neighborhood of the arc s
i
. Now the discussion
following Remark 10.1.2 shows that f
1
(V
0
(s
1
) V
1
) is dieomorphic to
D
2
F with a 2-handle H
1
attached along a circle
1
contained in a ber
pt. F S
1
F. Moreover, the 2-handle H
1
is attached with framing
(1) relative to the natural framing on
1
inherited from the the ber. (The
curve
1
was called the vanishing cycle.) In addition,
_
(D
2
F) H
1
_
is dieomorphic to an F-bundle over S
1
whose monodromy is equal to the
Dehn twist t
1
along
1
. Continuing counterclockwise around q
0
, we add
the remaining critical values to our description, yielding that
X
0

= f
1
_
V
0
_
m
_
i=1
(s
i
)
_
_
m
_
i=1
V
i
__
is dieomorphic to (D
2
F) (
m
i=1
H
i
), where each H
i
is a 2-handle
attached along a vanishing cycle
i
in a ber of S
1
F S
1
with relative
framing (1). Furthermore, the part of
X
0

=
_
(D
2
F)
_
m
_
i=1
H
i
__
which maps to D is an F-bundle over S
1
, whose monodromy is the product
of Dehn twists along the vanishing cycles. We will refer to this product as
the global monodromy of the bration. Suppose that an achiral Lefschetz
bration f : X admits k singular bers. The Euler characteristic of X
can be easily computed as (F)()+k since we add k 2-handles to a surface
bundle. The computation of (X), however, turns out be a nontrivial issue.
There is a signature formula for hyperelliptic Lefschetz brations [35] and
there is an algorithm to compute the signature for Lefschetz brations over
S
2
given in [130].
After these topological preparations we begin our discussion about the
relation between Lefschetz brations and symplectic/Stein manifolds. For
the rest of this chapter we assume that all Lefschetz brations are of the type
given by Denition 10.1.1(c), i.e., the complex coordinate charts respect
the orientation xed on X. We start with the case of closed symplectic
manifolds; Stein surfaces will be discussed in the next section. The most
important result of the subject is Donaldsons groundbreaking theorem:
Theorem 10.1.6 (Donaldson, [22]). If (X, ) is a closed symplectic 4-
manifold and [] H
2
(X; R) is integral then X admits a Lefschetz pencil
such that the generic ber is a smooth symplectic submanifold.
Exercise 10.1.7. Prove that every symplectic manifold (X, ) admits a
symplectic form
such that [
] H
2
(X; R) lifts to an integral cohomology
class, i.e., it is in the image of the map H
2
(X; Z) H
2
(X; R) induced by
the inclusion Z R.
Remark 10.1.8. The proof of this theorem is rather involved, here we
restrict ourselves merely to a quick indication of the main idea. Let L X
be the complex line bundle with c
1
(L) = h H
2
(X; Z) (where h maps
to [] under the map H
2
(X; Z) H
2
(X; R)). To prove Theorem 10.1.6,
Donaldson showed that if k is large enough, then L
k
X admits a section
s such that s
1
(0) X is a symplectic submanifold. Using the same basic
idea, he also showed that for k even larger there are linearly independent
sections s
0
, s
1
(L
k
) such that the submanifolds
_
(t
0
s
0
+t
1
s
1
)
1
(0) X [ [t
0
: t
1
] CP
1
_
are symplectic and form a Lefschetz pencil on X. The proof is based on a
technique of Kodaira for embedding K ahler manifolds in CP
N
, although the
analytical details are much more subtle in the symplectic case. Specically,
it was proved that the map x
_
s
0
(x) : s
1
(x)
CP
1
(dened on
X
_
s
1
0
(0) s
1
1
(0)
_
) provides a Lefschetz bration on some blow-up
of X. The proof of Donaldsons result, in fact, shows the following:
Corollary 10.1.9 (Donaldson, [22]). If X is a closed symplectic 4-manifold
then it decomposes as W D where W is a Stein domain and D is a D
2
-
bundle over a surface
g
.
Proof (sketch). Take a section (L
k
) as above and consider the
function log [[
2
away from the zero set s
1
(0). This provides a plurisubhar-
monic function on W = Xs
1
(0) for some appropriate complex structure.
Since s
1
(0) is a D
2
-bundle over the surface
g
= s
1
(0), the conclusion
follows.
Donaldsons theorem admits a converse (which is considerably simpler to
prove):
Theorem 10.1.10 (Gompf, [66]). If the smooth, closed 4-manifold X
admits a Lefschetz bration such that the homology class of the ber is
nonzero in H
2
(X; R) then X admits a symplectic structure with the bers
being symplectic submanifolds (at their smooth points).
Remark 10.1.11. The proof of the above theorem follows the idea pio-
neered by Thurston [166] providing symplectic structures on surface bun-
dles, cf. Theorem 3.1.13. The extra complication of having singular bers
can be taken care of by implementing the existing local models around the
critical points. For details see [66, Chapter 10]. The main idea in the proof
is that (by splicing forms together) we get a closed 2-form which is sym-
plectic along the bers and then we add a large multiple of the pull-back of
a symplectic structure from the base to it. This leaves the ber directions
intact and takes care for the orthogonal directions. In fact, by taking even
larger multiples we can arrange that nitely many (xed) sections of the
bration become symplectic as well. This leads us to the following:
Corollary 10.1.12 (Gompf, [66]). If a smooth, closed 4-manifold X admits
a Lefschetz pencil then it carries a symplectic structure such that the generic
ber is a smooth symplectic submanifold.
Proof. By blowing up X we get X#nCP
2
equipped with a Lefschetz -
bration, moreover the n exceptional curves (being sections) can be chosen
to be symplectic. Now the symplectic normal sum blows them back down,
providing a symplectic structure on X.
Note that the above corollary is just the converse of Donaldsons Theo-
rem 10.1.6. We just remark here that the assumption in Theorem 10.1.10
about the homology class of the ber is not very restrictive: if the ber
genus is not equal to one or the bration has at least one singular ber then
it is fullled, see [66]. (For torus brations the statement does not neces-
sarily hold, as the obvious torus bration S
1
S
3
S
2
coming from the
Hopf map S
3
S
2
shows.)
10.2. Lefschetz fibrations on Stein domains
In [73] Harer proved that if a smooth 4-manifold X is obtained by attaching
1- and 2-handles to D
4
then it admits an achiral Lefschetz bration over
D
2
. Notice that the boundary of an achiral Lefschetz bration f : X D
2
acquires a canonical open book decomposition induced from the bration:
compose the map f with the radial projection : D
2
0 D
2
to get
f :
_
X f
1
(0)
_
S
1
, providing an open book decomposition on
X with binding f
1
(0). An alternative proof of Theorem 9.1.3 follows
from this fact since every closed oriented 3-manifold Y is the boundary
of a smooth 4-manifold obtained by attaching 2-handles to D
4
. Loi and
Piergallini [104] (and later Akbulut and the rst author [7]) showed that a
Stein domain always admits a Lefschetz bration structure:
Theorem 10.2.1 (LoiPiergallini, [104]). If W is a Stein domain then it
admits a Lefschetz bration structure. In addition, we can assume that the
vanishing cycles in the resulting bration are homologically essential.
In fact, Loi and Piergallini proved that any Stein domain can be given as
an analytic branched cover of D
4
along a holomorphic curve, or of D
2
D
2
along a positive braided surface. The theorem above follows from this result.
Proof. We describe the proof of this theorem given by Akbulut and the
rst author [7]. The proof explicitly constructs the vanishing cycles of the
Lefschetz bration, and associates to every Stein domain innitely many
pairwise nonequivalent such Lefschetz brations. We say that a Lefschetz
10.2. Lefschetz brations on Stein domains 163
bration is allowable if and only if all its vanishing cycles are homologically
nontrivial in the ber F. Note that a simple closed curve on a surface with
at most one boundary component is homologically trivial if and only if it
separates the surface. Sometimes we will refer to a homologically trivial
(resp. nontrivial) curve as a separating (resp. nonseparating) curve. A
positive allowable Lefschetz bration over D
2
with bounded bers will be
abbreviated as PALF. (Here the adjective positive just emphasizes that
we are working with Lefschetz brations, that is, all singular bers give rise
to right-handed Dehn twists in the monodromy.)
In the following we digress to give the details of a construction which is
due to Lyon [106]. We say that a link in R
2
is in a square bridge position
with respect to the plane x = 0 if the projection onto the plane is regular
and each segment above the plane projects to a horizontal segment and
each one below to a vertical segment. Clearly any link can be put in a
square bridge position. (Notice that we require the horizontal segment to
pass over the vertical; therefore in putting a projection in square bridge
position we have to pay special attention to possible illegal crossings. For
these see Figure 10.2, cf. also Figure 4.3.) Suppose that the horizontal and
Figure 10.2. How to handle illegal crossings
vertical segments of the projection of the link in the yz-plane are arranged
by isotopy so that each horizontal segment is a subset of
0 [0, 1] z
i

z
y
Figure 10.3. Trefoil knot in a square bridge position
for some 0 < z
1
< z
2
< < z
p
< 1 and each vertical segment is a subset
of
0 y
j
[0, 1]
for some 0 < y
1
< y
2
< < y
q
< 1. Now consider the 2-disk
D
i
= [, 1] [0, 1] z
i
for each i = 1, 2, . . . , p and the 2-disk

E
j
= [1, ] y
j
[0, 1]
for each j = 1, 2, . . . , q, where is a small positive number. Attach these
disks by small bands (see Figure 10.4) corresponding to each point (0, y
i
, z
j
)
for i = 1, . . . , p and j = 1, . . . , q. If p and q are relatively prime then the
result is the minimal Seifert surface F for a (p, q) torus knot K such that
K L = and L F. It is easy to see that each component of the link L
is a nonseparating curve on the surface F. Moreover we can choose p and
q arbitrarily large by adding more disks of either type D or type E. This
concludes our digression.
x
D
i
E
j
L
z
y
Figure 10.4. Attaching disks
Let K be a torus knot in S
3
. It is well-known that K is a bered knot
and the corresponding bration induces an open book decomposition of
S
3
whose monodromy is a product of nonseparating positive Dehn twists,
cf. Example 9.1.4. This factorization of the monodromy denes a PALF
X D
2
such that the induced open book decomposition on S
3
= X is
equal to the open book decomposition given by the torus knot.
Exercise 10.2.2. Verify that for any torus knot K the 4-manifold (PALF)
K
underlying the corresponding Lefschetz bration is dieomorphic to D
4
.
(Hint: Consider the handlebody decomposition of (PALF)
K
and use Kirby
calculus; in particular, locate cancelling 1-handle/2-handle pairs.)
Returning to the proof of Theorem 10.2.1 suppose rst that W is a Stein
domain built by 2-handles only. This means that we attach Weinstein 2-
handles to D
4
along the components L
i
of a Legendrian link L in S
3
= D
4
with framing tb(L
i
) 1 to get the Stein domain W. Hence our starting
point is a Legendrian link diagram in (R
3
,
st
) (S
3
,
st
). First we smooth
all the cusps of the diagram and rotate everything counterclockwise to put L
into a square bridge position. See Figure 10.5 for an example. (Notice that
clockwise rotation results in a diagram with vertical segments passing over
horizontal ones, contradicting our convention for square bridge position.)
Then the construction of Lyon described above allows us to nd a torus
z
y
Figure 10.5. Rotation of a Legendrian knot into square bridge position
knot K with its Seifert surface F such that each L
i
is an embedded circle
on F for i = 1, 2, . . . , n. In Figure 10.6 we depicted the embedding of the
right-handed trefoil knot into the Seifert surface of the (5, 6) torus knot.
Let L
+
i
be a parallel copy of L
i
on the surface F, and let lk(L
i
, L
+
i
) be
the linking number of L
i
and L
+
i
computed with parallel orientations. This
linking number is called the surface framing of L
i
. We will denote it by
sf(L
i
). Then we observe that the surface framing of L
i
will pick up a 1
at each left corner of the link in square bridge position and will change by
the amount of writhe at each under/over-crossing. To see this, imagine a
parallel copy L
+
i
of L
i
on the surface F then cut out and straighten the
narrow band on the surface bounded by L
i
and L
+
i
. Notice, however, that
this is exactly the recipe how the ThurstonBennequin invariant of L
i
is
calculated in its Legendrian position (before rotating and smoothing its
corners): 1 for each left kink plus the writhe of the knot. Thus we get
tb(L
i
) = sf(L
i
).
This simple observation turns out to be crucial for the proof of the theorem.
The Stein domain W is obtained by attaching a Weinstein 2-handle H
i
to D
4
along L
i
with framing tb(L
i
) 1 = sf(L
i
) 1 for i = 1, 2, . . . , n.
By our discussion of the handle decomposition of a Lefschetz bration
z
y
x
Figure 10.6. Trefoil knot on the Seifert surface of the (5, 6) torus knot
in Section 10.1 we can extend the Lefschetz bration structure on the 4-
manifold D
4
= (PALF)
K
over the 2-handles to get a new PALF since
L = L
1
, . . . , L
n
is embedded in a ber F of (PALF)
K
= S
3
. Thus we
showed that W

= D
4
H
1
H
n

= (PALF)
K
H
1
H
n
admits
a PALF and the global monodromy of this PALF is the monodromy of the
torus knot K composed with positive Dehn twists along the L
i
s. Notice
that the Dehn twists along the L
i
s commute since they are pairwise disjoint
embedded curves on the surface F.
Now we turn to the general case. Suppose that W is a Stein domain obtained
by attaching 1- and 2-handles to D
4
. First of all, we would like to extend
(PALF)
K
on D
4
to a PALF on D
4
union 1-handles. Recall that attaching
a 1-handle to D
4
(with the dotted-circle notation) is the same as pushing
the interior of the obvious disk that is spanned by the dotted circle into the
interior of D
4
and scooping out a tubular neighborhood of its image fromD
4
.
To reach our goal, we represent the 1-handles with dotted-circles stacked
over the front projection of the Legendrian tangle which is in standard form
as it is described in [65]. Then we modify the handle decomposition by
twisting the strands going through each 1-handle negatively once. In the
new diagram the Legendrian framing will be the blackboard framing with
one left-twist added for each left cusp. This is illustrated by the second
diagram in Figure 10.7.
tangle
Legendrian
tangle
Legendrian
tangle
Legendrian
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Figure 10.7. Legendrian link diagram in square bridge position
Exercise 10.2.3. Verify that the twisting operation does not change the
topology of the underlying 4-manifold. (Hint: See Figure 10.8. In (b) we
introduce a cancelling 1-handle/2-handle pair, in order to get (c) we slide
the new dotted circle over the old one, then in (d) we slide the strands over
the (1)-framed 2-handle and nally cancel the 1-handle/2-handle pair.)
Determine the change of the surgery coecients on the components of the
link passing through the 1-handle depicted in Figure 10.8(a).
(a) (b)
(c)
(d)
(e)
1
1
1
1
1
Figure 10.8. Introduction of negative twists
Next we ignore the dots on the dotted circles for a moment and consider the
whole diagram as a link in S
3
. We put this link diagram in a square bridge
position as in the previous case (see Figure 10.7) and nd a torus knot K
such that all link components lie on the Seifert surface F of K. Before
attaching the 1-handles we isotope each dotted circle in the complement
of the rest of the link such that it becomes transverse to the bers of the
bration S
3
K S
1
, meeting each ber only once, see [106] for details.
Now for each 1-handle we push the interior of the disk spanned by the
dotted circle into D
4
and this becomes a section of (PALF)
K
. Thus by
attaching a 1-handle to D
4
we actually remove a small 2-disk D
2
from
each ber of (PALF)
K
, and hence obtain a new PALF on D
4
union a 1-
handle. In other words, we extend the open book decomposition on S
3
induced by the torus knot K to an open book decomposition on S
1
S
2
.
The boundary of the disk we remove from the ber becomes a component
of the binding of the open book decomposition on S
1
S
2
. Notice that
this circle becomes a longitudional curve after the surgery on the boundary
(induced by attaching the 1-handle) since we swap meridian and longitude
by a 0-surgery. After attaching all the 1-handles to D
4
we get an open book
decomposition on the connected sum of k copies of S
1
S
2
for some k N
and a new PALF on
k
S
1
D
3
such that the regular ber is obtained by
removing disjoint small disks from F. Then as in the previous case we can
extend our PALF on D
4
1-handles to a PALF on D
4
1-handles 2-
handles. The global monodromy of the constructed PALF is the product
of the monodromy of the torus knot K and right-handed Dehn twists along
vanishing cycles corresponding to the 2-handles. Finally, we note that
the (p, q) torus knot can be constructed using arbitrarily large p and q.
Therefore our construction yields innitely many pairwise nonequivalent
PALFs, since for chosen p and q the genus of the regular ber will be equal
to
1
2
(p 1)(q 1).
In the proof of Theorem 10.2.1 we constructed an explicit Lefschetz
bration on a Stein domain which is given by its handle decomposition.
The boundary of this PALF has an open book decomposition induced from
the bration and it also acquires a contact structure induced from the Stein
domain. It turns out that the induced open book and the contact structure
are compatible. In the following we will outline a proof of this fact due
to Plamenevskaya [146] which is obtained by a slight modication of the
proof of Theorem 10.2.1. Recall that in the proof of Theorem 10.2.1 we
smoothly isotoped the Legendrian link into a square bridge position in order
to put it on the Seifert surface of a torus knot and we forgot about the
contact structure. One can modify this construction as follows: For a given
Legendrian link L in (R
3
,
st
) there exists a surface F containing L such
that d is an area form on F (where = dz +xdy), F = K is a torus knot
which is transverse to
st
and the components of L do not separate F. The
construction of the surface F is identical to the one in the proof above except
that we rst isotope L by a Legendrian isotopy so that in the front projection
all the segments have slope (1) away from the points where L intersects the
yz-plane, see Figure 10.9. Then we use narrow strips around these segments
as in the proof of Theorem 10.2.1 and connect them by small twisted bands
(twisting along with the contact planes) to construct the Seifert surface F
of a torus knot K = F. By further isotopies we can ensure that L lies in
F, d is an area form on F and F = K is transverse to
st
. Now thicken
Figure 10.9. Legendrian link diagram
this one page F (which carries the Legendrian link L) into a handlebody
U
1
which is the union of an interval worth of pages (see Section 5.2) so that
d is an area form on every page. Now we can ber the complementary
handlebody U
2
in S
3
with binding K and pages dieomorphic to F since
K is a bered knot. So far we obtained an open book decomposition of
S
3
which is expressed as a union of two half open books, one of which
is compatible with
st
. We would like to extend the contact structure
st
to the bered handlebody U
2
(as some contact structure ) so that it is
compatible with the open book decomposition on U
2
. This can be achieved
(see [146]) by an explicit construction of a contact form on U
2
similar to
the one we described in Theorem 9.2.5. Hence we get a contact structure
on S
3
which is compatible with our open book decomposition. Since the
monodromy of this open book decomposition is a product of right-handed
Dehn twists, is Stein llable and therefore isotopic to
st
. Moreover, by
construction and
st
coincide on U
1
so that the isotopy between and
st
can be assumed to be the identity on U
1
. Now we need to show that
the contact structures and
st
are isotopic on U
2
relative to U
2
. Notice
that U
2
can be made convex and one can check that the binding K is the
dividing set on U
2
. Uniqueness (up to isotopy) of a tight contact structure
with such boundary conditions was shown in the proof of Theorem 5.2.1.
In summary we proved
Proposition 10.2.4 (Plamenevskaya, [146]). For a given Legendrian link
L in (S
3
,
st
) there exists an open book decomposition of S
3
satisfying the
following conditions:
(1) the contact structure compatible with this open book decomposition
is isotopic to
st
,
(2) L is contained in one of the pages and none of the components of L
separate F,
(3) L is Legendrian with respect to ,
(4) there is an isotopy which xes L and takes to
st
,
(5) the surface framing of L (induced by the page F) is the same as its
contact framing induced by (or
st
).
In fact item (5) in the theorem above follows from (1)-(4) by
Lemma 10.2.5. Let C be a Legendrian curve on a page of a compatible
open book ob
in a contact 3-manifold (Y, ). Then the surface framing of

C (induced by the page) is the same as its contact framing.
Proof. Let be the contact 1-form for such that > 0 on the binding
and d > 0 on the pages of ob
. Then the Reeb vector eld R
is transverse
to the pages (by Exercise 9.2.2) as well as to the contact planes. Hence R
denes both the surface framing and the contact framing on C.

The rest of Plamenevskayas argument (including the case with the 1-
handles) is the same as the proof of Theorem 10.2.1. In summary, we
have an algorithm which constructs an explicit PALF on a Stein domain
X. This algorithm also yields an open book decomposition on X which is
compatible with the contact structure induced from the Stein domain.
Exercise 10.2.6. Find a PALF on D
2
T
2
using the given algorithm. Also
nd an open book decomposition of T
3
which is compatible with the contact
structure induced from the Stein domain D
2
T
2
. (Hint: See Figure 12.8
for a Stein structure on D
2
T
2
.)
The converse of the above theorem also holds, namely
Theorem 10.2.7 (LoiPiergallini, [104]). Every PALF admits a Stein
structure.
10.3. Some applications 173
Proof. We describe the proof given in [7]. Let X be a PALF. We can assume
that the boundary of a regular ber is connected by plumbing Hopf bands if
necessary. It is clear that X is obtained by a sequence of steps of attaching
2-handles X
0
= D
2
F X
1
X
2
X
n
= X, where each
X
i1
is a PALF and X
i
is obtained from X
i1
by attaching a 2-handle to a
nonseparating curve C
i
lying on a ber F X
i1
with framing sf(C
i
)1.
Notice that D
2
F has a Stein structure since it is obtained from D
4
by
attaching 1-handles only. Inductively, we assume that X
i1
has a Stein
structure and thus X
i1
has an induced compatible contact structure. In
[53] it was shown that this induced contact structure agrees with Torisus
contact structure given in Section 5.2. Let denote the double of a page F
of the open book (as in Section 5.2) induced on X
i1
by the PALF. We can
assume that is a convex surface which is divided by the binding F. The
simple closed curve C
i
on the convex surface is nonisolating with respect
to the dividing curve F since we assumed that C
i
is a nonseparating curve.
Then we apply the Legendrian Realization Principle (cf. Lemma 5.1.18) to
make C
i
Legendrian such that the surface framing sf(C
i
) of C
i
is equal to
its ThurstonBennequin framing (see Exercise 5.1.14). The result follows
by Eliashbergs handle attachment Theorem 8.2.1.
Remark 10.2.8. The same proof is valid for homologically trivial (i.e.,
separating) vanishing cycles except that one needs to apply a fold (see [76])
in that case. A fold introduces convenient additional dividing curves so
that the nonisolating condition is quaranteed even for homologically trivial
curve.
10.3. Some applications
In this section we use ideas developed above to solve certain low dimensional
problems.
Theorem 10.3.1. If W is a Stein domain then we can embed it into a
minimal, closed, symplectic 4-manifold X.
Proof. We know that a Stein domain W admits a PALF. By plumbing Hopf
bands if necessary, we may also assume that the boundary of the regular
ber F is connected. The bration induces an open book decomposition of
W with connected binding F. First we enlarge W to

W by attaching a
2-handle along the binding F with framing 0 (with respect to the surface
framing) to get a Lefschetz bration over D
2
with closed bers. Hence

W
is an

F-bundle over S
1
, where

F denotes the closed surface obtained by
capping o the surface F by gluing a 2-disk along its boundary. Let
F
denote the mapping class group of the closed surface

F. Now we can easily
extend

W into a Lefschetz bration X over S
2
with regular ber

F. Let
t
c
1
t
c
2
t
c
k
be the global monodromy of the PALF on W, where c
i
denotes
a simple closed curve on F for i = 1, 2, . . . , n. Then this product (after
capping o the boundary component) can be viewed as a product in
F
.
We clearly have
t
c
1
t
c
2
t
c
k
t
1
c
k
t
1
c
k1
t
1
c
1
= 1.
By Lemma 15.1.16 we can replace every left-handed Dehn twist by a product
of right-handed Dehn twists to obtain a factorization of the identity into a
product of right-handed Dehn twists. This factorization gives a Lefschetz
bration X over S
2
(with closed bers) which admits a symplectic structure
by Theorem 10.1.10. We can assume that the genus of F (and therefore of

F)
is at least two so that the hypothesis [

F] ,= 0 H
2
(X; R) in Theorem 10.1.10
is automatically satised. Consequently, the Stein domain W is embedded
into a closed symplectic 4-manifold X. As we will see, by taking ber sum
if necessary we can assume that X is minimal, cf. Proposition 10.3.9.
Remark 10.3.2. The embeddability of a Stein domain into a 4-manifold
with some extra structure was rst noticed by Lisca and Matic [97]. They
proved that for any Stein domain W there is a minimal surface X of gen-
eral type such that W embeds into X, i.e., there is a K ahler embedding
f : W X. This observation was used to distinguish homotopic but non-
isotopic (Stein llable) contact structures. The above embedding of a Stein
domain into a minimal, closed, symplectic 4-manifold is due to Akbulut and
the rst author [8]. In another direction, we will show that any symplectic
lling embeds into a closed symplectic 4-manifold, see Theorem 12.1.7.
The following converse of Theorem 10.3.1 easily follows from Theorem 10.2.7
and Remark 10.2.8; compare also with Corollary 10.1.9.
Corollary 10.3.3. Let f : X S
2
be a Lefschetz bration which admits
a section. Then by removing a neighborhood of the section union a regular
ber we get a (positive) Lefschetz bration which admits a Stein structure.
A slightly weaker version of the next theorem is due to Loi and Pier-
gallini. This theorem provides a connection between llability properties of
a contact structure and the monodromy of a compatible open book decom-
position, cf. Corollary 9.2.15. For denitions of various llability notions
see Section 12.1.
Theorem 10.3.4 (Giroux). A contact 3-manifold (Y, ) is Stein llable
if and only if it admits a compatible open book decomposition with mon-
odromy h
g,r
such that h = t
a
1
. . . t
an
with t
a
i
right-handed Dehn twists
along homotopically nontrivial simple closed curves. Each Stein lling of
(Y, ) occurs as the Lefschetz bration corresponding to such a decomposi-
tion. The genus of the bration, however, might change from one lling to
another.
Proof. Suppose that h = t
a
1
. . . t
an
for some right-handed Dehn twists along
homotopically nontrivial simple closed curves. Then the (positive) Lefschetz
bration with total monodromy h is a Stein lling of Y . Conversely, any
Stein lling of (Y, ) admits a PALF and thus induces an open book decom-
position on the boundary which is compatible with the contact structure .
Notice, however, that in order to encounter all Stein llings we might need
to stabilize the open book decomposition.
Next we give an explicit constructions of some genus-2 Lefschetz bra-
tions. This construction will be used later in our study of Stein llings of
certain contact 3-manifold. Furthermore, these examples show that ber
sums of holomorphic Lefschetz brations do not necessarily admit complex
structures.
Theorem 10.3.5 ([131]). There are innitely many (pairwise nonhomeo-
morphic) 4-manifolds which admit genus-2 Lefschetz brations but do not
carry complex structure with either orientation.
Proof. Matsumoto [108] showed that S
2
T
2
#4CP
2
admits a genus-2
Lefschetz bration over S
2
with global monodromy (t
1
t
4
)
2
, where
1
, . . . ,
4
are the curves depicted by Figure 10.10. Let B
n
denote the
smooth 4-manifold obtained by the twisted ber sum of the Lefschetz -
bration S
2
T
2
#4CP
2
S
2
with itself, using the dieomorphism h
n
of
the ber
2
, where h denotes the right-handed Dehn twist about the curve
which is depicted in Figure 10.11. Then B
n
admits a genus-2 Lefschetz
bration over S
2
with global monodromy (t
1
t
4
)
2
(t
h
n
(
1
)
t
h
n
(
4
)
)
2
.
Standard theory of Lefschetz brations gives that
1
(B
n
) =
1
(
2
)/
1
, . . . ,
4
, h
n
(
1
), . . . , h
n
(
4
)
_
,
showing that
1
(B
n
) = Z Z
n
.
2
1

4 3
Figure 10.10. Vanishing cycles
Figure 10.11. The twisting curve

Exercise 10.3.6. Show that the Lefschetz bration B
n
S
2
admits a
section. (Hint: Verify the statement for S
2
T
2
#4CP
2
S
2
and splice
the sections together.)
The denition of B
n
provides a handlebody decomposition for it and shows,
in particular, that the Euler characteristic (B
n
) is equal to 12. Since B
n
is the ber sum of two copies of S
2
T
2
#4CP
2
, we get that the signature
(B
n
) = 8, consequently b
2
(B
n
) = 12, b
+
2
(B
n
) = 2 and b
2
(B
n
) = 10. Let
M
n
denote the n-fold cover of B
n
with
1
(M
n
)

= Z. Easy computation
shows that b
+
2
(M
n
) = 2n and b
2
(M
n
) = 10n. This allows us to show that
B
n
does not admit a complex structure (see [131] for details).
Next we study the problem of nding the minimal number of singular
bers a Lefschetz bration can have. (If we allow achiral brations as well,
then the answer becomes trivial.)
Lemma 10.3.7 ([155]). For a given Lefschetz bration f : X there
are almost-complex structures J and j on X and resp., such that f is
pseudoholomorphic, that is, df J = j df.
Using SeibergWitten theory, in particular Taubes results on Seiberg
Witten invariants of closed symplectic 4-manifolds, this observation quickly
leads us to the proofs of the following two results:
Proposition 10.3.8 ([155]). Suppose that f : X is a given Lefschetz
bration with g() > 0. Then the bration X is relatively minimal if
and only if X as a symplectic 4-manifold is minimal.
Proof. One direction of the theorem is obvious: if X contains no (1)-
sphere then the Lefschetz bration f : X is necessarily relatively
minimal. For the converse direction suppose that X is not minimal. Using
Taubes result, a (1)-sphere S can be displaced to be a J-holomorphic
submanifold, hence f[
S
: S is a holomorphic map. By the assumption
on the genus of it is therefore constant, hence S f
1
(p) for some p ,
contradicting relative minimality of the bration f : X .
A similar (but somewhat longer) chase for (1)-spheres proves
Proposition 10.3.9 ([154]). A Lefschetz bration X S
2
is relatively
minimal if and only if X#
f
X is minimal.
Here are a few corollaries of the above propositions:
Corollary 10.3.10. If X is a relatively minimal Lefschetz bration
and g() > 0 then c
2
1
(X) 0. If X S
2
is relatively minimal then
c
2
1
(X) 4 4g.
Proof. The rst statement follows from Proposition 10.3.8 and Taubes
result 13.1.10. For the second statement notice that (X#
f
X) = 2(X)
and (X#
f
X) = 2(X) + 4g 4, therefore by Proposition 10.3.9 we have
that
0 c
2
1
(X#
f
X) = 3(X#
f
X) + 2(X#
f
X)
= 6(X) + 4(X) + 8g 8 = 2c
2
1
(X) + 8g 8,
which implies the result.
Corollary 10.3.11. A genus-g Lefschetz bration X S
2
has at least
4
5
g
singular bers.
Proof (sketch). We can assume that X is relatively minimal. Let n and
s denote the number of homologically nontrivial and homologically trivial
vanishing cycles, respectively. Then 4 4g c
2
1
(X) = 3(X) + 2(X)
3(n s) + 2(4 4g + n + s) = 5n s + 8 8g, implying n
4
5
g. The
inequality (X) ns (see [130] for example) follows from the fact that a
2-handle attachment can change the signature of the 4-manifold by at most
1, and if the vanishing cycle is homologically trivial, such an attachment
reduces the signature.
If we allow the base space to have higher genus then the problem of nding
the minimal number of singular bers in a relatively minimal Lefschetz
bration is almost completely solved, see Chapter 15.
11. Contact Dehn surgery
Now we are in the position to describe the contact version of the smooth
surgery scheme we started our notes with. This method provides a rich
and yet to be explored source of all kinds of contact 3-manifolds. The
approach to 3-dimensional contact topology we outline here was initiated by
Ding and Geiges [16, 17], see also [18, 19]. Using contact surgery diagrams
and applying achiral Lefschetz brations we will make connection
to Girouxs theory on open book decompositions, and we will also show
a way to determine homotopic properties of the contact structures under
examination. We begin by reviewing the classication of tight structures on
S
1
D
2
due to Honda this is the result which allows us to dene contact
surgery diagrams.
1
D
2
The idea of Honda in the classication of tight contact structures on the
solid torus S
1
D
2
is roughly the following: there is a strong relationship
between contact structures on the solid torus S
1
D
2
with certain boundary
condition, the thickened torus T
2
[0, 1] with certain related boundary
conditions, and on the lens space L(p, q) where p and q depend on the above
boundary conditions. Legendrian surgery provides many tight, in fact, Stein
llable contact structures on lens spaces this gives a lower bound for the
number of structures on the solid torus. Using convex surface theory then
Honda gives an upper bound for that number, which matches with the
lower bound given by the surgeries. This concludes the proof. Here we will
show the lower bound in the general case by using Legendrian surgery, and
produce a (generally much weaker) upper bound for the number of tight
structures on S
1
D
2
. In one particular case, however, our two numbers
180 11. Contact Dehn surgery
will match up, giving the classication in that case and this is the case
our contact surgery construction will rely on. Let us start by stating the
result of Honda. For this, let us assume that for p q 1 the rational
number
p
q
is equal to [r
0
, . . . , r
k
] where [r
0
, . . . , r
k
] denotes the continued
fraction expansion of the rational number
p
q
, i.e.,
p
q
= r
0
1
r
1
1
r
2
...
1
r
k
with r
i
2 (i = 0, . . . , k) in case p > q. (For p = q = 1 we have k = 0 and
r
0
= 1.) Notice that any nonzero slope on the boundary of the solid torus
can be transformed into the form
p
q
with p q 1 and (p, q) = 1 by a self-
equivalence S
1
D
2
S
1
D
2
. The case p = 0 needs dierent treatment,
see the concluding remark of this section. Now x relative primes (p, q) with
p q 1.
Theorem 11.1.1 (Honda, [76]). The solid torus S
1
D
2
has exactly
(r
0
+ 1)(r
1
+ 1) (r
k1
+ 1)r
k
nonisotopic tight contact structures with

convex boundary having two dividing curves of slope
p
q
. Consequently,
any nonzero boundary slope can be given as the boundary of a tight contact
structure on S
1
D
2
.
Remark 11.1.2. Recall that the dividing set on a convex torus in a tight
contact structure consists of 2n parallel circles of some common slope r.
The above theorem provides the classication of tight contact structures on
S
1
D
2
with convex boundary having dividing set of two components. For
results regarding the general (i.e., n > 1) case see [76] those results will
not be used in this volume.
Example 11.1.3. If the boundary slope is
1
n
for some n Z then by a self-
dieomorphism we can transform it to
1
1
= 1, hence k = 0 and r
0
= 1,
consequently (up to isotopy) there is a unique tight contact structure on
S
1
D
2
with boundary slope
1
n
and two dividing curves, see Theorem 5.1.30.
Now we turn to the proof of the lower bound of tight structures on the
solid torus with xed boundary condition. As we already saw, L(p, q) can
be given as
p
q
-surgery on the unknot K S
3
and this is equivalent to
attaching 4-dimensional 2-handles to D
4
along a chain of (k + 1) unknots
with framings r
0
, . . . , r
k
, cf. Exercise 2.2.7(c). In fact, the unknots can be
put in Legendrian position, and since r
i
< 1, by adding zig-zags we can
1
D
2
181
arrange r
i
= tb(K
i
) 1 to hold. Note that there is a certain freedom in
adding the zig-zags to the Legendrian unknot shown by Figure 11.1: in
total the zig-zags can be positioned in
(r
0
+ 1) (r
k
+ 1)
dierent ways.
All these choices produce dieomorphic Stein domains with some induced
contact structures on the boundary. It is not hard to determine the spin
c
structures induced by these contact structures, and a direct computation
easily shows that the structures are not isotopic. Recall that a xed diagram
induces a Stein structure on the underlying smooth 4-manifold X with
complex structure J satisfying
(11.1.1)
c
1
(X, J), [
i
]
_
= rot(K
i
)
with [
i
] H
2
(X; Z) denoting the homology element corresponding to the
knot K
i
, cf. Chapter 2. Now c
1
(X, J) determines a spin
c
structure on X and
its restriction to X is the spin
c
structure induced by the contact structure
of the surgery diagram.
Exercise 11.1.4. Show that the contact structures given by the above
Legendrian surgery diagrams on L(p, q) all have dierent spin
c
structures.
There is another way, involving much deeper theory, to distinguish these
structures. According to Proposition 8.2.4 (or Equation 11.1.1) the c
1
-
invariants of these Stein domains are all dierent, hence Theorem 11.1.5 of
Lisca and Matic applies:
1
3
1 1
Figure 11.1. Adjusting contact framing by stabilization
Theorem 11.1.5 (LiscaMatic, [97]). Suppose that J
1
, J
2
are two Stein
structures on a xed smooth 4-manifold X and
1
,
2
are the induced contact
structures on X. If c
1
(X, J
1
) ,= c
1
(X, J
2
) then
1
and
2
are not isotopic.
Remark 11.1.6. The proof of this statement rests on the fact that a
Stein domain can be embedded into a minimal surface of general type, and
this 4-manifold has only two SeibergWitten basic classes. Two isotopic
contact structures with dierent c
1
-invariants would produce more basic
classes. An alternative proof was given by Kronheimer and Mrowka [86]
using Theorem 13.2.2 and SeibergWitten theory, and by Plamenevskaya
[146] using Heegaard Floer theory.
In conclusion, the
(r
0
+ 1) (r
k
+ 1)
tight contact structures on L(p, q)

are all distinct. In fact, the above mentioned relation between contact
structures on L(p, q) and on the solid torus with some xed boundary
condition, together with Theorem 11.1.1 nishes the classication of contact
structures on lens spaces:
Theorem 11.1.7 (Honda, [76]). Any tight contact structure on L(p, q) is
isotopic to one of the structures given as Stein boundaries above. Conse-
quently L(p, q) carries
(r
0
+1) (r
k
+1)
nonisotopic tight contact struc-

tures all are Stein llable.
Remark 11.1.8. In fact, for some specic contact structures on L(p, q)
all Stein llings can be described, see Section 12.3 and [96]. The contact
structures covered by the theorem of Lisca are the ones for which all the
zig-zags in the diagram are either on the left or on the right these two
structures are actually contactomorphic and universally tight.
The link description of the contact structures shows that all these structures
contain a Legendrian knot K, the Legendrian realization of the normal circle
to, say, the left-most surgery curve in the chain, such that L(p, q) int K
is a solid torus. We can assume that K is a convex torus with a two-
component dividing set, and by examining the gluing map we can easily see
that the slope of the dividing curves on this torus is
p
with pq
qp
= 1
when viewed from the complementary solid torus. From this equation we
get that
p
has continued fraction representation [r

0
, . . . , r
k1
, r
k
+ 1] if
p
q
= [r
0
, . . . , r
k
]. Since the neighborhood of K is standard, in this way
we found
(r
0
+ 1) (r
k
+ 1)
isotopy classes of tight contact structures

on S
1
D
2
with boundary slope
p
, giving the desired lower bound for

arbitrary
p
. Notice that this lower bound is equal to the number of tight

1
D
2
183
contact structures stated in Theorem 11.1.1 (where the statement was given
for
p
q
rather than for
p
), and is equal to 1 for slopes of the form

1
n
.
Now we turn to the derivation of the upper bound for the number of
tight structures on S
1
D
2
in general. In doing so we will follow the proof
of Theorem 5.1.30. To this end x a tight contact structure on S
1
D
2
with convex boundary and the xed boundary slope of the dividing set
(S
1
D
2
)
on it. Again, we only deal with the case when
(S
1
D
2
)
has two
components. Consider the meridional simple closed curve (S
1
D
2
)
which becomes homotopically trivial when viewed in S
1
D
2
. Put it into
Legendrian position, consider the spanning disk D S
1
D
2
with D =
and isotope this disk into convex position. Since is tight, the dividing set
D
on D contains no closed components, hence
D
is equal to a collection
of arcs with boundary on D = . From the fact that the contact planes
rotate in the same direction when travelling around it follows that the
intersection points
D
D and
(S
1
D
2
)
D follow each other in an
alternating manner, that is, for consecutive intersections x, y
D
D
there is a unique z
(S
1
D
2
)
D between x and y and vice versa,
cf. Figure 11.2. Since
(S
1
D
2
)
, and so
(S
1
D
2
)
D is given by the
boundary condition, the number [
D
D[ and so the number of arcs in
D
is also xed. Since there is an upper bound for the possible congurations
of the embedded arcs of
D
with these boundary conditions, this argument
provides an upper bound for the tight contact structures near D in terms
of the boundary slope
p
q
. Since by Eliashbergs Theorem there is a unique
(up to isotopy) tight contact structure on S
1
D
2
D = D
3
, the above
reasoning provides an upper bound for the number of tight structures on
S
1
D
2
with the given boundary condition encoded by the slope
p
q
of the
dividing set
(S
1
D
2
)
on the boundary. This upper bound is in general
far from being sharp. Isotoping the disk D in a xed contact structure
we might get dierent congurations for
D
, although the contact structure
has not been changed. Hondas method of manipulating the dividing sets
with bypasses yields an equivalence relation among possible congurations
of dividing sets on D and concludes in a sharp upper bound for the number
of tight structures on the solid torus, nishing the proof of Theorem 11.1.1.
Note that for p = 1 the above argument already gives 1 as an upper bound,
hence veries Theorem 11.1.1 in this simple case.
x y
z
T
2
D
Figure 11.2. Dividing sets

(S
1
D
2
)
and D
Remark 11.1.9. Throughout the argument above we assumed that the
boundary slope is dierent from zero. The reason is that there is no
tight contact structure on S
1
D
2
with boundary slope zero: in this case
_
pt. D
2
_
can be isotoped to be disjoint from the dividing curves of
the boundary, therefore pt.D
2
(after being isotoped to have Legendrian
boundary) provides an overtwisted disk.
11.2. Contact Dehn surgery 185
11.2. Contact Dehn surgery
Now we are in the position to dene a version of Dehn surgery on 3-manifolds
adapted to the contact category. The discussion presented here rests on the
work of Ding and Geiges [16, 17]. Suppose that K (Y, ) is a Legendrian
knot in a given contact 3-manifold. As we already saw, K comes with a
canonical framing, hence we can perform r-surgery on (Y, ) along K
the surgery coecient is measured with respect to the contact framing.
In order to see that the surgered manifold Y
r
(K) also admits a contact
structure, we have to describe the surgery procedure a little more carefully.
As the Legendrian neighborhood theorem shows, for some positive there
is a contact embedding f : (N
, ) (Y, ) with f(C) = K where

N
=
_
(, x, y) [ x
2
+y
2

_
S
1
R
2
,
= cos(2n) dx sin(2n) dy
and
C =
_
(, x, y) [ x = y = 0
_
,
see Example 5.1.23. Let
N
2
=
_
(, x, y) [ x
2
+y
2
2
_
S
1
R
2
.
Now we will cut out f(N
) Y and reglue N
2
by a dieomorphism
g: (N
2
int N
) (N
2
int N
)
which maps boundary to boundary and on N
2
int N

= T
2
I it maps
the meridian to p + q. Such a map obviously exists on T
2
, and this
can be trivially extended to N
2
int N
. Considering the contact structure
1
= (g
)
1
() on N
2
int N
we need the following

Proposition 11.2.1. For p ,= 0 the contact structure
1
extends to a tight
contact structure
on N
2
.
Proof. Using the identication given by
_
p p
q q
_
with pq
q = 1 to glue S
1
D
2
back in, we need to choose the slope on
the solid torus to match up with the old longitude, which is isotopic to the
dividing curve. Recall that the slope of the boundary of the neighborhood
of a Legendrian knot can be assumed to be equal to by choosing the
longitude given by the contact framing. The meridian of S
1
D
2
will map
to p + q. Computing the inverse of the above matrix, the inverse image
of the longitude turns out to be p
+ p, hence the slope of the tight

contact structure on S
1
D
2
we need should be equal to
p
p
. According to
Theorem 11.1.1 this boundary condition can be fullled by a tight contact
structure on S
1
D
2
once p ,= 0.
Now identifying w N
2
int N
with f
_
g(w)
_
Y f(N
) we glue N
2
to
Y f(N
) and get a manifold Y
with glued up contact structure
. From
the construction it is clear that
Y
= Y
p
q
(K).
The contact structure
depends on the choice of the extension of

1
to N
2
.
In general this extension is not unique, but as the classication given in
the previous section shows uniqueness holds for p = 1. (For p = 1 we can
choose p
= 0 and q
= 1, hence we have to understand tight structures on

S
1
D
2
with slope
1
0
= , which is equivalent to
1
1
= 1.) Notice also that
even if is tight, the resulting structure
might be overtwisted. In order

to have a well-dened construction one needs to check that the resulting
contact structure
is (up to isotopy) independent of all the choices made

throughout the above gluing process. This is the content of
Theorem 11.2.2 (DingGeiges, [17]). If the extension
1
in N
2
is xed then
the resulting contact structure
on Y
is uniquely dened up to isotopy.

In particular, if p = 1 then the contact structure
on Y
is specied up to
isotopy by the Legendrian knot K and q Z.
The construction above allows us to prove numerous classical results in
contact topology. For example
Corollary 11.2.3 (Martinet, [107]). Every 3-manifold admits a contact
structure.
Proof. Every closed 3-manifold can be given as rational surgery on a link
in S
3
. Put the link into Legendrian position in (S
3
,
st
) and recompute
the framing coecients with respect to the contact framing. The previous
procedure provides a contact structure on the desired 3-manifold. (By
adding zig-zags if necessary we can always avoid contact 0-framings.)
Remark 11.2.4. A rened version of this theorem will be discussed in
Proposition 11.3.15.
From the 4-dimensional point of view, integral surgeries are especially
important, since these correspond to 4-dimensional 2-handle attachments.
In this sense, contact (1)-surgery produces both a 4-manifold and a unique
contact structure on its boundary. Recall that for (1)-surgery the result-
ing cobordism admits a Stein structure as well. We have already considered
a surgery scheme producing (+1)-surgery (with respect to the contact fram-
ing) in Section 7.3. It is natural to ask which contact 3-manifold can be
presented as contact (1)-surgery along a Legendrian link in (S
3
,
st
).
Theorem 11.2.5 (DingGeiges, [17]). For any closed contact 3-manifold
(Y, ) there is a Legendrian link L = L
+
L
(S
3
,
st
) such that contact
surgery on L
with framings (1) relative to the contact framings provides

(Y, ).
In order to give a short proof of Theorem 11.2.5 we will rst sharpen our
observation of Lemma 7.3.1. Therefore suppose that (Y, ) is a given contact
manifold, L (Y, ) is a Legendrian knot and L
is its contact push-o.

Perform contact (1)-surgery on L and (+1)-surgery on L
, resulting in the
contact manifold (Y
).
Lemma 11.2.6 (The Cancellation Lemma, DingGeiges [16]). The contact
3-manifolds (Y, ) and (Y

,
) are contactomorphic. The contactomorphism

can be chosen to be the identity outside of a small tubular neighborhood of
the Legendrian knot L.
Proof. The complete proof of this useful lemma relies on the solution of
the following two exercises.
Exercises 11.2.7. (a) Computing contact OzsvathSzabo invariants verify
that the result of (+1)-surgery along the Legendrian unknot of Figure 4.2(a)
is tight. (Hint: For a possible solution see Lemma 14.4.10. A direct argu-
ment for the same statement is given in [18]. See also Proposition 11.3.4.)
(b) Show that the result of contact (+1)-surgery on the Legendrian unknot
and (1)-surgery on its Legendrian push-o gives a tight contact structure
on S
3
. (Hint: From (a) deduce that the result of the (+1)-surgery is Stein
llable, and conclude that is also Stein llable, hence tight.)
Returning to the proof of Lemma 11.2.6, the idea is as follows: consider a
neighborhood N of L containing L
. It is easy to see that the two surgeries do

not change the topological type (and the gluing map) of this neighborhood,
so we only need to see that after surgeries the contact structure on N is tight
this shows that the two surgeries amount to a contact -surgery along
L, verifying the statement. By the Legendrian neighborhood theorem this
tightness can be checked on a model case, for example if L is the Legendrian
unknot in (S
3
,
st
). In this case, however, Exercise 11.2.7 shows that the
result of (1)-surgery on L and (+1)-surgery on L
embeds into a tight

contact S
3
, hence it is tight, completing the proof.
Corollary 11.2.8. Suppose that L, L
are Legendrian knots in a surgery

diagram for (Y, ) such that L
is the contact push-o of L and there is

a neighborhood of L disjoint from the rest of the diagram and containing
L and L
only. If we do (1)-surgery on L and (+1)-surgery on L
then
the diagram given by the same link after deleting L and L
yields the same

contact 3-manifold (Y, ).
Now we can begin the proof of Theorem 11.2.5. We will prove this
theorem in two steps. First we reduce the problem to the case of an
overtwisted 3-sphere, and then in the next section we nish the proof by
explicit diagrams for those contact 3-manifolds.
Proof. (Reduction of Theorem 11.2.5 to an overtwisted S
3
.) Perform con-
tact (+1)-surgery on the Legendrian knot L (S
3
,
st
) pictured by Fig-
ure 11.3. It is not hard to see that the result is an overtwisted structure
1
+1
Figure 11.3. Overtwisted contact structure on S
3
on S
3
, see Exercise 11.2.10(a). Now consider (Y, ) and take the connected
sum with (S
3
,
1
). The result is an overtwisted (Y,
2
) which can be given
as contact (+1)-surgery along a copy of L in a Darboux chart on (Y, ).
By Theorem 2.2.5 the 3-manifold Y can be turned into S
3
by a topologi-
cal surgery along a link, and since the complement of each knot in (Y,
2
)
can be assumed to be overtwisted, this link can be isotoped to a Legen-
drian position with contact framing one less than the framing prescribed by
the topological surgery. In conclusion, a sequence of contact (+1)-surgeries
turns (Y, ) into (S
3
,
) with some contact structure
. Adding one more

copy of (S
3
,
1
) to the whole process, the resulting (S
3
,
) can be assumed
to be overtwisted. Now reversing the surgeries we get that contact (1)-
surgery on a Legendrian link in some overtwisted contact 3-sphere (S
3
,
)
yields (Y, ). In conclusion, once we have a surgery presentation for (S
3
,
),
we can combine it with the above argument to yield a proof for Theo-
rem 11.2.5. (Such diagrams will be given in Lemmas 11.3.10 and 11.3.11,
cf. Corollary 11.3.13.)
Remark 11.2.9. Combining the above proof with an argument of Etnyre
and Honda we can actually assume that the Legendrian link L (S
3
,
st
)
producing (Y, ) has only one component on which (+1)-surgery is per-
formed. Etnyre and Honda [46] noticed that for any contact 3-manifold
(Y, ) and overtwisted structure (N, ) there is a Legendrian link in (N, )
along which contact (1)-surgery provides (Y, ). Using this principle with
(N, ) = (S
3
,
1
) given by (+1)-surgery along the knot L of Figure 11.3 we
have the above sharpening of Theorem 11.2.5.
Exercises 11.2.10. (a) Show that the contact structure
1
we get by per-
forming (+1)-surgery on the Legendrian knot of Figure 11.3 is overtwisted.
(Hint: Consider the Legendrian knot L shown by Figure 11.4. Show that it
L
+1
Figure 11.4. Boundary of an overtwisted disk in the diagram
bounds a disk in the surgered manifold, and compare the contact framing
on L with the one induced by this disk, cf. [18].)
(b) Using the same idea, verify that contact (+1)-surgery on the stabiliza-
tion of a Legendrian knot results in an overtwisted contact structure.
The original proof of Ding and Geiges for Theorem 11.2.5 followed slightly
dierent lines. In [17] they worked out a way for turning contact rational
surgeries into contact (1)-surgeries. Since this method is very useful in
applications, below we describe the algorithm for the proof the reader
is advised to turn to [17, 18]. Let us rst assume that we want to perform
contact r-surgery on the Legendrian knot L with r < 0. In this case
the surgery can be replaced by a sequence of contact (1)-surgeries along
Legendrian knots associated to L as follows: suppose that r =
p
q
and the
continued fraction coecients of
p
q
are equal to [r
0
+ 1, r
1
, . . . , r
k
], with
r
i
2 (i = 0, . . . , k). Consider a Legendrian push-o of L, add [r
0
+ 2[
zig-zags to it and get K
0
. Push this knot o along the contact framing and
add [r
1
+ 2[ zig-zags to it to get K
1
. Do contact (1)-surgery on K
0
and
repeat the process with K
1
. After (k + 1) steps we end up with a diagram
involving only contact (1)-surgeries. According to [17, 18] the result of the
sequence of (1)-surgeries is the same as the result of the original r-surgery.
Remark 11.2.11. Recall that for generic r, contact r-surgery is not unique:
there is a nite set of tight structures on S
1
D
2
with the correct boundary
slope. This non-uniqueness is present in the sequence of (1)-surgeries as
well: we have a freedom in adding the zig-zags in each step either on the
right or on the left. It is not very hard to see that there are equally many
choices in both constructions.
The next proposition will guide us how to turn contact r-surgery with r > 0
into a sequence of contact (1)-surgeries.
Proposition 11.2.12 (DingGeiges, [17]). Fixt r =
p
q
> 0 and an integer
k > 0. Then contact r-surgery on the Legendrian knot K is the same as
contact
1
k
-surgery on K followed by contact
p
qkp
-surgery on the Legendrian
push-o K
of K.
By choosing k > 0 large enough, the above proposition provides a way to
reduce a contact r-surgery (with r > 0) to a
1
k
- and a negative r
-surgery.
This latter one can be turned into a sequence of (1)-surgeries, hence the
algorithm is complete once we know how to turn contact
1
k
-surgery into
(1)-surgeries.
Lemma 11.2.13 (DingGeiges, [17]). Let K
1
, . . . , K
k
denote k Legendrian
push-os of the Legendrian knot K. Contact
1
k
-surgery on K is then
isotopic to performing contact (+1)-surgeries on the k Legendrian knots
K
1
, . . . , K
k
.
Exercises 11.2.14. (a) Verify that the above algorithm is correct on the
topological level, that is, the algorithm provides a surgery presentation of a
3-manifold dieomorphic to the result of the given r-surgery.
11.3. Invariants of contact structures given by surgery diagrams 191
(b) Notice that in applying Proposition 11.2.12 the choice of k N is not
unique. Show that after applying the Cancellation Lemma 11.2.6 suciently
many times the resulting diagram will be independent of the choice of k.
(c) Show that for any contact 3-manifold (Y, ) there is (Y
) and a Stein
cobordism W between Y and Y
such that H
1
(Y

; Z) = 0. (Hint: Start with
a contact surgery diagram L of (Y, ) and for every knot L
i
in L consider a
Legendrian knot K
i
linking L
i
once, not linking the other knots in L, and
having tb(K
i
) = 1. Adding Weinstein handles along K
i
we get W; check
that the resulting 3-manifold Y
after the handle attachment is an integral

homology sphere. For more details see [159].)
(d) Find an open book decomposition of #
k
(S
1
S
2
) compatible with the
standard contact structure.
11.3. Invariants of contact structures given by surgery
diagrams
In this section we show how one can read o homotopic data of a contact
structure given by a contact surgery diagram. Suppose that (Y, ) is the re-
sult of contact (1)-surgery on the Legendrian link L = L
+
L
(S
3
,
st
).
Recall that integral surgery can also be regarded as (4-dimensional) 2-handle
attachment to D
4
, hence the diagram represents a compact 4-manifold X
with X = Y . There is, however, an additional structure on X. It is
fairly easy to see that the surgery diagram for (Y, ) gives an achiral Lef-
schetz bration on the 4-manifold X: just repeat the algorithm of Akbulut
and the rst author outlined in Section 10.2. (Also take the renement of
Plamenevskaya [146] given in Proposition 10.2.4 into account.) Recall that
an achiral Lefschetz bration on X naturally provides an open book de-
composition ob
L
on X = Y . Next we would like to show that (as the
result of contact (1)-surgeries) on Y is compatible with this open book
decomposition. Notice that this step will complete a portion of the proof
of Girouxs Theorem 9.2.11 about relating open book decompositions and
contact structures. Let
L
denote the contact structure (unique up to iso-
topy by Part(a) of Theorem 9.2.11, see also Proposition 9.2.7) compatible
with the open book decomposition ob
L
. Our main result is now
Theorem 11.3.1. The contact structures and
L
on Y are isotopic, hence
is compatible with the open book decomposition ob
L
dened above.
In the light of Theorem 5.2.1 of Torisu we would like to show that both
and
L
admit the properties listed under (i) and (ii) of that theorem. This
is obviously satised (and explicitly stated in [168]) for
L
, hence we only
need to verify them for .
Lemma 11.3.2 ([160]). The restrictions [
U
i
to the handlebodies U
i
(i = 1, 2) of the Heegaard decomposition induced by the open book de-
composition ob
L
are tight. The dividing set of the convex surface Y
with respect to is isotopic to the binding of the open book decomposition.
Proof. Consider the open book decomposition found on S
3
induced by
the Lefschetz bration D
4
D
2
in the course of the algorithm presented in
Section 10.2. Recall that this Lefschetz bration is given by the factorization
of the monodromy of the (p, q) torus knot dened by the knot in square
bridge position. Since the monodromy of this open book decomposition is
the product of right-handed Dehn twists only, the corresponding contact
structure is isotopic to
st
. In addition, this open book decomposition
induces a Heegaard decomposition of S
3
, and the contact handlebodies of
this Heegaard decomposition since they are contained by the tight S
3
are tight. The Heegaard decomposition S
3
= V
1
V
2
can be chosen in
such a way that L is contained in V
1
. Therefore (ii) of the assumptions of
Theorem 5.2.1 obviously holds, since surgery along L will not change the
convex surface V
1
= V
2
, and the binding of the open book decomposition
remains unchanged. We only need to check (i), that is, that the contact
structures [
U
i
are tight for i = 1, 2. By our choice U
2
= V
2
and [
U
2
=
st
[
V
2
,
hence we only need to deal with [
U
1
. Consider Legendrian push-os for all
Legendrian knots in L
+
in such a way that these push-os are in V
2
. This
can be done, since the contact framings of the knots in L coincide with the
page framing they inherit from the open book decomposition. Therefore
a contact push-o can be assumed to lie on a page, and this page can be
chosen to be in V
2
. Doing the prescribed surgeries along the knots of L
and contact (1)-surgeries on these push-os we get a contact 3-manifold
(Y
) which contains [
U
1
. It is easy to see that (Y
) is tight: by the
Cancellation Lemma 11.2.6 it can be given by doing (1)-surgery along
L
(S
3
,
st
), therefore (Y
) is Stein llable, hence tight. Since [

U
1
is
contained by a tight 3-manifold, it is tight, concluding the proof.
Proof (of Theorem 11.3.1). By [168] and Lemma 11.3.2 both
L
and
satisfy conditions (i) and (ii) of Theorem 5.2.1, hence the theorem implies
that and
L
are isotopic.
Remark 11.3.3. A similar theorem was proved by Gay [53] in the case
when no (+1)-surgeries are present in the picture.
Now Theorem 11.3.1 allows us to nd open book decompositions for all
contact structures given by contact (1)-surgery diagrams. Notice also that
we just proved that every 3-manifold admits an open book decomposition:
presenting Y as the boundary of a 4-dimensional handlebody with a unique
0-handle and some 2-handles, we get a contact surgery diagram of Y with
some contact structure. Turn this diagram into (1)-surgeries and apply the
above theorem to nd an open book decomposition on Y . (This operation
will change the 4-dimensional handlebody, though.) As an easy application
we show that
Proposition 11.3.4. Contact (+1)-surgery on the Legendrian unknot pro-
vides a tight structure on S
1
S
2
.
Proof. After performing the algorithm given in Section 10.2 we get an
achiral Lefschetz bration X D
2
with ber dieomorphic to the Seifert
surface of the (2, 2) torus knot, i.e., the annulus A. The 4-manifold X
is built from the Lefschetz bration D
4
D
2
by attaching a 2-handle
along the central circle C of this annulus, see Figure 11.5. Since the
C
A
Figure 11.5. The vanishing cycle C on the annulus A
monodromy of the (2, 2) torus knot is equal to the right-handed Dehn twist
t
C
along C, the total monodromy of the induced open book decomposition
on X = S
1
S
2
is equal to t
C
t
1
C
= 1. The reason for the negative
exponent on the second Dehn twist is that we need to do (+1)-surgery,
corresponding to a left-handed Dehn twist in the monodromy. Therefore,
according to Theorem 11.3.1 the contact structure we get by (+1)-surgery
on the Legendrian unknot is compatible with the open book decomposition
dened by the identity element 1
A
. Corollary 9.2.15 now implies that it
is Stein llable, hence the proof is complete. In addition, by the classication
of tight contact structures on S
1
S
2
the above argument also shows that
the surgery described above is the same as the boundary of the Stein 1-
handle.
Exercises 11.3.5. (a) Verify the Cancellation Lemma 11.2.6 using open
book decompositions. (Hint: Notice that the curve L
can be given as the

push-o of L on a page. Then the Dehn twists corresponding to L and L
cancel in the monodromy, giving the result.)

(b) Prove Proposition 11.3.4 using contact OzsvathSzabo invariants.
(Hint: Use Lemma 14.4.10.) Prove tightness for the contact structure given
by (+1)-surgery along the k-component Legendrian unlink.
(c) Show that any solid genus-g handlebody admits a contact structure
which can be embedded into a Stein llable structure on a closed 3-manifold.
Our next application concerns computability of homotopic invariants of
contact structures on a 3-manifold Y . Recall form Chapter 6 (cf. also [65])
that two oriented 2-plane elds
1
and
2
on Y are homotopic if and only
if their induced spin
c
structures t
i
and 3-dimensional invariants d
3
(
i
) are
equal. If c
1
(t
) is nontorsion then d
3
() does not admit a Q-lift, but for
c
1
(t
) torsion, this latter invariant can be lifted to Q and can be computed

as
1
4
_
c
2
1
(X
i
, J
i
) 3(X
i
) 2(X
i
)
_
where (X
i
, J
i
) are almost-complex 4-manifolds with X
i
= Y such that the
oriented 2-plane elds of complex tangencies of J
i
along X
i
are homotopic
to
i
. The surgery picture together with Theorem 11.3.1 easily provides such
a 4-manifold X: Suppose that (Y, ) is given by (1)-surgery on L = L
+
(S
3
,
st
), and let X
1
denote the 4-manifold dened by the diagram.
As explained in Section 10.2, X
1
admits an achiral Lefschetz bration
structure. Consider the oriented 2-plane eld of tangents of bers away
from the set C of critical points of the bration. By taking the orthogonal
complement for some metric, this oriented 2-plane eld provides an almost-
complex structure on X
1
C: dene J as counterclockwise 90
rotation
on these planes. This almost-complex structure obviously extends through
those points of C which admit orientation preserving complex charts
just use the local model. At points of C with oppositely oriented coordinate
charts (corresponding to contact (+1)-surgeries) the two branches of the
oriented singular ber provide an orientation for X incompatible with the
one originally xed. The obstruction for extending J through such points
can be computed using a local model, as explained in [66, Lemma 8.4.12] or
in [160]. In conclusion, for these points of C we need to take the connected
sum of X
1
with CP
2
with its standard complex structure for extending the
almost-complex structure dened on X
1
C. Consequently X = X
1
#qCP
2
with the extended almost-complex structure is a good choice of (X, J) for
the given contact structure (Y, ). Here q denotes the number of components
in L
+
. By repeating the proof of [65, Proposition 2.3] verbatim (see also
Proposition 8.2.4) we get
Theorem 11.3.6 (Gompf, [65]). The rst Chern class c
1
(X, J) H
2
(X; Z)
of the resulting almost-complex structure evaluates on the 2-homology de-
ned by the surgery curve K as its rotation number rot(K).
Since is isotopic to the oriented 2-plane eld of complex tangencies along
Y = X, the cohomology class c
1
() is equal to the restriction of the above
c
1
(X, J) to X. The class c
1
(X, J) is specied by Theorem 11.3.6, and the
description of H
1
(Y ; Z) in terms of a surgery diagram then provides c
1
().
Note that here X is simply connected, hence the spin
c
structure s
J
induced
by J is specied by c
1
(X, J). In this way the induced spin
c
structure t
is
given as s
J
[
X
. If c
1
() H
2
(Y ; Z) is torsion, then for appropriate n N the
class PD
_
nc
1
(X, J)
_
H
2
(X, X; Z) is the image of a class H
2
(X; Z),
hence c
2
1
(X, J) can be computed as
1
n
2
2
Q as discussed in Section 6.3.
Notice also that both c
1
(X
1
C, J) and the induced spin
c
structure s
J
extend uniquely through the points of C, hence for practical purposes we
can work with this extended cohomology class c H
2
(X
1
; Z), although it is
not the rst Chern class of any almost-complex structure. When computing
d
3
(), this fact results a correction term in the formula. In conclusion, for
with torsion induced spin
c
structure t
all terms in the formula for d

3
()
can be easily computed once is given by a surgery diagram. This leads to
Theorem 11.3.7 ([160]). Suppose that the contact 3-manifold (Y, ) is
given by contact (1)-surgery along the link L = L
+
L
(S
3
,
st
).
Let X
1
denote the 4-manifold dened by the diagram and suppose that
c H
2
(X
1
; Z) is given by c
_
[
K
]
_
= rot(K) on [
K
] H
2
(X
1
; Z), where
K
is the surface corresponding to the surgery curve K L. If the
restriction c[
X
1
to the boundary is torsion and L
+
has q components then
d
3
() =
1
4
_
c
2
3(X
1
) 2(X
1
)
_
+q.
Proof. Since (X
1
) =
_
X
1
x
1
, . . . , x
k
_
and (X
1
) = q +
_
X
1

x
1
, . . . , x
q
_
for the critical points x
1
, . . . , x
q
of the achiral Lefschetz
bration X
1
D
2
which lie on incorrectly oriented coordinate charts, the
formula easily follows.
Next we show an alternative way for verifying the above formula, cf. [18].
This method works only for knots with nonzero ThurstonBennequin in-
variants, but conceptually it is simpler for example, it makes no use of
the achiral Lefschetz bration or the open book decomposition provided
by the surgery diagram. As shown in Chapter 8, the complex structure of
D
4
(inherited from C
2
) extends to all the 2-handles attached with contact
framing 1. We do not have such an extension for the contact (+1)-framed
2-handles, but there is no obstruction to nding an appropriate almost-
complex structure on these handles away from a point. In conclusion, we
have an almost-complex structure on X
1
x
1
, . . . , x
q
where q is the car-
dinality of L
+
the knots on which we do contact (+1)-surgery. Since a
spin
c
structure (like a 2-cohomology element) extends through a point in a
4-manifold, we have a spin
c
structure s on X
1
extending the spin
c
structure
t
Spin
c
(Y ) induced by . We want to determine c
1
(s) on the homo-
logy classes given by the Legendrian knots in L. For this computation, x
L (S
3
,
st
), perform contact (+1)-surgery on it and consider the resulting
4-manifold X
L
with spin
c
structure s
L
Spin
c
(X). Let k denote the value
of c
1
(s
L
) on a generator for H
2
(X
L
; Z). (To be precise, we need to x an
orientation for L, which provides a canonical generator for H
2
(X
L
; Z)
= Z.)
Dene u as the obstruction to extending the almost-complex structure from
X
L
pt. to X
L
, i.e., the 3-dimensional invariant of the oriented 2-plane
eld induced on the boundary S
3
of the neighborhood of the point is u.
Proposition 11.3.8 ([18, 99]). If tb(L) ,= 0 then k = rot(L) and u =
1
2
.
Proof. Consider 2n Legendrian push-os of L and call them L
1
, . . . , L
n
and L
1
, . . . , L
n
. Do contact (1)-surgeries along L
i
and (+1)-surgeries
along L
i
. According to the Cancellation Lemma 11.2.6 the result is (S
3
,
st
)
again. On the other hand, simple homological computation shows that the
3-dimensional invariant of the result of the surgery is
1
4
(n
_
k
2
rot(L)
_
n
2
tb(L)
_
k rot(L)
_
2
2) +n
_
u
1
2
_
1
2
.
Since d
3
(S
3
,
st
) =
1
2
, the above expression implies u =
1
2
and k = rot(L)
provided tb(L) ,= 0.
Remark 11.3.9. In fact, we need to use the above expression for n = 1
and n = 2 only to draw that conclusion. Note that since u can be easily
shown to be independent of L, for tb(L) = 0 the above argument gives
k = rot(L); a more detailed study of the almost-complex structure on
X
L
pt. actually proves that k = rot(L) in this case as well, see also [18].
In most cases, however, the proof for the tb(L) ,= 0 case is sucient.
The formula of Theorem 11.3.7 above gives us a way to distinguish
contact structures given by surgery diagrams on a xed 3-manifold. For
example, let L be n unlinked copies of the knot given by Figure 11.3, and
take (Y,
n
) to be (+1)-surgery on L. Simple computation veries
Lemma 11.3.10. Y = S
3
and d
3
(
n
) = n
1
2
.
Proof. By turning the contact framing coecients to Seifert framings, we
see that Y is given by (1)-surgery on the n-component unlink, hence we can
blow all surgery curves down, showing that Y = S
3
. The corresponding 4-
manifold X
1
therefore has (X
1
) = n, (X
1
) = n+1 and since L = L
+
, we
have q = n. Easy computation shows that c
2
= n; by plugging these values
into the formula of Theorem 11.3.7 the proof of the lemma is complete.
A similar quick calculation shows
Lemma 11.3.11. n geometrically disjoint copies of the link of Figure 11.6
provide a sequence of contact structures
n
on S
3
with d
3
(
n
) = n
1
2
.
Proof. Figure 11.6 shows that the manifold we get after the surgery is
dieomorphic to S
3
. Application of the formula for the 3-dimensional
invariant d
3
now implies the result.
Since S
3
admits a unique tight contact structure
st
and d
3
(
st
) =
1
2
, the
contact structures
n
for n Z0 encountered above are all overtwisted.
Exercises 11.3.12. (a) By nding the overtwisted disks show directly that
the contact structures (S
3
,
n
)
_
n Z0
_
of the above two Lemmas are
overtwisted.
(b) Find an overtwisted contact structure
0
on S
3
homotopic to
st
. (Hint:
Take the connected sum of
1
and
1
.)
(c) Show that the contact structures on L(3, 1) given by Figures 11.7(a)
and (b) are not isotopic but contactomorphic. (Hint: Compute the spin
c
structures induced by the contact structures. Verify that reection induces
a contactomorphism.)
(d) Find open books compatible with the contact structures given by Fig-
ures 11.7(a) and (b).
+1
1
smoothly
5
1
1
1
Figure 11.6. The contact 3-manifold (S
3
, 1)
1 1
(a) (b)
Figure 11.7. Contactomorphic, nonisotopic contact structures
The map (S
3
, ) d
3
() +
1
2
gives a bijection between the space of over-
twisted contact structures and Z. To see this we only need to verify that
for any oriented 2-plane eld on S
3
the quantity d
3
() +
1
2
is an integer.
Recall that
d
3
() =
1
4
_
c
2
1
(X, J) (X)
_

1
2
_
(X) +(X)
_
for an appropriate simply connected almost-complex 4-manifold (X, J).
The expression
1
4
_
c
2
1
(X, J) (X)
_
is an even integer since c
1
(X, J) is
a characteristic vector, and
1
2
_
(X) + (X)
_
= b
+
2
(X)
1
2
. Therefore we
have
Corollary 11.3.13. The above lemmas together with Exercise 11.3.12(b)
show surgery diagrams for all overtwisted contact structures
n
(n Z) on
the 3-sphere.
Notice that this corollary concludes the proof of Theorem 11.2.5.
Exercise 11.3.14. Consider a Legendrian knot L (S
3
,
st
) and its Leg-
endrian push-o L
. Stabilize L
twice to get L
1
and perform contact (+1)-
surgery on L and L
1
. Determine the resulting 3-manifold Y and compute
d
3
() for the resulting contact structure . (Hint: See [19].)
Following similar lines, in fact, we can produce surgery diagrams for all
overtwisted contact structures on any 3-manifold presented by a surgery
diagram. This presentation (given in [18]) provides a new proof of a classical
result of Lutz and Martinet:
Proposition 11.3.15 (LutzMartinet, [105]; cf. also [18]). For a given 3-
manifold Y and oriented 2-plane eld (Y ) there is a contact structure
homotopic to . The contact structure can be chosen to be overtwisted.
Exercises 11.3.16. (a) Let L
0
(S
3
,
st
) be the Legendrian unknot and L
1
another Legendrian unknot linking it k times (k Z). Add two zig-zags to
the Legendrian push-o L
2
of L
1
and get L
2
. Perform contact (+1)-surgery
on L
0
, L
1
and L
2
. Prove that the resulting manifold is dieomorphic to
S
1
S
2
. Determine the spin
c
structure of the resulting contact structure .
(b) Using Exercise 11.3.14 and the above result verify Proposition 11.3.15
for S
1
S
2
.
(c) Prove Proposition 11.3.15 in general. (Hint: See [18].)
Recall that according to Eliashbergs result, isotopy classes of overtwisted
contact structures and homotopy classes of oriented 2-plane elds are in
one-to-one correspondence. Therefore the solution of Exercise 11.3.16(c)
provides a surgery diagram for any overtwisted contact structure on a closed
3-manifold.
1
.....
.....
+1
1 1 1
m
1
Figure 11.8. Contact structure
(2m+1)
on S
3
Exercise 11.3.17. Show that the contact surgery diagram depicted in
Figure 11.8 gives a contact structure on S
3
with d
3
=
1
2
2(m + 1),
where m 0 is the number of unknots in the gure with vanishing rotation
number. (Hint: Compute d
3
and use the classication of overtwisted contact
structures.) Notice that this surgery diagram represents some overtwisted
contact structures on S
3
using unknotted surgery curves and only one (+1)
surgery curve. This example also illustrates (Stein) cobordisms between
various contact structures.
12. Fillings of contact 3-manifolds
This chapter is devoted to the study of llability properties of contact 3-
manifolds. After having the necessary denitions we will see dierent types
of llings, and give a family of tight, nonllable contact structures. The
construction of these latter examples utilizes contact surgery, while tightness
is proved by computing contact OzsvathSzabo invariants (see Chapter 14).
In the last section we will concentrate on topological restrictions a contact
3-manifold imposes on its Stein llings.
12.1. Fillings
Denition 12.1.1. A given contact 3-manifold (Y, ) is weakly symplecti-
cally llable (or llable) if there is a compact symplectic manifold (W, )
such that W = Y (as oriented manifolds) and with this identication [
does not vanish. In this case we say that (W, ) is a symplectic lling.
(W is oriented by the volume form , while the orientation of Y is
the one compatible with .) (Y, ) is strongly symplectically llable if it is
the -convex boundary of a compact symplectic manifold (W, ). In other
words, is exact near the boundary and its primitive (i.e., a 1-form with
d = ) can be chosen in such a way that ker
_
[W
_
= . Yet another for-
mulation of strong lling is to require a transverse, symplectically dilating
vector eld for the boundary (dened near X) pointing outwards. (Y, )
is holomorphically llable if there is a compact complex surface (X, J) such
that the contact structure on X given by the complex tangencies is con-
tactomorphic to (Y, ). Finally, (Y, ) is Stein llable if it is the J-convex
boundary of a Stein surface.
202 12. Fillings of contact 3-manifolds
Remarks 12.1.2. (a) Without imposing the compactness condition on
W, the above denition of (weak or strong) symplectic llability would be
satised by all closed contact 3-manifold (Y, ): Consider simply Y (0, 1]
equipped with the symplectic structure it inherits from the symplectization
of (Y, ).
(b) According to a result of Bogomolov, the complex structure on a holo-
morphic lling can be deformed such that (X, J
) becomes the blow-up of a

Stein lling. Therefore the two last notions of llability in Denition 12.1.1
are the same.
(c) Notice that holomorphic/Stein llability implies strong llability, which
in turn implies weak llability. For a related discussion on various llability
notions see [30].
Notice that a symplectic 4-manifold (W, ) is by denition a strong sym-
plectic lling if its boundary W is -convex. Recall that by results of
Chapter 7 we can attach Weinstein handles to a strong symplectic lling
along Legendrian knots in a way that the symplectic structure extends to
the handle and the new symplectic 4-manifold strongly lls its boundary.
In this gluing process, however, the symplectically dilating vector eld is
used only in a neighborhood of the attaching circle. It turns out that if
L (Y, ) is Legendrian and (W, ) is a weak lling of (Y, ) then there is
always a symplectically dilating vector eld near L, implying
Theorem 12.1.3 ([16]). Suppose that (Y
) is given by contact (1)-

surgery along L (Y, ). If (Y, ) is weakly llable then so is (Y

,
).
It is known that there are weakly llable contact structures which are not
strongly llable: for example, the contact tori (T
3
,
n
) with n 2 all have
this property [29]. (For an even larger collection of such contact 3-manifolds
see [16].) It is still unknown whether strong llability implies Stein llability.
Of course one can modify a Stein lling in such a way that it does not admit
a Stein structure anymore, but such an operation does not aect llability
properties of the boundary contact 3-manifold.
Example 12.1.4. The Legendrian surgery diagram of Figure 12.1 gives a
strong (in fact, Stein lling) of the boundary of the nucleus N
n
with the
inherited contact structure.
Suppose that (W, ) is a weak lling of (Y, ). It is obvious that if is not
exact near W = Y then (W, ) is not a strong lling. The exactness of ,
however, enables us to modify near the boundary in such a way that it
12.1. Fillings 203
.
.
.
.
.
.
.
.
.
.
zigzags
nk
zigzags
k
1
1
Figure 12.1. Stein structure on the nucleus Nn
becomes a strong lling, see [30]. In the special case of rational homology
spheres therefore we have
Theorem 12.1.5 (OhtaOno, [127]). Suppose that b
1
(Y ) = 0. The sym-
plectic structure on a weak symplectic lling W of any contact structure
on Y can be extended to W Y [0, 1] to a strong lling of (Y, ). In
conclusion, a contact structure on a rational homology sphere Y is weakly
llable if and only if it is strongly llable.
According to a recent result of Eliashberg [30] a weak lling can be
symplectically embedded into a closed symplectic 4-manifold. This theorem
turned out to be of central importance in recent studies of contact invariants,
see [87, 143]. Here we prove this theorem in two steps.
Theorem 12.1.6. If (W, ) is a strong lling of (Y, ) then W can be
embedded into a closed symplectic 4-manifold.
Proof. Consider a surgery presentation L = L
+
L
(S
3
,
st
) of (Y, ),
and let K denote the Legendrian link we get by considering Legendrian push-
os of the knots of L
+
. Attaching Weinstein handles to (W, ) along the
knots of K we get a strong lling (W
) of a contact 3-manifold (Y

,
).
Notice that by the Cancellation Lemma 11.2.6 this latter contact manifold
can be given as Legendrian surgery along L
, consequently it is Stein llable

(although (W
) might not be a Stein lling of it). Consider a Stein lling

(X, J) of (Y
) and embed this lling into a closed symplectic 4-manifold

(Z,
Z
) as explained in Section 10.3. Performing symplectic cut-and-paste
(as in Theorem 7.1.9) along Y
Z we get a symplectic structure on the

closed 4-manifold U = (Z int X)
Y
W
. Since (W, ) is a symplectic

submanifold of (W
), this provides a symplectic embedding of (W, ) into

the closed symplectic 4-manifold U. Notice that by adding more Weinstein
handles we can make sure that b
+
2
(W
W) and so b
+
2
(U) is at least 2.
Surprisingly enough, from this point the embeddability of a weak symplectic
lling follows by a trivial argument.
Theorem 12.1.7 (Eliashberg, [30, 42]). If (W, ) is a weak symplectic
lling of (Y, ) then (W, ) embeds symplectically into a closed symplectic
4-manifold (U,
U
).
Proof. According to Exercise 11.2.14(c) the weak symplectic lling embeds
rst into a weak symplectic lling (W
) such that the boundary Y
= W
is an integral homology sphere. Now Theorem 12.1.5 provides a way to

modify
near W
to achieve that the new symplectic form

1
provides
a strong symplectic lling of (Y

,
). The application of Theorem 12.1.6

now provides a symplectic embedding of (W
,
1
) into a closed symplectic
4-manifold, and since (W, ) is a symplectic submanifold of (W
,
1
), the
proof is complete.
It is still a question of central importance in contact topology whether a
given contact structure is llable or not (in any of the above sense) and
which 3-manifolds support llable contact structures.
The previous chapters provided a very powerful topological tool for con-
structing Stein manifolds: attach 2-handles to
n
S
1
D
3
along a Leg-
endrian link with framing 1 relative to the contact framing. (Here
(
n
S
1
D
3
) = #
n
S
1
S
2
is equipped with its unique tight contact struc-
ture.) In fact, every Stein domain can be given in this way. This approach
has been systematically studied by Gompf in [65]; he showed, for example
12.1. Fillings 205
Theorem 12.1.8 (Gompf, [65]). Every Seifert bered 3-manifold M =
M(g, n; r
1
, . . . , r
k
) with one of its orientations admits a Stein llable contact
structure. If g 1 then M admits Stein llable contact structures with
either of its orientations.
According to a result of Eliashberg, Stein llability needs to be determined
only for prime 3-manifolds:
Proposition 12.1.9. The connected sum (Y
1
,
1
)#(Y
2
,
2
) is Stein llable
if and only if both (Y
i
,
i
) are Stein llable.
According to a theorem of Eliashberg and Gromov, a llable contact
structure (in any of the above sense) is tight.
Theorem 12.1.10 (EliashbergGromov). A weakly symplectically llable
contact 3-manifold (Y, ) is tight.
Proof (sketch). Let (W, ) be a symplectic lling of (Y, ) and suppose
that (Y, ) contains an overtwisted disk. Choose a disk D with Legendrian
boundary and the property that tb(D) = 2, and attach a Weinstein handle
along D to the weak lling (W, ). The resulting 4-manifold W
will be a
weak symplectic lling of the surgered contact 3-manifold (Y
) containing
a sphere with self-intersection (+1). Now embed (W
) into a closed
symplectic 4-manifold U with b
+
2
(U) > 1. The adjunction inequality of
Theorem 13.3.3 for the sphere of positive self-intersection now provides the
desired contradiction.
Remark 12.1.11. The rst proof of the above theorem is due to Eliashberg
and Gromov [32], and used completely dierent ideas and methods.
The above result might lead one to expect that all tight contact struc-
tures are llable in some sense. Until recently, however, it was very hard
to nd counterexample to this expectation, since the only tool for proving
tightness of a given (Y, ) was to show that it is llable. The state traversal
method tightness, and this method led to the discovery of the rst tight
but not llable contact structures [44]. The introduction of OzsvathSzabo
invariants then gave a very eective way for examining tightness properties
of contact structures on closed manifolds, leading to a plethora of examples
of tight nonllable contact structures.
12.2. Nonfillable contact 3-manifolds
Not all contact structures are llable and there are examples of 3-manifolds
which do not admit any symplectic llings.
Theorem 12.2.1 (Lisca, [94]). The Poincare homology 3-sphere with its
natural orientation reversed admits no llable contact structure.
Proof. The Poincare homology sphere is dieomorphic to the Brieskorn
sphere (2, 3, 5), hence the oriented 3-manifold of the theorem is equal to
(2, 3, 5). Suppose W is a lling of (2, 3, 5), and embed it into a
closed symplectic manifold (as it is given in Theorem 10.3.1). The fact that
(2, 3, 5) admits a positive scalar curvature metric implies that b
+
2
(W) = 0,
cf. Proposition 13.1.7(5.). Now if E stands for the positive denite E
8
-
plumbing given by the plumbing graph of Figure 1.5 then W (E) is
a negative denite closed 4-manifold with nonstandard intersection form,
contradicting Donaldsons famous diagonalizability result [20]. Therefore
W cannot exist.
Exercise 12.2.2. Show that if
_
Z
n
, (E
n
)
_
(n = 6, 7, 8) is a sublattice of a
negative denite lattice (Z
k
, Q) then Q cannot be diagonal. (Hint: Notice
that E
6
is contained in all these lattices. For a solution see [95].)
A similar argument shows that the boundary of the positive denite E
6
-
and E
7
-plumbing cannot be the boundary of a Stein domain. Notice that
in the light of Proposition 12.1.9 we have many 3-manifolds which are
not boundaries of Stein domains just take connected sum with one of
the above mentioned nonllable manifolds. For example, the 3-manifold
(2, 3, 5)#
_
(2, 3, 5)
_
is not a Stein boundary with either orientation.
The result of Theorem 12.2.1 was not sucient for producing a tight, non-
llable contact structure, since by a result of Etnyre and Honda [45] the
oriented 3-manifold (2, 3, 5) actually does not support any tight struc-
ture are all.
Probably the simplest tight, nonllable contact 3-manifold (Y, ) is given
by the contact surgery diagram of Figure 1.6. Notice that as a smooth 3-
manifold Y is just (+2)-surgery on the right-handed trefoil (=(2, 3, 4)).
In the light of Exercise 12.2.2 the proof of Theorem 12.2.1 shows that
Y supports no llable contact structures, hence Figure 1.6 must dene a
nonllable structure. In the proof of tightness we will make use of the
contact OzsvathSzabo invariants. For an overview of these invariants and
12.2. Nonllable contact 3-manifolds 207
OzsvathSzabo homology see Appendix 14; here we will freely use the result
discussed there. Recall that

HF(Y ) denotes the OzsvathSzabo homology
group of the closed, oriented 3-manifold Y , while c(Y, )

HF(Y ) is the
contact invariant of (Y, ).
Proposition 12.2.3. The contact 3-manifold (Y, ) given by Figure 1.6 has
nonvanishing contact OzsvathSzabo invariants, hence is tight.
Proof. Since (Y, ) is dened as contact (+1)-surgery along a single knot,
according to Theorem 14.4.5 the contact invariant c(Y, ) can be given as
F
W
_
c(S
3
,
st
)
_
, where W is the cobordism of the handle attachment with
reversed orientation. Therefore injectivity of F
W
gives the nonvanishing
of the invariant. The cobordism W can be given by a single 2-handle
attachment along the left-handed trefoil knot with framing 2. Denote
the left-handed trefoil by T. Then the surgery exact triangle reads as
HF(S
3
)

HF
_
S
3
2
(T)
_
HF
_
S
3
1
(T)
_
F
W
Since

HF
_
S
3
n
(T)
_
=

HF
_
S
3
n
(T)
_
, the genus of T is 1 and S
3
5
(T) is a
lens space, Propostion 14.3.5 implies that dim

HF(S
3
n
(T)) = n, hence the
above triangle translates to
Z
2
Z
2
Z
2
Z
2
F
W
therefore exactness implies the injectivity of F
W
, concluding the proof.
Exercises 12.2.4. (a) Show that S
3
5
(T) is a lens space. (Hint: Use the
presentation of S
3
1
(T) as plumbing on the positive denite E
8
-diagram and
truncate its long leg, cf. also Exercise 2.3.5(f).)
(b) Using the result of the above proposition nd a tight contact structure
on the boundary of the positive denite E
6
-plumbing. (Hint: Take the dia-
gram of Figure 1.6 with surgery coecient (+1) on the right-handed trefoil,
consider the Legendrian push-o of it, add a zig-zag and perform contact
(1)-surgery on the resulting knot. Verify that the resulting manifold is the
boundary of the positive denite E
6
-plumbing using Kirby calculus.)
(c) Show that if (Y
K
,
K
) is given by contact (+1)-surgery on (Y, ) along
a Legendrian knot K and (Y, ) is not llable then (Y
K
,
K
) is not llable
either. (Hint: Remember that contact (+1)-surgery along a Legendrian
knot can be cancelled by contact (1)-surgery along its Legendrian push-
o, so (Y, ) can be given as (1)-surgery along some Legendrian knot in
(Y
K
,
K
), cf. Theorem 12.1.3.)
This observation leads us to a family of nonllable contact structures. Con-
sider k Legendrian push-os of the right-handed trefoil and perform con-
tact (+1)-surgery on each component, resulting in the contact 3-manifold
(Y
k
,
k
). According to the above exercise these structures are all nonllable.
Exercise 12.2.5. Show that Y
k
can be given by the surgery diagram of
Figure 12.2. Conclude that
H
1
(Y
k
; Z)
= k + 1. (Hint: Convert the

copies
k
.
.
.
.
.
.
k
1
+1
+1
.
.
.
Figure 12.2. Tight, nonllable contact 3-manifold (Y
k
,
k
)
Legendrian surgery diagram into a smooth diagram and slide the trefoils
over each other.)
Proposition 12.2.6 ([101]). The contact OzsvathSzabo invariant of
(Y
k
,
k
) is nonzero, hence it is a tight contact structure for any k N.
Proof. The proof proceeds by induction; for k = 0 the contact structure is
just (S
3
,
st
) and for k = 1 we can apply Proposition 12.2.3. Notice that
(Y
k+1
,
k+1
) is given as contact (+1)-surgery on (Y
k
,
k
), giving rise to a map
F
W
:

HF(Y
k
)

HF(Y
k+1
) with the property that F
W
_
c(Y
k
,
k
)
_
=
c(Y
k+1
,
k+1
). As the surgery diagram of Figure 12.3 shows, the third
manifold in the corresponding surgery triangle is S
3
1
(T) again. Since
dim

HF(Y
k
)
H
1
(Y
k
; Z)
= k + 1 and

HF(Y
0
) = Z
2
, the triangle
HF(Y
k
)

HF(Y
k+1
)
Z
2
F
W
shows that

HF(Y
k
) = Z
k+1
2
for all k N and F
W
is injective. Therefore
by induction c(Y
k+1
,
k+1
) ,= 0, nishing the proof. For related results see
[101].
The above results might give the impression that nonllability can follow
only from some strong topological properties of the underlying 3-manifold,
and nonllability must hold for all contact structures on a given manifold
at the same time. Below we discuss a family of examples of 3-manifolds
admitting both llable and tight nonllable structures. Let Y
n,g

g
denote the circle bundle with Euler number n over the genusg surface
g
. Honda [77] gave a complete classication of tight contact structures on
these 3-manifolds. He showed that all tight structures are llable, with the
exception of one for n = 2g > 0 and two for n > 2g > 0. Using Seiberg
Witten gauge theory it has been veried that these exceptional structures
are, in fact, nonllable:
Theorem 12.2.7 ([99]). Suppose that g > 0 and n 2g. Then the
virtually overtwisted contact circle bundles Y
n,g
given in [77] are not sym-
plectically llable.
Extending the classication results of Honda to Seifert bered 3-manifolds,
Ghiggini [60] classied tight contact structures on the Seifert bered 3-
manifolds of type M(1, n; r) (cf. Chapter 2 for conventions). Through a
2
2
2
2
2
2
2
2
2
2
2
1
1
2 2
2
= Y
k1
k+1
= Y
Y =
k
k
2
2
2
1 k+
Figure 12.3. Kirby calculus in the surgery triangle
sequence of exercises we show a proof of Theorem 12.2.7 in the simplest
possible case: when g = 1 and n = 2. Then we show some examples of tight
nonllable structures on the type of Seifert bered 3-manifolds for which
the classication result of Ghiggini holds.
Exercises 12.2.8. (a) Show that the surgery diagram of Figure 12.4 gives
a contact structure on Y
2,1
. (Hint: Turn contact surgery coecients into
Seifert framings, put dots on the 0-framed unknots and compare the result
with the diagram of Figure 2.11. In doing so one might need to apply the
transformation of Figure 10.8.)
+1
+1
+1
Figure 12.4. Tight nonllable contact circle bundle
(b) Verify that is nonllable. (Hint: Consider the diagram without
the two Legendrian unknots. Verify that it gives a contact structure on
(2, 3, 4). Finally show that contact (+1)-surgery on a nonllable contact
structure produces a nonllable structure, cf. Exercise 12.2.4(c).)
(c) By computing the contact Ozsv athSzabo invariants of the contact
structure dened by the surgery diagram of Figure 12.4, show that it is
tight. (Hint: Use Lemma 14.4.10 and the result of Exercise 14.3.11(c).)
(d) Verify that Figure 12.5 gives the same contact structure as dened
by Figure 12.4. (Hint: Show that the neighborhood of K in Figure 12.5
containing the linking Legendrian knots K
1
and K
2
remains tight after the
surgeries on K
1
and K
2
. Since it is glued with the same framing as (+1)-
surgery on K, the solution follows from uniqueness of contact surgery with
coecient of the form
1
k
.)
K
2
1
K
K
+1
+1
+1
+1
1
Figure 12.5. Another surgery diagram for the same structure as in Figure 12.4
(e) Show that the contact structures
(2 integer) dened by Fig-

ure 12.6 on the 3-manifolds Y
are all tight. Notice that there are many

choices to turn the ()-surgery into a (1)-surgery by adding zig-zags to
the knot.
The proof of nonllability of the contact structures encountered in Exer-
cise 12.2.8(e) above requires more theory, and relies on the following theo-
rem:
Theorem 12.2.9 ([98]). Suppose that is a contact structure on Y with
induced spin
c
structure t
Spin
c
(Y ) such that the SeibergWitten moduli
space /
Y
(t
) is a smooth manifold consisting of reducible solutions only.

Then any weak symplectic lling W of (Y, ) satises b
+
2
(W) = 0 and the
map H
2
(W; R) H
2
(W; R) induced by the inclusion W W is zero.
Remark 12.2.10. The crux of the argument is that with such moduli space
the SeibergWitten equations over the 3-manifold admit a perturbation
with no solutions, and such perturbation can be extended to the symplectic
lling unless the topological properties listed in the theorem hold for the
lling W. But an extension would imply vanishing SW
(W,)
-invariants for a
weak symplectic lling, contradicting Theomem 13.2.2 of Kronheimer and
Mrowka.
Exercise 12.2.11. Determine the spin
c
structure induced by
.
By applying results of Mrowka, Ozsv ath and Yu [123] the solution of the
above exercise can be used to verify that the assumptions of Theorem 12.2.9
do hold for the contact structures
. Notice that the surgery description

involves several contact (1)-surgeries and one contact ()-surgery. This
latter surgery, however, is not unique. By introducing zig-zags on the
corresponding Legendrian unknot it can be turned into Legendrian (1)-
surgery, but there are many dierent ways to put these zig-zags on the knot.
Dierent choices can be distinguished by the resulting rotation numbers.
Exercises 12.2.12. (a) Show that the 3-manifold of Figure 12.6 is dif-
feomorphic to the Seifert bered 3-manifold M(1, 2;
1
1
). (Hint: Recall
denitions from Section 2.3 and perform handleslides. Notice that all the
surgery coecients are given with respect to the contact framing; rst con-
vert those into surgery coecients with respect to the Seifert framing.)
(b) Verify that for any n there exists N such that among the contact
structures of Figure 12.6 with that xed there are at least n noncontac-
tomorphic. (Hint: Determine c
1
of the resulting contact structures with
+1
+1
+1
+1
1
Figure 12.6. Tight nonllable stuctures on Seifert bered manifolds

the help of the diagram and compute the order of the rst Chern class, see
[100].)
12.3. Topology of Stein llings 215
(c) Show that the manifold Y
is the boundary of a negative denite

manifold with intersection form containing E
8
. (Hint: Embed the 4-
manifold given by the surgery diagram into a blown-up CP
2
and compute
the intersection form of the complement.)
Theorem 12.2.13. The contact structures
dened by Figure 12.6 are

tight and nonllable.
Proof. The same idea as in the proof of Theorem 12.2.1 now shows that
the contact structures
i
(i = 1, . . . , 1) given by Figure 12.6 on Y
are
nonllable. Since these structures can be given as contact (1)-surgery
on the contact structure given by Figure 12.5 and this latter structure has
nonvanishing contact OzsvathSzabo invariants, tightness of (Y
,
i
) follows
from Corollary 14.4.8.
From the solution of Exercise 12.2.12(b) now follows
Corollary 12.2.14 ([100]). For any n N there is a 3-manifold Y
n
with at
least n pairwise noncontactomorphic tight contact structures, none of them
weakly symplectically llable.
Notice that by the work of Gompf all these manifolds admit Stein llable
contact structures. For related results see [101].
12.3. Topology of Stein fillings
We switch perspective now, and instead of examining llability properties of
3-manifolds, we study topological properties of the llings. The motivating
problem of this section can be summarized as:
Problem 12.3.1. Fix a contact 3-manifold (Y, ) and describe all Stein
llings of (Y, ).
Remark 12.3.2. Similar questions for weak (or strong) llings are not
expected to have nice answers in general. The reason is that a weak (strong)
lling can be blown up without destroying the lling property. In addition,
if the lling contains symplectic submanifolds (e.g., a symplectic torus with
self-intersection 0) then by taking symplectic normal sums we can change the
topology of the lling drastically. In some cases (when no such submanifolds
are present) we might be able to describe the classication of weak llings
(up to blow-up), as it is given for lens spaces and links of certain surface
singularities, see Remark 12.3.8 below.
Let us begin the study of Stein llings by a simple observation. If W
is a Stein lling of Y then
1
(Y )
1
(W) is surjective since W can be
built on Y [0, 1] by attaching 2-, 3- and 4-handles only; in particular
b
1
(W) b
1
(Y ). In the following we will list contact 3-manifolds for which
Stein llings have been determined (up to dieomorphism). We start with
a famous result of Eliashberg.
Theorem 12.3.3 (Eliashberg). A Stein lling of S
3
with its standard
contact structure
st
is dieomorphic to D
4
.
Proof. Let us x a Stein lling W of S
3
. By considering a neighborhood
of a point p CP
2
together with a Liouville vector eld and using the
symplectic cut-and-paste operation we get that Z = W
S
3 (CP
2
D
4
) is a
symplectic 4-manifold. Notice that CP
2
D
4
and so Z contains a symplectic
sphere with square (+1). Standard gauge theory (cf. Proposition 13.1.7(5.)
and (6.)) shows that b
+
2
(W) = b
2
(W) = 0. By our observation above we
also get that
1
(W) = 1, therefore Z is homotopy equivalent, hence (by
a theorem of McDu) symplectomorphic to CP
2
. In CP
2
, however, two
symplectic spheres representing the generator of H
2
(CP
2
; Z) are isotopic,
showing that W is dieomorphic to CP
2
CP
1
= D
4
.
Using roughly the same line of reasoning as in the proof of Theorem 12.3.3,
McDu showed that the lens space L(p, 1) with the contact structure
st
it
inherits from (S
3
,
st
) admits a unique (up to dieomorphism) Stein lling
for p ,= 4, which can be given as (1)-surgery on a Legendrian unknot with
tb = p1 and rot = 2p. For other L(p, q)s (still with the quotient of the
standard contact structure (S
3
,
st
)) Lisca [96] gave a complete description
of Stein llings in general, however, uniqueness fails to hold.
Exercises 12.3.4. (a) Verify that the boundary of the Kirby diagram of
Figure 12.7 is L(4, 1). (Hint: Blow down the (1)-framed unknot.)
(b) Replace the 0-framing in Figure 12.7 by a dot and verify that the
resulting 4-manifold is the complement of a quadric in CP
2
. Equip this
4-manifold with a Stein structure.
(c) Show that both the disk bundle over S
2
with Euler class 4 and the
complement of the quadric in CP
2
provide Stein llings of L(4, 1) for some
contact structures. By determining their homotopy types, show that the
two contact structures coincide. In fact, the above two distinct examples
comprise a complete list of Stein llings for (L(4, 1),
st
) up to dieomor-
phism.
1
0
Figure 12.7. 4-manifold with lens space boundary
Remark 12.3.5. For the xed lens space L(p, q) consider the continued
fraction expansion [b
1
, . . . , b
k
] of
p
pq
. Elements of the set
Z
p,q
=
_
(n
1
, . . . , n
k
) Z
k
[ [n
1
, . . . , n
k
] = 0 and 0 n
i
b
i
_
give rise to Stein llings of (L(p, q),
st
) as follows: First consider a linear
chain of k unknots framed by n
i
(providing W with W = S
1
S
2
) and on
that do Legendrian surgery on unknots linking the circles of the chain
there are b
i
n
i
such circles linking the circle with framing n
i
. Surgering
out the 4-manifold W given by the linear chain, i.e., replacing W with a
0-handle and a 1-handle we get a Stein lling of L(p, q). The determination
of the homotopy type of the contact structure on the boundary shows that
what we constructed are llings of
st
(cf. the classication result of Honda
in Section 11.1). Now Lisca proves that any Stein lling of this contact lens
space is dieomorphic to one of the manifolds constructed above. This last
step is carried out by embedding a lling into a rational surface and showing
that the complement is standard, similar to the argument presented in the
proof of Theorem 12.3.3. For further details see [96].
Using the same main ideas as above, Ohta and Ono described Stein llings
of links of simple and simple elliptic singularities (again with specic contact
structures). All these llings happened to have b
+
2
= 0 and could be
embedded into rational or ruled surfaces. In particular:
Theorem 12.3.6 (OhtaOno, [127]). The Poincare homology 3-sphere
(2, 3, 5) (with its contact structure inherited from S
3
) admits a unique (up
to dieomorphism) Stein lling which is the negative denite E
8
-plumbing.
The same uniqueness holds for the boundary of the negative denite E
6
-
and E
7
-plumbings.
Remark 12.3.7. In [127] it was shown that simple (or ADE) singularities
with the contact structure given by the link of the singularity admit a unique
Stein lling (up to dieomorphism). For a simple elliptic singularity L
k

which is topologically a circle bundle of Euler class k < 0 over the 2-
torus T
2
with the contact structure given by the link of the singularity
it has been proved [128] that (i) L
k
admits a strong symplectic lling X
with c
1
(X) = 0 if and only if 0 < k 9 and such X (which can be regarded
as the generalization of smoothing) is unique up to dieomorphism unless
k = 8, when there are two possibilities, and (ii) for k 10 a minimal
strong symplectic lling is unique up to dieomorphism, and such a lling
is dieomorphic to the minimal resolution. Finally for k 9 a minimal
lling is dieomorphic either to the minimal resolution or to a smoothing
(i.e., a lling with c
1
= 0). The proofs of the above statements given in
[127, 128] use SeibergWitten theory.
Remark 12.3.8. Above we considered only Stein llings of the given con-
tact 3-manifold. This is, however, not the greatest generality for most of
the cases discussed: if Y is a rational homology 3-sphere then any weak ll-
ing can be deformed into a strong lling by Theorem 12.1.5, and for strong
llings the same cut-and-paste argument works.
The key common feature of the above results is that in each case any
lling can be embedded into a closed symplectic 4-manifold with = .
In particular, all the above llings have b
+
2
= 0. Next we will describe
some particular cases when the above approach fails (for example, because
of the existence of llings with b
+
2
> 1). Using ad hoc arguments of
embedding Stein llings of T
3
and (2, 3, 11) into homotopy K3-surfaces
one gets strong constraints on the topology of such Stein manifolds [158].
These methods, however, seem to be insucient in greater generality. For
example, the contact structures on the 3-torus T
3
have been classied by
Kanda and Giroux [62, 79] by showing that any (T
3
, ) is contactomorphic
to one of (T
3
,
n
) where
n
= ker
_
cos(2nt) dx + sin(2nt) dy
_
(n 1).
Using delicate results of Gromov, Eliashberg showed [29] that (T
3
,
n
) is not
strongly llable once n 2. For n = 1, Figure 12.8 gives a Stein lling of
(T
3
,
1
). Using a version of the cut-and-paste argument outlined above, one
can show
Proposition 12.3.9 ([158]). If W is a Stein lling of (T
3
,
1
) then W is
homeomorphic to T
2
D
2
.
0
Figure 12.8. Stein structure on D
2
T
2
For the understanding of Stein llings of (2, 3, 11) the OzsvathSzabo
homology groups of these manifolds seem to play a crucial role. First we
discuss a slightly more general result, since by embedding Stein llings into
symplectic 4-manifolds and applying product formulae for OzsvathSzabo
invariants, one can show
Theorem 12.3.10. Suppose that Y is a rational homology sphere with
HF(Y, t) = Z
2
for t Spin
c
(Y ). If W is a Stein lling of (Y, ) such that
t
= t then b
+
2
(W) = 0. If

HF(Y, t) = Z
3
2
then for any Stein lling W with
b
+
2
(W) > 0 we have c
1
(W) = 0.
Remark 12.3.11. The proof of this statement falls aside from the main
topic of these notes, and we do not present it here. We just note that the
proof rests on the embeddability of Stein llings into Lefschetz brations.
The fact

HF(Y, t) = Z
2
is equivalent to HF
red
(Y, t) = 0, while

HF(Y, t) =
Z
3
2
is the same as HF
red
(Y, t) = Z
2
.
Computation shows that

HF
_
(2, 3, 11)
_

= Z
3
2
, hence a Stein lling of
it with b
+
2
> 0 has c
1
= 0. Surgery on the K3-surface together with the
homeomorphism characterization of the K3-surface using SeibergWitten
invariants due to Morgan and Szab o given in Theorem 3.3.11 provides:
Proposition 12.3.12. If W is a Stein lling of (2, 3, 11) then
b
2
(W) = 2. If W is a Stein lling of (2, 3, 11) then either b
+
2
(W) = 0
or b
2
(W) = 20.
Proof (sketch). It is not very hard to prove that (2, 3, 11) does not
bound negative denite 4-manifold: there exists a compact 4-manifold N
2
with N
2
= (2, 3, 11) such that three disjoint copies of N
2
are embedded
in K3 in a way that the intersection form of K33 N
2
is 2E
8
. (For a Kirby
diagram of N
2
see Figure 2.14.) If X = (2, 3, 11) and X is negative
denite then the closed negative denite 4-manifold (K3 3 N
2
) 3 X
would contradict Donaldsons diagonalizability theorem [20]. If W is a
Stein lling of (2, 3, 11) then for Z = (K3 N
2
) W a suitable product
formula of the SeibergWitten invariants and Theorem 3.3.11 implies that
Z is homeomorphic to K3. This concludes the proof of the rst statement.
The same reasoning shows that if W is Stein with W = (2, 3, 11) and
b
+
2
(W) > 0 then the intersection form Q
W
is equal to 2E
8
2H. (For more
details see [158].)
In fact, it is reasonable to conjecture that if W is a Stein lling of (2, 3, 11)
then it is dieomorphic to N
2
, and a Stein lling of (2, 3, 11) is dieomor-
phic either to the smoothing or to the resolution of the isolated singularity
x
2
+y
3
+z
11
= 0 C
3
similar to the case of simple and simple elliptic
singularities. Notice that in the above arguments we did not make use of
the particular choice of the contact structures on (2, 3, 11). A recent re-
sult of Ghiggini and Sch onenberger [59] classies tight contact structures on
certain Seifert bered spaces including (2, 3, 11). According to these
results, (2, 3, 11) admits (up to isotopy) a unique tight contact structure,
which can be given as the boundary of the Stein domain of Figure 12.9. The
Seifert bered space (2, 3, 11) admits exactly two (nonisotopic) tight con-
tact structures, both Stein llable.
Returning to Problem 12.3.1, we might ask what can we say about Stein
llings in general. According to the next result we cannot expect a nite
list as a solution of Problem 12.3.1, since
Proposition 12.3.13 ([132]). For g 2 the element
2
g

g,1
admits
innitely many decompositions into right-handed Dehn twists with the cor-
responding Lefschetz brations having distinct rst homologies. Conse-
quently, the contact 3-manifold given by
2
g
through the corresponding
open book decomposition admits innitely many distinct Stein llings.
1
1
Figure 12.9. Stein structure on the nucleus N2
In the theorem
g

g,1
denotes the right-handed Dehn twist along a
simple closed curve parallel to the unique boundary component of
g,1
, cf.
also discussion in Chapter 15.
Proof. Consider the Lefschetz bration we get by taking the desingu-
larization of the double branched cover of
h
S
2
along two copies of
h

h
S
2
and two (four for even g) copies of S
2

h
S
2
.
The bration map can be given by perturbing the composition of the
branched cover map with the projection to the second factor. It is easy
to see that the resulting bration has a section of square 1, hence gives a
factorization of
g

g,1
, cf. Section 15.2. Taking a twisted ber sum of
two copies of this bration we get factorizations of
2
g

g,1
. The twisting
can be chosen in such a way that the resulting 4-manifolds have dierent
torsion in their rst homologies, cf. the proof of Theorem 10.3.5. Now
taking the complement of a section and a regular ber we get Lefschetz -
brations over D
2
with nonclosed bers, hence innitely many Stein llings
of the contact 3-manifold given by
2
g
. The llings are distinguished by the
torsion of their rst homologies.
We close this section with a general result concerning the topology of
Stein llings:
Theorem 12.3.14 ([159]). For a given contact 3-manifold (Y, ) there
exists a constant K
(Y,)
such that if W is a Stein lling of (Y, ) then
K
(Y,)
3(W) + 2(W).
In other words, the number c(W) = 3(W) + 2(W) for a Stein lling W
of (Y, ) which resembles the c
2
1
-invariant of a closed complex surface
is bounded from below.
Remark 12.3.15. The idea of the proof is roughly as follows: Suppose
that W
1
, . . . , W
n
, . . . is the possibly innite list of Stein llings of (Y, ).
Consider the K ahler embeddings W
i
X
i
where X
i
are minimal surfaces
of general type. Our aim is to control the topology of W
i
. So x a lling
W
1
and consider T
1
= X
1
int W
1
. Now for any other Stein lling W of
(Y, ) we can form Z = T
1
W, and according to Theorem 7.1.9 this is a
symplectic 4-manifold (with b
+
2
(Z) > 1). Therefore minimality of Z would
imply c
2
1
(Z) 0, giving the desired lower bound for 3(W) + 2(W) in
terms of invariants of the xed 4-manifold T
1
. Minimality of Z is, however,
hard to prove although it seems to be true , so rather we have to use
a larger (still nite) set of test manifolds X
i
int W
i
to compare the Stein
lling W with. Also we may relax the minimality requirement by trying to
prove that the number of blow-ups contained in the symplectic 4-manifold
Z is bounded by some number depending only on (Y, ). In the computation
the mod 2 reduced version of SeibergWitten theory is used; for details see
[159].
Notice that K
(Y,)
3(W) + 2(W) can be rewritten as
b
2
(W) +C
(Y,)
5b
+
2
(W)
where C
(Y,)
is another constant depending only on the contact 3-manifold
(Y, ). In other words if b
+
2
(W) is bounded for all Stein llings of a given
contact 3-manifold then all the characteristic numbers form a bounded set.
For many 3-manifolds a Stein lling has to have vanishing b
+
2
-invariant. Such
3-manifolds are, for example, the ones carrying positive scalar curvature, or
having vanishing reduced Floer homologies, e.g. lens spaces or boundaries of
certain plumbings along negative denite plumbing diagrams [124, 139, 143].
This observation leads us to the following conjecture.
Conjecture 12.3.16. The set
(
(Y,)
=
_
(W) [ W is a Stein lling of (Y, )
_
is nite.
13. Appendix: SeibergWitten invariants
In this chapter we recall basic denition, notions and results of Seiberg
Witten gauge theory. The introduction is not intended to be complete,
we rather describe arguments most frequently used in the text. We also
review a variant of the theory for 4-manifolds with contact type boundary,
which setting turns out to be very useful in the study of contact topological
problems. The last section is devoted to a discussion centering around
the adjunction inequality. For a more complete discussion of the topics
appearing in this chapter the reader is advised to turn to [21, 119, 126, 149].
13.1. SeibergWitten invariants of closed 4-manifolds
Let us assume that X is a closed (i.e., compact with X = ), ori-
ented, smooth 4-manifold. Suppose furthermore that b
+
2
(X) > 1 and
b
+
2
(X) b
1
(X) is odd. Below we outline the construction of a map
SW
X
: Spin
c
(X) Z, the SeibergWitten invariant of X, which turns out
to be a dieomorphism invariant, that is, for a dieomorphism f : X
1
X
2
and spin
c
structure s Spin
c
(X
2
) we have SW
X
2
(s) = SW
X
1
(f
s). The
value SW
X
(s) counts solutions of a pair of equations for pairs of connec-
tions and sections of bundles naturally associated to the spin
c
structure s.
In the following we will assume that the reader is familiar with basic notions
of dierential geometry, such as connections, covariant dierentiation and
LeviCivita connections.
Fix a metric g on X and suppose that the spin
c
structure s Spin
c
(X)
is given by the hermitian spinor bundles W
X with Cliord multi-

plication c: T
X Hom
C
(W
+
, W
) satisfying c(v)
c(v) = [v[
2
id
W
+.
The xed metric induces a connection, the LeviCivita connection on
TX and on all bundles associated to it. By xing the connection A on
224 13. Appendix: SeibergWitten invariants
L = det W
+
(
= det W
) we get a coupled connection on W
and hence a
covariant dierentiation
A
: (W
) (W
X).
Composing this with the Cliord multiplication
c: (W
X) (W
)
we get
Denition 13.1.1. The operator /
A
: (W
) (W
) given as /
A
=
c
A
is called the Dirac operator associated to the connection A on L.
Formally, for (W
) we have
A
() = w with w (W
) and
(T
X), and then /

A
= c()w (W
).
We recall that the metric g on X gives rise to the Hodge star operator
g
:
i
(X)
4i
(X). On two forms
2
g
= id
+
(X)
, and a 2-form
2
(X)
is self-dual (anti-self-dual, or ASD) if
g
= (resp.
g
= ). The
self-dual part of , which is equal to
1
2
( +
g
), is denoted by
+
. The
Clior multiplication naturally extends to 2-forms and provides a bundle
isomorphism :
+
(X) su(W
+
). Moreover, for any section (W
+
)
we can consider the action of
on W
+
. The traceless part of this
endomorphism will be denoted by q(). Now we are in the position to write
down the SeibergWitten equations. Let
+
(X) be a xed self-dual
2-form. For a spinor (W
+
) and connection A on L = det W
+
the
-perturbed SeibergWitten equations read as follows:
/
A
= 0
(F
+
A
+i) = q()
where F
+
A
is the self-dual part of the curvature of the connection A. The
gauge group G = Map(X, S
1
) acts on the space A(L) (W
+
) =
_
U(1)-
connections on L
_
(W
+
) by g(, A) = (g, A2
dg
g
), and it is not hard
to see that this action maps SeibergWitten solutions to SeibergWitten
solutions. The conguration space
_
A(L) (W
+
)
_
/G will be denoted by
B.
Denition 13.1.2. The set of gauge equivalence classes of solutions of the
-perturbed SeibergWitten equations is called the SeibergWitten moduli
space M
(s). The union
+
(X)
M
(s) is the parameterized moduli space

M(s).
13.1. SeibergWitten invariants of closed 4-manifolds 225
These spaces admit suitable topologies; for technical reasons we actually
consider solutions in some Sobolev completions, but this subtlety will be
ignored in the following.
Denition 13.1.3. The pair (A, ) B is reducible if 0 and irreducible
otherwise. The set of irreducible elements in B is denoted by B
. We dene
M
(s) as B
M(s).
Recall that the parameterized moduli space admits a map : M(s)
+
(X) by associating the perturbation parameter to every solution. The
structure of the moduli space can be summarized in the following
Theorem 13.1.4. The map : M(s)
+
(X) is a smooth, proper Fred-
holm map of index d(s) =
1
4
_
c
2
1
(s) 3(X) 2(X)
_
. The subspace M
(s)
is a smooth innite dimensional manifold.
By studying reducible solutions, it can be shown that for generic and
b
+
2
(X) > 0 the moduli space M
(s) consists of irreducible solutions only.

The SardSmale theorem implies that for generic the moduli space is
a compact smooth manifold of dimension d(s). After xing a homology
orientation on X (that is, an orientation for H
2
+
(X; R) H
1
(X; R)) the
moduli space admits a canonical orientation. Therefore for generic the
oriented, compact submanifold M
(s) gives rise to a homology class in

H
(B
X
; Z), which provides us a way to turn it into a number. When
d(s) = 0, this simply means that we (algebraically) count the number of
solutions modulo gauge equivalence of the SeibergWitten equations. For
d(s) > 0 we evaluate suitable cohomology classes of the cohomology ring
H
(B
X
; Z) on M
(s). In this way we produce a number SW

X
(s) Z for
which the following result holds:
Theorem 13.1.5 (SeibergWitten, [174]). If b
+
2
(X) > 1 then the value
SW
X
(s) Z is independent of the chosen metric g and perturbation ,
providing a smooth invariant of X.
For manifolds with b
+
2
(X) = 1 the proof of independence from the chosen
metric and perturbation does not apply, since in that case a 1-parameter
family of moduli spaces might contain reducible solutions. Such solutions
are xed points of a nontrivial subgroup of the gauge group and therefore
require special attention. For a thorough discussion the reader is advised to
turn to [119, 149]. Recall that an element K H
2
(X; Z) is characteristic
if for all x H
2
(X; Z) we have that K(x) Q
X
(x, x) (mod 2). The set of
characteristic elements in H
2
(X; Z) will be denoted by (
X
.
Denition 13.1.6. A class K (
X
is a basic class if SW
X
(s) ,= 0 for
s Spin
c
(X) with c
1
(s) = K. The set of basic classes will be denoted
by B
X
. The manifold X is of simple type if K B
X
implies that K
2
=
3(X) + 2(X).
The following proposition summarizes some of the basic properties of
B
X
and SW
X
. Here (for simplicity) we assume that X is of simple type.
Recall that in the denition we assumed that b
+
2
(X) b
1
(X) is odd and
b
+
2
(X) is greater than 1.
Proposition 13.1.7.
1. The set B
X
of basic classes is nite and K B
X
if and only if
K B
X
. In fact,
SW
X
(K) = (1)
1
4
((X)+(X))
SW
X
(K).
2. If B
X
,= and X is an embedded surface representing the
homology class [] with []
2
0 and ,= S
2
, then
2g() 2 []
2
+[K
_
[]
_
[
for all K B
X
. If B
X
,= and X is an embedded sphere
then []
2
< 0. The above inequality is usually called the adjunction
inequality, since it generalizes the formula of Theorem 3.1.9.
3. If X is a symplectic manifold then c
1
(X, ) B
X
. For a minimal
surface of general type B
X
=
_
c
1
(X)
_
. Moreover, in both cases
SW
X
_
c
1
(X)
_
= 1.
4. If X admits a positive scalar curvature metric, or decomposes as
X = X
1
#X
2
with b
+
2
(X
1
), b
+
2
(X
2
) > 0 then B
X
= .
5. More generally, if X = X
1

N
X
2
with b
+
2
(X
1
), b
+
2
(X
2
) > 0 and N
admits a metric of positive scalar curvature then B
X
= . Saying this
property in a dierent way, if X = X
1
N
X
2
, N admits positive scalar
curvature metric and B
X
,= then either b
+
2
(X
1
) = 0 or b
+
2
(X
2
) = 0.
6. If X = Y #CP
2
then B
X
= L E [ L B
Y
, where H
2
(X; Z)
is identied with H
2
(Y ; Z) H
2
(CP
2
; Z) and E is the generator of
H
2
(CP
2
; Z).
According to the following theorem, the assumption on the simple type
property of X is not too restrictive for our purposes, since
13.1. SeibergWitten invariants of closed 4-manifolds 227
Theorem 13.1.8 (Taubes). If X is a symplectic 4-manifold then X is of
simple type. If there is an embedded surface X such that 2g() 2 =
[]
2
0 then X is of simple type.
Exercise 13.1.9. Show that if
1
,
2
X are two embedded surfaces with
genera g(
1
) and g(
2
) in the symplectic 4-manifold X with [
1
] = [
2
],
[]
2
0 and
1
is a symplectic submanifold then g(
2
) g(
1
). (This
inequality is usually referred to as the Symplectic Thom Conjecture. For
the history of this problem see [133], cf. also Theorem 13.3.8.)
The next theorem describes a relation between SeibergWitten invari-
ants and J-holomorphic submanifolds in symplectic 4-manifolds. In order to
state the result, let us assume that (X, ) is a given symplectic 4-manifold
and J is a compatible almost-complex structure. Suppose furthermore that
b
+
2
(X) > 1.
Theorem 13.1.10 (Taubes, [162], [163]; see also [84]). Suppose that (X, )
is a symplectic 4-manifold with b
+
2
(X) > 1 and SW
X
(K) ,= 0 for a given
cohomology class K H
2
(X; Z). Assume furthermore that the class c =
1
2
_
K c
1
(X, )
_
is nonzero in H
2
(X; Z). Then for a generic compatible
almost-complex structure J on X the class PD(c) H
2
(X; Z) can be
represented by a pseudo-holomorphic submanifold.
In fact, Taubes proved much more. By dening a rather delicate way
of counting pseudo-holomorphic submanifolds representing a xed homo-
logy class PD(c) H
2
(X; Z), he proved that this number and the value
SW
X
(c
1
(X, ) +2c) are equal. In many applications only the direction that
a nonvanishing SeibergWitten invariant implies the existence of pseudo-
holomorphic curves is used. Note that the curve representing PD(c) is
not given to be connected. This observation becomes important if one wants
to apply the adjunction formula to compute the genus of . By Proposi-
tion 13.1.7 we have that c
1
(X, ) B
X
, consequently Theorem 13.1.10
implies, in particular, that the Poincare dual of c
1
(X, ) can be represen-
ted by a pseudo-holomorphic submanifold C (assuming it is nonzero). Since
a pseudo-holomorphic submanifold is always symplectic, the above reason-
ing shows that c
1
(X, ) [] =
_
C
> 0 for manifolds with b
+
2
(X) > 1
and c
1
(X, ) nonzero. Furthermore, it can be shown that if b
+
2
(X) > 1,
then a class e H
2
(X; Z) with e
2
= 1, c
1
(X, ) PD(e) = 1 and
SW
X
_
c
1
(X, ) + 2PD(e)
_
,= 0 can be represented by a symplectic sphere;
consequently X is nonminimal. (The fact c
1
(X, ) + 2PD(e) B
X
guar-
antees the existence of a pseudo-holomorphic representative for e. The two
other assumptions together with the adjunction formula ensure that
this representative is a sphere, see the proof of Corollary 13.1.13.) As a
further application of Theorem 13.1.10, one can show that a symplectic 4-
manifold with b
+
2
> 1 has SeibergWitten simple type, cf. Theorem 13.1.8
and [84]. Theorem 13.1.10 also proves the inequality in Theorem 3.1.12: If
K is a basic class, then c =
1
2
_
Kc
1
(K, )
_
can be represented by a pseudo-
holomorphic (in particular symplectic) submanifold (unless c = 0), hence
c [] 0. Reversing the sign of K if necessary, we can assume K [] 0,
so c
1
(X, ) [] K [] 0, which proves the inequality. Note that equal-
ity implies c [] = 0, hence c = 0, and consequently, K = c
1
(X, ) (or
K = c
1
(X, )).
Remark 13.1.11. Above we only dealt with the case of b
+
2
(X) > 1;
recall that for manifolds with b
+
2
(X) = 1 the SeibergWitten invariants
depend on the chosen metric and perturbation. After the appropriate
modications, the theorems and properties discussed above extend to the
case of b
+
2
(X) = 1. For the sake of brevity, however, here we will omit the
discussion of these extensions; see [145] for a nice review of the b
+
2
(X) = 1
case.
Corollary 13.1.12. Suppose that the symplectic 4-manifold X satisfying
b
+
2
(X) > 1 is minimal. Then c
2
1
(X) 0.
Proof. According to Theorem 13.1.10 the Poincare dual of the class
c
1
(X, ) can be represented by an embedded J-holomorphic submanifold
C =
n
i=1
C
i
; here C
i
are the connected components of C. Now the adjunc-
tion formula for C
i
reads as 2g(C
i
) 2 = [C
i
]
2
c
1
(X, )[C
i
] = 2[C
i
]
2
. Now
[C
i
]
2
0 holds, since [C
i
]
2
< 0 implies g(C
i
) = 0 and [C
i
]
2
= 1 contradict-
ing minimality. Therefore c
2
1
(X, ) =
[C
i
]
2
0, concluding the proof.
Corollary 13.1.13. If the symplectic 4-manifold X smoothly decomposes
as Y #CP
2
then it contains a symplectic (1)-sphere, i.e., it is not minimal
as a symplectic 4-manifold.
Proof. According to Proposition 13.1.7(6.) we know that B
X
= L E [
L B
Y
. Therefore c
1
(X) = (L E) for some L B
Y
; now apply
Theorem 13.1.10 for K = L + E. We get that
1
2
_
K c
1
(X)
_
= E
can be represented by a J-holomorphic (hence symplectic) submanifold,
furthermore by the adjunction formula E
2
= 1 and c
1
(X) E = 1 give
2g(E) 2 = E
2
c
1
(X) E = 2,
so g(E) = 0, therefore the representative is a sphere.
13.2. SeibergWitten invariants of 4-manifolds with contact boundary 229
Remark 13.1.14. In fact, we only need the existence of a basic class
K B
X
with the property that
_
K c
1
(X)
_
2
= 4 in order to deduce
that the symplectic 4-manifold (X, ) is not minimal.
By studying the SeibergWitten equations on 4-manifolds of the form
Y
3
R, SeibergWitten Floer homologies can be dened for closed oriented
3-manifolds. This theory has been developed in [88], see also [89].
13.2. SeibergWitten invariants of 4-manifolds with
contact boundary
In the study of llings of contact 3-manifolds a variant of the original
SeibergWitten equations developed by Kronheimer and Mrowka [86]
turns out to be extremely useful. Here we restrict ourselves to a quick
review of the invariants, for a more complete discussion see [86, 89]. Let X
be a given compact 4-manifold with nonempty boundary and x a contact
structure on X. Dene Spin
c
(X, ) to be all spin
c
structures on X which
restrict to the spin
c
structure t
induced by .
Remark 13.2.1. Recall that the set of spin
c
structures on X forms a
principal H
2
(X; Z)-space and for a 4-manifold it is never empty. As it
follows from the long exact sequence of cohomologies of the pair (X, X),
the above dened set Spin
c
(X, ) is a principal H
2
(X, X; Z)-space.
The invariant SW
(X,)
will map from Spin
c
(X, ) to Z and is roughly de-
ned as follows. Consider the symplectization of (X, ) and glue it to X
along X 1 to get X
+
. By choosing an almost-complex structure for
, by the symplectic form on Symp(X, ) we get a metric on X
+
X;
extend it to a metric g dened on X
+
. On X
+
X the canonical spin
c
structure denes a spinor
0
and a spin connection A
0
, for a spin
c
struc-
ture s Spin
c
(X, ) extend these over X
+
. Take the space of pairs (A, )
spin connections and spinors for the xed spin
c
structure s which
solve the usual (perturbed) SeibergWitten equations on the noncompact
Riemannian manifold (X
+
, g) and are close to (A
0
,
0
) in an appropriate
L
2
-sense. After dividing with the appropriate gauge group G we get the
moduli space M
X
+
,g
(s) of SeibergWitten solutions. The rest of the de-
nition is fairly standard now: one needs to show compactness, smoothness,
orientability of the (appropriately perturbed) moduli space, and SW
(X,)
is
dened by counting the number of elements (with sign) in M
X
+
,g
(s). To get
a well-dened invariant, we need to show independence of the choices (met-
ric, almost-complex structure, extensions, perturbation) made throughout
the denition. This argument follows the usual cobordism method applied
in the closed 4-manifold case. The two notable dierences from the closed
case are:
There are no reducible solutions (i.e., points in the moduli space with
vanishing spinor component) since
0
is nonzero on X
+
X and for
any element (A, ) in the moduli space is close to
0
. Therefore
the gauge group acts freely, the index formula provides the actual
dimension of a (smoothly cut out) moduli space and there is no need
to assume anything about b
+
2
(X).
In the case dimM
X
+
,g
(s) > 0 the invariant SW
(X,)
(s) is dened to
be zero, since there is no reasonable constraint with which one could
cut down the dimension. (The cohomology class used in the closed
case vanishes for (X, ).)
The main result of [86] concerning SW
(X,)
is the the generalization of
Theorem 13.1.10 of Taubes to the manifold-with-boundary case.
Theorem 13.2.2 (Kronheimer-Mrowka, [86]). If (X, ) is a weak sym-
plectic lling of (X, ) and s
is the spin
c
structure induced by an
almost-complex structure compatible with the symplectic form then
SW
(X,)
(s
) = 1. Moreover, if (X, ) is as above and SW

(X,)
(s) ,= 0
then [] (s s
) 0 with equality only if s = s
.
Notice that the last assertion implies that if is exact (for example, if
(X, ) is a Stein lling of (X, )) then s
is the only spin

c
structure with
nonzero SW
(X,)
-invariant. In addition, these invariants can be used to
prove the adjunction inequalities of the type of Proposition 13.1.7(2.) for
weak symplectic llings.
Let us take a contact 3-manifold (Y, ) and consider the noncompact
4-manifold X = Y (, 0] with contact type boundary. The ideas
outlined above now produce a contact invariant c
SW
(Y, ) of the contact
manifold in the appropriate SeibergWitten Floer cohomology of Y . This
invariant can be shown to share many properties with the contact invariant
c(Y, )

HF(Y ) to be discussed in Section 14.4. In this volume we will
restrict our attention to the Heegaard Floer theoretic contact invariants, for
the exact denition and some basic properties of c
SW
see [89].
13.3. The adjunction inequality 231
13.3. The adjunction inequality
Recall the adjunction equality from complex geometry:
Theorem 13.3.1. If C is a smooth, complex curve in the complex 4-
manifold X then (C) = [C]
2
c
1
(X)[C].
Remark 13.3.2. In complex algebraic geometry it is customary to use
the canonical bundle K
X
instead of c
1
(X). Since in H
2
(X; Z) we have
c
1
(X) = K
X
(one originates from the tangent, while the other from the
cotangent bundle), the formula reads as (C) = C
2
+K
X
C.
It is not hard to see that the above formula holds for a J-holomorphic
submanifold of an almost complex 4-manifold (X, J). In particular,
() = []
2
c
1
(X, )[]
holds for a symplectic submanifold of a symplectic 4-manifold (X, ). This
equality admits the following generalization for smoothly embedded sub-
manifolds in symplectic 4-manifolds:
Theorem 13.3.3. Suppose that is a smoothly embedded, closed, ori-
ented 2-dimensional submanifold in the symplectic 4-manifold (X, ) with
b
+
2
(X) > 1. If g() > 0 then
[]
2
+
c
1
(X, )[]
().
If g() = 0 and [] is nontrivial in homology then []
2
1.
Corollary 13.3.4. If (W, ) is a weak symplectic lling of the contact
3-manifold (Y, ) and W is a homologically nontrivial surface with
g() > 0 then
[]
2
+
c
1
(W, )[]
().
Proof. Embed the symplectic lling into a closed symplectic 4-manifold X
with b
+
2
(X) > 1 and apply Theorem 13.3.3.
Theorem 13.3.3 follows from the fact that c
1
(X, ) of a symplectic 4-
manifold is a basic class and
Theorem 13.3.5 (The adjunction inequality). Suppose that X is a smooth,
closed 4-manifold. If K H
2
(X; Z) is a basic class of the 4-manifold X
with b
+
2
(X) > 1 and g() > 0 then
[]
2
+[K
_
[]
_
[ ().
The theorem was proved in the case of []

2
0 by KronheimerMrowka [85]
and by MorganSzab oTaubes [121] in their proof for the Thom conjecture.
A more involved argument allows []
2
to be negative in the above formula.
This extension rests on the following result.
Theorem 13.3.6 (OzsvathSzabo, [133]). Suppose that is a smooth,
embedded, closed 2-dimensional submanifold in the smooth 4-manifold X
and for a basic class K we have () []
2
K
_
[]
_
= 2n < 0. Let
denote the sign of K
_
[]
_
. Then the cohomology class K + 2PD
_
[]
_
is
also a basic class.
The most spectacular application of these results is the proof of the Sym-
plectic Thom Conjecture due to Ozsvath and Szab o, which improves the
result of Exercise 13.1.9 by dropping the assumption on the self-intesection
of the surface.
Theorem 13.3.7 (OzsvathSzabo, [133]). If
1
,
2
X are two 2-
dimensional connected submanifolds of the symplectic 4-manifold (X, ),
the homology classes [
i
] are equal and
1
is a symplectic submanifold,
then the genus of
2
is not smaller than the genus of
1
.
In addition, the form of the adjunction inequality given in Theorem 13.3.5
implies an improved version of Corollary 13.3.4, already encountered in the
introduction:
Theorem 13.3.8. If is a smoothly embedded closed, oriented 2-dimen-
sional submanifold in the Stein surface S then
[]
2
c
1
(S)[] ()
unless is a nullhomologous sphere.
Proof. Recall that a Stein surface can always be symplectically embedded
into a symplectic 4-manifold X, therefore for []
2
0 the statement follows
from the usual adjunction inequality (together with the fact that c
1
(X) of
13.3. The adjunction inequality 233
a symplectic 4-manifold is a basic class). In the case of negative []
2
we use
the embedding of S into a minimal surface X of general type. Assuming
g() > 0 the relation of Theorem 13.3.6 implies that either the inequality is
satised or c
1
(X) 2PD([]) is a basic class. (The sign here is determined
by the sign of c
1
(X)
_
[]
_
.) But for a minimal surface of general type there
are only two basic classes, which are c
1
(X). Therefore we have either
[] = 0 or the dierence of the two basic classes c
1
(X) and c
1
(X) (which
is 2c
1
(X)) is equal to 2PD
_
[]
_
. This latter case, however, provides a
contradiction since it implies that c
2
1
(X) = []
2
is negative, which cannot
hold for a minimal surface of general type. Finally if g() = 0 then the
above principle provides []
2
2 since a sphere with self-intersection 1
would violate minimality of X.
14. Appendix: Heegaard Floer theory
The topological description of contact structures as open book decom-
positions provides the possibility of dening contact invariants which (at
least partially) can be computed from surgery diagrams. In this appendix
we outline the construction of such invariants for a complete discus-
sion the reader is referred to the original papers of Ozsvath and Szab o
[135, 136, 137, 138]. To set up the stage, rst we discuss OzsvathSzabo
homology groups of oriented, closed 3-manifolds (together with maps in-
duced by oriented cobordisms). The denition of the group

HF(Y ) for a
3-manifold Y will rely on some standard constructions in Floer homology.
After presenting the surgery triangles for this theory, we outline the deni-
tion of the contact OzsvathSzabo invariants and verify some of the basic
properties of this very sensitive invariant. A few model computations are
also given.
14.1. Topological preliminaries
Recall that a closed, oriented 3-manifold Y can be decomposed as a union of
two solid genus-g handlebodies Y = U
0
g
U
1
: consider a Morse function on
Y and dene U
0
as the union of the 0- and 1-handles while U
1
= 2-handles
3-handle. In fact, the 1-handles can be recorded on the genus-g surface
g
by their cocores, while the 2-handles by their attaching circles. Hence
the handlebody decomposition can be presented on
g
by two g-tuples of
embedded simple closed curves
1
, . . . ,
g
and
1
, . . . ,
g
which satisfy
that the s (and the s) are disjoint among themselves and form a linearly
independent system in H
1
(
g
; Z). Of course, the -curves might intersect
the -curves. In conclusion, a 3-manifold can be described by a Heegaard
236 14. Appendix: Heegaard Floer theory
diagram
_
g
,
i
g
i=1
,
i
g
i=1
_
with the and -curves satisfying the above
conditions.
It is not hard to nd a Heegaard diagram of a 3-manifold given by a
surgery diagram. As we saw, any rational surgery can be transformed into a
sequence of integral surgeries; in the following we will describe an algorithm
(given in [135]) for nding a Heegaard diagram of a 3-manifold given by
integral surgery on a knot. (The general case of surgery on a link follows
similar ideas.) Suppose that Y is given by an integral surgery on K S
3
and consider the following Heegaard diagram of S
3
K: Fix a projection
of K to some plane. Consider a tubular neighborhood of K in R
3
and
add vertical tubes for every crossing of the given projection, as it is shown
by the upper diagrams of Figure 14.1. By an isotopy, the resulting subset
U
K
R
3
can be regarded as an -neighborhood of the knot projection, cf.
the lower diagrams in Figure 14.1. In fact, U
K
is a genus-g handlebody (cf.

Figure 14.1. The -curves of the Heegaard decomposition of the knot complement
Figure 14.2 for the case of the trefoil knot) with the complement in S
3
also
a genus-g handlebody. This last statement can be easily veried by adding
g 3-dimensional 2-handles along the curves
i
encircling the holes of U
K
14.1. Topological preliminaries 237
(see Figure 14.2 again) turning U
K
into a solid ball. Notice that the -
curves are the cocores of the 1-handles of the complementary handlebody
S
3
U
K
. This species the -curves of the diagram for S
3
K. On
1
4
Figure 14.2. The -curves of a Heegaard decomposition of the complement for the trefoil
knot
the other hand the meridians of the vertical tubes that we attach (as it is
shown by Figure 14.1) give rise to the -curves since we can think of them
as the attaching circles of the 2-handles in S
3
K. By attaching handles
along the -curves we ll the complement of U
K
(minus a point), while
by attaching 2-handles along the -curves we ll the vertical tubes inside
U
K
. Therefore the Heegaard diagram
_
g
,
i
g
i=1
,
j
g1
j=1
_
provides the
knot complement S
3
K. Now taking a simple closed curve dening any
(integral) surgery along K as
g
we get a Heegaard diagram for the surgered
manifold. For example, if we choose the meridian of K as
g
then this choice
corresponds to a trivial surgery along K so that we get a Heegaard diagram
of S
3
. For another example see Figure 14.3.
Exercise 14.1.1. Determine the knot and compute the surgery coecient
of the surgery corresponding to the Heegaard diagram of Figure 14.3.
It is natural to wonder when do two Heegaard diagrams represent the
same 3-manifold. It is fairly easy to list moves which do not change the
resulting 3-manifold: isotoping the - and the -curves (by keeping the dis-
jointness property), or sliding -curves over -curves (and -curves over
4
Figure 14.3. The -curves of a Heegaard diagram of a surgery on the trefoil knot
-curves) obviously changes only the handle decomposition, not the 3-
manifold. Similarly, by stabilizing the Heegaard decomposition by tak-
ing the connected sum of the original diagram with the 2-torus T
2
and
, as shown by Figure 14.4 does not change the 3-manifold. In fact,
Figure 14.4. A cancelling pair of - and -curves

it can be shown that these moves are all, more precisely if two diagrams
represent dieomorphic 3-manifolds then one diagram can be transformed
into the other by a nite sequence of isotopies, handle slides and stabiliza-
tions/destabilizations [135]. This observation can be used to show that a
quantity dened for a Heegaard diagram is, in fact, an invariant of the cor-
responding 3-manifold: one only has to check that it does not change under
the moves listed above. See also Remark 14.2.3.
It is a little more complicated to present 4-manifolds in a similar fashion.
First of all notice that a 4-dimensional cobordism W from Y
1
to Y
2
can be
decomposed as a sequence of attaching 1-, 2- and 3-handles. By assuming
orientability of W, the gluing of 1-handles (and so of 3-handles) is essentially
unique, and so we only need to deal with 2-handle attachments, where all
14.2. Heegaard Floer theory for 3- and 4-manifolds 239
the interesting topology happens. Suppose that K Y
1
is a framed knot,
W is the cobordism given by the 2-handle attachment along K with the
given framing and consider a Heegaard diagram
_
g
,
i
g
i=1
,
j
g1
j=1
_
for
Y
1
K. (This can be given by implementing the algorithm described
above.) Let
j
=
j
for j = 1, . . . , g 1,
g
= meridian of K and
g
=the
curve representing the framing of K xed before. Then the Heegaard
diagrams
_
g
,
i
g
i=1
,
j
g
j=1
_
,
_
g
,
i
g
i=1
,
j
g
j=1
_
represent Y
1
and Y
2
.
Exercise 14.1.2. Verify that the Heegaard diagram
_
g
,
i
g
i=1
,
j
g
j=1
_
denes #
g1
(S
1
S
2
). (Hint: Displace
j
by an isotopy to make it disjoint
from
j
(j = 1, . . . , g 1). Destabilize (
g
,
g
) and use induction.)
Therefore, the 4-manifold X we get by attaching [0, 1] U
, [0, 1] U
and
[0, 1] U
to
g
solid triangle C along the sides I
g
, has three
boundary components: Y
1
, Y
2
and #
g1
(S
1
S
2
). (Here U
, U
and U
stand for the handlebodies dened by the corresponding sets of curves.)

Now lling the boundary component #
g1
(S
1
S
2
) with
g1
(S
1
D
3
) we
get a cobordism from Y
1
to Y
2
, which can be proved to be dieomorphic
to the given cobordism W we started with. Therefore the cobordism W
built on Y by attaching a 2-handle along K Y can be represented by the
Heegaard triple
_
g
,
i
g
i=1
,
j
g
j=1
,
k
g
k=1
_
,
where
_
g
,
i
g
i=1
,
j
g1
j=1
_
is a Heegaard diagram for Y K and
k
g
k=1
is given from
k
g
k=1
, the surgery curve K and the framing as described
above.
14.2. Heegaard Floer theory for 3- and 4-manifolds
Let Y be a given closed, oriented 3-manifold and x a Heegaard diagram
_
g
,
i
g
i=1
,
j
g
j=1
_
for Y . Without loss of generality we can assume that
each
i
intersects each
j
transversely. Let us consider the tori
T
=
1
. . .
g
, T
=
1
. . .
g
in the g-fold symmetric power Sym
g
(
g
). This symmetric power (which
is a smooth manifold of dimension 2g) can be equipped with a symplectic
structure and the Floer homology group

HF(Y ) is supposed to measure
how the above two (totally real) tori intersect each other in the symplectic
sense. More precisely, dene

CF(Y ) as the free Abelian group generated
by the intersection points T
. (It is easy to see that since the curves

i
and
j
intersect transversely, so do the tori T
and T
.) We consider two
intersection points to be removable if there is a holomorphic Whitney
disk showing how to get rid of them. More formally, x an -tame almost-
complex structure J on Sym
g
(
g
) and dene a dierential :

CF(Y )
CF(Y ) as follows: for x, y

CF(Y ) the matrix element x, y) counts the
J-holomorphic maps u:
2
Sym
g
(
g
) from the unit disk
2
C (up to
reparametrization) with
u(i) = x, u(i) = y,
u(z) T
if z
2
and e z < 0,
u(z) T
if z
2
and e z > 0.
In order to get a sensitive invariant, we need to choose a base point z
0

g
(
i

i
j
) and require u(
2
) z
0
Sym
g1
(
g
) = , that is,
the holomorphic disk should avoid the divisor z
0
Sym
g1
(
g
) dened
by the base point. If the space of these maps (up to reparametrization) is
not 0dimensional, we dene x, y) to be zero, otherwise
x, y) =
#
_
holomorphic disks from x to y disjoint from z
0
Sym
g1
(
g
)
_
.
Remark 14.2.1. Using a delicate construction (and xing some auxiliary
data) a sign can be attached to any map of the above type in a 0dimensional
space, and in the denition of x, y) we count the holomorphic maps with
those signs. Alternatively, we can use Z
2
-coecients, which turns out to
be sucient for our present purposes, therefore we will always restrict our
attention to this special case.
The complex
_
CF(Y ),
_
splits as a sum
tSpin
c
(Y )
_
CF(Y, t),
_
of sub-
complexes: an intersection point x T
and the xed base point

z
0

g
naturally determines a spin
c
structure s
z
0
(x) Spin
c
(Y ) in the
following way: Suppose that the Heegaard diagram is induced by a Morse
function f : Y R and x a Riemannian metric g
0
on Y . Then an in-
tersection point x T
can be regarded as a choice of gradient lines

for f (with respect to g
0
) connecting index-2 and index-1 critical points of
f: choose those gradient ow lines which pass through the coordinates of
x = (x
1
, . . . , x
g
) T
in
g
. The base point z
0
species a gradient
line connecting the minimum and maximum of f, therefore on the comple-
ment of the neighborhood of these paths the gradient f denes a nowhere
vanishing vector eld. Since along any of these paths the indices of the
critical points have opposite parity, the resulting vector eld extends to Y ,
giving rise to a well-dened spin
c
structure on the 3-manifold. It is not very
hard to verify that there is a topological Whitney disk connecting x and
y if and only if s
z
0
(x) = s
z
0
(y). Using Gromovs compactness theorem it
can be shown that for generic choices
2
= 0, hence the Floer homology
HF(Y, t) = H
CF(Y, t),
_
can be dened for all t Spin
c
(Y ).
Theorem 14.2.2 (OzsvathSzabo, [136]). Let Y be a given closed oriented
3-manifold equipped with a spin
c
structure t Spin
c
(Y ). The Ozsvath
Szab o homology group

HF(Y, t) is a topological invariant of the spin
c
3-
manifold (Y, t).
Remarks 14.2.3. (a) In the proof of the above theorem one needs to
show that

HF(Y, t) is independent of the chosen Heegaard decomposition,
almost-complex structure J on Sym
g
(
g
) and base point z
0

g
. The in-
dependence from the chosen almost-complex structure is essentially built in
the denition: it is a general feature of Floer homology groups associated to
intersecting Lagrangian submanifolds in symplectic manifolds. (Although
T
and T
are not Lagrangian in Sym

g
(), the general theory still ap-
plies because of special features of this particular case.) The independence
from the Heegaard decomposition requires to show that the groups do not
change under isotopies, handle slides and stabilization. The independence
of isotopies is again a consequence of some general facts regarding Floer ho-
mologies: any isotopy can be decomposed into a Hamiltonian isotopy and
another one which can be represented by the change of the almost-complex
structure on
g
. By a good choice of the base point, independence from
stabilization is a fairly easy exercise, while handle slide invariance requires
to work out a special case and a way to implement this special case under
general circumstances. Finally, the change of base point can be reduced to a
sequence of handle slides. For the details of the arguments indicated above,
the reader is advised to turn to the original papers [135, 136].
(b) In the case b
1
(Y ) > 0 one also has to assume a certain admissibility
of the Heegaard diagram, which can always be achieved by appropriate
isotopies of the - and the -curves. This condition is needed for having
nite sums in the denition of the boundary operator and in the proof of
independence of choices. For details see [135].
Proposition 14.2.4. The set
_
t Spin
c
(Y ) [

HF(Y, t) ,= 0
_
is nite for
any 3-manifold Y . In particular, the vector space
HF(Y ) =
tSpin
c
(Y )
HF(Y, t)
is nite dimensional.
Proof. After xing an admissible Heegaard diagram, there are only nitely
many intersection points in T
, hence the chain complex

CF(Y ) is
nite dimensional, implying the result.
Examples 14.2.5. (a) Consider the lens space L(p, q). It admits a genus-
1 Heegaard decomposition with
1
and
1
intersecting each other in p
points. These points all correspond to dierent spin
c
structures, therefore
the boundary map vanishes for any t Spin
c
_
L(p, q)
_
, and so we have
that

HF(L(p, q), t) = Z
2
. In particular,

HF(S
3
) = Z
2
.
(b) It is not hard to see that

HF(Y, t)

=

HF(Y, t). If (Y, t) decomposes
as a connected sum (Y
1
, t
1
)#(Y
2
, t
2
) then
HF(Y, t) =

HF(Y
1
, t
2
)
Z
2
HF(Y
2
, t
2
).
(c) It can be shown that if t Spin
c
(Y ) is torsion, that is, c
1
(t) H
2
(Y ; Z)
is a torsion element, then

HF(Y, t) is nontrivial. In particular, if Y is a
rational homology sphere (i.e., b
1
(Y ) = 0) then

HF(Y, t) is nonzero for
all t Spin
c
(Y ). Since any 3-manifold admits torsion spin
c
structure, the
above nontriviality statement implies that

HF(Y ) ,= 0 for any 3-manifold Y .
(d) The 3-manifold S
1
S
2
admits a genus-1 Heegaard decomposition
with two parallel circles as - and -curves. This Heegaard decomposition,
however, is not admissible. The diagram of Figure 14.5 gives an admissible
Heegaard diagram for S
1
S
2
. By analyzing the possible holomorphic
disks we get that for the spin
c
structure t
0
with vanishing rst Chern
class

HF(S
1
S
2
, t
0
)

= Z
2
Z
2
holds, while for all other spin
c
structures
the OzsvathSzabo homology group vanishes. Consequently

HF
_
#
k
(S
1
S
2
), t
_
is zero unless c
1
(t) = 0, and for c
1
(t
0
) = 0 we have

HF
_
#
k
(S
1
S
2
), t
0
_

= Z
2
k
2
_

= H
(T
k
; Z
2
)
_
.
Figure 14.5. Admissible Heegaard diagram for S

1
S
2
As we saw in Proposition 14.2.4, the groups are nontrivial only for nitely
many spin
c
structures in Spin
c
(Y ). In fact, the particular geometry of Y
provides a constraint for the nontriviality of the OzsvathSzabo homology
groups:
Theorem 14.2.6 (Adjunction formula, [136]). Suppose that Y is an
oriented surface of genus g in the 3-manifold Y with g > 0. If

HF(Y, t) is
nontrivial for a spin
c
structure t then [
c
1
(t), []
_
[ (). If

= S
2
then

HF(Y, t) ,= 0 implies that
c
1
(t), []
_
= 0.
Similar ideas provide invariants for 4-dimensional manifolds. Suppose
that (W, s) is a spin
c
cobordism between (Y
1
, t
1
) and (Y
2
, t
2
). Standard
manifold topology implies that W can be decomposed as W = W
1
W
2
W
3
,
where W
i
can be built using 4-dimensional i-handles only (i = 1, 2, 3). A
homomorphism F
W,s
:

HF(Y
1
, t
1
)

HF(Y
2
, t
2
) can be given as follows (for
simplicity we drop the spin
c
stucture from the notation): dene F
W
as the
composition F
W
3
F
W
2
F
W
1
, where the homomorphisms F
W
1
and F
W
3
are
standard maps, since the cobordisms W
1
and W
3
depend only on Y
1
and
Y
2
, and the number of 1-handles (3-handles) involved in the cobordism. For
example, W
1
is a cobordism between Y
1
and Y
1
= Y
1
#
k
(S
1
S
2
) (where k
is the number of 1-handles in W
1
), and so F
W
1
,s
is a map
HF
_
Y
1
, s[
Y
1
_

HF
_
Y
1
, s[
Y
1
_
=

HF
_
Y
1
, s[
Y
1
_

HF
_
#
k
(S
1
S
2
), t
0
_
sending x

HF
_
Y
1
, s[
Y
1
_
to x
k
where
k
is the highest degree element
in

HF
_
#
k
(S
1
S
2
), t
0
_

= H
(T
k
; Z
2
). Similar formula describes F
W
3
. The
cobordism W
2
, on the other hand, can be presented by a Heegaard triple,
i.e., three g-tuples of curves , and as we discussed it in the preceding
section. Counting specic holomorphic triangles in Sym
g
(
g
) with appro-
priate boundary conditions (in a similar spirit as was dened) we get
F
W
2
. As before, a long and tedious proof shows that F
W,s
is independent of
the choices made (i.e., the decomposition of W, the chosen almost-complex
structure, the base point, etc.).
Theorem 14.2.7 (OzsvathSzabo, [137]). The resulting map F
W,s
depends
only on the oriented 4-dimensional spin
c
cobordism (W, s) and is indepen-
dent of the choices made throughout the denition.
Once again, F
W,s
,= 0 holds only for nitely many spin
c
structures s
Spin
c
(W), moreover Theorem 14.2.6 can be used to show
Theorem 14.2.8 (Adjunction formula, [136]). If W is a closed,
oriented, embedded surface with 0 2g() 2 < []
2
+ [
c
1
(s), []
_
[ or
with g() = 0 and []
2
0 then F
W,s
= 0.
(For the detailed proof of a special case of this theorem see [101].) Simi-
lar ideas result a variety of OzsvathSzabo invariants of closed (oriented)
3-manifolds and oriented cobordisms between them. For the detailed dis-
cussion of these variants of the theory we advise the reader to turn to
[135, 136, 137].
14.3. Surgery triangles
homologies lies in the fact that there is a scheme for computing them
once the 3-manifold is given by a surgery diagram. The key step in such
computations is the application of an appropriate surgery exact sequence,
which relates OzsvathSzabo homologies of three 3-manifolds we get by
doing surgeries on some knots. As we will see, the scheme does not produce
the OzsvathSzabo homology group of the 3-manifold given by surgery
directly, but rather gives it as part of several exact sequences. In addition,
maps in the sequences are usually induced by cobordisms, hence exactness
provides information about the maps as well. Below we give the most
important surgery exact sequence proved for the

HF-theory.
To state the theorems, let us assume that Y is a given 3-manifold with
a knot K Y in it. Fix a framing f on K and suppose that Y
1
is the
result of an integral surgery on K with the given framing. Let X
1
denote
14.3. Surgery triangles 245
the resulting cobordism. Suppose that Y
2
is the result of a surgery along K
with framing we get by adding a right twist to the framing f xed on K.
Equivalently, Y
2
can be given by doing surgery on K with framing f and
(1)-surgery on a normal circle N to K. This alternative viewpoint also
provides a cobordism X
2
from Y
1
to Y
2
given by the second surgery. Let t
be a xed spin
c
structure on Y K, and let t(Y ), t(Y
1
) and t(Y
2
) denote
the set of extensions of t to Y, Y
1
and Y
2
, resp. Denote the homomorphism
HF
_
Y, t(Y )
_

HF
_
Y
1
, t(Y
1
)
_
induced by the cobordism given by the
rst surgery on K by F
1
. Here F
1
is the sum of F
X
1
,s
for all spin
c
structure
s Spin
c
(X
1
) extending elements of t(Y ) and t(Y
1
). We dene F
2
for the
cobordism X
2
in a similar fashion.
Exercise 14.3.1. Perform a surgery along a (1)-framed normal circle N
to N Y
1
and denote the resulting cobordism from Y
2
by X
3
. Show that
X
3
is a cobordism from Y
2
to Y . (Hint: Blow down N
and put a dot on

the image of N. Finally cancel the resulting 1-handle/2-handle pair, see
Figure 14.6.)
K K K
n
n n
N
N
0
1
1
Figure 14.6. Identication of a 3-manifold in the surgery exact triangle
Let F
3
denote the homomorphism

HF
_
Y
2
, t(Y
2
)
_

HF(Y, t) induced by
the cobordism X
3
given by the 2-handle attachment along N
as it is given in
Exercise 14.3.1. Consider the triangle of cobordisms as given by Figure 14.7.
Theorem 14.3.2 (Surgery exact triangle, OzsvathSzabo [136]). Under
the above circumstances the surgery triangle induces an exact triangle
HF
_
Y, t(Y )
_

HF
_
Y
1
, t(Y
1
)
_
HF
_
Y
2
, t(Y
2
)
_
F
1
F
3
F
2
for the corresponding homology groups.
Y
1
Y
2
X
1
X
3
X
2
< > n < > n
K
Y
K
n
N
N
1 <1>
K
1
N
Figure 14.7. Cobordisms in the surgery exact triangle
Remark 14.3.3. With Z
2
-coecients the map F
W
induced by a cobordism
W is simply the sum
F
W,s
for all spin
c
structures extending the xed
ones on the boundaries of W. With Z-coecients, however, signs have to
be attached to the various maps F
W,s
for exactness to hold. For a complete
argument see [136].
By summing over all spin
c
structures on Y K and denoting
tSpin
c
(Y )
HF(Y, t)
by

HF(Y ) as usual, we get
Corollary 14.3.4. The triangle
HF(Y )

HF(Y
1
)
HF(Y
2
)
F
1
F
3
F
2
induced by the surgery triangle of Figure 14.7 is exact.
14.3. Surgery triangles 247
As an example, we show
Proposition 14.3.5. Suppose that the 4-ball genus of the knot K S
3
is
equal to g
s
. Then for n 2g
s
1 > 0
HF
_
S
3
n
(K)
_

=

HF
_
S
3
2gs1
(K)
_
Z
n2gs+1
2
.
Proof. For n = 2g
s
1 the proposition obviously holds. The general case
now follows by induction. To see this, consider the surgery triangle for
Y = S
3
, and knot K with framing n:
HF(S
3
)

HF
_
S
3
n
(K)
_
HF
_
S
3
n+1
(K)
_
F
1
F
3
F
2
Since the rst cobordism contains a surface of genus g
s
with square n,
the adjunction formula of Theorem 14.2.8 implies that F
1
= 0, hence
HF
_
S
3
n+1
(K)
_

=

HF
_
S
3
n
(K)
_
Z
2
, concluding the proof.
To see a more complicated example, suppose that Y bers over S
1
with ber
F of genus 2, and consider the canonical spin
c
structure t
can
Spin
c
(Y )
induced by the oriented 2-plane eld formed by the tangencies of the bers
of Y S
1
. Obviously
c
1
(t
can
), [F]
_
= (F). The surgery exact triangle
and the adjunction formula together imply
Proposition 14.3.6. Under the above circumstances

HF(Y, t
can
)

=
Z
2
Z
2
.
The proof of the proposition involves two lemmas, only one of which will be
proved below.
Lemma 14.3.7. If Y
1
, Y
2
both ber over S
1
with equal ber genus 2
then for the canonical spin
c
structures t
i
Spin
c
(Y
i
) we have

HF(Y
1
, t
1
)
HF(Y
2
, t
2
).
Proof (sketch). Let m
i
be the monodromy of the bration Y
i
S
1
(i = 1, 2), and factor m
1
m
1
2
into the product of k right-handed Dehn twists
along homologically nontrivial simple closed curves. This factorization gives
rise to a Lefschetz bration over the annulus, which is a cobordism between
Y
1
and Y
2
. The proof will proceed by induction on k. By composing the
cobordisms it is enough to deal with the case of k = 1. In that case we get
Y
2
from Y
1
by doing surgery on the vanishing cycle of the singular ber of
the Lefschetz bration over the annulus. Writing down the surgery triangle
for that surgery, we get
HF
_
Y
1
, t(Y
1
)
_

HF
_
Y
2
, t(Y
2
)
_
HF
_
Y
0
, t(Y
0
)
_
F
1
F
3
F
2
The third group vanishes by the adjunction formula of Theorem 14.2.6:
Since Y
0
is the result of a surgery along the vanishing cycle with coecient
0 relative to the framing induced by the ber, it contains a surface of
genus (g 1) in the homology class of the (old) ber. Therefore F
1
is
an isomorphism and it is not hard to see that the nonzero terms belong to
the canonical spin
c
structures t
i
.
Now the following lemma (which we give without proof) concludes the
argument for Proposition 14.3.6.
Lemma 14.3.8. For S
1

g
with the canonical spin
c
structure t
can
we
have

HF(S
1
g
, t
can
) = Z
2
Z
2
.
We close this section with an observation which will be useful in our ap-
plications. A rational homology sphere Y is called an L-space if

HF(Y ) =
tSpin
c
(Y )
HF(Y, t) has dimension
H
1
(Y, Z)
. Since for a rational ho-

mology sphere

HF(Y, t) never vanishes, being an L-space is equivalent to
HF(Y, t) = Z
2
for all spin
c
structures t Spin
c
(Y ). For example, lens
spaces are all L-spaces. As an application of the above surgery exact trian-
gles, we show a useful criterion for being an L-space.
Proposition 14.3.9. Suppose that K S
3
is a knot of 4-ball genus g
s
> 0.
If there is n > 0 such that S
3
n
(K) is an L-space then all S
3
m
(K) with
m min (2g
s
1, n) is an L-space.
Proof. Recall from Proposition 14.3.5 that if S
3
n
(K) is an L-space and
n 2g
s
1 then S
3
2gs1
(K) is also an L-space. (Use the fact that
[H
1
_
S
3
n
(K); Z
_
[ = [n[ for all n ,= 0.) In addition, by applying the surgery
exact sequence for Y = S
3
, the knot K and framing m it is easy to see
that if S
3
m
(K) is an L-space then so is S
3
m+1
(K) (m 1). This observation
concludes the proof.
14.4. Contact OzsvathSzabo invariants 249
Example 14.3.10. If K denotes the right-handed trefoil knot then S
3
n
(K)
is an L-space for all n 1. This can be seen by the computation of
Proposition 14.3.5 together with the fact that S
3
5
(K) is a lens space.
Exercises 14.3.11. (a) Extend Proposition 14.3.9 to all rational m with
m min(2g
s
1, n). (Hint: Cf. [101].)
(b) Using the fact that

HF(Y ) ,= 0 holds for any 3-manifold (cf. Exam-
ple 14.2.5(c)) verify that with K denoting the right-handed trefoil knot,
HF
_
S
3
0
(K)
_
= Z
2
2
holds. (Hint: Use the surgery exact triangle and the
fact that

HF
_
S
3
1
(K)
_
= Z
2
.)
(c) Let Y
n
denote the circle bundle over the torus T
2
with Euler number
n > 0. Show that

HF(Y
n
) = Z
4n
2
. (Hint: Apply the surgery exact triangle
induced by the cobordisms of Figure 14.8. Find a torus of self-intersection
n in the coboridsm X and use induction on n. Find another triangle to
handle the case of n = 1.)
14.4. Contact Ozsv athSzab o invariants
One of the main applications of Heegaard Floer theory is in contact topol-
ogy. Contact OzsvathSzabo invariants can be fruitfully applied in deter-
mining tightness of structures given by contact surgery diagrams, hence
these invariants t perfectly in the main theme of the present notes. The
denition of the invariant of a contact structure given by Ozsvath and Szab o
is based on a compatible open book decomposition with connected binding.
According to Girouxs result discussed earlier, well-denedness of such an
invariant requires the verication that the quantity does not change under
positive stabilization. The construction of Ozsvath and Szab o goes in the
following way: Suppose that a compatible open book decomposition with
connected binding is xed on (Y, ). Then 0-surgery on the binding of this
open book decomposition produces a bered 3-manifold Y
B
and a cobordism
W between Y and Y
B
. Notice that the contact structure induces a spin
c
structure t
on Y , and Y
B
admits a natural spin
c
structure t
can
induced by
the oriented 2-plane eld tangent to the bers.
Exercise 14.4.1. Show that W admits a unique spin
c
structure s such that
s[
Y
= t
and s[
Y
B
= t
can
.
n
0
0
0
0
n+1
X
0
0
Figure 14.8. 3-manifolds in a particular surgery triangle
Turning W upside down to get W, we have a map
F
W,s
:

HF(Y
B
, t
can
)

HF(Y, t
),
and

HF(Y
B
, t
can
) has been computed to be isomorphic to Z
2
Z
2
. By
making use of the corresponding homology theory HF
+
, a nontrivial el-
ement h

HF(Y
B
, t
can
) can be distinguished: There is a long ex-
act sequence connecting the related theories HF
+
(Y, t) and

HF(Y, t) for
any spin
c
3-manifold (Y, t), and for a bered 3-manifold Y and t = t
can
we have (similarly to Proposition 14.3.6) that HF
+
(Y, t
can
) = Z
2
. Now
h

HF(Y
B
, t
can
) is the element mapping to the nontrivial element in
HF
+
(Y
B
, t
can
).
Denition 14.4.2. The contact Ozsv athSzab o invariant c(Y, ) of the
contact structure (Y, ) is dened to be equal to F
W,s
(h)

HF(Y, t
).
The fundamental theorem concerning c(Y, ) is
Theorem 14.4.3 (OzsvathSzabo, [140]). The OzsvathSzabo homology
element c(Y, )

HF(Y, t
) does not depend on the chosen compatible

open book decomposition, hence is an invariant of the isotopy class of the
contact 3-manifold (Y, ).
Remark 14.4.4. The denition given in [140] involves the OzsvathSzabo
knot invariant of the binding of a compatible open book decomposition
using this denition Ozsvath and Szab o veries independence of the open
book decomposition and then proves that the two denitions (one relying on
surgery along the binding and the one originating from the knot invariants)
are the same. Since we will not make any use of the knot invariants, we do
not discuss the details of the denition here.
The main properties of the invariant c(Y, ) are summarized in the following
statements
Theorem 14.4.5 (OzsvathSzabo, [140]; cf. also [100]). If (Y
K
,
K
) is given
as contact (+1)-surgery along the Legendrian knot K (Y, ) and W is the
corresponding cobordism then by reversing the orientation on W and using
the resulting cobordism W we get
F
W
_
c(Y, )
_
= c
_
Y (K), (K)
_
.
Again, F
W
stands for the sum
F
W,s
for all spin
c
structures on W.
Proof. The proof below is an adaptation of [140, Theorem 4.2], cf. also [100].
Present (Y, ) by a contact (1)-surgery diagram along the Legendrian link
L (S
3
,
st
) and add K to L. Applying the algorithm of Akbulut and
the rst author [7] for L K we get an open book decomposition of Y
compatible with such that K lies on a page of it. Denote the results of
the 0-surgeries along the bindings on Y and Y (K) with Y
B
and
_
Y (K)
_
B
respectively. The cobordism W
B
of the handle attachment along the knot
K gives rise to a map F
W
B
:

HF(Y
B
)

HF(
_
Y (K)
_
B
), which
ts into an exact triangle of the type encountered in Lemma 14.3.7. The
same argument now provides that F
W
B
is an isomorphism, resulting in a
commutative diagram
HF(Y
B
)

HF(
_
Y (K)
_
B
)
HF(Y )

HF
_
Y (K)
_
F
W
B
=
F
W
F
W
Y
F
Y (K)
Since the distinguished generator h
Y

HF(Y
B
) maps to the distinguished
generator h
Y (K)

HF(
_
Y (K)
_
B
), the statement of the theorem follows
from the commutativity of the above diagram and the denition of the
contact invariant.
Example 14.4.6. The contact OzsvathSzabo invariant of the overtwisted
structure (S
3
,
) depicted by Figure 11.3 vanishes. This can be veried by

applying the above principle for (S
3
,
st
) and K as in Figure 11.3. The co-
bordism W inducing the map F
W
with the property F
W
_
c(S
3
,
st
)
_
=
c(S
3
,
) contains a sphere of self-intersection (+1), hence F

W
= 0, there-
fore c(S
3
,
) = 0 as claimed.
This example can be generalized as
Theorem 14.4.7 (OzsvathSzabo, [140]). If (Y, ) is overtwisted then
c(Y, ) = 0.
Proof. Consider the oriented 2-plane eld
1
on Y with the property that
the oriented 2-plane eld (Y,
1
)#(S
3
,
) is homotopic to the oriented 2-

plane eld induced by (Y, ). (Here
is the oriented 2-plane eld induced by

the contact structure of Example 14.4.6.) By the classication of overtwisted
contact structures, there is a contact structure representing the oriented 2-
plane eld
1
. Consequently, the above argument shows that contact (+1)-
surgery along the knot of Figure 11.3 located in a Darboux chart of some
contact structure
1
on Y provides an overtwisted structure homotopic,
hence isotopic to (Y, ). Therefore c(Y, ) can be given as F
W
_
c(Y,
1
)
_
.
Since W contains a 2-sphere of self-intersection (+1), the adjunction
formula provides F
W
= 0 and therefore c(Y, ) = 0.
Corollary 14.4.8. If c(Y, ) ,= 0 for (Y, ) and (Y
K
,
K
) is given as contact
(1)-surgery along the Legendrian knot K (Y, ) then c(Y
K
,
K
) ,= 0,
therefore it is tight.
Proof. Let K
be a Legendrian push o of K in (Y, ), giving rise to a

Legendrian knot (also denoted by K
) in (Y
K
,
K
). By the Cancellation
Lemma 11.2.6, contact (+1)-surgery on K
gives (Y, ) back, therefore The-

orem 14.4.5 shows that for the cobordism W of the contact (+1)-surgery
we have F
W
_
c(Y
K
,
K
)
_
= c(Y, ). If c(Y, ) ,= 0, then this shows that
c(Y
K
,
K
) ,= 0. In the light of Theorem 14.4.7 this implies tightness of
(Y
K
,
K
).
Proposition 14.4.9 (OzsvathSzabo, [140]). For (S
3
,
st
) the contact
invariant c(S
3
,
st
) generates

HF(S
3
) = Z
2
.
Proof (sketch). Recall that (S
3
,
st
) admits an open book decomposition
with the unknot as binding. Therefore the map dening the invariant ts
into the exact triangle
HF(S
1
S
2
)

HF(S
3
)
HF(S
3
)
F
G
Since we know that

HF(S
3
) = Z
2
and

HF(S
1
S
2
) = Z
2
Z
2
, it follows that
G = 0, and F is onto. Now by using a certain grading on OzsvathSzabo
homologies (cf. [138]) it is not hard to see that the element h

HF(S
1
S
2
)
used in the denition of the contact invariant maps into the nonzero element
of

HF(S
3
), concluding the proof.
Lemma 14.4.10. Consider the contact structure
k
on #
k
(S
1
S
2
) given by
contact (+1)-surgery on the k-component Legendrian unlink. The contact
invariant c
_
#
k
(S
1
S
2
),
k
_
does not vanish.
Proof. The lemma will be proved by induction on k. For k = 0 we
have the standard contact 3-sphere (S
3
,
st
) which has nonzero invariant by
Proposition 14.4.9. By denition,
k
is given as contact (+1)-surgery along a
knot in
k1
, therefore F
W
(c
_
#
k1
(S
1
S
2
),
k1
_
) = c
_
#
k
(S
1
S
2
),
k
_
for the cobordism we get by the handle attachment. Therefore the injectivity
of F
W
immediately provides the result. Now writing down the surgery
exact triangle for the above handle attachment, for the OzsvathSzabo
homology groups we get
HF
_
#
k1
(S
1
S
2
)
_

HF
_
#
k
(S
1
S
2
)
_
HF
_
#
k1
(S
1
S
2
)
_
F
W
Since dim
Z
2
HF
_
#
k
(S
1
S
2
)
_
= 2
k
, injectivity of F
W
follows from exact-
ness and simple dimension count.
Notice that the nonvanishing of the contact invariant shows that the con-
tact 3-manifold (#
k
S
1
S
2
,
k
) is tight. It is known that #
k
(S
1
S
2
)
carries a unique isotopy class of tight contact structures, which is the Stein
llable boundary of D
4
k 1-handles. In conclusion, (+1)-surgery on the
k-component Legendrian unlink produces a contact 3-manifold contacto-
morphic to the boundary of the Stein surface we get by attaching k 1-handles
to D
4
, cf. Exercise 11.2.7.
Exercise 14.4.11. Show that if (Y, ) is a Stein llable contact 3-manifold
then c(Y, ) ,= 0. (Hint: Recall that any Stein llable contact structure can
be given as Legendrian surgery along a link in
_
#
k
(S
1
S
2
),
k
_
for some
k. Use Lemma 14.4.10 and Corollary 14.4.8.)
Making use of the Embedding Theorem 12.1.7 of weak symplectic llings
and the nonvanishing of the mixed OzsvathSzabo invariants for closed
symplectic 4-manifolds [137, 141], the above exercise was generalized for a
version of contact invariants in some twisted coecient system as follows:
Proposition 14.4.12 (OzsvathSzabo, [143]). If (Y, ) is a weakly sym-
plectically llable contact 3-manifold then by using an appropriate twisted
coecient system the contact invariant c(Y, ) does not vanish.
15. Appendix: Mapping class groups
In this appendix we summarize some basic facts regarding algebraic prop-
erties of mapping class groups. After discussing the presentation of these
groups we recall the equivalence between certain words in some mapping
class groups and geometric structures discussed in earlier chapters. We
close this chapter with some theorems making use of those connections.
15.1. Short introduction
Let
n
g,r
denote an oriented, connected genus-g surface with n marked points
and r boundary components.
Denition 15.1.1. The mapping class group
n
g,r
is dened as the quotient
of the group of orientation preserving self-dieomorphisms of
n
g,r
(xing
marked points and boundaries pointwise) by isotopies (xing marked points
and boundaries pointwise). For n = 0 (r = 0, resp.) we use the notation
g,r
(
n
g
, resp.), and in case n = r = 0 we write
g
. For F =
g,r
we will
denote
g,r
by
F
.
Simple closed curves in the surface give rise to special mapping classes:
Denition 15.1.2. A right-handed Dehn twist t
a
:
n
g,r

n
g,r
on an em-
bedded simple closed curve a in an oriented surface
n
g,r
is a dieomorphism
obtained by cutting
n
g,r
along a, twisting 360
to the right and regluing.

More formally, we identify a regular neighborhood a of a in
n
g,r
with
S
1
I, set t
a
(, t) = ( +2t, t) on a and smoothly glue into id
n
g,r
a
. A
left-handed Dehn twist is the inverse of a right-handed Dehn twist.
256 15. Appendix: Mapping class groups
Remark 15.1.3. Notice that in the denition of the Dehn twist along a
curve a we do not need to orient a even though the surface
n
g,r
has to be
oriented.
It is well-known that Dehn twists generate
n
g,r
in fact we can choose
a nite (fairly simple) set of generators, see [171] and Theorem 15.1.12.
First we discuss relations which hold in
n
g,r
. In the following we will use
the usual functional notation for products in
n
g,r
.
Lemma 15.1.4. If f :
n
g,r

n
g,r
is an orientation preserving dieomor-
phism and a
n
g,r
is a simple closed curve then ft
a
f
1
= t
f(a)
.
Proof. Let a
= f(a). Since f maps a to a
we can assume that (up to

isotopy) it also maps a neighborhood N of a to a neighborhood N
of a
.
Let us examine the eect of applying ft
a
f
1
. The homeomorphism f
1
takes N
to N, then t
a
maps N to N, twisting along a, and nally f takes
N back to N
. Since t
a
is supported in N, the composite map is supported
in N
and is a Dehn twist about a
.
We say that a simple closed curve a
n
g,r
is separating if
n
g,r
a has two
connected components otherwise a is called nonseparating. Lemma 15.1.4
together with the classication of 2-manifolds provides
Lemma 15.1.5. Suppose that a and b are nonseparating simple closed
curves in
n
g,r
. Then there is an orientation preserving dieomorphism
f :
n
g,r

n
g,r
which takes a to b. Consequently t
a
and t
b
are conjugate
in
n
g,r
. In particular, if a and b are homologically essential simple closed
curves in a surface with at most one boundary component then t
a
and t
b
are conjugate.
Exercise 15.1.6. Verify that if a intersects b transversely in a unique point
then t
a
t
b
(a) = b. (Hint: Use Figure 15.1.)
a
b
t
a
t (a)
b
t (a)
b
b
Figure 15.1. An identity for right-handed Dehn twists
15.1. Short introduction 257
Lemma 15.1.7. If a, b
n
g,r
are disjoint then t
a
t
b
= t
b
t
a
. If a intersects
b in a unique point then t
a
t
b
t
a
= t
b
t
a
t
b
.
Proof. The commutativity relation t
a
t
b
= t
b
t
a
is obvious. To prove the braid
relation t
a
t
b
t
a
= t
b
t
a
t
b
we observe that t
a
t
b
(a) = b (see Exercise 15.1.6). By
Lemma 15.1.4 we get t
a
t
b
t
a
= t
a
t
b
t
a
t
1
b
t
1
a
t
a
t
b
= t
tat
b
(a)
t
a
t
b
= t
b
t
a
t
b
.
Lemma 15.1.8. Let a
1
, a
2
, , a
k
be a chain of curves, i.e., the consecutive
curves intersect once and nonconsecutive curves are disjoint. Let N denote
a regular neighborhood of the union of these curves. Then the following
relations hold:
The commutativity relation: t
a
i
t
a
j
= t
a
j
t
a
i
if [i j[ > 1.
The braid relation: t
a
i
t
a
j
t
a
i
= t
a
j
t
a
i
t
a
j
if [i j[ = 1.
The chain relation: If k is odd then N has two boundary components
d
1
and d
2
, and (t
a
1
t
a
2
t
a
k
)
k+1
= t
d
1
t
d
2
. If k is even then N has one
boundary component d and (t
a
1
t
a
2
t
a
k
)
2k+2
= t
d
.
d
2
1
d
a
a
a
3
2
1
Figure 15.2. Chain relation for k = 3: (ta
1
ta
2
ta
3
)
4
= t
d
1
t
d
2
The next lemma was rst observed by Dehn and then rediscovered by
Johnson [78] who called it the lantern relation.
Lemma 15.1.9. Let U be a disk with the outer boundary a and with 3
inner holes bounded by the curves a
1
, a
2
, a
3
. For 1 i 3, let b
i
be the
simple closed curve in U depicted in Figure 15.3. Then the lantern relation
t
a
t
a
1
t
a
2
t
a
3
= t
b
1
t
b
2
t
b
3
holds.
3
3
2
2
1
1 a
b
a
b
a
b
a
U
Figure 15.3. The lantern relation
Lemma 15.1.10. If i denotes the hyperelliptic involution (i.e., rotation of
the standard embedded
g
R
3
by 180
around the x-axis, see Figure 15.4)

and a is a curve in
g
xy plane then [i, t
a
] = 1.
Remark 15.1.11. The idea of the proofs of Lemmas 15.1.8, 15.1.9 and
15.1.10 is the following: We split the surface into a union of disks by cutting
along a nite number of simple closed curves and properly embedded arcs.
We prove that the given product of Dehn twists takes each one of these
curves (arcs, resp.) onto an isotopic curve (arc, resp.). Then the product is
isotopic to a homeomorphism pointwise xed on each curve and arc. But
Alexanders lemma says that a homeomorphism of a disk xing its boundary
is isotopic to the identity, relative to boundary. Thus the given product is
isotopic to the identity.
Now a presentation of
g
(and
g,1
) can be given using the relations
described above. It turns out that the mapping class groups
g
and
g,1
are
Figure 15.4. The hyperelliptic involution i
generated by t
a
0
, . . . , t
a
2g
with curves a
0
, . . . , a
2g
depicted in Figure 15.5.
Let A
ij
= [t
a
i
, t
a
j
] for all pairs (i, j) with a
i
a
j
= . Let B
i
denote
a
2g+1 a
1
a
g 2
a
0
a a
a
2
3 5
a
4
a
6
Figure 15.5. The simple closed curves inducing a generating system
the braid relation t
a
i
t
a
i+1
t
a
i
t
1
a
i+1
t
1
a
i
t
1
a
i+1
for i = 1, . . . , 2g 1 and B
0
=
t
a
0
t
a
4
t
a
0
t
1
a
4
t
1
a
0
t
1
a
4
. Finally C, D and E = [i, t
a
2g+1
] denote the appropriate
chain, lantern and hyperelliptic relations, cf. Figures 15.6 and 15.7. Notice
that there are a number of relations of type A and B, but the relations
C, D and E are unique (as shown by the gures). Write all these relations
in terms of the generators t
a
0
, . . . , t
a
2g
and consider the normally generated
subgroups R
1
= A
ij
, B
i
, C, D)
No
and R = A
ij
, B
i
, C, D, E)
No
in the free
group F
2g+1
on 2g + 1 letters corresponding to the generators t
a
0
, . . . , t
a
2g
.
Now the presentation theorem of Wajnryb (see also [81]) reads as follows:
Theorem 15.1.12 (Wajnryb, [171]). For g 3 the sequences
1 R F
2g+1

g
1 and
1 R
1
F
2g+1

g,1
1
are exact; in other words, the above generators and relations provide a
presentation of
g
and
g,1
.
a
2
a
3
a
0
a
a
1
Figure 15.6. The chain relation in the presentation
a
1
a a
3 5
a
Figure 15.7. The a-curves in the lantern relation of the presentation
Remark 15.1.13. For g = 2 omit the lantern relation to get the correct
result. If we denote t
a
i
by a
i
for simplicity, an alternative presentation of
2
can be given by generators a
1
, a
2
, a
3
, a
4
, a
5
, the braid and commutativity
relations for them (i.e., a
i
a
i+1
a
i
= a
i+1
a
i
a
i+1
and a
i
a
j
= a
j
a
i
for [ij[ 2),
requiring that i = a
1
a
2
a
3
a
4
a
2
5
a
4
a
3
a
2
a
1
is central, i
2
= 1, and nally that
(a
1
a
2
a
3
a
4
a
5
)
2
= 1.
Next we would like to discuss two exact sequences relating various map-
ping class groups. By collapsing a boundary component to a point (or gluing
a disk with marked center to a boundary component) we get an obviously
surjective map
n
g,r

n+1
g,r1
. It is easy to see that the Dehn twist = t
along a curve parallel to the boundary we collapsed becomes trivial. In

fact,
1 Z
n
g,r

n+1
g,r1
1
turns out to be an exact sequence, where Z is generated by t
. Forgetting
the marked point we get a map
n
g,r

n1
g,r
, and now the sequence
1
1
(
n1
g,r
)
n
g,r

n1
g,r
1
is exact (here
1
(
n1
g,r
) is the fundamental group of the (n 1)-punctured
surface with r boundary components). Using these exact sequences, pre-
sentations for all
n
g,r
can be derived by starting with Wajnrybs result and
knowing presentations for the kernels in the above short exact sequences;
for such results see [58]. It follows that
n
g,r
is generated by nitely many
nonseparating Dehn twists plus Dehn twists along boundary-parallel curves.
In fact, if g 2 then for each boundary component of
n
g,r
we can embed a
lantern relation (as shown in Figure 15.9) in
n
g,r
in such a way that one of
the boundary curves in the lantern relation is mapped onto that boundary
component of
n
g,r
and all the other curves in the lantern relation are non-
separating in
n
g,r
. It follows that
n
g,r
is generated by nitely many Dehn
twists along nonseparating curves for g 2.
Proposition 15.1.14 (Powell, [147]). For g 3 the commutator subgroup
[
g
,
g
] is equal to
g
, i.e.,
g
is a perfect group.
Proof. Let a be any nonseparating curve on
g
. For g 3, there is an
embedding of a sphere with 4-holes (one of which is bounded by a) into
g
where all seven curves in the lantern relation
t
a
t
a
1
t
a
2
t
a
3
= t
b
1
t
b
2
t
b
3
a
a a
b
b b
a
2
1 3
2
3 1
Figure 15.8. Appropriate lantern relation involving a with a nonseparating
a
a
b
b
a
a
b
3
2
2
3
1
1
Figure 15.9. Appropriate lantern relation involving a with a separating
of Lemma 15.1.9 are nonseparating, see Figure 15.8. Since the a
i
s are
disjoint from the b
j
s we have
t
a
= t
b
1
t
1
a
1
t
b
2
t
1
a
2
t
b
3
t
1
a
3
.
By Lemma 15.1.5, on the other hand, there are dieomorphisms h
i
such
that t
b
i
= h
i
t
a
i
h
1
i
for i = 1, 2, 3. Substituting these expressions into the
relation above we get
t
a
= h
1
t
a
1
h
1
1
t
1
a
1
h
2
t
a
2
h
1
2
t
1
a
2
h
3
t
a
3
h
1
3
t
1
a
3
= [h
1
, t
a
1
][h
2
, t
a
2
][h
3
, t
a
3
].
We showed that a nonseparating Dehn twist is a product of (three) commu-
tators. This nishes the proof since
g
is generated by Dehn twists along
nonseparating curves (for g 3) and any two Dehn twists along nonsepa-
rating curves are conjugate by Lemma 15.1.5. Note that the conjugate of a
commutator is a commutator.
Remark 15.1.15. In fact, any mapping class group
n
g,r
is perfect, i.e.,
n
g,r
/[
n
g,r
,
n
g,r
] = 0 for g 3 (see [81], for example). For g = 1, 2 it is
impossible to embed a lantern relation into
g
with nonseparating boundary
components, and hence the above proof breaks down from the beginning.
Using the presentations of
n
1,r
and
n
2,r
, however, one can derive that

n
1,0
/[
n
1,0
,
n
1,0
] = Z
12
,

n
1,r
/[
n
1,r
,
n
1,r
] = Z
r
for r 1, and

n
2,r
/[
n
2,r
,
n
2,r
] = Z
10
.
Lemma 15.1.16. Any element in
g
can be expressed as a product of
nonseparating right-handed Dehn twists.
Proof. The following is a standard relation in the mapping class group
g
:
(t
a
1
t
a
2
t
a
2g
)
4g+2
= 1,
where the curves a
i
are depicted in Figure 15.5. We deduce that t
1
a
1
is
a product of nonseparating right-handed Dehn twists. Therefore any left-
handed nonseparating Dehn twist being conjugate to t
1
a
1
is a product
of nonseparating right-handed Dehn twists. This nishes the proof of the
lemma combined with the fact that
g
is generated by (right and left-
handed) nonseparating Dehn twists.
Exercises 15.1.17. (a) Show that any element in
g,1
can be expressed as
a product of nonseparating right-handed Dehn twists plus left-handed Dehn
twists along a boundary-parallel curve. (Hint: Use the same argument as
above with the relation (t
a
1
t
a
2
t
a
2g
)
4g+2
= t
in
g,1
where denotes a
curve parallel to the boundary.)
(b) Show that a separating right-handed Dehn twist in
g,1
can be expressed
as a product of nonseparating right-handed Dehn twists.
15.2. Mapping class groups and geometric structures
As our earlier discussion indicated, the geometric objects we discussed in
the preceding chapters have counterparts in various mapping class groups.
To clarify the situation, below we summarize these relations.
A product
k
i=1
[a
i
, b
i
] of k commutators in
g
gives a
g
-bundle over
the surface
k,1
with one boundary component. The mapping classes
a
i
and b
i
specify the monodromy along the obvious free generating
system
1
,
1
, . . . ,
k
,
k
) of
1
(
k,1
). If
k
i=1
[a
i
, b
i
] = 1 in
g
, we
get a
g
-bundle X
k
. (The bundle is uniquely determined by the
word once g 2.) In case
k
i=1
[a
i
, b
i
] = 1 holds in
1
g
, the bundle
X
k
admits a section. In this case
k
i=1
[a
i
, b
i
] = (t
)
n
in
g,1
for
some n Z, and it is not hard to see that the self-intersection of the
section given by this word is exactly n.
An expression
k
i=1
t
i

g
with t
i
right-handed Dehn twists provides a
genus-g Lefschetz bration X D
2
over the disk with ber
g
closed.
If
k
i=1
t
i
= 1 in
g
then the bration closes up to a bration over the
sphere S
2
and the closed up manifold is uniquely determined by the
word
k
i=1
t
i
once g 2. Once again, a lift of the relation
k
i=1
t
i
= 1
to
1
g
shows the existence of a section, and its self-intersection is n
if
k
i=1
t
i
= (t
)
n
in
g,1
for the Dehn twist t
along the boundary-

parallel simple closed curve
g,1
.
By combining the above two constructions, a word
w =
k
i=1
t
i
k
j=1
[a
i
, b
i
]
gives a Lefschetz bration over
k,1
and if w = 1 in
g
we get a
Lefschetz bration X
k
. Sections can be captured in the same
way as above.
An expression
n
i=1
t
i
= t
1
t
k
in
g,k
(where all t
i
stand for right-
handed Dehn twists and t
i
are right-handed Dehn twists along circles
parallel to the boundary components of the Riemann surface at hand)
naturally describes a Lefschetz pencil: The relation determines a Lef-
schetz bration with k section, each of self-intersection (1), and after
blowing these sections down we get a Lefschetz pencil. Conversely, by
blowing up the base locus of a Lefschetz pencil we arrive to a Lef-
schetz bration which can be captured (together with the exceptional
divisors of the blow-ups, which are all sections now) by a relator of
the above type.
15.3. Some proofs 265
If we allow the Dehn twists t
i
to have negative exponents in the previ-
ous constructions, we can also encounter achiral Lefschetz brations
in this way.
An element h
g,r
(r > 0) species a 3-manifold equipped with
an open book decomposition by considering the mapping cylinder
of h and collapsing the boundaries to the core circles. Notice that
the binding has r components. Through the equivalence discussed
in Section 9 the mapping class h
g,r
determines a contact 3-
manifold. All closed contact 3-manifolds can be given in this way;
h fails to be unique though, since by positively stabilizing the open
book decomposition (and so leaving the contact structure unchanged)
we can change g and r.
Since
n
i=1
t
i

g,r
gives a Lefschetz bration with nonclosed bers
over the disk D
2
, and these manifolds can be equipped with Stein
structures, a factorization h =
n
i=1
t
i
in
g,r
into right-handed Dehn
twists gives a Stein lling of the contact 3-manifold determined by
h
g,r
. All Stein llings arise in this manner, although we might
need to pass to a stabilization of h to recover certain llings of the
contact 3-manifold specied by h.
15.3. Some proofs
We close this chapter with a few results which show an interesting bridge
between Lefschetz brations, contact structures and mapping class groups.
For g 3, Proposition 15.1.14 shows that
g
is a perfect group, i.e.,
every element of
g
is a product of commutators. The minimal number
of commutators one has to use to express an element as a product in a
group is called the commutator length of that element.
Theorem 15.3.1 ([83]). The commutator length of a Dehn twist in
g
(g 3) is equal to two.
Proof. Consider a sphere X with four holes with boundary components
a, a
1
, a
2
, a
3
. Since the genus of
g
is at least three, X can be embedded in
g
in such a way that a
1
, a
2
, a
3
, b
1
, b
2
, b
3
are all nonseparating. The simple
closed curve a can be chosen either nonseparating or separating bounding
a subsurface of arbitrary genus (cf. Figures 15.8 and 15.9). Furthermore,
the complement of a
1
b
1
and of a
2
b
2
are connected. Hence, there is
an orientation preserving dieomorphism f of
g
such that f(a
1
) = b
2
and
f(b
1
) = a
2
. Let h be another orientation preserving dieomorphism of
g
such that h(a
3
) = b
3
. Then the lantern relation combined with the above
choices implies
t
a
= t
b
1
t
1
a
1
t
b
2
t
1
a
2
t
b
3
t
1
a
3
= t
b
1
t
1
a
1
t
f(a
1
)
t
1
f(b
1
)
ht
a
3
h
1
t
1
a
3
= t
b
1
t
1
a
1
ft
a
1
f
1
ft
1
b
1
f
1
ht
a
3
h
1
t
1
a
3
= [t
b
1
t
1
a
1
, f][h, t
a
3
].
Next we show that the commutator length of a Dehn twist is not equal to
one. Suppose that a right-handed Dehn twist is equal to a single commu-
tator. Then there is a 4-manifold X which admits a (relatively minimal)
genus-g Lefschetz bration over the torus T
2
with only one singular ber. It
is easy to see that (X) = 1. Since the bration is relatively minimal, and so
by Proposition 10.3.8 the 4-manifold X is a minimal symplectic 4-manifold,
we have the inequality
0 c
2
1
(X) = 3(X) + 2(X)
which implies that (X)
2
3
. This gives (X) 0 since (X) is an
integer. Recall that the holomorphic Euler characteristic is dened by
h
(X) =
1
4
_
(X) +(X)
_
and it is an integer for any closed almost-complex, hence for any closed
symplectic 4-manifold. Rewriting the above equality we get
h
(X) =
1
4
_
(X) + 1
_
.
Therefore (X) = 4
h
(X) 1 and so
c
2
1
(X) = 3(X) + 2(X) = 12
h
(X) 1.
On the other hand, by [155] it follows that
c
2
1
(X) 10
h
(X)
since X admits a Lefschetz bration over T
2
. Since the holomorphic Eu-
ler characteristic
h
(X) is an integer, it follows that
h
(X) 0 imply-
ing (X) + 1 = 4
h
(X) 0. This last inequality, however, contradicts
(X) 0, which has been shown earlier.
Recall from Proposition 15.1.14 that
g
is a perfect group for g 3.
The mapping class group
g
is, however, not uniformly perfect, that is,
there is no constant K such that any element of
g
can be written as a
product of at most K commutators. This statement can be proved by
using the correspondence between certain words in mapping class groups
and Lefschetz brations. (For a dierent proof see [13].)
Theorem 15.3.2 (EndoKotschick [36], Korkmaz [82]). Let c
g
be
a separating simple closed curve. If t
n
c
=
kn
i=1
_
i
(n),
i
(n)
then the
sequence k
n
cannot be bounded. In conclusion, the mapping class group
g
is not uniformly perfect.
Proof. Notice that a commutator expression of the type of the theorem gives
a relator which gives rise to a Lefschetz bration X
n

kn
. Suppose that
k
n
is bounded, say k
n
K. By adding trivial monodromies if necessary,
this assumption provides a sequence f
n
: X
n

K
(n N) of Lefschetz
brations over the xed base
K
. It is easy to see that
(X
n
) = (
g
)(
K
) +n = 4(K 1)(g 1) +n,
while by Novikov additivity and the signature calculation for a separating
vanishing cycle (cf. [130]) we get
(X
n
) = n +
_
X
n
i=1
f
1
n
(q
i
)
_
.
On the other hand one can show that
_
X
n
i=1
f
1
n
(q
i
)
_
C
for some constant C depending on K and g only. (The points q
i
denote the
critical values of the Lefschetz bration f
n
.) This implies that
c
2
1
(X
n
) = 3(X
n
) + 2(X
n
) 3n + 2n +C
= n +C
,
where C
= 3C+8(K1)(g1) and hence for n large enough the expression

c
2
1
(X
n
) will be negative. This observation contradicts the result of [155]
where it is proved that a relatively minimal Lefschetz bration over a
base of positive genus is minimal, hence its c
2
1
invariant is nonnegative,
cf. Corollary 10.3.10. The contradiction shows that the sequence k
n
is
unbounded, verifying the statement of the theorem.
Remark 15.3.3. In fact, using the exact sequences in Section 15.1 one
can show that the mapping class group
n
g,r
is not uniformly perfect. As
discussed in [36, 82], the fact that
n
g,r
is not uniformly perfect has inter-
esting corollaries regarding the second bounded cohomology of
n
g,r
. Also,
the proof given above can be rened to get explicit lower bounds for the
commutator lengths for certain elements in
n
g,r
; for details see [36, 82].
Similar question can be raised for the length of expressions writing a given
element as product of right-handed Dehn twists. Since by Lemma 15.1.16
1
g
can be written as a nontrivial product of right-handed Dehn twists
there is no bound for the length of such expression for h
g
. The situation,
however, is dierent in
g,r
once r > 0.
Theorem 15.3.4 ([7, 157]). If r 1 then 1
g,r
admits no nontrivial
factorization 1 = t
1
t
n
into a product of right-handed Dehn twists.
Proof. Suppose that 1
g,r
admits a nontrivial factorization 1 = t
1
t
n
into a product of right-handed Dehn twists. Now cap o all but one of the
boundary components with disks to get a relation in
g,1
where identity is
expressed as a nontrivial product of right-handed Dehn twists. Thus we re-
duce the problem to show that 1
g,1
admits no nontrivial factorization
1 = t
1
t
n
into a product of right-handed Dehn twists. Clearly we can as-
sume that g 1. Moreover we can assume that all the t
i
s are nonseparating
Dehn twists since any separating right-handed Dehn twist in
g,1
is a prod-
uct of nonseparating right-handed Dehn twists. Then we can express t
1
1
,
and hence any nonseparating left-handed Dehn twist, as a product of right-
handed Dehn twists. We know that any element in
g,1
can be expressed
as a product of nonseparating Dehn twists. Now replace every left-handed
Dehn twist in this expression by a product of right-handed Dehn twists to
conclude that any element in
g,1
can be expressed as a product of (non-
separating) right-handed Dehn twists. We will show that this is impossible
using contact geometry. For any given g 1 we can construct a genus-g
surface with one boundary component by plumbing left-handed Hopf bands.
This would give us an open book with monodromy
g,1
whose compat-
ible contact structure is overtwisted by construction. But if could be
expressed as a product of right-handed Dehn twists then would be Stein
llable, leading to a contradiction. For a dierent proof see [157].
This observation leads us to
Conjecture 15.3.5. For any mapping class h
g,r
with r > 0 there is a
constant C
h
such that if h = t
1
t
n
factors as a product of right-handed
Dehn twists in
g,r
then n C
h
.
The armative solution of this conjecture would provide a bound for Euler
characteristics of Stein llings of xed open book decompositions a weaker
version of the statement given in Conjecture 12.3.16.
Bibliography
[1] C. Abbas and H. Hofer, Holomorphic curves and global questions in contact geo-
metry, book in preparation.
[2] B. Aebisher et al., Symplectic geometry, Progress Math. 124, Birkhauser, Boston,
MA, 1994.
[3] S. Akbulut, On 2-dimensional homology classes of 4-manifolds, Math. Proc. Cam-
bridge Philos. Soc. 82 (1977), 99106.
[4] S. Akbulut, An exotic 4-manifold, J. Dierential Geom. 33 (1991), 357361.
[5] S. Akbulut and R. Matveyev, A convex decomposition theorem for 4-manifolds, Int.
Math. Res. Notices 7 (1998), 371381.
[6] S. Akbulut and R. Matveyev, Exotic structures and adjunction inequality, Turkish
J. Math. 21 (1997), 4753.
[7] S. Akbulut and B. Ozbagci, Lefschetz brations on compact Stein surfaces, Geom.
Topol. 5 (2001), 319334.
[8] S. Akbulut and B. Ozbagci, On the topology of compact Stein surfaces, Int. Math.
Res. Not. 15 (2002), 769782.
[9] J. Alexander, A lemma on systems of knotted curves, Proc. Nat. Acad. Sci. USA
9 (1923), 9395.
[10] W. Barth, C. Peters and A. van de Ven, Compact complex surfaces, Ergebnisse der
Mathematik, Springer-Verlag Berlin, 1984.
[11] D. Bennequin, Entrelacements et equations de Pfa, Asterisque 107108 (1983),
87161.
[12] R. Berndt, An introduction to symplectic geometry, Graduate Studies in Mathe-
matics, vol. 26, American Math. Society, Providence 2000.
[13] M. Bestvina and K. Fujiwara Bounded cohomology of subgroups of mapping class
groups, Geom. Topol. 6 (2002), 6889.
[14] V. Braungardt and D. Kotschick, Einstein metrics and the number of smooth
structures on a fourmanifold, arXiv:math.GT/0306021
[15] Y. Chekanov, Dierential algebra of Legendrian links, Invent. Math. 150 (2002),
441483.
[16] F. Ding and H. Geiges, Symplectic llability of tight contact structures on torus
bundles, Algebr. Geom. Topol. 1 (2001), 153172.
[17] F. Ding and H. Geiges, A Legendrian surgery presentation of contact 3-manifolds,
Math. Proc. Cambridge Philos. Soc. 136 (1004), 583598.
272 Bibliography
[18] F. Ding, H. Geiges and A. Stipsicz, Surgery diagrams for contact 3-manifolds,
Turkish J. Math. 28 (2004), 4174.
[19] F. Ding, H. Geiges and A. Stipsicz, Lutz twist and contact surgery,
to appear in Asian J. Math., arXiv:math.SG/0401338
[20] S. Donaldson, An application of gauge theory to 4-dimensional topology, J. Dier-
ential Geom. 18 (1983) 279315.
[21] S. Donaldson, The SeibergWitten equations and 4-manifold topology, Bull. Amer.
Math. Soc. 33 (1996), 45-70.
[22] S. Donaldson, Symplectic submanifolds and almost-complex geometry, J. Dieren-
tial Geom. 44 (1996), 666705.
[23] S. Donaldson, Lefschetz brations in symplectic geometry, Proc. Internat. Cong.
Math. (Berlin, 1998), Vol II, Doc. Math. Extra Volume ICMII (1998), 309314.
[24] Y. Eliashberg, Classication of overtwisted contact structures on 3-manifolds, In-
vent. Math. 98 (1989), 623637.
[25] Y. Eliashberg, Topological characterization of Stein manifolds of dimension > 2,
International J. of Math. 1 (1990), 2946.
[26] Y. Eliashberg, Contact 3-manifolds twenty years since J. Martinets work, Ann.
Inst. Fourier, 42 (1992), 165192.
[27] Y. Eliashberg, Filling by holomorphic disks and its applications, Geometry of Low-
Dimensional Manifolds: 2, Proc. Durham Symp. 1989, London Math. Soc. Lecture
Notes, 151, Cambridge Univ. Press, 1990, 4567.
[28] Y. Eliashberg, On symplectic manifolds with some contact properties, J. Dierential
Geom. 33 (1991), 233238.
[29] Y. Eliashberg, Unique holomorphically llable contact structure on the 3-torus, Int.
Math. Res. Not. 2 1996, 7782.
[30] Y. Eliashberg, Few remarks about symplectic lling, Geom. Topol. 8 (2004), 277
293.
[31] Y. Eliashberg and M. Fraser, Classication of topologically trivial Legendrian knots,
Geometry, topology, and dynamics (Montreal, PQ, 1995), 1751, CRM Proc. Lec-
ture Notes, 15, Amer. Math. Soc., Providence, RI, 1998.
[32] Y. Eliashberg and M. Gromov, Convex symplectic manifolds, Several complex
variables and complex geometry (Santa Cruz, CA, 1989) 135162.
[33] Y. Eliashberg and N. Mishachev, Introduction to the h-principle, Graduate Studies
in Mathematics 48 AMS, 2002.
[34] Y. Eliashberg and L. Traynor, Symplectic Geometry and Topology, IAS/Park City
Math. Series vol. 7, 1999.
[35] H. Endo, Meyers signature cocycle and hyperelliptic brations, Math. Ann. 316
(2000), 237257.
[36] H. Endo and D. Kotschick, Bounded cohomology and non-uniform perfection of
mapping class groups, Invent. Math. 144 (2001), 169175.
Bibliography 273
[37] J. Epstein, D. Fuchs and M. Meyer, ChekanovEliashberg invariants and transverse
approximation of Legendrian knots, Pacic J. Math. 201 (2001), 89106.
[38] J. Etnyre, Symplectic convexity in low-dimensional topology, Topology and its Appl.
88 (1998), 325.
[39] J. Etnyre, Introductory lectures on contact geometry, arXiv:math.SG/0111118
[40] J. Etnyre, Legendrian and transversal knots, arXiv:math.SG/0306256
[41] J. Etnyre, Transversal torus knots, Geom. Topol. 3 (1999), 253268.
[42] J. Etnyre, On symplectic llings, Algebr. Geom. Topol. 4 (2004), 7380.
[43] J. Etnyre, Convex surfaces in contact geometry: class notes,
http://www.math.upenn.edu/ etnyre/preprints/notes.html
[44] J. Etnyre and K. Honda, Tight contact structures with no symplectic llings, Invent.
Math. 148 (2002), 609626.
[45] J. Etnyre and K. Honda, On the non-existence of tight structures, Ann. of Math.
153 (2001), 749766.
[46] J. Etnyre and K. Honda, On symplectic cobordisms, Math. Ann. 323 (2002), 3139.
[47] J. Etnyre and K. Honda, Knots and contact geometry. I. Torus knots and the gure
eight knot, J. Symplectic Geom. 1 (2001), 63120.
[48] F. Forstneric, Complex tangents of real surfaces in complex surfaces, Duke Math. J.
67 (1992), 353376.
[49] F. Forstneric, Stein domains in complex surfaces, J. Geom. Anal. 13 (2003), 7794.
[50] F. Forstneric and J. Kozak, Strongly pseudoconvex handlebodies, J. Korean Math.
Soc. 40 (2003), 727745.
[51] M. Freedman and F. Quinn, Topology of 4-manifolds, Princeton Mathematical
Series 39, Princeton University Press, 1990.
[52] D. Gabai, Detecting bered links in S
3
, Comment. Math. Helv. 61 (1986), 519-555.
[53] D. Gay, Explicite concave llings of 3-manifolds, Math. Proc. Cambridge Philos.
Soc. 133 (2002), 431441.
[54] D. Gay and R. Kirby, Constructing symplectic forms on 4-manifolds which vanish
on circles, Geom. Topol. 8 (2004), 743777.
[55] H. Geiges, Symplectic structures on T
2
-bundles over T
2
, Duke Math. J. 67 (1992),
539555.
[56] H. Geiges, h-Principles and exibility in geometry, Mem. Amer. Math. Soc. 164
(2003), no. 779.
[57] H. Geiges, Contact geometry, to appear in the Handbook of Dierential Geometry,
vol. 2.
[58] S. Gervais, A nite presentation of the mapping class group of a punctured surface,
Topology 40 (2001), 703725.
[59] P. Ghiggini and S. Schonenberger On the classication of tight contact structures,
Topology and Geometry of manifolds (Athens, GA, 2001), Proc. Sympos. Pure
Math. 71 (2003), 121151.
274 Bibliography
[60] P. Ghiggini, Tight contact structures on Seifert manifolds over T
2
with one singular
ber, arXiv:math.GT/0307340
[61] E. Giroux, Convexite en topologie de contact, Comment. Math. Helv. 66 (1991),
637677.
[62] E. Giroux, Structures de contact sur les varietes brees en cercles dune surface,
Comment. Math. Helv. 76 (2001), 218262.
[63] E. Giroux, Contact geometry: from dimension three to higher dimensions, Proceed-
ings of the International Congress of Mathematicians (Beijing 2002), 405414.
[64] R. Gompf, A new construction of symplectic manifolds, Ann. of Math. 142 (1995),
527595.
[65] R. Gompf, Handlebody construction of Stein surfaces, Ann. of Math. 148 (1998),
619693.
[66] R. E. Gompf and A. I. Stipsicz, 4-manifolds and Kirby calculus, Graduate Studies
in Mathematics, vol. 20, American Math. Society, Providence 1999.
[67] J. Gonzalo, Branched covers and contact structures, Proc. Amer. Math. Soc. 101
(1987), 347352.
[68] N. Goodman, Contact structures and open books, Ph.D thesis, UT Austin, 2003.
[69] H. Grauert, On Levis problem, Ann. of Math. 68 (1958), 460472.
[70] H. Grauert and R. Remmert, Theory of Stein spaces, Springer-Verlag, 1979.
[71] J. Gray, Some global properties of contact structures, Ann. of Math. 69 (1959),
421450.
[72] V. Guillemin and A. Pollack, Dierential topology, Prentice-Hall, 1974.
[73] J. Harer, Pencils of curves on 4-manifolds, Ph.D thesis, UC Berkeley, 1974.
[74] J. Harer, How to construct all bered knots and links, Topology 21 (1982), 263280.
[75] H. Hilden, Every closed orientable 3-manifold is a 3-fold branched covering space
of S
3
, Bull. Amer. Math. Soc. 80 (1974), 12431244.
[76] K. Honda, On the classication of tight contact structures, I., Geom. Topol. 4
(2000), 309368.
[77] K. Honda, On the classication of tight contact structures, II., J. Dierential Geom.
55 (2000), 83143.
[78] D. Johnson, Homeomorphisms of a surface which act trivially on homology, Proc.
Amer. Math. Soc. 75 (1979), 119125.
[79] Y. Kanda, The classication of tight contact structures on the 3-torus, Comm.
Anal. Geom. 5 (1997), 413438.
[80] R. Kirby, A calculus for framed links in S
3
, Invent. Math. 45 (1978), 3556.
[81] M. Korkmaz, Low-dimensional homology groups of mapping class groups: a survey,
Turkish J. Math. 26 (2002), 101114.
[82] M. Korkmaz, Stable commutators length of a Dehn twist, Michigan Math. J. 52
(2004), 2331.
Bibliography 275
[83] M. Korkmaz and B. Ozbagci, Minimal number of singular bers in a Lefschetz
bration, Proc. Amer. Math. Soc. 129 (2001), 15451549.
[84] D. Kotschick, The SeibergWitten invariants of symplectic four-manifolds [after
C. H. Taubes], Seminare Bourbaki 48`eme annee (1995-96) n
o
812, Asterisque 241
(1997), 195220.
[85] P. Kronheimer and T. Mrowka, The genus of embedded surfaces in the projective
plane, Math. Res. Lett. 1 (1994), 797808.
[86] P. Kronheimer and T. Mrowka, Monopoles and contact structures, Invent. Math.
130 (1997), 209255.
[87] P. Kronheimer and T. Mrowka, Wittens conjecture and Property P, Geom. Topol.
8 (2004), 295310.
[88] P. Kronheimer and T. Mrowka, Floer homology for SeibergWitten monopoles, book
in preparation.
[89] P. Kronheimer, T. Mrowka, P. Ozsvath and Z. Szabo, Monopoles and lens space
surgeries, arXiv:math.GT/0310164
[90] H. Lai, Characteristic classes of real manifolds immersed in complex manifolds,
Trans. Amer. Math. Soc. 172 (1972), 133.
[91] F. Laudenbach and V. Poenaru, A note on 4-dimensional handlebodies, Bull. Soc.
Math. France 100 (1972), 337344.
[92] C. LeBrun, Dieomorphisms, symplectic forms and Kodaira brations, Geom.
Topol. 4 (2000), 451456.
[93] R. Lickorish, A representation of orientable combinatorial 3-manifolds, Ann. of
Math. 76 (1962), 531540.
[94] P. Lisca, Symplectic llings and positive scalar curvature, Geom. Topol. 2 (1998),
103116.
[95] P. Lisca, On symplectic llings of 3-manifolds, Proceedings of the 6
th
Gokova
Geometry-Topology Conference, Turkish J. Math. 23 (1999), 151159.
[96] P. Lisca, On lens spaces and their symplectic llings, arXiv: math.SG/0203006
[97] P. Lisca and G. Matic, Tight contact structures and SeibergWitten invariants,
Invent. Math. 129 (1997), 509525.
[98] P. Lisca and A. Stipsicz, An innite family of tight, not llable contact three-man-
ifolds, Geom. Topol. 7 (2003), 10551073.
[99] P. Lisca and A. Stipsicz, Tight not semillable contact circle bundles, Math. Ann.
328 (2004), 285298.
[100] P. Lisca and A. Stipsicz, Seifert bered contact 3-manifolds via surgery, Algebr.
Geom. Topol. 4 (2004), 199217.
[101] P. Lisca and A. Stipsicz, OzsvathSzabo invariants and tight contact three-mani-
folds I., Geom. Topol. 8 (2004), 925945.
[102] P. Lisca and A. Stipsicz, OzsvathSzabo invariants and tight contact three-mani-
folds II., arXiv:math.SG/0404136.
276 Bibliography
[103] T.-J. Li and A. Liu, Symplectic structure on ruled surfaces and a generalized
adjunction formula, Math. Res. Lett. 2 (1995), 453471.
[104] A. Loi and R. Piergallini, Compact Stein surfaces with boundary as branched covers
of B
4
, Invent. Math. 143 (2001), 325348.
[105] R. Lutz, Sur quelques proprietes des formes dierentielles en dimension trois,
These, Strasbourg, 1971.
[106] H. Lyon, Torus knots in the complements of links and surfaces, Michigan Math. J.
27 (1980), 39-46.
[107] J. Martinet, Formes de contact sur les vari
tes de dimension 3, in: Proc. Liverpool

Singularity Sympos. II, Lecture Notes in Math. 209, Springer, Berlin (1971), 142
163.
[108] Y. Matsumoto, Lefschetz brations of genus two a topological approach, Topol-
ogy and Teichm uller spaces (Katinkulta, 1995), 123148, World Sci. Publishing,
River Edge, NJ, 1996.
[109] D. McDu, The local behaviour of holomorphic curves in almost-complex 4-mani-
folds, J. Dierential Geom. 34 (1991), 311358.
[110] D. McDu, Symplectic manifolds with contact type boundaries, Invent. Math. 103
(1991), 651671.
[111] D. McDu and D. Salamon, Introduction to symplectic topology, Oxford Math.
Monographs, 1995.
[112] C. McMullen and C. Taubes, 4-manifolds with inequivalent symplectic forms and
3-manifolds with inequivalent brations, Math. Res. Lett. 6 (1999), 681696.
[113] M.-L. Michelsohn and B. Lawson, Spin geometry, Princeton Mathematical Series
38, Princeton University Press, 1989.
[114] J. Milnor, Morse theory, Ann. Math. Studies 51, Princeton University Press, 1963.
[115] J. Milnor, Lectures on the h-Cobordism Theorem, notes by L. Siebenmann and
J. Sondow, Princeton University Press, 1965.
[116] J. Milnor and J. Stashe, Characteristic classes, Ann. Math. Studies 76, Princeton
University Press, 1974.
[117] J. Montesinos-Amilibia, A representation of closed orientable 3-manifolds as 3-fold
branched covering of S
3
, Bull. Amer. Math. Soc. 80 (1974), 845846.
[118] J. Montesinos and H. Morton, Fibred links from closed braids, Proc. London Math.
Soc. (3) 62 (1991), 167201.
[119] J. Morgan, The SeibergWitten equations and applications to the topology of smooth
four-manifolds, Math. Notes 44, Princeton University Press, Princeton NJ, 1996.
[120] J. Morgan and Z. Szabo, Homotopy K3 surfaces and mod 2 SeibergWitten inva-
riants, Math. Res. Lett. 4 (1997), 1721.
[121] J. Morgan, Z. Szabo and C. Taubes, A product formula for the SeibergWitten
invariants and the generalized Thom conjecture, J. Dierential Geom. 44 (1996),
706788.
Bibliography 277
[122] A. Mori, A note on ThurstonWinkelnkempers construction of contact forms on
3-manifolds, Osaka J. Math. 39 (2002), 111.
[123] T. Mrowka, P. Ozsvath and B. Yu, SeibergWitten monopoles on Seifert bered
spaces, Comm. Anal. Geom. 5 (1997), no. 4, 685791.
[124] A. Nemethi, On the OzsvathSzabo invariants of negative denite plumbed 3-man-
ifolds, arXiv:math.GT/0310083.
[125] S. Nemirovski, Complex analysis and dierential topology on complex surfaces,
Russian Math. Surveys 54 (1999), 729752.
[126] L. Nicolaescu, Notes on SeibergWitten theory, Graduate Studies in Mathematics
28 AMS, Providence, RI, 2000.
[127] H. Ohta and K. Ono, Simple singularities and topology of symplectically lling
4-manifolds, Comment. Math. Helv. 74 (1999), 575590.
[128] H. Ohta and K. Ono, Symplectic llings of the link of simple elliptic singularities,
J. Reine Angew. Math 565 (2003), 183205.
[129] B. Owens and S. Strle, Rational homology spheres and four-ball genus,
arXiv:math.GT/0308073
[130] B. Ozbagci, Signatures of Lefschetz brations, Pacic J. Math. 202 (2002), 99118.
[131] B. Ozbagci and A. Stipsicz, Noncomplex smooth 4-manifolds which admit genus-2
Lefschetz brations, Proc. Amer. Math. Soc. 128 (2000), 31253128.
[132] B. Ozbagci and A. Stipsicz, Contact 3-manifolds with innitely many Stein llings,
Proc. Amer. Math. Soc. 132 (2004), 15491558.
[133] P. Ozsvath and Z. Szabo, The symplectic Thom conjecture, Ann. of Math. 151
(2000), 93124.
[134] P. Ozsvath and Z. Szabo, On embedding of circle bundles in 4-manifolds, Math.
Res. Lett. 7 (2000) 657669.
[135] P. Ozsvath and Z. Szabo, Holomorphic disks and topological invariants for rational
homology three-spheres, Ann. of Math., to appear, arXiv:math.SG/0101206
[136] P. Ozsvath and Z. Szabo, Holomorphic disks and three-manifold invariants: prop-
erties and applications, Ann. of Math., to appear, arXiv:math.SG/0105202
[137] P. Ozsvath and Z. Szabo, Holomorphic triangles and invariants of smooth 4-man-
ifolds, arXiv:math.SG/0110169
[138] P. Ozsvath and Z. Szabo, Absolutely graded Floer homologies and intersection forms
of four-manifolds with boundary, Adv. Math. 173 (2003), 179261.
[139] P. Ozsvath and Z. Szabo, On the Floer homology of plumbed three-manifolds, Geom.
Topol. 7 (2003), 185224.
[140] P. Ozsvath and Z. Szabo, Heegaard Floer homologies and contact structures,
arXiv:math.SG/0210127
[141] P. Ozsvath and Z. Szabo, Holomorphic triangle invariants and the topology of
symplectic 4-manifolds, arXiv:math.SG/0201049
278 Bibliography
[142] P. Ozsvath and Z. Szabo, Holomorphic disks and knot invariants,
[143] P. Ozsvath and Z. Szabo, Holomrphic disks and genus bounds, Geom. Topol. 8
(2004), 311334.
[144] J. Park, The geography of irreducible 4-manifolds, Proc. Amer. Math. Soc. 126
(1996), 24932503.
[145] J. Park, Simply connected symplectic 4-manifolds with b
+
2
= 1 and c
2
1
= 2,
[146] O. Plamenevskaya, Contact structures with distinct HeegaardFloer invariants,
arXiv:math.SG/0309326
[147] J. Powell. Two theorems on the mapping class group of a surface, Proc. Amer.
Math. Soc. 68 (1978), 347350.
[148] D. Rolfsen, Knots and links, Publish or perish, 1976.
[149] D. Salamon, Spin geometry and SeibergWitten invariants, book in preparation.
[150] I. Smith, On moduli space of symplectic forms, Math. Res. Lett. 7 (2000), 779788.
[151] J. Stallings, Construction of bered knots and links, A.M.S. Proc. Symp. in Pure
Math. 32 (1978), 55-60.
[152] A. Stipsicz, A note on the geography of symplectic manifolds, Turkish J. Math. 20
(1996), 135139.
[153] A. Stipsicz, Simply connected 4-manifolds near the BogomolovMiyaokaYau line,
Math. Res. Lett. 5 (1998), 723730.
[154] A. Stipsicz, On the number of vanishing cycles in a Lefschetz bration, Math. Res.
Lett. 6 (1999), 449456.
[155] A. Stipsicz, Chern numbers of certain Lefschetz brations, Proc. Amer. Math. Soc.
128 (2000), 18451851.
[156] A. Stipsicz, The geography problem of 4-manifolds with various structures, Acta
Math. Hungar. 87 (2000), 267278.
[157] A. Stipsicz, Sections of Lefschetz brations and Stein llings, Proceedings of the
Gokova Geometry-Topology Conference (2000), 97101.
[158] A. Stipsicz, Gauge theory and Stein llings of certain 3-manifolds, Proceedings of
the Gokova Geometry-Topology Conference (2001), 115131.
[159] A. Stipsicz, On the geography of Stein llings of certain 3-manifolds, Michigan
Math. J. 51 (2003), 327337.
[160] A. Stipsicz, Surgery diagrams and open book decompositions on 3-manifolds, Acta
Math. Hungar., to appear.
[161] C. Taubes, The SeibergWitten invariants and symplectic forms, Math. Res. Lett.
1 (1994), 809822.
[162] C. Taubes, Counting pseudo-holomorphic submanifolds in dimension 4, J. Dier-
ential Geom. 44 (1996), 818893.
Bibliography 279
[163] C. Taubes, SeibergWitten and Gromov invariants, Geometry and Physics (Aarhus,
1995) (Lecture Notes in Pure and Applied Math., 184) Dekker, New York (1997),
591601.
[164] C. Taubes, SW =Gr: From the SeibergWitten equations to pseudo-holomorphic
curves, Journal of the Amer. Math. Soc. 9 (1996), 845918.
[165] C. Taubes, Hermann Weyl lecture at IAS, Princeton NJ, 2002.
[166] W. Thurston, Some simple examples of symplectic manifolds, Proc. Amer. Math.
Soc. 55 (1976), 467468.
[167] W. Thurston and H. Winkelnkemper, On the existence of contact forms, Proc.
Amer. Math. Soc. 52 (1975), 345347.
[168] I. Torisu, Convex contact structures and bered links in 3-manifolds, Int. Math.
Res. Not. 9 (2000), 441454.
[169] V. Turaev, Torsion invariants of spin
c
structures on 3-manifolds, Math. Res. Lett.
4 (1997), 679695.
[170] S. Vidussi, Homotopy K3s with several symplectic structures, Geom. Topol. 5
(2001), 267285.
[171] B. Wajnryb, An elementary approach to the mapping class group of a surface,
Geom. Topol. 3 (1999), 405466.
[172] A. Weinstein, Symplectic manifolds and their Lagrangian submanifolds, Adv. Math.
6 (1971), 32946.
[173] A. Weinstein, Contact surgery and symplectic handlebodies, Hokkaido Mathemati-
cal Journal 20 (1991), 24151.
[174] E. Witten, Monopoles and four-manifolds, Math. Res. Lett. 1 (1994), 769796.
[175] W.-T. Wu, Sur le classes caracteristique des structures brees spheriques, Actu-
alites Sci. Industr. 1183 (1952).
Index
achiral Lefschetz bration, 155
adjunction
equality, 51
formula, 243
inequality, 13, 226
almost-complex structure, 50, 106
almost-Kahler structure, 50
basic class, 226
Bennequin inequality, 21, 77
binding, 131
blackboard framing, 29
botany, 53
boundary connected sum, 26
branch set, 150
bypass, 183
characteristic foliation, 78, 87
classication, 179
compatible, 191
complex point, 127
connected sum, 26
boundary, 26
contact
1-form, 63
Dehn surgery, 185
framing, 68
invariant, 230
structure, 63
coorientable, 67
llable, 254
isotopic, 66
overtwisted, 21, 76
positive, 67
standard, 66
tight, 21, 76
universally tight, 76
virtually overtwisted, 77
type, 111
type boundary, 229
vector eld, 85
contactomorphic, 66
continued fraction, 35
convex, 86
Darboux theorem, 70
Dehn
surgery, 31
contact, 185
twist, 158, 193, 255
destabilization
negative, 137
positive, 137
Dirac operator, 224
dividing set, 86
dotted circle, 39, 169
elimination lemma, 82
elliptic, 128
singularity, 80
bered link, 132
llable
holomorphically, 201
Stein, 201, 254
strongly symplectically, 201
weakly symplectically, 201
lling
Stein, 230, 265
Floer homology, 241
four-ball genus, 18
framing, 27
282 Index
blackboard, 29, 74
contact, 68
Seifert, 29
ThurstonBennequin, 68
Fredholm map, 225
Frobenius theorem, 64
front projection, 72
gauge group, 224
geography, 53
gordian number, 18
handle, 28
handlebody, 28
relative, 28
h-cobordism theorem, 11
Heegaard
decomposition, 27, 96
diagram, 30, 236
Hirzebruch signature theorem, 104
Hodge star operator, 224
holomorphic convex hull, 121
holomorphically convex, 121
Hopf link, 133
hyperbolic, 128
singularity, 80
hyperelliptic involution, 258
Kirby calculus, 38
Kodaira dimension, 59
Lagrangian
neighborhood theorem, 57
submanifold, 51
Lefschetz
bration, 156, 264
achiral, 155
allowable, 163
relatively minimal, 155
pencil, 156, 264
achiral, 155
Legendrian
isotopy, 72
knot, 68
realization principle, 90
unknot, 76
lens space, 34, 216
LeviCivita connection, 223
Liouville vector eld, 113
longitude, 32
mapping class group, 131, 255
presentation, 259
meridian, 32
minimal model, 58
monodromy, 131
Mosers method, 55
Murasugi sum, 134
neighborhood theorem
contact, 70
Lagrangian, 57
symplectic, 56
nonisolating, 90
normal connected sum, 114
-concave, 111
-convex, 111
open book decomposition, 96, 131
binding, 131
compatible, 138, 191
monodromy, 131
page, 131
standard, 133
overtwisted
contact structure, 21, 76
disk, 76
OzsvathSzabo invariant, 22
contact, 205, 249
page, 131
PALF, 163
perfect, 267
uniformly, 267
plumbing, 134
plurisubharmonic, 122
pseudo-holomorphic submanifold, 51
pseudoconvex, 123
rational surgery, 31
real point, 127
Reeb vector eld, 67
regular ber, 155
relation
Index 283
braid, 257
chain, 257
commutativity, 257
hyperelliptic, 258
lantern, 257
relatively minimal, 155
Rolfsen twist, 35
rotation number, 74
SardSmale theorem, 225
SeibergWitten
invariant, 223
moduli space, 224
parametrized, 224
simple type, 226, 228
Seifert
bered manifold, 45
framing, 29
self-linking number, 82
simple
(ADE) singularity, 218
cover, 150
elliptic singularity, 218
type, 226
singular ber, 155
singularity
elliptic, 80
hyperbolic, 80
slam-dunk, 35
slice genus, 18
slope, 91
sobering arc, 148
spin
group, 99
structure, 99
induced, 101
spin
c
group, 100
structure, 100, 240
induced, 101
stabilization
negative, 137
positive, 137
state traversal, 205
Stein
cobordism, 124
domain, 124, 162
manifold, 121
surface, 122
surface bundle, 264
surgery, 27
Dehn, 31
exact triangle, 244
rational, 31
symplectic
cut-and-paste, 111
dilation, 111, 123
form, 49
manifold, 49
minimal, 58, 177
neighborhood theorem, 56
structure, 49
deformation equivalent, 54
equivalent, 54
singular, 54
standrad, 49
submanifold, 51
Symplectic Thom conjecture, 227
symplectization, 71
3-dimensional invariant, 105
Thom conjecture, 232
ThurstonBennequin
framing, 68
invariant, 166
tight contact structure, 21
totally real submanifold, 51
transverse knot, 68
2-plane eld, 102
twisted coecient system, 254
twisting, 138
unknotting number, 18
vanishing cycle, 156
Vitushkins conjecture, 17
Weinstein handle, 115
Whitney
disk
holomorphic, 240
trick, 11
writhe, 73

B. Ozbagci and A. I. Stipsicz - Surgery On Contact 3-Manifolds and Stein Surfaces

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

B. Ozbagci and A. I. Stipsicz - Surgery On Contact 3-Manifolds and Stein Surfaces

Hochgeladen von

Copyright:

Verfügbare Formate

Surgery

ANOS BOLYAI MATHEMATICAL SOCIETY

ANOS BOLYAI MATHEMATICAL SOCIETY

(K) can be dened as

(K) g(K), and as the proof of Corollary 1.2.3

(K) can be equal to 0 for a nontrivial knot K, e.g. for K

are called smoothly slice.) The

(K), obtained by gluing the four-ball genus minimizing surface to

(K). It is not hard to nd

(K) for any knot K S

(K) is not an equality in general;

(L) Z denote the framing induced by the contact

-) manifolds, possibly with nonempty boundary. The gen-

Figure 2.1. Flipping X2 with Y2

of a null-homologous knot K Y denes

) is invertible, the matrix

= 1. Consequently, after xing a basis of H

is invertible, the chosen class

, since it maps the meridian of (S

with rational coecients in

by a nite sequence of Rolfsen twists, isotopies

(X, X; Z) can also be read o from the picture. We restrict ourselves

with surgery coecient k. Let H

= T (as complex bundles) gives

L of the zero-section in the cotangent bundle of L and a

L). For more details of the proofs see [111]

) are horizontal along the z-axis

d(u, Jv) is a Riemannian metric. (Here J denotes multiplication by

is the induced map on T

X.) Then the 1-form

) are called con-

, this is equivalent to the existence of a nowhere

= g. Two contact structures and

on a manifold Y are said to be isotopic if there is a contactomorphism

) which is isotopic to the identity.

) = 1 is called the Reeb vector eld of . In other

is transverse to the contact planes and depends

for a nowhere vanishing

= (r sin r +cosr +r cot r cos r)

(Y ) with f > 0 or f < 0.)

= . Using the same simple idea verify that on (Y )

is a small vertical push-o of L. Now Figure 4.5 shows

(L) is the ThurstonBennequin invariant of L with respect to

with the trivialization along given by

() for any closed embedded

. Notice that the planes

along its boundary are no longer horizontal. It is clear that

becomes a closed leaf of the characteristic foliation with

, see Figure 4.7.

can be oriented as follows: If p is a

is generic, i.e., the singular points are isolated and

denote the number of elliptic/hyperbolic

. Fix a vector eld w which directs T

. Now it easily follows

with the zero section by

. We choose the section of [

given by w which also gives a

to derive the formula

. We will call the new surface

is oriented, there is no positive elliptic point on

, no two hyperbolic points are adjacent. Suppose that there is no posi-

() of as the linking number of and

from the front projection of a Legendrian knot L: For L